Datasets:

Hendrik-a

/

arxiv

like 0

supermassive black hole binaries ( smbhbs ) are the products of frequent galaxy mergers . the coalescence of the smbhbs is a distinct source of gravitational wave ( gw ) radiation . the detections of the strong gw radiation and their possible electromagnetic counterparts are essential . numerical relativity suggests that the post - merger supermassive black hole ( smbh ) gets a kick velocity up to due to the anisotropic gw radiations . here we investigate the dynamical co - evolution and interaction of the recoiling smbhs and their galactic stellar environments with one million direct-body simulations including the stellar tidal disruption by the recoiling smbhs . our results show that the accretion of disrupted stars does not significantly affect the smbh dynamical evolution . we investigate the stellar tidal disruption rates as a function of the dynamical evolution of oscillating smbhs in the galactic nuclei . our simulations show that most of stellar tidal disruptions are contributed by the unbound stars and occur when the oscillating smbhs pass through the galactic center . the averaged disruption rate is , which is about an order of magnitude lower than that by a stationary smbh at similar galactic nuclei . our results also show that a bound star cluster is around the oscillating smbh of about the black hole mass . in addition , we discover a massive cloud of unbound stars following the oscillating smbh . we also investigate the dependence of the results on the smbh masses and density slopes of the galactic nuclei .

1201.3407

supermassive black hole binaries ( smbhbs ) are predicted by the hierarchical galaxy formation model in @xmath4 cold dark matter ( @xmath4cdm ) cosmology @xcite . for merging galaxies , their two smbhs with galactic cores will firstly approach each other by dynamical friction , and then get close enough to form a bound , compact binary system . after that , the smbhb may stall at a so called `` hard binary separation '' for a time even longer than the hubble time @xcite . however , recent investigations suggest that the hardening rates of smbhbs can be boosted and they may coalesce within the hubble time either due to various stellar dynamical processes other than spherical two - body relaxation @xcite , or gas dynamics ( * ? ? ? * ; * ? ? ? * and references therein ) . since the evolution of smbhbs deeply impacts the evolution of host galaxies , it is very important for us to find observational evidences to constrain evolution models of smbhbs . a statistic way is the calculation of tidal disruption rates . a dormant smbh could be temporarily activated by tidally disrupting a star passing by and accreting the disrupted stellar debris @xcite , which may have been observed in several non - active galaxies @xcite . @xcite and @xcite calculated the tidal disruption rate in smbhb systems at different evolutionary stages , and found that is significantly different from the typical rate in a single smbh for several orders of magnitude . the most certain proof for detecting smbhb individually may come from gravitational wave ( gw ) radiation observation . coalescing smbhbs are important sources of gw radiation @xcite , and can be detected within the next decades by the laser interferometer space antenna ( lisa ) and the pulsar timing array ( pta ) program @xcite . because of the poor spatial resolution of both lisa and pta for locating gw radiation sources , it is of key importance to detect electromagnetic counterparts ( emcs ) of gw radiation sources . besides , identifying smbhbs by their emcs is also essential to constrain the poorly understood galaxy - merger history . several emcs have been suggested in the literature to probe smbhbs , ( 1 ) precession of jet orientation and its acceleration in radio galaxies during the in - spiraling of smbhbs @xcite , ( 2 ) optical periodic outbursts in blazars due to the interaction of smbhbs and accretion disk @xcite , ( 3 ) jet reorientation in x - shaped radio galaxies due to the exchange of angular momentum between smbhbs and accretion disk @xcite , ( 4 ) interruption of tidal disruption flares in smbhbs systems @xcite . besides , there are also some emcs to probe the coalescence of smbhbs and its remnant , ( 1 ) intermittent activity in double - double radio galaxies at binary coalescence @xcite , ( 2 ) x - ray , uv , optical , and ir afterglow following binary coalescence @xcite , ( 3 ) systematically shifted broad emission lines relative to narrow emission lines @xcite and off - center active galactic nuclei ( agns ) @xcite because of smbh gw radiation recoil , ( 4)tidal disruption flares and hypercompact stellar systems for recoiling black holes @xcite . the breakthrough on numerical relativity in the past few years reveals the final stage of smbhb s coalescence @xcite . the coalescence remnant smbh can be recoiled due to the anisotropically gw emission during the inspiral and final coalescence @xcite . for nonspinning smbhs , the recoil velocity is @xmath5 @xcite , which is just as the same order of magnitude as the stellar dispersion velocity in the galactic center . however , if both of two smbhs are rapidly spinning , the recoil velocity can be as large as thousands kilometers per second . this recoil velocity depends sensitively on the intersection angle between spin vectors of two smbhs and their linear momenta . for some extreme cases , with maximally spinning equal - mass smbhs and antialigned spins in orbital plane , the recoiling velocity can achieve to @xmath6 @xcite . a recoiling smbh with high velocity has significant displacement relative to the galactic center . for most of massive galaxies with escape velocity @xmath7 @xcite , the recoiling smbh has opportunity to escape from its host galaxy . those recoiling smbhs which are still bound to host galaxies will oscillate around the galactic centers and change the core density profiles of the host galaxies due to dynamical friction @xcite . in addition , the recoiling smbh with a fraction of its accretion disk can produce sets of emission lines separated by relative velocity @xcite or off - center active galactic nuclei ( agns ) @xcite . in gas poor environment , a recoiling smbh might be observed as a `` hypercompact stellar system '' ( hcss ) . that is a kind of star cluster which is bound to the recoiling smbh , with similar luminosity as a globular cluster or ultracompact dwarf galaxy , and very high velocity dispersion @xcite . the other signature is an offset tidal flare due to the recoiling smbh tidally disrupting surrounding stars . @xcite have used both analysis and @xmath1-body simulations to calculate the tidal disruption rates of the recoiling smbh for bound and unbound stars . they found that the tidal disruption rates , in most cases , are smaller than in stationary systems . recently , @xcite have investigated the bound stellar density profile and tidal disruption evolution around the recoiling smbh . they found that the tidal disruption rate will monotonically fall as a power - law after the recoil . there is also an analytical work focusing on the tidal disruption rate during a short period just after coalescence @xcite , which predicts multiple tidal disruption flares in few years or decades after the coalescence . most of the above works about tidal disruption of the recoiling smbh have focused on theoretical analysis and neglected the impact of background stars . since the real evolution process in such kind of system is very complex , a detailed study on the recoiling smbh co - evolving with the galactic center is essential . unlike previous work , we evolve the entire system before and after the recoil with unbound stars included . our work focuses on investigating the co - evolution between the recoiling smbh and surrounding stars with tidal disruption and accretion processes . to it come true , we use a special high - accuracy , parallel direct @xmath1-body code accelerated by many - core hardware ( @xmath8-grape and @xmath8-gpu ) @xcite , including a simplified tidal disruption scheme @xcite . most of our simulations are calculated on the @xmath9 gpu cluster in national astronomical observatories of china ( naoc ) . with up to million of particles in simulation , we can find out the dynamical evolution of stars near the recoiling smbhs and estimate their tidal disruption rates , which may give us some hints for observation . our results show that the accretion of disrupted stars does not significantly impact the dynamical evolution of recoiling smbh . however , it changes the tidal disruption rate compared to a stationary smbh system in a galactic center . most of tidal disruption events are contributed by unbound stars when the smbh passing through the galactic center , and the disruption rate in average is @xmath2 , which is roughly an order of magnitude lower than the stationary case , @xmath10 . besides , the tidal disruption rate for oscillating smbh far from the galactic center is roughly one or two orders of magnitude lower than stationary smbh . we also find a bound stellar system around the recoiling smbh with a mass of @xmath3 black hole mass , which is consistent with @xcite . except for bound stellar systems , we find that there is a polarization cloud " composed of stars dynamically perturbed by the oscillating smbh . this polarization cloud " is axisymmetrically diffuse in a large spatial area which is beyond the gravitational influence region of the oscillating smbh . for this reason , most of the stars in the cloud are not permanently bound to the smbh . this kind of clouds are the echo of dynamical friction . the calculations for parameter dependence show that the variation of initial mass of the recoiling smbh can speed up or slow down the dynamical evolution , and impact the tidal disruption rate . in addition , the variation of density slope can obviously change the dynamical evolution and tidal disruption rate of the recoiling smbh . we give our galactic model and simulation method in section [ method ] . some results about bound / unbound stellar systems and tidal disruption rates for the stationary / recoiling smbh are showed in section [ results ] . the dependence of our results on other parameters are also investigated . section [ dissum ] gives discussion about the observational implications and a short summary . for simplicity , we adopt a spherical dehnen model to describe the ellipticals or bulges of galaxies @xcite , which is different from a srsic model used in @xcite . in a dehnen model , the space density profile follows ( r)= , [ dehnen rho ] where @xmath11 is the scaling radius , @xmath12 is the total mass of galaxies , and the slope index @xmath13 is a constant between interval [ 0,3 ) . the mass distribution is proportional to @xmath14 for @xmath15 and @xmath16 for @xmath17 . in this model , it is easy for us to derive the galactic potential , cumulative mass , and half - mass radius respectively @xcite , \(r ) = , [ eq : pot ] m(r)=m()^3- , [ eq : mass ] r_1/2=a [ 2 ^ 1/(3-)-1]^-1 . [ eq : halfr ] where @xmath18 is the gravitational constant . from equation ( [ eq : pot ] ) the escape velocity at radius @xmath19 for @xmath20 is @xmath21 thus the escape velocity from the galactic center for @xmath20 is v_esc(0)=()^1/2 . [ eq : vesc0 ] if there is a smbh with mass @xmath22 at the galactic center , we can estimate its influence radius @xmath23 by using the definition with stellar mass @xmath24 as in @xcite to estimate @xmath23 at an order of magnitude . ] r_inf=. [ eq : rinf ] for simplicity , we adopt model units @xmath25 thereafter . in this new unit , the quantities relate to physical quantities with the scale factor of time , velocity , and length , respectively @xmath26 & = & \left(\frac{gm}{a^3}\right)^{-1/2 } \nonumber \\ & = & 1.491\times 10 ^ 6(2^{\frac{1}{3-\gamma}}-1)^{3/2}\left ( \frac{m}{10^{11}\msun}\right)^{-1/2}\left(\frac{r_{1/2}}{1\kpc}\right)^{3/2 } \yr , \\ \label{eq : scalingv } [ v ] & = & \left(\frac{gm}{a}\right)^{1/2 } \nonumber \\ & = & 655.8\times ( 2^{\frac{1}{3-\gamma}}-1)^{-1/2}\left ( \frac{m}{10^{11}\msun}\right)^{1/2}\left(\frac{r_{1/2}}{1\kpc}\right)^{-1/2 } \kms , \\ \label{eq : scalingr } [ r ] & = & a=(2^{\frac{1}{3-\gamma}}-1)\left(\frac{r_{1/2}}{1\kpc}\right ) \kpc.\end{aligned}\ ] ] with the model units , equations ( [ eq : halfr ] ) , ( [ eq : vesc0 ] ) , and ( [ eq : rinf ] ) can be rewritten r_1/2=[2 ^ 1/(3-)-1]^-1 , [ eq : halfrd ] v_esc(0)=()^1/2 , [ eq : vesc0d ] r_inf=(2)^. [ eq : rinfd ] for a fiducial model adopted in the following calculations with @xmath27 , @xmath28 , @xmath29 and @xmath30 , we can derive @xmath31 and @xmath32 . compared to a scattering experiment , direct @xmath1-body simulation can well represent the dynamical co - evolution of the recoiling smbh and surrounding stars , especially for the stellar interactions . however , it can not deal with other process well , for example , the tidal disruption by smbh . to solve this problem , we implement a special tidal disruption scheme in our simulation . a star with mass @xmath33 and radius @xmath34 will be tidally disrupted if it approaches a bh within the tidal radius @xcite @xmath35 here @xmath36 is a dimensionless parameter of order unity . it is roughly @xmath37 for the real conditions in galactic nuclei @xcite . based on equation ( [ eq : rt ] ) , to make sure that the tidal radius is larger than the event horizon of the bh , there is a limitation for @xmath22 . for a disrupted star with solar radius @xmath38 and mass @xmath39 , a schwarzschild bh should have @xmath40 . this limitation mass can be heavier when the bh has rapid spin . in order to investigate the tidal disruption of the recoiling smbh with @xmath1-body simulation , we have to find a proper way to represent the tidal radius in our simulations . as shown in equation ( [ eq : rt ] ) , the tidal radius is proportional to the radius of the disrupted star . however , in our @xmath1-body simulations without stellar evolution scheme , stars are point like particles , whose radii are not defined . we can use the approximate scaling relation between @xmath23 and @xmath41 discussed above to set a proper value for @xmath41 , but this value is so small that there will be only few tidal disruption events recorded in the simulation . that is because of the limitation on particle resolution . our simulation particle number is not enough even with @xmath42 . to solve this problem , a simplified strategy is adopted . we assume a larger @xmath41 in our @xmath1-body simulation to represent the tidal radius . the particle which get close to smbh within @xmath41 will be tidally disrupted and removed from the system with its mass added onto the smbh . with this method , we can directly follow the growth of @xmath22 and calculate the tidal disruption rate . however , the large @xmath41 we adopted here will overestimate the tidal disruption rate . the solution is to carry out a series of simulations with decreasing @xmath43 and extrapolate the results to the regime corresponding to real galaxies . our extrapolation scheme is explained in section [ calc num ] in detail . we do not consider the growth of @xmath41 for accreting smbh or the momentum transfer from tidal disrupted stars to the recoiling smbh , because the total mass of disrupted stars is negligible compared to the bh . a test simulation taking into account momentum transfer shows no significant difference in either the dynamical evolution of recoiling smbh or the stellar disruption rate . to investigate the evolution of the recoiling smbh with tidal disruption carefully , we make a series of integrations with different parameters which are listed in table [ tab : para ] . the first column is the sequence number for different models . columns ( 2 ) - ( 4 ) give particle number @xmath1 , tidal radius @xmath41 , and initial stellar density slope @xmath13 respectively . column ( 5 ) is the initial mass ratio between @xmath22 and the total mass of system . column ( 6 ) gives initial recoil velocity in the unit of escape velocity at the galactic center . all of our integrations adopt a parallel direct @xmath1-body @xmath8-grape/@xmath8-gpu code with fourth - order hermite integrator and simplified tidal disruption scheme . they all have time - step accuracy parameter @xmath44 and a softening length @xmath45 . most of our models adopt @xmath42 equal mass particles , with initial mass of smbh @xmath46 to represent the ellipticals / bulges and central smbh respectively . to find out the dependency of the results on particle number , several sets of integrations with @xmath47 and @xmath48 particles have been included . there is also a calculation with @xmath49 to check the dependency on @xmath22 . as mentioned before , for the purposes of extrapolating to the smaller tidal radius in the real galactic center conditions , we have varied the @xmath43 from @xmath50 to @xmath51 , which is roughly corresponding to @xmath52 to @xmath53 for @xmath54 . since the density slope prior to the recoil is under debate , we choose @xmath54 as our fiducial value , and also investigate other cases with @xmath55 and @xmath56 . the recoil velocity @xmath57 is set to @xmath58 for most of the integrations , and there are calculations without recoiling velocity for comparison . besides , an integration with @xmath59 and @xmath55 is included for comparing with @xcite . since dehnen model provides the stellar distribution in a galaxy without central smbh , we have to first to set up consistent initial conditions for our system . at first , we put a smbh into the center of a stellar system with dehnen profile , and evolve the entire system for 50 simulation time units to relax the core region . after that the system is dynamically and thermally relaxed in relation to the smbh gravitational potential . at this stage , we do not consider stellar tidal disruption process . however , we want to make sure that stars inside the tidal disruption loss cone of the central smbh have already been removed by tidal accretion before the recoil of smbh . therefore , as a second step , we switch on the tidal disruption scheme in the program and run it again with the smbh still in equilibrium in the center - this costs a few dynamical timescales at the influence radius . thus all particles within the tidal disruption loss cone can be removed before we apply a kick to the smbh . after setting up the initial configuration of the stellar distribution of the systems , we artificially give a recoil velocity to the smbh and follow its dynamical evolution . to carefully investigate the dynamical evolution of recoiling smbh , we have a long time integration with model @xmath60 to @xmath61 , as shown in figure [ fig : osc ] . it is clear that the evolution of the recoiling smbh can be easily divided into two different phases . the phase i has obviously damping trajectory from @xmath62 to @xmath63 . after that , the smbh evolves into phase ii , where the trajectory damped very slow for a long time . in phase i , the trajectory of recoiling smbh can be well predicted by chandrasekhar s dynamical friction theory @xcite . however , this stage will just keep till the smbh s oscillation decayed to the size of core , where a slowly decayed phase ii begins . that is because the stellar mass interior to smbh s orbit is roughly equal to @xmath22 at the end of phase i , where the assumptions of chandrasekhar s dynamical friction theory are invalid . we have tried to fit the trajectory analytically in phase i with the same method as @xcite used , and got a similar result even though a dehnen model is adopted here instead of their core - srsic model . if we continue the integration , the orbital oscillation of the smbh will slowly decay and finally achieve to a thermal equilibrium with surrounding stars , which is the so - called phase iii in @xcite . here we have not continue our integrations into that stage because phase iii is actually similar to the case of a stationary central smbh , which is out of the scope of this paper . we will just focus on phase i and phase ii in the following integrations . thus all of our models integrate to @xmath64 in order to make sure that the evolution can achieve to phase ii . with regard to the orbital motion of the smbh , our results agree qualitatively well with those of @xcite . that is because the growth of the black hole due to tidal accretion , which did not take into account in @xcite , is relatively small . a recoiling smbh can carry a retinue of bound stars . @xcite predict that there will be a hypercompact stellar system " ( hcss ) bound to the recoiling smbh . we can analytically estimate the bound population of the hcss with equations ( 1a ) and ( 6a ) in @xcite . for example , the roughly boundary radius @xmath65 of hcss should be @xmath66 with our model @xmath60 , where @xmath67 , we have @xmath68 in our simulation unit . thus we can define a factor @xmath69 to describe the fraction of bound stars , @xmath70 where @xmath71 is total mass of bound stars . based on equations ( [ eq : rej ] ) and ( [ eq : rinfd ] ) , we have @xmath72 . for the case of @xmath54 , with @xmath73 in figure 1a of @xcite , we have @xmath74 . that means the ratio for hcss mass to @xmath22 should be @xmath75 , corresponding to just few particles for @xmath76 simulation . it should be aware that this fraction is just a lower limit because our two times @xmath22 " assumption actually overestimate @xmath23 . to confirm the conclusion above , we investigate this problem with @xmath1-body simulations . in the simulation , it is very difficult to distinguish whether a star is bound to smbh or not . that is because most of star particles with negative total energies relative to the smbh are loosely bound . only few of them could orbit the smbh for more than one orbital period . for this reason , we identify a star to be bound to the recoiling smbh if following two criteria are both satisfied : the particle ( 1 ) has negative total energy relative to the smbh and ( 2 ) remains bound for at least one orbital oscillation period of the recoiling smbh . we find that most of the bound " particles around smbh particle will be evaporated before the end of simulation . this effect is more significant for our escaping smbh calculation with model 10 , where the number of bound particles decreases monotonically . for other oscillating smbh calculations , there are many unbound particles can be captured by the smbh . that means the interaction between the hcss and stellar background is very important . figure [ fig : realbd ] shows the evolution of bound stars around a recoiling smbh in model @xmath60 . at @xmath62 , when @xmath77 , there are five bound particles , roughly consistent with the estimation from equation ( [ eq : fb ] ) . the number of bound particles , @xmath78 , slightly increases during the evolution from phase i to phase ii . because in phase ii ( 1 ) the low relative velocity between stars and the recoiling smbh increases the probability of dynamical capture and ( 2 ) the stellar density around the smbh in phase ii is higher . our results suggest that a hcss may grow via capturing unbound single stars . however , the low particle resolution and large statistical error prevent us from further investigation of the growth of hcss . the hcss may be detected as an off - nuclear compact system of similar size and stellar mass of a globular clusters but of having very high internal velocities @xcite . however , that is not the whole story . as shown in the figures [ fig : traceapo ] and [ fig : trace ] , we find that there is a stellar cloud composed of stars which are strongly impacted by the recoiling smbh at the first apocenter . the total mass of this cloud is comparable to @xmath22 . they form a quasi - axisymmetric stellar cloud and always have a dense region around smbh . this kind of clouds gives an evidence for the echo of dynamical friction . for this reason , we name that stellar cloud as polarization cloud " . in our simulation of model 06 , we choose a time snapshot when the smbh arrives at its first apocenter , and pick up those polarization cloud members from the @xmath79 phase space which includes all of star particles . all of polarization cloud members in the @xmath79 phase space are outliers around smbh particle and can be easily distinguished . thus we can collect those polarization cloud particles and trace their evolution . however , it is very difficult to find an automatical solution to efficiently distinguish those polarization cloud members for every time snapshot . to deeply investigate this kind of polarization clouds , an efficient method to extract those outlier particles is needed . this problem should be solved in our following work . in figure [ fig : traceapo ] , we plot that polarization cloud at time @xmath80 , when the recoiling smbh arrives at its first apocenter . all the parameters here are obtained from model @xmath60 . in order to investigate the origin and evolution of those cloud members , we look back the data and follow the evolution of the stars from time @xmath62 . figure [ fig : trace ] gives the evolution of those stars and the recoiling smbh from @xmath62 until @xmath81 , when our computation stopped . the snapshots with @xmath82 in figure [ fig : trace ] show the evolutions of the cloud members from beginning toward the first apocenter of the recoiling smbh . the following three snapshots correspond to the time when the recoiling smbh comes back from the first apocenter and passes through the density center for the second time . the snapshot at @xmath83 shows the smbh passing through galactic center for the third time , and the last snapshot is for the end of computation . our calculation results and figure [ fig : trace ] imply that the shape of polarization cloud varies with the oscillations of the recoiling smbh . the comparison between the first and the last two panels vividly shows the changes of the cloud members during the oscillation . as shown in figure [ fig : realbd ] , there are only small fraction of stars really bound to the smbh , whereas the mass of the polarization cloud is roughly equal to @xmath22 . therefore most of these cloud members are unbound to the recoiling smbh . after the recoil of smbh , those polarization cloud members are accelerated and fall behind the oscillating smbh . during the oscillation , some of the stars gradually expand to larger area while others continuously follow the smbh to oscillate around the galactic center , and form an axisymmetrical distribution along the velocity direction of the recoiling smbh . after multiple interactions with oscillating smbh , these stars obtain energy from oscillating smbh and diffuse to a large area . this is a vivid example that an oscillating smbh transfer its orbital energy to surround stars through dynamical friction . in order to investigate the observational properties of this polarization cloud , we focus on its most compact stage , when the recoiling smbh arrive at the apocenter for the first time . the distribution of those stars are shown in figure [ fig : traceapo ] . here we scale the @xmath1-body system to our physical fiducial galaxy model with stellar mass @xmath84 and half mass radius @xmath30 . for such system , the polarization cloud has mass of @xmath85 , and a size of diameter @xmath86 . further calculation shows that , its half mass radius is @xmath87 . this size and mass are much larger than a globular cluster and comparable to an ultracompact dwarf galaxy . we also calculate the velocity dispersion of this polarization cloud . the line - of - sight velocity dispersion along x , y and z axes are @xmath88 , @xmath89 and @xmath90 individually . that is significantly higher than a typical ultracompact dwarf galaxy . comparing with bound hcss , this polarization cloud has larger size , heavier mass and similar velocity dispersion . since this cloud is mixed with hcss , it may bring troubles to the detection of hcss . we also estimate the average surface density inside the half mass radius of the polarization cloud . the overdensity of the polarization cloud relative to the stellar background is only @xmath91 4% . this may be because of that the apocenter of the recoiling smbh here is not very far from the galactic center , where the surface density of the stellar background is still relatively high in model 06 with @xmath54 . for a smbh with a higher recoil velocity and a larger apocenter , the situation may be improved . to investigate this problem , more simulations are needed . we will investigate this problem in our future work . it should be noted that our example above is just for one time snapshot of the evolving smbh . every snapshot could have a similar stellar cloud . the composition of these clouds may form an interesting structure which can trace the trajectory of the smbh . the evolution and observation characters of this kind of structures will be discussed in our future work . for a recoiling smbh with highly enough velocity to escape from the galaxy , that kind of polarization cloud may not be distinguishable because many stars will be dropped behind the escaping smbh . without the multiple interactions between oscillating smbh and surrounding stars , the polarization cloud can not survive for a long time . for a solar type star disrupted by a smbh with mass @xmath92 , based on equation ( [ eq : rt ] ) and ( [ eq : scalingr ] ) , the tidal radius is @xmath93 , corresponding to @xmath94 in simulation units with our fiducial model . as mentioned in section [ gala model ] , because of the limitation on particle resolution , it is practically impossible with present computing capabilities to calculate the tidal disruption rate directly ( with a realistically small tidal radius ) through direct @xmath1-body simulations . to solve this problem , we perform a parameterized study , which treats both the particle number @xmath1 and the size of the tidal disruption radius @xmath41 as free parameters . all of the simulations have been done with these varying parameters , and physical conclusions are drawn from extrapolating to the real galactic conditions . @xcite have shown in fokker - planck models that such scaling is reliable and provides physically correct results . besides , tidal disruption events in our simulations , as well as the case in real galactic center , do not occur frequently , which means that the increase of @xmath22 is discrete . thus it is difficult to compute the tidal disruption rate through calculating accreted masses in time bins . instead , we measure the tidal disruption rate in a cumulative , averaged way - fitting the mass growth of the black hole and derive its time derivative . in order to find out the extrapolation relation , the dependence of tidal disruption rate on particle numbers and tidal radius should be investigated . figure [ fig : tdnrt ] shows the disruption rates changes with different @xmath1 and @xmath41 , with parameters are @xmath95 , @xmath96 , @xmath54 , @xmath46 and @xmath77 . the left panel shows that the dependence of averaged tidal disruption rates on @xmath1 and @xmath41 . here we calculate the averaged disruption rates for phase i , phase ii and entire process separately with different @xmath1 and @xmath41 . these integrations show that the dependence of tidal disruption rate on particle number is very weak . further discussions are showed in section [ nd ] . based on the loss cone feeding theory @xcite , contrary to empty loss cone case , the tidal disruption rate for full loss cone does not depend on @xmath1 . it should be notice that the tidal disruption rate here is for mass disruption rate , and the disrupted particles indeed depend on @xmath1 . the slight different average rates between @xmath76 and @xmath48 cases means that the loss cones in our integrations are nearly full and the changes of particle numbers do not bring significant impact . opposed to particle numbers , the changes on @xmath41 strongly influence the disruption rates . a detailed study is in the right panel of figure [ fig : tdnrt ] , which shows the evolution of disruption rate - @xmath41 dependence for several special points with integrations for @xmath97 . five special snapshot have been picked out . here 1st . apo . " , 1st . back " , 1st . down " , 2nd apo . " and phase tran . " refer respectively to the snapshot that the oscillating smbh the first time arrives at the apocenter , the first time returns to the density center , the first time reaches the opposite apocenter , the second time return to the apocenter and the time of transition from phase i to phase ii . our results show that the disruption rates increase for every @xmath41 value during the evolution of the recoiling smbh . that is because a decayed orbit of smbh leads to a growing stellar background around it . besides , the similar linear disruption rate - @xmath41 relation as left panel has been found . however , the relation here for the first four snapshots are not as good as the average results in the left panel . that may because our particle resolution for small @xmath41 is not good enough in phase i. based on our result data , it seems that there is an approximate linear relation between @xmath41 and disruption rates . as a result of full loss cone condition , the disruption rates will be proportional to cross section @xmath98 and density @xmath99 near the recoiling smbh , s(r ) _ rela , [ eq : tdr ] where @xmath100 is average relative velocity between the recoiling smbh and the surrounding stars . it is difficult for us to analytically derive stellar distribution around a recoiling smbh . however , the cross section can be wrote as sr_t^2(1 + ) , [ eq : cs ] which can write the tidal disruption rate as r_t^2(r ) _ rela(1 + ) . [ eq : mdot ] for @xmath101 that is @xmath102 ( even if one takes into account that @xmath103 contains a contribution from the smbh kick velocity ) , so we are in the gravitational focusing regime , where sr_t . [ eq : csp ] thus we have ( r)r_t . [ eq : mdotsimp ] if @xmath104 and @xmath100 are constant with time , the tidal accretion rate will scale linearly with @xmath41 . this relation is clearly only correct in a time averaged sense , because when the smbh oscillating in the galaxy , all quantities will be strongly changed within a single orbit ( density , stellar velocity dispersion and smbh velocity , the two latter quantities determining @xmath100 ) . however , in a running time averaged over several orbits , our results show that there is not a large variation in these quantities . so we can define a dimensionless tidal accretion parameter @xmath105 by the following relation : t_dyn = . [ eq : alpha ] here @xmath106 is the dynamic time scale around the position of the recoiling smbh . our simulations imply that across all of our model families @xmath105 does not depend strongly on @xmath1 or @xmath41 , but only on the evolutionary phase . thus @xmath105 can be used in extrapolations to the real system with very small @xmath41 and large @xmath1 . based on the analysis above , though it is complicated to get an accurate scaling relation , we still can estimate the tidal disruption rate through linear extrapolation of tidal radius . for comparison purposes , we calculate the tidal disruption rate of a stationary smbh without recoil in the galactic center . a series of simulations with different @xmath41 values are carried out . figure [ fig : tdr ] shows the static disruption rate for @xmath96 , corresponding to model @xmath107 , @xmath108 , @xmath109 and @xmath110 respectively . as mentioned in section [ calc num ] , the tidal radius for our fiducial model is @xmath111 , corresponding to @xmath94 in our simulation units . our simulation results with different particle numbers for stationary case show that the tidal disruption rates weakly depend on the particle numbers . thus we can loosely accept a full loss cone condition for the stationary case . as discussed in section [ calc num ] , there is a roughly linear correlation between tidal disruption rate and @xmath41 for full loss cone case . based on equation ( [ eq : mdotsimp ] ) , we can extrapolate our simulation results to the real galaxy conditions with smaller tidal radii . figure [ fig : tdr ] shows that the tidal disruption rate for @xmath112 is @xmath113 in our simulation units . that can be extrapolated to @xmath114 , and gives us the tidal disruption rate @xmath115 , which corresponds to the order of magnitude @xmath10 for our fiducial model . in a recoiling smbh system , there are two populations of stars contributed to the tidal disruption rate . one is the bound stars around the smbh , and the other is those unbound stars encountered with the recoiling smbh . if the bound stars dominate , as demonstrated in @xcite for an escaping smbh , the tidal disruption rate should roughly keep a constant level at the beginning , and finally drop down because of the consumption of bound stars . on the contrary , if the tidal disruption rate is dominated by unbound stars , the results will depend on stellar density , relative velocity of smbh , and the size of tidal radius , which described by equation ( [ eq : mdot ] ) . if we just consider about the contribution of the unbound stars , for an oscillating smbh , its tidal disruption rate near the galactic center should be higher than the apocenter region because of the different stellar density . however , @xmath100 can also change the result . for instance , we have an estimation in our model @xmath116 with @xmath117 , @xmath54 , @xmath46 and @xmath118 . our simulation result shows that the first apocenter of the recoiling smbh is around @xmath119 . with the assumptions that @xmath120 and @xmath121 , we can estimate the disruption rate ratio from the galactic center to the first apocenter region . by substituted into equations ( [ dehnen rho ] ) and ( [ eq : mdot ] ) , that ratio is @xmath122 . when scaling to the real galactic center conditions with @xmath114 , it should be @xmath123 . moreover , this ratio also depends on @xmath100 . if @xmath100 near apocenters can be larger than @xmath124 , that ratio may be higher . unfortunately , we can not obtain an accurate value for @xmath100 . for estimation , it can be seen that the tidal disruption rates near the galactic center are several times or even an order of magnitude higher than the apocenter region . that conclusion is confirmed by our simulation results . figure [ fig : tde ] gives the distribution of the tidal disruption events count @xmath125 relative to distance @xmath19 in model @xmath116 . it shows that most of the tidal disruption events appear around the density center , which indicates a boosted tidal disruption rate when the oscillating smbh pass through the galactic center . a smbh around the apocenter with a nearly zero velocity always corresponds to a low @xmath100 , which should give us a high disruption rate . however , as equation ( [ eq : mdot ] ) shown , the low stellar density environment makes the tidal disruption events less than at galactic center . the higher disruption rate near the galactic center means that , for an oscillating smbh , most of tidal disruption events are contributed by unbound stars . for this reason , unbound stars are very important for calculating tidal disruption rate . figure [ fig : tdrk ] shows the tidal disruption rates for a recoiling smbh with different @xmath126 , which belong to model @xmath127 and @xmath128 respectively . to increase the accuracy of our fitting results for disruption rate in phase i and phase ii , we neglect the transition boundary between the two phases in the fitting . therefore , we fit disruption rate for phase i and phase ii separately . as a result , the fitted results are not continuous between two phases . the left part of figure [ fig : tdrk ] shows that the tidal disruption rates in phase i are linearly increasing . for different @xmath41 values , there is also a similar linear relation to stationary smbh case , even though there is an increasing tidal disruption rate instead of a constant value . here the calculation with model @xmath128 ( @xmath129 ) is not included because the particle resolution is too low to estimate disruption rate . with the same extrapolating relation described in section [ calc num ] , we estimate that the averaged tidal disruption rate for phase i should around the order of magnitude @xmath2 , which is about an order of magnitude lower than stationary smbh in the galactic center . for the phase ii , when the orbit of the recoiling smbh damped to relative small region , the variations of stellar density and velocity of smbh are smaller than phase i. thus the difference of disruption rate between apocenters and galactic center is not so intensive . since the tidal disruption events during entire phase i are few in our simulation , the accuracy for calculating tidal disruption rate is not very good . however , there are plenty of tidal disruption events in phase ii , which provides a better accuracy to calculate the tidal disruption rate in phase ii . actually , as shown in the right part of figure [ fig : tdrk ] , there are constant disruption rates as similar as the stationary case . that is because ( 1 ) the stellar densities around smbhs in phase ii and in the stationary smbh case are similar and ( 2 ) their loss cones are both full . since @xmath76 particles do not have enough high resolution to represent a real galaxy , it is important for us to find out the dependence of our results on the particle number . only with this considered , the direct @xmath1-body simulation results for dynamical properties of the recoiling smbhs and surrounding stars can be extrapolated to real conditions in galactic nuclei . to bring this forth , a series of integrations with various @xmath1 and same other parameters have been done , which relative to models @xmath130 , @xmath131 and @xmath60 with @xmath132 respectively . figure [ fig : ndep ] shows the particle number dependence of the recoiling smbh orbital oscillation and mass increasing for model @xmath130 , @xmath131 and @xmath60 respectively . it can be seen that the @xmath48 and @xmath76 cases have similar oscillate amplitudes in phase ii , whereas the @xmath47 run can not well represent the damped evolution of the recoiling smbh . besides , only the calculations with @xmath48 and @xmath76 particles can give relatively smooth mass increasing curve , and they all achieve to the similar final mass . limited by the particle resolution , we can not obtain good enough mass evolution data to fit the disruption rates in phase i both for @xmath47 and @xmath48 integrations with model @xmath130 and @xmath131 . the only thing can be done is to calculate average tidal disruption rates for phase i and ii , as discussed in section [ td ] with figure [ fig : tdnrt ] . however , those problems do not exist for all of @xmath76 integrations and other @xmath48 integrations . for this reason , most of our calculations adopt @xmath76 particles . as shown in figure [ fig : tdnrt ] , the difference of average disruption rates between @xmath48 and @xmath76 is very small , which also indicates that the loss cones are nearly full . with @xmath76 particles simulation , the mass ratio of smbh to stars should be @xmath133 . this ratio is smaller than the real galaxy condition which should be @xmath134 . the small mass ratio between smbh and stars may lead to a higher brownian velocity comparing with real galaxy condition @xcite . however , most of our simulations are terminated before phase iii , which makes the amplitude of oscillating smbh is greater than brownian amplitude . detailed discussion in @xcite with series of calculations for different particle numbers shows that the particle mass can not significantly impact the dynamical evolution for both phase i and ii . the limitation of particle number can also influnece the relaxation time @xmath135 . for a homogenous isotropic distribution with equal - mass stars , the two - body relaxation time in our simulation units is @xcite t_r(r ) = , [ eq : tr ] where @xmath136 is coulomb logarithm and we can estimate it as = . [ eq : cl ] when @xmath1 is large enough , for instance @xmath137 , the relaxation time will be roughly proportional to @xmath138 . it means that our @xmath1-body simulation with @xmath76 particles always give us a shorter relaxation time comparing with real galaxy condition . this fast relaxation in @xmath1-body simulation may bring troubles to our results . if the core region in our simulation has relaxed before the recoiling smbh return , our results will departure from the real galaxy condition , because the relaxation time in a real galactic core is always very long . to solve this problem , we calculate the relaxation time of the central core region for both the simulation model and real galaxy condition . we choose @xmath139 , @xmath42 and @xmath140 in model @xmath60 to calculate the relaxation time @xmath141 . since we know little about the exactly value of velocity dispersion @xmath142 around @xmath139 , we range the @xmath142 from @xmath143 to 1 . for the real galaxy condition , our estimations show that , the @xmath141 is always longer than the half period of the oscillating smbh . that means the core region with a real galaxy condition can not relax before the smbh return . the same result applied to our simulations with @xmath144 for @xmath42 . based on equation ( [ eq : scalingv ] ) , @xmath145 corresponds to @xmath146 , which is smaller than most of the galactic center cases . thus we can conclude that the core region can not relax before the recoiling smbh return both for simulation and real galaxy condition . the limitation of particle number in @xmath1-body simulation may also impacts the results for the stationary smbh . in recoil case , the oscillating smbh always corresponds to the full loss cone status both for the simulation and the real galaxy condition . for a stationary smbh , things will be different . both the analytical calculations and the numerical simulations for long term evolution of tidal disruption from stationary smbh show that the loss cone in real galaxy is empty for most of the cases @xcite . however , our calculations are just focus on a relative short period , which correspond to @xmath147 for our fiducial model . during this short period , the loss cone could be roughly full . this prediction has been confirmed by our simulation results . based on discussions above , there are some conclusions : ( 1 ) both @xmath48 and @xmath76 integrations are good enough to represent dynamical evolution of the recoiling smbh , ( 2 ) in order to investigate the tidal disruption and co - evolution of the recoiling smbh with surrounding stars , it is better to use integrations with @xmath97 , ( 3 ) the small difference on average disruption rates between @xmath48 and @xmath76 integrations indicates that the loss cones are nearly full , otherwise the results should change obviously with various particle numbers , ( 4 ) the particle number we adopt in simulation is less than the real galaxy condition , which can lead to a shorter relaxation time scale . however , it can not significantly impact our results . based on the chandrasekhar s dynamical friction theory @xcite , the dynamical friction time scale is inversely proportional to the mass of the recoiling smbh . for this reason , a heavier smbh should has shorter evolution time scale in phase i. in order to investigate the dependence of evolution time scale on black hole mass @xmath22 , we run a test model @xmath148 with @xmath149 . comparing with model @xmath60 , we find that model @xmath148 gives a shorter oscillation period in phase i , which is roughly two times smaller than that in model @xmath60 . that is consistent with the prediction of dynamical friction theory . besides , the tidal disruption rate in model @xmath148 is roughly two times higher than that in model @xmath60 for both of the two phases , which is roughly consistent with equation ( [ eq : mdotsimp ] ) . for a galaxy with lighter smbh , which has @xmath150 , there will be a longer dynamical evolution time in phase i and smaller tidal disruption rate comparing with our results above . limited by the calculation capability , simulations with the lower mass smbh will cost so much time that we can not investigate them carefully . we note that in the simulations with very large tidal radius , for example , @xmath151 in model @xmath152 and @xmath153 , rapid increase on @xmath22 lead to the declined disruption rates . this significant growth of @xmath22 is artificial , which makes the declination of disruption rate is unphysical . when considering different density slopes , the results are quite different . the higher of @xmath13 value is , the faster of the system evolved . our integration with model @xmath154 shows that the recoiling smbh will enter phase ii around @xmath155 , which is more faster than evolution of smbh in model @xmath60 with @xmath156 . thus our models for @xmath55 and @xmath56 actually reach phase iii at the end of integration . that can be easily understood because a larger @xmath13 means a denser cusp in the galactic center . for the same reason , a larger @xmath13 also leads to a higher disruption rate . figure [ fig : rtgamma ] shows the disruption rates with different @xmath13 values for both the stationary and the recoiling cases . the left panel is for stationary integrations and the right panel is for the recoiling smbh with time duration @xmath157 . here we can not compare the same phase for different @xmath13 models because of their different evolving times . the results in the right panel are phase ii for model @xmath60 and phase iii for model @xmath158 and @xmath154 . it can be seen that , for both panels , there are several times differences between every two adjacent @xmath13 values . in order to compare with the work in the literature with higher @xmath159 , a simulation with similar initial parameters comparing with model @xmath60 has been done in model @xmath160 , where @xmath161 and @xmath162 . our result shows that there are just five particles tidally disrupted at the beginning of recoil , and following with zero event recorded for the rest of time . because of the low particle resolution , we can only give a lower limit to the tidal disruption rate . for a galaxy with mass @xmath163 , @xmath164 and @xmath165 , the kick velocity should be @xmath166 . thus our lower limit for tidal disruption rate is @xmath167 , which is lower than @xmath168 in figure @xmath169 of @xcite with similar conditions . for the bound stars , we find that there are only @xmath170 stars still bound to smbh after the recoil , which is roughly as the same order of magnitude as the result in @xcite . the small amount of bound stars , low stellar density outside and poor particle resolution result in the rare disruption events in computation . to solve this problem , a series of simulations with larger particle numbers are needed . supermassive binary black holes in galactic nuclei may under certain conditions get very close and coalesce due to strong emission of gravitational waves . because of the anisotropy of gw emission , the coalescence remnant smbh experiences a strong recoil velocity depending on spins of the original pair of black holes . in this paper we investigate the interaction and co - evolution of the recoiling smbhs with their galactic stellar environments , using very large direct @xmath1-body simulations with a simplified tidal disruption scheme . it is the first numerical investigation for the evolution of the tidal disruption by an oscillating smbhs with unbound stars included . moreover , we discover a polarization cloud of stars surrounding the oscillating smbhs , which is not mentioned in the literature . as argured by @xcite , the dynamical evolution of the recoiling smbh can be divided into three phases . during the phase i , the trajectory of the recoiling smbh can be well predicted by chandrasekhar s dynamical friction theory @xcite . the phase ii begins when the smbh s oscillation decayed to the size of galactic core , and the motion of the smbh damps slowly . at phase iii , the orbital oscillation of smbh slowly decay to achieve a thermal equilibrium with surrounding stars . in our simulation , we obtain similar phase i and ii as @xcite have got . besides , our results show that the mass fraction of bound stars relative to smbh is roughly @xmath171 at the beginning . this value is consistent with @xcite and @xcite . in addition to following the dynamical evolution of the recoiling smbh , we estimate the stellar tidal disruption rates by the recoiling smbhs using direct @xmath1-body simulation together with our simple tidal disruption scheme . our results show that the stellar tidal disruptions do not significantly change the dynamical evolution of the recoiling smbh . the stellar disruption rate by the recoiling smbh in phase ii is @xmath10 , which is about the same order of magnitude of the stellar disruption rate by the stationary smbh at a spherical galactic center . however , in phase i , the stellar tidal disruption rate is about an order of magnitude lower . in our simulation , a spherical stellar distribution is assumed . this simplified model is not accurate to represent a real galaxy . for most of real bulges / ellipticals , their stellar distributions are non - spherical . as argued by @xcite , the orbital decay time scales of the recoiling smbhs are prolonged in triaxial galaxies comparing with spherical galaxies . that is because most of the recoiling smbhs in triaxial galaxies can not return directly to their galactic centers , where the influences of dynamical friction are the strongest . for the same reason , a non - spherical dark matter halo can also significantly increase the decay time of a recoiling smbh @xcite . since most of the tidal disruption events happened during the smbhs passing through the galactic centers , the triaxiality may significantly reduce the tidal disruption rate from unbound stars . to investigate the impacts of triaxiality , a good solution is to have series of simulations with triaxial stellar distributions . however , it is very complicated to generate a stationary triaxial distribution for @xmath1-body simulation , and the prolonged orbital decay time needs more calculation resources . here we just follow previous works to investigate the tidal disruptions from the recoiling smbhs in spherical stellar distribution . further works with triaxial stellar distribution will be included in our next paper . actually , the triaxiality can not change our results significantly because the influence of triaxiality is moderate in galaxy models with shallow central density profiles @xcite . thus our calculations with @xmath27 will not significantly departure from the triaxial model . besides , in most of our simulations , the maximal displacement of the recoiling smbh relative to the density center is no more than @xmath172 . that keeps the trajectory of the smbh inside the bulge , where the impact of dark matter halo are very weak comparing with baryonic matter . for this reason , we neglect the influence of the triaxial dark matter halo which is mentioned by @xcite . our simulation for escaping smbh does not have enough particle resolution to estimate disruption rates . that needs much larger @xmath1-body particle number in the future simulations . here our integrations are focus on relatively low recoiling velocity . for the assumption that @xmath28 and @xmath30 , recoil velocity in our fiducial model is @xmath173 . this velocity produces the oscillating smbh an amplitude less than the effective radius of elliptical / bulge . our results show that most of tidal disruption events occur when the recoiling smbh passes through density center , but not exact the original center because of the orbital precession . the tidal disruption rates around each apocenter are much lower . an estimate for average tidal disruption rate at about apocenters in phase i over several periods for model @xmath60 has been done . we calculate the average disruption rate when smbh have large distant relative to density center . the disruption rate at large distance is several times or even an order of magnitude lower than the average value over whole phase i , which implies that the rate for apocenters is one or two order of magnitude lower than that by stationary smbhs . @xcite argued that off - nuclear x - ray tidal flares could be one of the observational signatures for an escaping smbh . unfortunately , our simulation of model @xmath160 with @xmath76 particles and the escapable recoil velocity does not have sufficient resolution to estimate the disruption rate . we can just give a lower limit @xmath167 , which does not conflict to their work with similar configurations . besides , our model @xmath158 with @xmath55 and recoil velocity @xmath174 corresponds to a tidal disruption rate @xmath175 . this value is similar to the result of model 1 with @xmath176 in @xcite . based on our results , an oscillating smbh with enough large recoil velocity may also contribute to off - nuclear x - ray flares around its apocenters with the low disruption rates . an hcss around the recoiling smbh in our integrations has similar mass as predicted by @xcite . however , we can not give more information about the star clusters limited by particle resolution . in addition , we find a larger cloud of stars around the recoiling smbhs , which are not permanently bound to the black hole but keep a kinematic correlation to it for more than an orbital time . it can be described as polarization cloud reacting to the gravitational disturbance by the recoiling smbh , in other words , the echo of dynamical friction . the size and mass of the polarization clouds are comparable to ultracompact dwarf galaxies , except significantly higher velocity dispersion . the existence of such kind of system may bring difficulties for distinguishing the hcss . more detailed properties of the polarization cloud will be studied in our following work . our simulations for different particle numbers show that both @xmath48 and @xmath76 integrations are sufficient in following the dynamical evolution of the recoiling smbh , but the former do not have enough resolution for investigating the tidal disruption and co - evolution of smbh with surrounding stars . different initial masses of the recoiling smbhs can change the results moderately . we find that a higher galactic density slope can significantly speed up the dynamical evolution of the system , and the tidal disruption rate is also boosted . support for this work was provided by the chinese national @xmath177 program ( @xmath178 ) , the national natural science foundation of china ( @xmath179 ) , the research fund for the doctoral program of higher education ( rfdp ) , and the china scholarship council for financial support ( @xmath180 ) . rs and pb acknowledge support by naoc cas through the silk road project and ( rs ) through the chinese academy of sciences visiting professorship for senior international scientists , grant number @xmath181 . we thank for partial support by global networks and mobility program of the university of heidelberg ( zuk 49/1 tp14.8 spurzem ) . the special supercomputer at the center of information and computing at national astronomical observatories , chinese academy of sciences , funded by ministry of finance of people s republic of china under the grant @xmath182 , has been used for some of the largest simulations . the titan cluster has been used , funded by the volkswagen foundation under grant no . some numerical computation has also be done on the sgi altix @xmath184 system at the astronomy department , peking university . we are grateful to x .- b . wu , zhaoyu li and heling yan for useful discussions . many thanks are due to the referee for valuable comments . . here the bottom horizontal axis and the left vertical axis give the evolved time and the displacement of bh @xmath185 respectively , with the units @xmath25 . the right and the top axis are transformed to real physical units with the assumption that @xmath28 and @xmath30 . it can be easily distinguished for the phase i and ii which are similar to the result in @xcite . [ fig : osc ] , scaledwidth=90.0% ] . the horizontal and vertical axis are the same as figure [ fig : traceapo ] , the right horizontal axis is the number count of the bound particles . the grey dotted line gives the trajectory of the recoiling smbh , and the black solid line gives the variation of the bound particles . [ fig : realbd ] , scaledwidth=90.0% ] for time snapshot @xmath186 . we project the stellar distribution onto @xmath187 plane , with the physical units under the assumption that @xmath28 and @xmath30 . the large black spot marks the position of the recoiling smbh . it can be seen that all of the impacted star particles form a cloud structure around the smbh at that moment . [ fig : traceapo ] , scaledwidth=90.0% ] . the units and definition of axis are the same as figure [ fig : traceapo ] . the black circle and the grey dots mark the recoiling smbh and the stars respectively . the black arrow represents the size and orientation of bh s velocity . it can be seen that many unbound stars are influenced by the smbh . [ fig : trace ] , scaledwidth=90.0% ] and @xmath41 . the left panel shows the dependence of the average tidal disruption rates on @xmath1 and @xmath41 . the solid line and dashed line are the average disruption rates for @xmath76 and @xmath48 integrations respectively . the square , triangle , and circles respect to the average disruption rates for phase i , phase ii and the entire process respectively . the integrations are based on model @xmath60 with different particle numbers and tidal radius . the integration for @xmath129 with @xmath188 is not included because the amount of resulted tidal disruption events is too small to calculate an average rate . the right panel shows the disruption rate - @xmath41 dependence for several key points with integrations for @xmath97 . here 1st . apo . " , 1st . back " , 1st . down " , 2nd apo . " and phase tran . " are corresponding to the moment which smbh the first time arrive at its apocenter , the first time come back to the density center , the first time arrive at apocenter on the other side , the second time arrive at its apocenter and the transition from phase i to phase ii respectively . for both of two panels , horizontal and vertical axis give the tidal radius and tidal disruption rate in the units of @xmath25 respectively.[fig : tdnrt ] , scaledwidth=90.0% ] . here the horizontal and vertical axis give the evolved time and tidal disruption rate in the units of @xmath25 respectively . the dotted , dashed , solid and dash dotted lines are calculations with different tidal radii @xmath189 , which correspond to model @xmath190 and @xmath110 respectively . [ fig : tdr ] , scaledwidth=90.0% ] relative to @xmath19 with model @xmath116 . here the horizontal and vertical axis give the cumulative tidal disruption events count and @xmath19 respectively . the shaded histogram marks the cumulative tidal disruption events count for the different distance . [ fig : tde ] , scaledwidth=90.0% ] . the left part and right part are the fitted results for phase i and phase ii respectively . the shaded region corresponds to a transition area from phase i to phase ii . axis are labeled as same as figure [ fig : tdr ] . the dotted , dashed , solid and dash dotted lines are the calculations with different tidal radii @xmath96 , corresponding to model @xmath127 and @xmath128 respectively . since the model @xmath128 with @xmath129 does not have enough particle resolution to calculate the disruption rate during phase i , it is absent in the left part . [ fig : tdrk ] , scaledwidth=90.0% ] , @xmath131 and @xmath60 . axis are labeled as same as in figure [ fig : osc ] . the dotted , dashed and solid lines are the calculations with particle numbers @xmath191 respectively . left panel is the damped orbital evolutions of recoiling smbhs with various @xmath1 . it can be seen that the @xmath48 and @xmath76 cases have a similar oscillate amplitude in phase ii , while the @xmath47 is different from others because of the poor particle resolution . the right panel shows the mass increasing for the recoiling smbh . here the vertical axis represents to the mass of the recoiling smbh . the different initial mass of the recoiling smbh is due to the special relaxation scheme described in section [ num method ] . only the calculations with @xmath48 and @xmath76 particles can give us relatively smooth mass increasing curve , and all the calculations achieve to similar order of magnitude for final mass . [ fig : ndep ] , scaledwidth=100.0% ] values and different models . the left panel is the disruption rates of stationary smbhs with model @xmath109 , @xmath192 and @xmath193 . the right panel shows disruption rates of the recoiling smbhs for the time interval @xmath157 , in model @xmath60 , @xmath158 and @xmath154 . here the horizontal and vertical axis stand for the evolving time and the disruption rates respectively , with the units of @xmath25 . the solid , dashed and dotted lines stand for the integrations with @xmath194 respectively . [ fig : rtgamma ] , scaledwidth=90.0% ] 01 & @xmath47 & @xmath195 & 0.5 & 0.001 & 0.7 + 02 & @xmath48 & @xmath195 & 0.5 & 0.001 & 0.7 + 03 & @xmath48 & @xmath196 & 0.5 & 0.001 & 0.7 + 04 & @xmath48 & @xmath197 & 0.5 & 0.001 & 0.7 + 05 & @xmath76 & @xmath195 & 0.5 & 0.001 & 0.0 + 06 & @xmath76 & @xmath195 & 0.5 & 0.001 & 0.7 + 07 & @xmath76 & @xmath195 & 0.5 & 0.002 & 0.7 + 08 & @xmath76 & @xmath195 & 1.0 & 0.001 & 0.0 + 09 & @xmath76 & @xmath195 & 1.0 & 0.001 & 0.7 + 10 & @xmath76 & @xmath195 & 1.0 & 0.001 & 1.1 + 11 & @xmath76 & @xmath195 & 1.5 & 0.001 & 0.0 + 12 & @xmath76 & @xmath195 & 1.5 & 0.001 & 0.7 + 13 & @xmath76 & @xmath51 & 0.5 & 0.001 & 0.0 + 14 & @xmath76 & @xmath51 & 0.5 & 0.001 & 0.7 + 15 & @xmath76 & @xmath197 & 0.5 & 0.001 & 0.0 + 16 & @xmath76 & @xmath197 & 0.5 & 0.001 & 0.7 + 17 & @xmath76 & @xmath196 & 0.5 & 0.001 & 0.0 + 18 & @xmath76 & @xmath196 & 0.5 & 0.001 & 0.7 + 19 & @xmath76 & @xmath50 & 0.5 & 0.001 & 0.0 + 20 & @xmath76 & @xmath50 & 0.5 & 0.001 & 0.7 +

the complete expression for the intensity in pseudo - scalar meson photoproduction with a polarized beam , target , and recoil baryon is derived using a density matrix approach that offers great economy of notation . a cartesian basis with spins for all particles quantized along a single direction , the longitudinal beam direction , is used for consistency and clarity in interpretation . a single spin - quantization axis for all particles enables the amplitudes to be written in a manifestly covariant fashion with simple relations to those of the well - known cgln formalism . possible sign discrepancies between theoretical amplitude - level expressions and experimentally measurable intensity profiles are dealt with carefully . our motivation is to provide a coherent framework for coupled - channel partial - wave analysis of several meson photoproduction reactions , incorporating recently published and forthcoming polarization data from jefferson lab .

1010.4978

the production amplitude for pseudo - scalar meson photoproduction involves eight complex amplitudes which depend on the spin states of the photon and target and final - state baryons . the reaction can be simplified by considering the parity invariance of the strong and electromagnetic interactions , reducing the number of independent complex amplitudes to four . barker , donnachie , and storrow ( bds ) @xcite showed that there exist fifteen experimentally observable single- and double - polarization observables which , in addition to the differential cross - section , can be expressed as bilinears in the four independent amplitudes . several ambiguities originate with the bds work . the bds article treats reactions with the four amplitudes in the helicity basis ( non - flip , @xmath0 , double - flip , @xmath1 , and two single - flip amplitudes , @xmath2 and @xmath3 ) , but does not clearly specify to which helicity configurations the amplitudes @xmath2 and @xmath3 refer . other authors @xcite follow different schemes for enumerating the four amplitudes . the construction of the helicity amplitudes presents a separate problem , as choice of phase conventions in wigner rotation matrices can also potentially lead to sign ambiguities . chiang and tabakin ( ct ) @xcite later showed that to completely characterize the full production amplitude , measurements of the differential cross - section and a carefully chosen set of only seven polarization observables is required ; that is , there is redundancy in the full set of sixteen bilinear observables . the ct study assumes `` measurements '' are made with infinite precision , a situation obviously unattainable by any experiment . more recently , the effects of uncertainty in polarization measurements on constraining amplitudes has been studied from an information theory perspective @xcite . sandorfi _ et al . _ @xcite have pursued descriptions of the polarization observables in terms of the chew - goldberger - low - nambu ( cgln ) amplitudes @xcite . they have tested configurations of production amplitudes which reproduce available polarization data for the reaction @xmath4 to within experimental uncertainties by randomly sampling the amplitude space and projecting observables from the amplitudes . their work shows that the currently observed set of polarization observables and experimental uncertainties do not provide enough constraint to distinguish between production models containing different resonance contributions , thus suggesting that measurement of a larger number of observables than prescribed by ct will be required to fully extract the four complex amplitudes . the scope of the present work is three - fold . first , we derive the general expression for the reaction intensity with all three polarizations ( beam , target , and recoil ) . our motivation is the density matrix approach of fasano , tabakin and saghai ( fts ) @xcite , the power of which is compactness of notation . the full expression consists of @xmath5 terms . invariance under mirror symmetry transformations ( a parity inversion followed by a rotation , see sec . [ sec : mirror_symmetry ] ) removes half of these terms . in the remaining terms , each of the sixteen physically measurable observables occurs twice . all results herein follow simply from the properties of the pauli matrices and the mirror symmetry operator acting on the spin density matrices of the photon and baryons . second , we provide amplitude - level expressions for the polarization observables corresponding to measurable particle momentum distributions , carefully keeping track of the relative signs between experimental measurements and amplitude - level expressions . our amplitudes are constructed in the longitudinal basis , that is , with spin projections for _ all _ particles quantized along a single direction , the beam direction . for reactions with multiple decays and non - zero spins for the the final - state particles , a single spin - quantization axis enables one to write the full production amplitude in a manifestly lorentz - invariant fashion . a relevant example of such a process is the reaction chain @xmath6 . finally , we list and numerically validate the various `` consistency relations '' connecting the different spin observables . these consistency relations provide important checks for both theoretical analyses and constraints in the case of future experiments which will have access to polarizations of the beam and target and recoil baryons . we aim to establish a consistent partial - wave - analysis formalism for recent @xcite and future @xcite meson photoproduction data from clas , as well as all presently available polarization data from other experiments such as graal @xcite and leps @xcite . to ensure correct interpretation of these observables , one must be confident of the relationship between measurements extracted from momentum distribution asymmetries and the polarization observables as treated in theoretical studies . until now , most of the world data on polarizations has been limited by statistics and kinematic coverage . furthermore , experimental limitations have restricted measurement to only a few of the fifteen polarization observables . however , with a new generation of experiments at jefferson lab on the horizon @xcite , much activity is anticipated in the field in the coming years . this article will provide a self - contained and comprehensive description of the formalism for pseudo - scalar meson photoproduction from the fundamental derivations and a careful treatment of the connection between theory and measurement of polarization observables . -axis is along the photon momentum direction and the @xmath7-axis is normal to the reaction plane . momenta of the incoming ( outgoing ) particles are shown in bold red ( blue ) arrows . @xmath8 is the polar meson - production angle in the c.m . see text for details.[fig : axis_setup],scaledwidth=48.0% ] in the case of single pseudo - scalar meson photoproduction , let @xmath9 , @xmath10 , @xmath11 and @xmath12 be the momenta of the incoming photon , target baryon , outgoing meson , and outgoing baryon , respectively , in the overall center - of - mass ( c.m . ) frame ( see fig . [ fig : axis_setup ] ) . the beam direction defines the @xmath13-axis , @xmath14 . the @xmath7-axis is taken to be normal to the reaction plane established by the photon and meson momenta , @xmath15 . the @xmath16-axis is then simply @xmath17 . here the subscript `` evt '' denotes that these axes , with @xmath18 and @xmath19 parallel and perpendicular to the reaction plane , are defined on an event - by - event basis . there is some disparity between the optics and particle - physics community in the nomenclature of the right- and left - handed polarization states . particle physicists define the right - handed polarization state following the right - hand rule for the transverse electric polarization vector . the spin of the photon points along its momentum for the right - handed polarization state ( or positive helicity state ) . the left - handed polarization state has the photon spin anti - parallel to its direction of motion . the optics community swaps the definitions for the right- and left - handed states , though the notions of positive- and negative - helicity states are the same in both treatments . here , we adhere to the particle - physics convention . we will define the polarization basis states for the photon as @xmath20 is the right - handed ( positive - helicity ) state , @xmath21 is the left - handed ( negative - helicity ) state , and @xmath22 and @xmath23 are states of transverse polarization along @xmath18 and @xmath19 , respectively . looking _ into _ the incoming beam , the @xmath7-component phase leads ( trails ) the @xmath16-component phase for the positive ( negative ) helicity states and the polarization vector rotates counter - clockwise ( clockwise ) for the positive ( negative ) helicity states , in accordance with the right - hand rule . for a general mixed state , it is useful to switch to the density matrix notation for describing the polarization state of the photon . we follow the work of adelseck and saghai ( as ) @xcite and fts @xcite and write the photon spin density matrix as @xmath24_{\scriptsize \mbox{as } } \nonumber \\ \!\!&=&\!\ ! \frac{1}{2 } \left[\begin{array}{cc } 1 + p^s_z & p^s_x -i p^s_y \\ p^s_x + i p^s_y & 1 - p^s_z \end{array}\right]_{\scriptsize \mbox{fts}}.\end{aligned}\ ] ] in the as prescription , the quantities @xmath25 and @xmath26 denote the degree of linear ( @xmath27 ) and circular ( @xmath28 ) polarization . in the fts treatment , @xmath29 is the stokes vector common in optics , with @xmath16- , @xmath7- , and @xmath13-components indicating the amount of polarization along each spatial direction . the kinematic variable @xmath30 is the azimuthal angle between the reaction plane and the laboratory @xmath31 for a linearly polarized photon for a given event . the linear polarization vector relative to the laboratory frame is @xmath32 defines the fixed laboratory axis of the experiment . the polarization direction ( @xmath33 ) is at an angle @xmath34 to the laboratory @xmath16-axis . for a given event , the reaction plane is at angle @xmath30 to the laboratory @xmath16-axis . therefore , the polarization vector is at an angle @xmath35 relative to the reaction plane.[fig : pol_angles ] , scaledwidth=50.0% ] for the circularly polarized photon ( or unpolarized beam ) case , the production amplitude is azimuthally symmetric about the beam direction . however , for the linearly polarized case , @xmath31 and @xmath36 define a preferred transverse coordinate system . the experimental conditions define @xmath37 and @xmath38 ( @xmath39 ) correspond to parallel ( perpendicular ) plane polarizations according to the experimentalist . for a given event , relative to the reaction plane , the polarization vector is @xmath40 and the stokes vector @xmath41 may be related to the circular polarization quantities by @xmath42 in the basis formed by @xmath43 and @xmath35 . the angles @xmath30 , @xmath34 , and @xmath44 are shown schematically in fig . [ fig : pol_angles ] and the connection between @xmath41 and the different polarization states are given in table [ table : pol_stokes ] . apart from the right- ( @xmath45 ) and left - handed ( @xmath46 ) circular polarizations , which are our basis states , there are the perpendicular ( @xmath47 ) and parallel ( @xmath48 ) states , corresponding to photons linearly polarized along the @xmath7- and @xmath16-axes , respectively , and two linearly polarized states at @xmath49 to the @xmath16-axis ( in the @xmath50-@xmath51 plane ) , labelled as @xmath52 . .stokes vector @xmath41 for different photon polarization configurations ( adapted from ref . the right- ( @xmath45 ) and left - handed ( @xmath46 ) circular polarizations are our basis states . the different configurations for the linearly polarized states can be expressed in terms of these basis states . @xmath44 is the angle the linear polarization direction makes with @xmath18 . see text for details . [ cols="<,^,^,^",options="header " , ] we first define the sixteen observables as the various @xmath53 elements ( see table [ table : obs ] ) noting that each observable occurs twice in the expansion given by eq . [ eqn : profile_lmn ] . our definitions for the observables follow those in fts @xcite . it is to be noted that there are minus signs in front of the defining @xmath53 elements for the four double - polarization observables that use a linearly polarized beam ( @xmath54 , @xmath55 , @xmath56 , and @xmath57 ) . these extra sign flips are needed to preserve the definitions of these variables as _ physical asymmetries _ , as given in fts @xcite . the signs for @xmath54 , @xmath55 , @xmath56 , and @xmath57 , in terms of the _ density matrix trace calculations _ , as given in appendix a in the fts article will therefore acquire sign flips ( see sec . [ sec : beam_target_profile ] and sec . [ sec : beam_recoil_profile ] for further details ) . we now write the full intensity profile in terms of the polarization observables as : @xmath58 we note that the three single polarizations ( @xmath59 , @xmath60 and @xmath61 ) occur again as double correlations and the twelve double polarizations ( @xmath62 , @xmath63 , @xmath54 , @xmath55 , @xmath64 , @xmath65 , @xmath56 , @xmath57 , @xmath66 , @xmath67 , @xmath68 and @xmath69 ) occur again as triple correlations . the expansion in eq . [ eqn : profile_lmn ] has 64 terms , while eq . [ eqn : full_profile_expanded ] has only 32 terms . the rest of the 32 terms vanish under @xmath70 invariance . @xmath71 and @xmath72 are examples of such terms ( they do not occur in table [ table : obs ] ) . physically , these two elements correspond to recoil polarizations ( with unpolarized beam and target ) along the @xmath50 and @xmath73 directions , which are required by @xmath70 invariance to be zero . the general structure of these vanishing terms can be understood from the following example . from eq . [ eqn : pi_reshuffle_photon ] for the photon , under a @xmath70 transform , @xmath74 is connected to the the identity matrix ( @xmath75 ) . similarly , from eqs . [ eqn : pi_reshuffle_baryon ] and [ eqn : pi_reshuffle_outgoingbaryon ] for the baryons , it is @xmath76 that is connected to the identity matrix . we group @xmath75 and the pauli matrix connected to @xmath75 by the @xmath70 operator as `` e '' ( type + 1 ) , and the rest ( @xmath76 and @xmath77 for the photon , and @xmath74 and @xmath77 for the baryons ) as of the `` o '' ( type -1 ) . a general correlation @xmath53 vanishes if the product of the `` types '' of @xmath46 , @xmath78 , and @xmath79 is -1 , since these are not invariant under the mirror symmetry transformation . we will show that our expressions for the intensity profiles as measured by the experimentalist conform to the definition of these observables as asymmetries . following the notation set up in fts @xcite we will denote the cross - section for any configuration of the beam , target and recoil polarizations as @xmath80 . for beam - target type experiments , @xmath81 , and eq . [ eqn : full_profile_expanded ] becomes @xmath82 the beam asymmetry is defined as @xmath83 where @xmath48 and @xmath47 correspond to a beams with polarizations along the @xmath18 ( @xmath84 ) and @xmath19 ( @xmath85 ) directions , respectively , and @xmath86 denotes an unpolarized configuration . the target asymmetry is defined as @xmath87 and the four double polarizations are defined as [ eqn : beam_target_defs ] @xmath88 where `` @xmath45 '' denotes a right - handed circularly polarized beam ( all photons in the state @xmath89 ) , and `` @xmath90 '' denotes a linearly polarized beam with @xmath91 with respect to @xmath18 . the full expression for the cross - section in beam - target experiments reads @xmath92 where we have added a subscript `` theory '' to remind the reader that this is for the theoretical formalism only . it can easily be checked that the definitions in eqs . [ eqn : beam_asym]-[eqn : beam_target_defs ] are consistent with the intensity profile given by eq . [ eqn : beam_target_profile ] . recall that @xmath91 corresponds to @xmath93 ( see table [ table : pol_stokes ] ) , explaining the extra minus signs for @xmath54 and @xmath55 in the definitions of the corresponding @xmath53 elements in table [ table : obs ] . it is to be noted that one has access to an `` extra '' single - polarization observable , the recoil polarization @xmath59 , even though the polarization of the recoiling baryon is not measured here . this is again due to the @xmath70 transform relations . in fact , any double - polarization experiment has access to all the three single polarization observables . the definition of @xmath59 as an asymmetry is given in the next sub - section . for an experiment with beam and recoil baryon polarization information , we follow a similar logic . here , @xmath94 , and the beam - recoil expression is @xmath95 where the recoil polarization @xmath59 is defined as @xmath96 the four beam - recoil double polarizations are [ eqn : beam_recoil_obs ] @xmath97 the `` extra '' single polarization observable accessible here is the target asymmetry @xmath61 , defined in eq . [ eqn : targ_asym ] . as in the case of @xmath54 and @xmath55 , the definitions of @xmath56 and @xmath57 as asymmetries use @xmath91 that corresponds to @xmath93 . this explains the extra minus signs in the defining @xmath53 elements in table [ table : obs ] for @xmath56 and @xmath57 . the target - recoil expression is @xmath98 where the four target - recoil double polarizations are [ eqn : target_recoil_obs ] @xmath99 until now , we have been careful to distinguish the `` theoretical '' intensity profiles from what experimentalists will actually measure . the only difference lies in the case of a linearly polarized beam where the laboratory analyzing direction set by the choice of the angle @xmath34 ( see fig . [ fig : pol_angles ] ) can vary . [ eqn : target_recoil_profile ] remains the same between the theory and experimental formalisms , since it is independent of @xmath34 . for eqs . [ eqn : beam_target_profile ] and [ eqn : beam_recoil_profile ] , however , we need to get back to the relation @xmath35 in fig . [ fig : pol_angles ] . the easiest choice is to measure everything with respect to @xmath31 , which usually represents the experimentalist s choice of photon polarization axis . therefore , we take @xmath100 ( also called `` para '' setting ) , so that @xmath101 and eqs . [ eqn : beam_target_profile ] and [ eqn : beam_recoil_profile ] become @xmath102 and @xmath103 respectively . similarly , for @xmath104 ( also called `` perp '' setting ) , we get @xmath105 and @xmath106 it is important to note that the sine and cosine terms in eqs . [ eqn : beam_target_profile ] and [ eqn : beam_recoil_profile ] alter signs differently in going from the `` theory '' expressions to the `` para '' and `` perp '' settings , and this directly affects the signs of the extracted polarization observables . therefore , care must be taken by the experimentalist to conform to a definition of the `` para '' and `` perp '' settings that matches with the theoretical definitions . finally , we also note that it is beneficial to measure the intensity profiles for both the `` para '' and `` perp '' settings and extract the polarizations from the asymmetries between the two settings . this removes the overall normalization factor ( the unpolarized cross - section ) , and therefore , any dependence on the detector acceptance ( see ref . @xcite for details ) . we list some basic caveats that will be useful during the computations . 1 . the matrix representation of an operator is @xmath107 ( note order of subscripts ) . for the pauli matrices for example , @xmath108 , @xmath109 , etc . 2 . fts uses @xmath110 and @xmath111 for outgoing baryon and incoming ( target ) spins . we adopt @xmath112 and @xmath113 as the final baryon and initial proton spins , respectively , and denote the photon spin by @xmath114 . any other index will be a dummy index for summation purposes . also , unless otherwise mentioned , it is understood that repeated indices are to be summed over . a useful relation is that for the conjugate operator @xmath115 , the matrix elements are @xmath116 . 4 . there are two types of traces in the fts paper . `` tr '' implies a trace over all spins , while `` tr '' implies a trace over the baryon spins , assuming that the photon spins have been traced over . to go from `` tr '' to `` tr '' , that is , the procedure of doing the photon spin trace , is as follows . let @xmath117 and @xmath118 be any operator in the final baryon and initial target proton spin space respectively . for the three pauli matrices @xmath119 , @xmath120 and @xmath121 , the photon traces are computed as follows : + @xmath122 \!&=&\ ! \displaystyle \sum_{\lambda \lambda ' } \mbox{tr}[\omega_b j_\lambda \omega_i ( \sigma^\gamma_x)_{\lambda \lambda ' } ( j^\dagger)_{\lambda ' } ] \nonumber \\ & = & \mbox{tr}[\omega_b j_+ \omega_i ( j^\dagger)_- + \omega_b j_- \omega_i ( j^\dagger)_+ ] \nonumber \\ \\ \mbox{tr}[\omega_b j \omega_i \sigma^\gamma_y j^\dagger ] \!&=&\ ! \displaystyle \sum_{\lambda \lambda ' } \mbox{tr}[\omega_b j_\lambda \omega_i ( \sigma^\gamma_y)_{\lambda \lambda ' } ( j^\dagger)_{\lambda ' } ] \nonumber \\ & = & -i\;\ ; \mbox{tr}[\omega_b j_+ \omega_i ( j^\dagger)_- - \omega_b j_- \omega_i ( j^\dagger)_+ ] \nonumber \\ \\ \mbox{tr}[\omega_b j \omega_i \sigma^\gamma_z j^\dagger ] \!&= & \!\displaystyle \sum_{\lambda \lambda ' } \mbox{tr}[\omega_b j_\lambda \omega_i ( \sigma^\gamma_z)_{\lambda \lambda ' } ( j^\dagger)_{\lambda ' } ] \nonumber \\ & = & \mbox{tr}[\omega_b j_+ \omega_i ( j^\dagger)_+ - \omega_b j_- \omega_i ( j^\dagger)_- ] . \nonumber \\\end{aligned}\ ] ] + note that these expressions for the summations over the photon states are equivalent to those from the more conventional forms that can be found in ref . @xcite , for example . the trace notation is simply a more compact way of expressing the spin sums . 5 . overall normalization factor . all 15 polarization observables will be normalized by the intensity factor @xmath123 $ ] . this is given as @xmath124 = \displaystyle \sum_{\lambda m_i m_b } |\mathcal{a}_{\lambda m_i m_b}|^2.\ ] ] this will not appear in our expressions below , but it is understood that this normalization always goes into the computations . the detailed computation of the 15 polarizations are given below : @xmath125 \nonumber \\ \!\!&=\!\ ! & \langle m_b| \sigma_y | m_b'\rangle \langle m_b ' |j_\lambda | m_i \rangle \langle m_i |j^\dagger_\lambda | m_b \rangle \nonumber \\ \!\!&=\!\ ! & \langle m_b| \sigma_y | m_b'\rangle \langle m_b ' |j_\lambda | m_i \rangle ( \langle m_b |j_\lambda | m_i \rangle)^\ast \nonumber \\ \!\!&=\!\ ! & -i \ ; \langle - |j_\lambda | m_i \rangle ( \langle + |j_\lambda | m_i \rangle)^\ast + i \langle+ |j_\lambda | m_i \rangle ( \langle - |j_\lambda | m_i \rangle)^\ast\nonumber \\ \!\!&=\!\ ! & \displaystyle \sum_{\lambda m_i } -2\ , \mbox{im}\left ( \mathcal{a}_{\lambda m_i + } \mathcal{a}^\ast_{\lambda m_i -}\right)\end{aligned}\ ] ] @xmath126 \nonumber \\ \langle m_b |j_+ | m_i \rangle \langle m_i |j^\dagger_- | m_b \rangle + \langle m_b |j_- | m_i \rangle \langle m_i |j^\dagger_+ | m_b \rangle \nonumber \\ \!\!&=&\!\ ! \langle m_b |j_+ | m_i \rangle ( \langle m_b |j_- | m_i \rangle)^\ast\ ! + \ ! \langle m_b |j_- | m_i \rangle ( \langle m_b |j_+ | m_i \rangle)^\ast \nonumber \\ \!\!&=&\!\ ! \displaystyle \sum_{m_i m_b } 2\ , \mbox{re}\left ( \mathcal{a}_{+ m_i m_b } \mathcal{a}^\ast_{- m_i m_b}\right)\end{aligned}\ ] ] @xmath127 \nonumber \\ & = & \!\!-i \langle m_b |j_\lambda | + \rangle \langle - |j^\dagger_\lambda | m_b \rangle + i \langle m_b |j_\lambda | - \rangle \langle + |j^\dagger_\lambda | m_b \rangle \nonumber \\ & = & \!\!-i \langle m_b |j_\lambda | + \rangle ( \langle m_b |j_\lambda | - \rangle)^\ast + i \langle m_b |j_\lambda | - \rangle ( \langle m_b |j^\dagger_\lambda | + \rangle)^\ast \nonumber \\ & = & \!\!\displaystyle \sum_{\lambda m_b } -2\ , \mbox{im}\left ( \mathcal{a}_{\lambda - m_b } \mathcal{a}^\ast_{\lambda + m_b}\right)\end{aligned}\ ] ] @xmath128 \nonumber \\ \!\!&= & \langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle \langle m_i ' | j_+^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5cm}- \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle \langle m_i ' | j_-^\dagger | m_b \rangle \nonumber \\ \!\!&=&\!\ ! \langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle ( \langle m_b | j_+ | m_i ' \rangle)^\ast \nonumber \\ & & \hspace{0.5cm}- \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle ( \langle m_b | j_- | m_i ' \rangle)^\ast \nonumber \\ \!\!&=&\!\ ! \langle m_b| j_+ | + \rangle ( \langle m_b | j_+ | + \rangle)^\ast - \langle m_b| j_+ | - \rangle ( \langle m_b | j_+ | - \rangle)^\ast \nonumber \\ \!\ ! & & \!\ ! - \langle m_b| j_- | + \rangle ( \langle m_b | j_- | + \rangle)^\ast + \langle m_b| j_- | - \rangle ( \langle m_b | j_- | - \rangle)^\ast \nonumber \\ \!\!&=&\!\ ! \displaystyle \sum_{m_b } \left(|\mathcal{a}_{++m_b}|^2 - |\mathcal{a}_{+-m_b}|^2 - |\mathcal{a}_{-+m_b}|^2 + |\mathcal{a}_{--m_b}|^2 \right ) \nonumber \\\end{aligned}\ ] ] @xmath129 \nonumber \\ \!\!&=&\!\ ! \langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle \langle m_i ' | j_+^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5cm}- \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle \langle m_i ' | j_-^\dagger | m_b \rangle \nonumber \\ \!\!&=&\!\ ! \langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle ( \langle m_b | j_+ | m_i ' \rangle)^\ast \nonumber \\ & & \hspace{0.5cm}- \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle ( \langle m_b | j_- | m_i ' \rangle)^\ast \nonumber \\ \!\!&=&\!\ ! \langle m_b| j_+ | + \rangle ( \langle m_b | j_+ | - \rangle)^\ast + \langle m_b| j_+ | - \rangle ( \langle m_b | j_+ | + \rangle)^\ast \nonumber \\ \!\ ! & & \!\!\ ! - \langle m_b| j_- | + \rangle ( \langle m_b | j_- | - \rangle)^\ast - \langle m_b| j_- | - \rangle ( \langle m_b | j_- | + \rangle)^\ast \nonumber \\ \!\!&=&\!\ ! \displaystyle \sum_{m_b } 2 \mbox{re } \left ( a_{++m_b } a^\ast_{+-m_b } - a_{-+ m_b } a^\ast_{--m_b } \right)\end{aligned}\ ] ] @xmath130 \nonumber \\ & = & \!\ ! i \langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle \langle m_i ' | j_-^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5 cm } -i \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle \langle m_i ' | j_+^\dagger | m_b \rangle \nonumber \\ & = & \!\ ! i\langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle ( \langle m_b | j_- | m_i ' \rangle)^\ast \nonumber \\ & & \hspace{0.5cm}- i \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_z | m_i ' \rangle ( \langle m_b | j_+ | m_i ' \rangle)^\ast \nonumber \\ & = & \!\ ! i\langle m_b| j_+ | + \rangle ( \langle m_b | j_- | + \rangle)^\ast -i \langle m_b| j_+ | - \rangle ( \langle m_b | j_- | - \rangle)^\ast \nonumber \\ & & \!\!\!\ ! -i \langle m_b| j_- | + \rangle ( \langle m_b | j_+ | + \rangle)^\ast + i \langle m_b| j_- | - \rangle ( \langle m_b | j_+ | - \rangle)^\ast \nonumber \\ & = & \displaystyle \sum_{m_b } 2 \mbox{im } \left ( \mathcal{a}_{+-m_b } \mathcal{a}^\ast_{--m_b } + \mathcal{a}_{-+m_b}\mathcal{a}^\ast_{++m_b } \right)\end{aligned}\ ] ] @xmath131 \nonumber \\ & = & \!\ ! i \langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle \langle m_i ' | j_-^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5 cm } -i \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle \langle m_i ' | j_+^\dagger | m_b \rangle \nonumber \\ & = & \!\ ! i\langle m_b| j_+ | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle ( \langle m_b | j_- | m_i ' \rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } - i \langle m_b| j_- | m_b ' \rangle \langle m_b ' | \sigma_x | m_i ' \rangle ( \langle m_b | j_+ | m_i ' \rangle)^\ast \nonumber \\ & = & \!\ ! i\langle m_b| j_+ | + \rangle ( \langle m_b | j_- | - \rangle)^\ast + i \langle m_b| j_+ | - \rangle ( \langle m_b | j_- | + \rangle)^\ast \nonumber \\ & & \!\ ! -i \langle m_b| j_- | + \rangle ( \langle m_b | j_+ | - \rangle)^\ast -i \langle m_b| j_- | - \rangle ( \langle m_b | j_+ | + \rangle)^\ast \nonumber \\ & = & \!\!\!\ ! \displaystyle \sum_{m_b } 2 \mbox{im } \left ( \mathcal{a}_{--m_b } \mathcal{a}^\ast_{++m_b } + \mathcal{a}_{-+m_b}\mathcal{a}^\ast_{+-m_b } \right)\end{aligned}\ ] ] @xmath132 \nonumber \\ & = & \!\ ! \langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle \langle m_i | j_+^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5 cm } - \langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_- | m_i \rangle \langle m_i | j_-^\dagger | m_b \rangle \nonumber \\ & = & \!\ ! \langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle ( \langle m_b | j_+ | m_i \rangle)^\ast \nonumber \\ & & \hspace{0.5cm}- \langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_- | m_i \rangle ( \langle m_b | j_- | m_i \rangle)^\ast \nonumber \\ & = & \langle - | j_+ | m_i \rangle ( \langle + |j_+ | m_i \rangle ) ^\ast - \langle - | j_- | m_i \rangle ( \langle + |j_- | m_i \rangle ) ^\ast \nonumber \\ & & \!\ ! + \langle + | j_+ | m_i \rangle ( \langle - |j_+ | m_i \rangle ) ^\ast - \langle + | j_- | m_i \rangle ( \langle - |j_- | m_i \rangle ) ^\ast \nonumber \\ & = & \!\!\ ! \displaystyle \sum_{m_i } 2 \mbox{re } \left ( \mathcal{a}_{+m_i- } \mathcal{a}_{+m_i+}^\ast - \mathcal{a}_{-m_i- } \mathcal{a}_{-m_i+}^\ast \right)\end{aligned}\ ] ] @xmath133 \nonumber \\ & = & \langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle \langle m_i | j_+^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5 cm } - \langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_- | m_i \rangle \langle m_i | j_-^\dagger | m_b \rangle \nonumber \\ & = & \!\ ! \langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle ( \langle m_b | j_+ | m_i \rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } - \langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_- | m_i \rangle ( \langle m_b | j_- | m_i \rangle)^\ast \nonumber \\ & = & \!\ ! \langle + | j_+ | m_i \rangle ( \langle + |j_+ | m_i \rangle ) ^\ast - \langle + | j_- | m_i \rangle ( \langle + |j_- | m_i \rangle ) ^\ast \nonumber \\ & & - \langle - | j_+ | m_i \rangle ( \langle - |j_+ | m_i \rangle ) ^\ast + \langle - | j_- | m_i \rangle ( \langle - |j_- | m_i \rangle ) ^\ast \nonumber \\ & = & \!\!\ ! \displaystyle \sum_{m_i } \left ( |\mathcal{a}_{+m_i+}|^2 - |\mathcal{a}_{-m_i+}|^2 -|\mathcal{a}_{+m_i-}|^2 + |\mathcal{a}_{-m_i-}|^2 \right ) \nonumber \\\end{aligned}\ ] ] @xmath134 \nonumber \\ & = & \!\!i\langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle \langle m_i | j_-^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5 cm } - i \langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_- | m_i \rangle \langle m_i | j_+^\dagger | m_b \rangle \nonumber \\ & = & \!\!i\langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle ( \langle m_b | j_- | m_i \rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } -i \langle m_b | \sigma_x | m_i ' \rangle \langle m_i'| j_- | m_i \rangle ( \langle m_b | j_+ | m_i \rangle)^\ast \nonumber \\ & = & \!\ ! i\langle - | j_+ | m_i \rangle ( \langle + |j_- | m_i \rangle ) ^\ast -i \langle - | j_- | m_i \rangle ( \langle + |j_+ | m_i \rangle ) ^\ast \nonumber \\ & & \!\!\ ! + i\langle + | j_+ | m_i \rangle ( \langle - |j_- | m_i \rangle ) ^\ast -i \langle + | j_- | m_i \rangle ( \langle - |j_+ | m_i \rangle ) ^\ast \nonumber \\ & = & \!\ ! \displaystyle \sum_{m_i } 2 \mbox{im } \left ( \mathcal{a}_{-m_i- } \mathcal{a}_{+m_i+}^\ast + \mathcal{a}_{-m_i+ } \mathcal{a}_{+m_i-}^\ast \right)\end{aligned}\ ] ] @xmath135 \nonumber \\ & = & \!\ ! i\langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle \langle m_i | j_-^\dagger | m_b \rangle \nonumber \\ & & \hspace{0.5 cm } - i \langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_- | m_i \rangle \langle m_i | j_+^\dagger | m_b \rangle \nonumber \\ & = & \!\ ! i\langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_+ | m_i \rangle ( \langle m_b | j_- | m_i \rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } -i \langle m_b | \sigma_z | m_i ' \rangle \langle m_i'| j_- | m_i \rangle ( \langle m_b | j_+ | m_i \rangle)^\ast \nonumber \\ & = & \!\ ! i\langle + | j_+ | m_i \rangle ( \langle + |j_- | m_i \rangle ) ^\ast -i \langle + | j_- | m_i \rangle ( \langle + |j_+ | m_i \rangle ) ^\ast \nonumber \\ & & \!\!\ ! -i \langle - | j_+ | m_i \rangle ( \langle - |j_- | m_i \rangle ) ^\ast + i \langle - | j_- | m_i \rangle ( \langle - |j_+ | m_i \rangle ) ^\ast \nonumber \\ & = & \!\ ! \displaystyle \sum_{m_i } 2 \mbox{im } \left ( \mathcal{a}_{-m_i+ } \mathcal{a}_{+m_i+}^\ast + \mathcal{a}_{+m_i- } \mathcal{a}_{-m_i-}^\ast \right)\end{aligned}\ ] ] @xmath136 \nonumber \\ & = & \!\ ! \langle m_b|\sigma_x| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle \langle m_b ' |j_\lambda^\dagger | m_b\rangle \nonumber \\ & = & \!\ ! \langle m_b|\sigma_x| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle ( \langle m_b |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & = & \!\ ! \langle -| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle ( \langle + |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } + \langle + | j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle ( \langle - |j_\lambda | m_b'\rangle)^\ast \nonumber \\ & = & \!\ ! \langle -| j_\lambda | + \rangle ( \langle + |j_\lambda| -\rangle)^\ast + \langle + | j_\lambda | + \rangle ( \langle - |j_\lambda| -\rangle)^\ast \nonumber \\ & & \!\!\ ! + \langle -| j_\lambda | -\rangle ( \langle + |j_\lambda | + \rangle)^\ast + \langle + | j_\lambda | -\rangle ( \langle - |j_\lambda | + \rangle)^\ast \nonumber \\ & = & \!\ ! \displaystyle \sum_\lambda 2 \mbox{re}\left ( \mathcal{a}_{\lambda + -}\mathcal{a}_{\lambda -+}^\ast + \mathcal{a}_{\lambda + + } \mathcal{a}_{\lambda --}^\ast \right)\end{aligned}\ ] ] @xmath137 \nonumber \\ & = & \langle m_b|\sigma_z| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle \langle m_b ' |j_\lambda^\dagger | m_b\rangle \nonumber \\ & = & \langle m_b|\sigma_z| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle ( \langle m_b |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & = & \langle + | j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle ( \langle + |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } - \langle -| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_x| m_b'\rangle ( \langle - |j_\lambda | m_b'\rangle)^\ast \nonumber \\ & = & \langle + | j_\lambda | + \rangle ( \langle + |j_\lambda| -\rangle)^\ast - \langle -| j_\lambda | + \rangle ( \langle - |j_\lambda| -\rangle)^\ast \nonumber \\ & & + \langle + | j_\lambda | -\rangle ( \langle + |j_\lambda | + \rangle)^\ast - \langle -| j_\lambda | -\rangle ( \langle - |j_\lambda | + \rangle)^\ast \nonumber \\ & = & \displaystyle \sum_\lambda 2 \mbox{re}\left ( \mathcal{a}_{\lambda + + } \mathcal{a}_{\lambda -+}^\ast - \mathcal{a}_{\lambda + -}\mathcal{a}_{\lambda --}^\ast \right)\end{aligned}\ ] ] @xmath138 \nonumber \\ & = & \langle m_b|\sigma_x| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle \langle m_b ' |j_\lambda^\dagger | m_b\rangle \nonumber \\ & = & \langle m_b|\sigma_x| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle ( \langle m_b |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & = & \langle -| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle ( \langle + |j_\lambda| m_b'\rangle)^\ast + \langle \nonumber \\ & & \hspace{0.5 cm } + | j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle ( \langle - |j_\lambda | m_b'\rangle)^\ast \nonumber \\ & = & \langle -| j_\lambda | + \rangle ( \langle + |j_\lambda| + \rangle)^\ast + \langle + | j_\lambda | + \rangle ( \langle - |j_\lambda| + \rangle)^\ast \nonumber \\ & & - \langle -| j_\lambda | -\rangle ( \langle + |j_\lambda | -\rangle)^\ast - \langle + | j_\lambda | -\rangle ( \langle - |j_\lambda | -\rangle)^\ast \nonumber \\ & = & \displaystyle \sum_\lambda 2 \mbox{re}\left ( \mathcal{a}_{\lambda + -}\mathcal{a}_{\lambda + + } ^\ast - \mathcal{a}_{\lambda --}\mathcal{a}_{\lambda -+}^\ast \right)\end{aligned}\ ] ] @xmath139 \nonumber \\ & = & \langle m_b|\sigma_z| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle \langle m_b ' |j_\lambda^\dagger | m_b\rangle \nonumber \\ & = & \langle m_b|\sigma_z| m_i \rangle \langle m_i| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle ( \langle m_b |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & = & \langle + | j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle ( \langle + |j_\lambda| m_b'\rangle)^\ast \nonumber \\ & & \hspace{0.5 cm } - \langle -| j_\lambda | m_i'\rangle \langle m_i ' |\sigma_z| m_b'\rangle ( \langle - |j_\lambda | m_b'\rangle)^\ast \nonumber \\ & = & \langle + | j_\lambda | + \rangle ( \langle + |j_\lambda| + \rangle)^\ast - \langle -| j_\lambda | + \rangle ( \langle - |j_\lambda| + \rangle)^\ast \nonumber \\ & & - \langle + | j_\lambda | -\rangle ( \langle + |j_\lambda | -\rangle)^\ast + \langle -| j_\lambda | -\rangle ( \langle - |j_\lambda | -\rangle)^\ast \nonumber \\ & = & \displaystyle \sum_\lambda \left ( |\mathcal{a}_{\lambda++}|^2 -|\mathcal{a}_{\lambda+-}|^2 - |\mathcal{a}_{\lambda-+}|^2 + |\mathcal{a}_{\lambda--}|^2\right ) \nonumber \\\end{aligned}\ ] ] a summary of the expressions for the 16 observables in terms of the @xmath140 amplitudes as given below : [ eqn : long_basis_obs ] @xmath141 where it is understood that all the 15 polarization observables are to be normalized by @xmath142 . it is well known that the fifteen polarization observables occurring as bilinears in eq . [ eqn : long_basis_obs ] can be connected by various identities . these are also called constraint equations , because they interconnect and place restrictions on the physical values these observables can take . simply put , these equations are nothing but identities in the four independent amplitudes @xmath140 . chiang and tabakin @xcite have showed that these identities can be derived in a more sophisticated fashion by considering the complex space spanned by the four amplitudes . the observables can then be expanded in terms of the sixteen @xmath143 dirac gamma matrix bilinears @xmath144 and the constraint relations emerge from the various fierz identities connecting products of the dirac bilinears . we list the set of relations ( eqs . l.0-s.r ) that we find to be valid ( we maintain the equation - numbering as in chiang - tabakin @xcite ) . these relations have been numerically verified by assigning random values to the four complex @xmath140 amplitudes and calculating the polarizations employing eqs . [ eqn : long_basis_obs]a - p . the relations consisting of only squares of the observables ( eqs . l.0 and s.bt-s.r ) have no sign ambiguities . however , the signs in the remaining set of relations depend on the conventions adopted while defining the polarizations . there appears to be some disagreement between different groups in the sign conventions for the polarizations , most likely arising from differences in the physics motivation . for example , in the clas @xmath145 measurements for @xmath146 photoproduction @xcite , it was found that @xmath147 at @xmath148 and @xmath65 was seen as the _ spin_-transfer from a right - handed circularly polarized photon to the recoiling baryon . other groups @xcite prefer to have @xmath147 at @xmath149 , with the interpretation that @xmath65 is the transfer of _ helicity _ from a right - handed circularly polarized photon to the @xmath150 . indeed , hadron - helicity - conservation is a feature of perturbative qcd at high enough energies @xcite . whatever be the choice of convention , the important issue is that the intensity profile the experimentalist uses must match with the asymmetry definitions that give the amplitude - level expressions . this point was detailed in sec . [ sec : experimental_profiles ] . @xmath151 @xmath152 we provide a detailed and self - contained description of the intensity profiles and amplitude - level expressions for the 15 polarization observables in pseudo - scalar meson photoproduction . our calculations are based on the density matrix approach of fasano , tabakin , and saghai and our spin amplitudes have a universal spin - projection direction along the incident beam direction for all particles . we have also stressed the preservation of consistency between the sign - conventions of experiment and theoretical amplitude - level expressions . the current work is geared towards performing a mass - independent partial - wave analysis on the recently published clas data @xcite and forthcoming results from jlab @xcite . the authors thank professor reinhard schumacher for his helpful comments during the preparation of this article . this work was supported in part by the us department of energy under grant no . de - fg02 - 87er40315 . d. ireland also acknowledges the support of the united kingdom s science and technology council . eugenio _ et al . _ ( clas collaboration ) , _ search for missing nucleon resonances in the photoproduction of hyperons using a polarized photon beam and a polarized target _ , jefferson lab experiment e02 - 112 .

a hypergraph is called an _ grid _ if it is isomorphic to a pattern of horizontal and vertical lines , i.e. , a family of sets such that for and for . three sets form a _ triangle _ if they pairwise intersect in three distinct singletons , , . a hypergraph is _ linear _ , if holds for every pair of edges . in this paper we construct large linear-hypergraphs which contain no grids . moreover , a similar construction gives large linear-hypergraphs which contain neither grids nor triangles . for our constructions are almost optimal . these investigations are also motivated by coding theory : we get new bounds for optimal superimposed codes and designs .

1103.1691

in this section we first present some previous investigations in extremal set theory on the topic described in the abstract . then we state our main theorem . this is followed by motivations in coding theory and corollaries where we improve the previously known bounds for so called _ optimal _ superimposed codes and designs . to prove the main theorem we are using tools from combinatorial number theory and discrete geometry given in section 2 . in section 3 we present constructions proving the stated theorems , followed by remarks on union - free and cover - free graphs and triple systems . speaking about a hypergraph @xmath12 we frequently identify the vertex set @xmath13 by the set of first integers @xmath14:=\ { 1,2 , \dots , n\}$ ] , or points on the plane @xmath15 , or elements of a @xmath16-element finite field @xmath17 . to shorten notations we frequently say hypergraph @xmath18 ( or set system @xmath18 ) thus identifying @xmath19 to its edge set @xmath18 . @xmath19 is _ linear _ if for all @xmath20 , @xmath21 we have @xmath22 . the _ degree _ , @xmath23 , of an element @xmath24 $ ] is the number of hyper - edges in @xmath25 containing @xmath26 . @xmath27 is _ regular _ if every element @xmath24 $ ] has the same degree . it is _ uniform _ if every edge has the same number of elements , @xmath1-uniform means @xmath28 for all @xmath29 . an @xmath30-_packing _ is a linear @xmath1-uniform hypergraph @xmath31 on @xmath32 vertices . obviously , @xmath33 . if here equality holds , then @xmath31 is called an @xmath34 _ steiner system_. a set system @xmath35 contains an @xmath36 _ grid _ if there exist two disjoint subfamilies @xmath37 such that * @xmath38 , @xmath39 , @xmath40 , * @xmath41 for all @xmath42 , @xmath43 , @xmath44 , @xmath45 , and * @xmath46 for all @xmath47 , @xmath48 . thus an @xmath1-uniform @xmath0 grid , @xmath49 , is a disjoint pair @xmath50 of the same sizes @xmath1 such that they cover exactly the same set of @xmath51 elements . [ th:1 ] for @xmath11 there exists a real @xmath52 such that there are linear @xmath1-uniform hypergraphs @xmath25 on @xmath32 vertices containing no grids and @xmath53 the proof is postponed to section [ ss:32 ] . the _ turn number _ of the @xmath1-uniform hypergraph @xmath54 , denoted by @xmath55 , is the size of the largest @xmath54-free @xmath1-graph on @xmath32 vertices . if we want to emphasize @xmath1 , then we write @xmath56 . let @xmath57 be ( more precisely @xmath58 ) the class of hypergraphs of two edges and intersection sizes at least two . this class consists of @xmath59 non - isomorphic hypergraphs , @xmath60 , @xmath61 , @xmath62 such that @xmath63 , @xmath64 . using these notations the above theorem can be restated as follows . @xmath65 holds for every @xmath66 . in the case of @xmath67 we only have @xmath68 see in section [ ss : prob_lower ] . the case of graphs , @xmath69 , is different , see later in section [ ss:35 ] . [ c:1 ] the asymptotic ( [ eq:2 ] ) holds for @xmath67 , too . + even more , for any given @xmath70 there are infinitely many steiner systems avoiding @xmath49 . + probably there exists an @xmath71 such that , for every admissible @xmath72 ( this means that @xmath73 and @xmath74 are both integers ) there exists a grid - free @xmath34 . there are many problems and results concerning subfamilies of block designs , see , e.g. , colbourn and rosa @xcite . a steiner triple system @xmath75 is called @xmath76-_sparse _ if it contains no set of @xmath76 distinct triples spanning at most @xmath77 points . every steiner triple system is 3-sparse . a longstanding conjecture of erds @xcite is that for every @xmath78 there exists an @xmath79 such that if @xmath80 and @xmath32 is admissible ( i.e. , @xmath81 or @xmath82 ( mod @xmath83 ) ) , then there exists an @xmath76-sparse @xmath84 . systems that are 4-sparse are those without a pasch configuration ( 4 blocks spanning 6 points , @xmath85 , @xmath86 , @xmath87 , @xmath88 , @xmath89 ) . completing the works of brouwer @xcite , murphy @xcite , ling and colbourn @xcite and others , finally grannell , griggs and whitehead @xcite proved that 4-sparse @xmath84 s exist for all admissible @xmath32 except 7 and 13 . a 5-sparse system is precisely one lacking pasch , @xmath85 , and mitre configurations , @xmath90 , the latter comprising five blocks of the form @xmath91 , @xmath92 , @xmath93 , @xmath94 @xmath95 . in a sequence of papers ( e.g. , colbourn , mendelsohn , rosa , and ir @xcite ) culminating in y. fujiwara @xcite and wolfe @xcite it was established that systems having no mitres exist for all admissible orders , except for @xmath96 . concerning the even more difficult problem of constructing 5-sparse systems ( see ling @xcite ) wolfe @xcite proved that such systems exist for almost all admissible @xmath32 . more precisely , let @xmath97 or 3 ( mod 6 ) , @xmath98 and @xmath99 there exists a 5-sparse @xmath84 with @xmath98 , then @xmath100 . forbes , grannell and griggs @xcite constructed infinite classes of 6-sparse @xmath84 s . as teirlinck @xcite writes in his 2009 review of @xcite `` currently no nontrivial example of a 7-sparse steiner triple system is known '' . our conjecture [ c:1 ] is related to but not a consequence of erds problem . colbourn @xcite has checked ( using a computer ) all the 80 different @xmath101 s and each contained at least 11 copies of @xmath102 . blokhuis @xcite reformulated ( a weaker version of ) conjecture [ c:1 ] as follows : are there latin squares without the following subconfiguration ? @xmath103 although the evidence is scarce one is tempted to generalize . an @xmath34 is @xmath76-_sparse _ if the union of any @xmath76 blocks exceeds @xmath104 . [ c:2 ] for every @xmath78 there exists an @xmath105 such that if @xmath106 and @xmath32 is admissible , then there exists an @xmath76-sparse @xmath34 . brown , erds and ss @xcite introduced the function @xmath107 to denote the maximum number of edges in an @xmath1-uniform hypergraph on @xmath32 vertices which does not contain @xmath76 edges spanned by @xmath108 vertices . such hypergraphs are called @xmath109-_free _ ( more precisely @xmath110-_free _ ) . they showed that @xmath111 for every @xmath112 and @xmath113 , especially @xmath114 . the upper bound @xmath74 is easy , and this was the source of erds conjecture concerning sparse steiner systems . on the other hand , if we forbid @xmath76 edges spanned by one more vertex this problems becomes much more difficult . brown , erds and ss conjectured that @xmath115 one of the most famous results of this type is the @xmath116-theorem of ruzsa and szemerdi @xcite , the case @xmath117 , saying that if no six points contain three triples then the size of the triple system is @xmath118 , on the other hand @xmath119 . this was extended by erds , frankl , and rdl @xcite for arbitrary fixed @xmath70 , @xmath120 the case @xmath121 was further extended by alon and shapira @xcite @xmath122 even the case @xmath123 , @xmath124 , is still open . nearly tight upper bounds were established by srkzy and selkow @xcite : @xmath125 and for the case @xmath126 , @xmath127 @xmath128 three sets @xmath7 form a _ triangle _ , @xmath129 , if they pairwise intersect in three distinct singletons , @xmath8 , @xmath9 . an @xmath1-uniform triangle is frequently denoted by @xmath130 . a _ perfect matching _ is a subfamily @xmath131 of the set system @xmath132 such that the members of @xmath131 cover every element of @xmath133 exactly once . the main result of this paper is a construction . [ th : main ] for @xmath11 there exist an @xmath134 and @xmath135 such that @xmath136 holds for @xmath137 . in other words , there exists a linear @xmath1-uniform hypergraph @xmath25 which contains neither grids nor triangles and @xmath138 $ ] . in addition , if @xmath1 divides @xmath32 , then @xmath25 can be decomposed into perfect matchings , especially it is regular . for the case @xmath67 we have the same with a much weaker lower bound @xmath139 again , the proof is postponed , to section [ ss:33 ] , and the cases @xmath140 to section [ ss:35 ] . note that @xmath141 by ( [ eq : efr ] ) so the lower bound ( [ eq : th ] ) is almost optimal . this result slightly improves the erds - frankl - rdl ( [ eq : efr ] ) construction in two ways . we make the hypergraph regular , and avoid not only triangles but grids , too . almost all of the problems discussed in this paper can be formulated as a forbidden substructure question , i.e. , as a turn type problem . here we present the standard probabilistic lower bound for the turn number due to erds , in a slightly stronger form as usual . an @xmath1-uniform hypergraph @xmath142 is called @xmath1-_partite _ if there exists an @xmath1-partition of @xmath143 , @xmath144 , such that @xmath145 for all @xmath29 , @xmath146 $ ] . [ le : erdos ] ( erds lower bound on the turn number)suppose that @xmath54 is a ( finite ) family of @xmath1-graphs each of them having at least two edges , and let @xmath147 then there exists a @xmath148 such that one can find an @xmath32-vertex @xmath1-partite @xmath1-graph @xmath149 of size at least @xmath150 avoiding each member of @xmath54 . hence @xmath151 sketch of the proof : choose independently each of the @xmath152 edges of the complete @xmath1-partite hypergraph on @xmath32 vertices with probability @xmath153 . leave out an edge from this random selection of each copy of @xmath154 . the expected size of the remaining edges is at least @xmath155 the rest is an easy calculation . the ratio @xmath156 for @xmath58 , @xmath49 , @xmath129 are @xmath157 , @xmath158 , and @xmath159 , resp . , so lemma [ le : erdos ] implies ( [ eq : r=3 ] ) , i.e. , @xmath160 . however , if @xmath129 is among the forbidden substructures then the probabilistic lower bound fails miserably , it gives only @xmath161 which is very far from the truth . to prove the slightly better lower bound ( [ eq : th_r=3 ] ) we are going to use a version of the original ruzsa - szemerdi method . the three forbidden configurations , @xmath58 , @xmath49 , @xmath130 , have 7 non - empty combinations . the cases @xmath162 and @xmath163 were discussed above in theorems [ th:1 ] and [ th : main ] , respectively . it is easy to see that @xmath164 so the ruzsa and szemerdi @xcite and the erds , frankl , and rdl @xcite theorems , see ( [ eq : efr ] ) , determine the right order of magnitude , @xmath165 . it was conjectured by chvtal and erds and proved by frankl et al . @xcite that @xmath166 for @xmath70 and @xmath167 . the only extremal @xmath1-graph consists of all @xmath1-tuples sharing a common element . this hypergraph has no grid either , so we have @xmath168 for the same range of @xmath1 and @xmath32 . we have @xmath169 if and only if a steiner system @xmath34 exists , which problem was solved for @xmath167 by wilson @xcite and the exact packing number was determined for all large enough @xmath32 by caro and yuster @xcite . the grid can not be covered by @xmath170 vertices , it has @xmath1 disjoint edges . so the @xmath1-graph having all edges meeting an @xmath171-element set is grid free . this gives the lower bound for the last case out of the seven . @xmath172 the classical result concerning the turn number of the complete @xmath1-partite graph on @xmath0 vertices by erds @xcite gives only an upper bound @xmath173 with @xmath174 . the truth should be much closer to the lower bound . determine the order of magnitude of @xmath175 . union free families were introduced by kautz and singleton @xcite . they studied binary codes with the property that the disjunctions ( bitwise @xmath176s ) of distinct at most @xmath1-tuples of codewords are all different . in information theory usually these codes are called _ superimposed _ and they have been investigated in several papers on multiple access communication ( see , e.g. , nguyen quang a and zeisel @xcite , dyachkov and rykov @xcite , johnson @xcite ) . alon and asodi @xcite , and de bonis and vaccaro @xcite studied this problem in a more general setup . small values of generalized superimposed codes and and their relation to designs were considered by kim , lebedevin and oh @xcite . the same problem has been posed in different terms by erds , frankl and fredi @xcite in combinatorics , by ss @xcite in combinatorial number theory , and by hwang and ss @xcite in group testing . one can find short proofs of the best known upper bounds of these codes in the papers by the present authors in @xcite and @xcite . in @xcite the connection of these codes to the big distance ones is shown . a geometric version has been posed by ericson and gyrfi @xcite and later investigated in @xcite . for a direct geometry application , notice that a union - free family defines a set of points of exponential size in @xmath177 such that arbitrary three of them span an acute triangle @xcite . a family @xmath178}$ ] is @xmath76-_union - free _ if for arbitrary two distinct subsets @xmath179 and @xmath180 of @xmath181 with @xmath182 @xmath183 let @xmath184 ( @xmath185 ) be the maximum size of an @xmath76-union - free @xmath32 vertex hypergraph ( @xmath1-uniform hypergraph , resp . ) . the order of magnitude of @xmath186 was determined by frankl et al . @xcite . a family @xmath178}$ ] is @xmath76-_cover - free _ if for arbitrary distinct members @xmath187 @xmath188 let @xmath189 ( @xmath190 ) be the maximum size of an @xmath76-cover - free @xmath32 vertex hypergraph ( @xmath1-uniform hypergraph , resp . ) . an @xmath76-cover - free hypergraph is @xmath76-union - free and an @xmath76-union - free is @xmath191-cover - free . ( indeed , the existence of an @xmath191-cover @xmath192 gives @xmath193 ) . therefore , @xmath194 and @xmath195 we have @xmath196 ( for @xmath197 ) . indeed , every member of an @xmath1-uniform @xmath1-cover - free family has a vertex of degree one . in this section , based on theorem [ th : main ] , we determine the next two terms of the sequence ( [ eq : ur ] ) . first , observe that @xmath198 indeed , an @xmath1-uniform , @xmath171-cover - free family either has a vertex of degree one ( and then we use induction on @xmath32 ) , or it is a linear hypergraph . [ co:15 ] there exists a @xmath199 such that for all @xmath200 @xmath201 in addition , if @xmath1 divides @xmath32 , then our @xmath32-vertex , @xmath1-uniform , @xmath1-union - free family yielding the lower bound can be decomposed into perfect matchings , especially it is regular . * proof . * the upper bound follows from ( [ eq : ur ] ) and ( [ eq:8 ] ) , i.e. , @xmath202 on the other hand , we claim that @xmath203 hence the lower bound for @xmath204 follows from ( [ eq : th ] ) . we have to show that a linear @xmath1-graph without triangle and grid is @xmath1-union - free . suppose , on the contrary , that @xmath205 , @xmath206 , @xmath207 and @xmath208 form a linear @xmath1-uniform hypergraph . then @xmath209 . since @xmath210 , to cover the elements of @xmath211 the family @xmath212 must contain @xmath1 sets , i.e. , @xmath213 . moreover , the sets @xmath214 meet @xmath211 in distinct elements . if @xmath215 consists of disjoint sets only , then @xmath216 , and to cover these @xmath51 elements @xmath217 must consist of disjoint sets , too , and @xmath218 form a grid @xmath49 . otherwise , @xmath219 , such that @xmath220 . then @xmath211 , @xmath221 , and @xmath222 form a triangle . [ pr:10 ] in the case @xmath67 the probabilistic lower bound ( [ eq : erdos_formula2 ] ) implies @xmath223 the details are postponed to section [ ss:34 ] . let @xmath224 be an @xmath1 uniform hypergraph with edges @xmath225 , @xmath226 and @xmath227 as follows . the @xmath1-sets @xmath228 are pairwise disjoint , @xmath229 are distinct elements , @xmath230 and @xmath231 and @xmath232 . [ c:6 ] if @xmath18 is an @xmath32-vertex , @xmath1-uniform ( @xmath70 ) , linear hypergraph not containing @xmath224 , then its size @xmath233 . in other words , @xmath234 this would imply the conjecture of erds ( [ eq : econj ] ) in the case @xmath123 , @xmath235 . if it is true , then it implies the following more modest conjecture @xmath236 [ pr:1 ] suppose that @xmath237 , @xmath238 , @xmath239 and @xmath167 . then there exists a @xmath240-regular , @xmath171-cover - free @xmath1-graph . thus , in this case ( [ eq:8 ] ) gives @xmath241 * proof . * to obtain the @xmath240-regular construction one can apply a classical theorem of ray - chaudhuri and wilson @xcite : for any given @xmath242 there exists an @xmath134 such that , if @xmath243 and @xmath244 , then there exists a _ resolvable _ , @xmath1-uniform , @xmath32-vertex steiner system @xmath245 . this means that @xmath245 can be decomposed into @xmath246 perfect matchings , ( also called _ parallel classes _ ) @xmath247 , where @xmath248 and @xmath249 . taking @xmath240 of these parallel classes gives the desired @xmath171-cover - free family . dyachkov and rykov @xcite introduced the concept of _ optimal superimposed codes and designs_. recall an easy observation . ( dyachkov , rykov @xcite ) [ opt ] if @xmath178}$ ] is @xmath171-cover - free , @xmath250 , it has maximum degree @xmath240 and @xmath251 then @xmath252 holds . by ( [ eq : un ] ) a similar statement holds for @xmath1-union - free families , too . note that , from coding theory point of view , it is reasonable to assume that @xmath253 since a collection of singletons @xmath181 is @xmath171-cover - free for arbitrary @xmath254 with @xmath255 and usually the goal is to get a code as large as possible . * proof of [ opt].*let @xmath256 , @xmath257 . clearly , @xmath258 for every @xmath259 otherwise the union of some other @xmath171 members of @xmath181 cover @xmath225 . we obtain @xmath260 } \deg ( x)\leq k(n - t_0)+t_0 \ , . \qed\ ] ] it follows that in case of @xmath261 the family @xmath18 should be @xmath240-regular and @xmath1-uniform . [ def : opti ] ( see @xcite)the @xmath32-vertex family @xmath18 is called an _ optimal @xmath171-superimposed code _ if it is an @xmath171-cover - free , @xmath1-uniform , @xmath240-regular , linear hypergraph . it is called an _ optimal @xmath1-superimposed design _ if in addition it is @xmath1-union - free , too . in both cases @xmath261 holds . let @xmath262 @xmath263 denote the maximum @xmath240 that such a @xmath240-regular optimal @xmath171-superimposed code ( optimal @xmath1-superimposed design ) exits . they have showed for every @xmath242 and @xmath32 that @xmath264 they and macula @xcite gave a lower bound for every @xmath265 for a few special but infinitely many values of @xmath32 . @xmath266 our results , corollary [ co:15 ] and proposition [ pr:1 ] , imply that [ co : th32:fo ] if @xmath11 , @xmath267 and @xmath137 , then there exists an optimal @xmath240-regular , @xmath1-superimposed design for every @xmath268 , especially @xmath269 [ pr:16 ] for the case @xmath67 we have the same statement with a weaker lower bound @xmath270 the details are postponed to section [ ss:34 ] . [ co : th33 ] suppose that @xmath237 , @xmath238 and @xmath167 . then there exists an optimal @xmath171-superimposed code for every @xmath271 , especially @xmath272 [ le : mink ] ( minkowksi s theorem of simultaneous approximation @xcite)let @xmath16 be a prime and @xmath273 an integer point . then there exist an integer @xmath274 and residues @xmath275 such that @xmath276 and @xmath277 for all @xmath278 . sketch of the proof : consider all vectors of the form @xmath279 mod @xmath16 , @xmath280 . there will be two of them @xmath281 and @xmath282 ` close ' to each other . take @xmath283 . let @xmath284 be the maximum number of integers which can be selected from @xmath285 containing no @xmath240-term arithmetic progression . this function has been extensively studied in the last six decades by leading mathematicians see , e.g. , ruzsa @xcite . the major important known bounds ( appart from some recent minor improvements ) are due to behrend @xcite , heath - brown @xcite and szemerdi @xcite : there are positive constants @xmath286 and @xmath287 such that @xmath288 call a set @xmath289 $ ] @xmath1-_sum - free _ if the equation @xmath290 has no solutions with @xmath291 and @xmath292 are positive integers with @xmath293 except the one with @xmath294 . we will need the following lower bound used by erds , frankl and rdl @xcite , also see ruzsa @xcite . its proof requires only a slight modification of behrend s @xcite argument . [ le : behrend ] ( generalized behrend)for arbitrary positive integer @xmath1 there exists a @xmath295 such that for any integer @xmath16 one can find an @xmath1-sum - free subset @xmath296 such that @xmath297 . the case @xmath69 ( and @xmath298 ) is the original statement of behrend @xcite . ruzsa also notes that an upper bound @xmath299 for the general case can be proved by the methods of @xcite and @xcite . call a set of numbers _ @xmath300-free _ if it does not contain a subset of the form @xmath301 for some @xmath302 , @xmath303 . call it @xmath304_-free _ if it does not contain a fourtuple of the form @xmath305 for some @xmath306 , and call it @xmath307_-free _ if it contains no @xmath240-term arithmetic progression . let @xmath308 denote the maximum number of integers which can be selected from @xmath309 avoiding the patterns @xmath310 . with this notation @xmath311 . since an @xmath304-free set has no 5-term arithmetic progression we get @xmath312 by szemerdi s @xcite theorem ( [ eq : sz ] ) . a 4-sum - free sequence is @xmath304-free as well ( one has , e.g. , @xmath313 ) , thus lemma [ le : behrend ] gives a lower bound showing @xmath314 [ le : a6 ] @xmath315 sketch of the proof : similar to the proof of lemma [ le : erdos ] . take an @xmath316-free set @xmath317 of maximum size . choose independently each element of @xmath318 with probability @xmath153 , and leave out an element from this random selection of each copy of the arising configurations we want to avoid . the expected size of the remaining elements is at least @xmath319 define @xmath153 as @xmath320 . starting with @xmath321 $ ] , the same process gives @xmath322 concerning the upper bounds we only have @xmath323 . the random method notoriously gives a weak lower bound of sidon type problems ( for definitions , see , e.g. , babai and ss @xcite ) , e.g. , the above argument gives only @xmath324 , although the truth is @xmath325 ( erds and turn @xcite ) and it gives @xmath326 although the truth is @xmath327 . so we believe that there are much better lower bounds . there is an @xmath328 such that @xmath329 holds for large enough @xmath32 . possibly the order of magnitude of this function is @xmath327 . a set of _ pseudolines _ is usually a set of ( infinite ) planer curves pairwise meeting in at most one point with crossing , no two pseudolines are tangent . the main result of this subsection is in fact deals with pseudoline arrangements but we formulate it in a simpler way . let @xmath330 be parallel vertical lines on the plane , @xmath331 , @xmath242 . let @xmath332 be an @xmath1-set of points on the line @xmath333 , @xmath334 , @xmath335 , such that @xmath336 for every @xmath337 . the points of @xmath338 can be arranged in a matrix form @xmath339 where the elements of the @xmath340th column lie on @xmath333 and are ordered from the top to the bottom . the vertical distances @xmath341 are positive , but can be distinct from each other . a @xmath342-_polygon _ @xmath343 consists of @xmath170 segments of the form @xmath344 $ ] ( @xmath345 ) with @xmath346 . there are @xmath347 such polygonal arcs . such a @xmath343 can be considered as a piecewise linear continuous function @xmath348\to\br$ ] . two sets of @xmath342-polygons @xmath31 and @xmath349 are called _ crossing _ if ( c1 ) @xmath350 , ( c2 ) @xmath31 is covering all vertices of @xmath342 , i.e. , @xmath351 and the same holds for @xmath349 , ( c3)@xmath352 is almost disjoint , i.e. , @xmath353 ( and @xmath354 ) imply that @xmath355 , and ( c4)@xmath352 behaves like pseudolines , i.e. , @xmath353 , @xmath356 imply that they are truly crossing , i.e. , if @xmath357 then either @xmath358 for all @xmath359 together with @xmath360 for @xmath361 , or @xmath360 for all @xmath359 together with @xmath358 for @xmath361 . note that ( c1 ) and ( c2 ) imply that every member of @xmath342 belongs to a unique @xmath362 and also to a unique @xmath363 . then ( c3 ) yields that each @xmath362 meets each @xmath364 at an element of @xmath342 , and only there . the members of @xmath31 ( and @xmath349 ) can be disjoint among themselves . a _ very special _ crossing system is depicted below in ( [ pattern ] ) using thin and thick lines to indicate the segments of @xmath365 and @xmath366 . its properties described in ( vs1)(vs7 ) below . @xmath367 there are two types of segments defining @xmath368 , ( vs1 ) all edges of the upper and lower envelops @xmath369 $ ] , @xmath370 $ ] , @xmath345 , and ( vs2)all diagonal edges @xmath371 . ( vs3)the edges on the upper ( lower ) envelops alternate between @xmath31 and @xmath349 . ( vs4)the crossing diagonal edges @xmath372 $ ] and @xmath373 $ ] simultaneously belong to either @xmath365 or @xmath366 . @xmath368 consists of the following three types of polygonal paths : ( vs5 ) the two diagonals @xmath374 , and @xmath375,(vs6)for @xmath376 a _ top _ path consisting of an increasing part ( on figure ( [ pattern ] ) ) @xmath377 , @xmath378 a horizontal edge @xmath379 and a decreasing part @xmath380 , @xmath381,(vs7)for each @xmath382 a _ bottom _ path starting at @xmath377 and consisting of a decreasing part having vertices @xmath383 @xmath384 , a horizontal edge @xmath385 $ ] and an increasing part @xmath386 @xmath387 . [ le : cross ] suppose that the two sets of @xmath342-polygons @xmath31 and @xmath349 form a crossing @xmath342-polygon system ( i.e. , satisfy ( c1)(c4 ) ) . then they have the unique structure described by ( vs1)(vs7 ) . the proof is postponed to the next subsection . from this lemma we can read out the intersection structure . we obtain @xmath388{8lines.eps}\ ] ] [ co:1 - 4 ] suppose that the two sets of @xmath342-polygons @xmath389 and @xmath390 form a crossing @xmath342-polygon system with @xmath391 @xmath392 and with @xmath393\subset \pi_1 $ ] . then the vertices of @xmath394 are @xmath395 , the vertices of @xmath396 are @xmath397 , the vertices of @xmath398 are @xmath399 , the vertices of @xmath400 are @xmath401 , the vertices of @xmath402 are @xmath403 , the vertices of @xmath404 are @xmath405 , the vertices of @xmath406 are @xmath407 , the vertices of @xmath408 are @xmath409 . here we prove lemma [ le : cross ] with a series of propositions . assume @xmath410 for @xmath411 . for @xmath412 let @xmath413 be the bipartite graph ( a matching ) with parts @xmath332 and @xmath414 and with edges defined by the corresponding parts of the polygons from @xmath31 , i.e. , @xmath415\in e(g_j^{\cp})$ ] if and only if there is a @xmath416 with @xmath417 and @xmath418 . ( thus , to simplify notations , we identify the graph @xmath413 with its geometric representation . ) @xmath419 is defined similarly . finally , @xmath420 is the union of @xmath413 and the graph @xmath421 is having all the edges of @xmath420 and @xmath422 . [ egy ] suppose @xmath423 for some @xmath362 . then @xmath424 . similarly , @xmath363 , @xmath345 , @xmath425 and @xmath426 imply @xmath427 . * proof.*we prove the second statement . assume to the contrary that @xmath428 ( the case @xmath429 is similar ) . consider the @xmath430 edges of @xmath431 with vertices @xmath432 . since there is not enough room to match these vertices to @xmath433 @xmath434 there exits a @xmath435\in e(g^\cp_j)$ ] with @xmath436 and @xmath437 . then this segment intersects @xmath415\in e(g_j^\cr)$ ] inside the open strip @xmath438 . this contradicts to the fact that a @xmath416 and a @xmath363 meet only in the points of @xmath342 . in the same way we obtain the following . [ pr : endpoint ] suppose that @xmath439 , the first point of @xmath440 is @xmath441 , the last one is @xmath442 . then @xmath443 . * proof.*suppose @xmath444 . every @xmath445 starts above @xmath446 on @xmath447 , they meet @xmath446 in a point of @xmath342 , so their endpoints on @xmath448 lie below or on the endpoint of @xmath440 , @xmath442 . hence @xmath449 . considering @xmath450 the same argument gives @xmath451 . now we prove lemma [ le : cross ] by checking ( vs1)(vs7 ) . * proof of ( vs5).*the paths @xmath394 and @xmath402 end at either @xmath452 or at @xmath453 by the previous proposition . they can not meet in a second point , so one of them finishes at @xmath453 . then proposition [ egy ] implies that this path is the diagonal of the @xmath0 array , @xmath454 . by symmetry , the other diagonal @xmath455 also belongs to @xmath368 . * proof of ( vs1).*we show that each top edge , @xmath456 $ ] , belongs to @xmath457 . according to proposition [ egy ] the neighbor of @xmath458 in @xmath459 ( in @xmath460 ) is either @xmath380 or @xmath461 . the @xmath31 and @xmath349 edges are distinct , so both @xmath456 $ ] and @xmath462 $ ] must belong to @xmath457 . similar argument gives that both @xmath456 $ ] and @xmath463 $ ] belong to @xmath457 and we got the edges of the top layer of ( [ pattern ] ) . * proof of ( vs3).*we show that the top edges alternate between @xmath464 and @xmath465 . if two consecutive of them , @xmath456 $ ] and @xmath466 $ ] , both belong to @xmath464 then they are part of the same @xmath416 and the diagonal edges @xmath462 $ ] and @xmath467 $ ] are necessarily @xmath465 edges , belonging to the same @xmath363 . then @xmath343 and @xmath468 cross twice , violating ( c3 ) . * proof of ( vs6).*call a path @xmath439 a _ top _ ( _ bottom _ ) path if it contains a top edge @xmath369 $ ] ( bottom edge @xmath370 $ ] ) . the pseudoline structure implies that these paths are all distinct , so @xmath368 consists of the two diagonals and @xmath170 top and @xmath170 bottom paths . again proposition [ egy ] implies that if @xmath369\subset \gamma$ ] and the first point of @xmath440 is @xmath441 , the last is @xmath442 then @xmath469 and @xmath470 so @xmath471 . this and proposition [ pr : endpoint ] give that @xmath472 and @xmath440 must have the shape as described in ( vs6 ) . by symmetry , we have the same for the bottom paths , property ( vs7 ) . properties ( vs67 ) imply that all diagonal edges @xmath473 $ ] are in @xmath457 , this is property ( vs2 ) . property ( vs4 ) follows from ( vs2 ) and from the fact that a @xmath31 and an @xmath349 edge can meet only at the vertices of @xmath342 . this completes the proof of lemma [ le : cross ] . here we apply the results of the previous subsections when the sets of crossing polygons @xmath31 and @xmath349 are actually straight lines . let @xmath474 , @xmath475 , @xmath476 , @xmath477 with @xmath478 , @xmath479 , @xmath480 , @xmath389 , @xmath390 and @xmath481\}= \{\ell_j\cap \rho_i : i\in [ r ] \}$ ] with @xmath482 , @xmath483 . [ le:28 ] if @xmath31 and @xmath349 are forming a crossing pair of straight lines , then @xmath140 . * proof.*@xmath31 and @xmath349 satisfy ( c1)(c4 ) so lemma [ le : cross ] can be applied . consider the first four lines @xmath484 and also @xmath485 . according to corollary [ co:1 - 4 ] these lines meet at @xmath486 , @xmath487 , @xmath488 , and @xmath489 and at 12 further points of @xmath342 ( see ( [ eq:8x8 ] ) ) . these 12 points yield 12 equations for @xmath490 , e.g. , considering @xmath491 we get the equation @xmath492 . so the points @xmath491 , @xmath493 , @xmath494 , @xmath495 , @xmath496 , @xmath497 , and @xmath498 , @xmath499 , @xmath452 , @xmath500 , @xmath501 , and @xmath453 give the following 12 equations . @xmath502 the solution set of this homogeneous linear equation system has dimension at least two because every vector @xmath503 generated by @xmath504 and @xmath505 is a solution . we claim that these are all the solutions . this implies @xmath506 and @xmath507 , contradicting the fact that the lines @xmath508 are distinct . let @xmath509 be the @xmath510 coefficient matrix of ( [ eq:12 ] ) . let us denote the characteristic polynomial of @xmath509 by @xmath511 . we obtain @xmath512 here the coefficient of @xmath513 is not @xmath514 for @xmath515 ( since @xmath516 ) . thus the rank of @xmath509 is @xmath517 and the solution set of ( [ eq:12 ] ) has dimension 2 , as stated . the calculations have been verified by both the _ maple _ and the _ mathematica _ programs . when @xmath67 there are crossing families of straight lines . let @xmath474 , @xmath475 , @xmath519 , as before with @xmath520 , such that @xmath521 , @xmath522 . then @xmath523 , @xmath524 may form a crossing pair of families , i.e. , @xmath525 with @xmath482 , @xmath526 . for example , for any @xmath527 ( with @xmath528 ) we have such a system with values @xmath529 the corresponding crossing systems are @xmath530 and @xmath531 [ le:210 ] if the set of slopes @xmath532 is @xmath304-free and @xmath300-free , then @xmath31 and @xmath349 can not be a set of @xmath518 crossing euclidean lines . * proof.*from lemma [ le : cross ] we know that a grid for @xmath67 has the following intersection pattern : @xmath533 considering the intersection points @xmath491 , @xmath493 , @xmath495 , @xmath496 , @xmath497 , and @xmath534 , we derive the following system of linear equations . @xmath535 it is easy to see that all solutions of ( [ eq:17 ] ) are of the form of ( [ eq:3 ] ) . in this section we prove our main results , theorems [ th:1 ] and [ th : main ] . given integers @xmath536 , @xmath537 define the hypergraph @xmath538 as follows . define the vertex set on the euclidean plane @xmath539 thus @xmath540 , and let @xmath541 . for integers @xmath542 define the @xmath1-set @xmath543 where the second coordinates are taken modulo @xmath16 . for @xmath544 and @xmath545 or for @xmath546 the points of @xmath547 are collinear . for a subset of _ slopes _ @xmath548 let @xmath549 obviously , this hypergraph is @xmath1-uniform , @xmath550-regular . even more , it can be decomposed into @xmath550 perfect matchings . [ le:31 ] suppose that @xmath551 , and for all slopes @xmath552 we have @xmath553 ( taken modulo @xmath16 ) . then @xmath554 is @xmath49-free . * proof.*suppose that given a grid @xmath555 . the hypergraph @xmath554 is shift invariant , so we may assume that @xmath556 and @xmath557 which start in @xmath558 . in other words , replace each @xmath547 by @xmath559 ( the second coordinates are always taken modulo @xmath16 ) . the shifted system of hypergedges have the same intersection structure as @xmath31 and @xmath349 . so from now on , we may suppose that @xmath560 . since the slopes @xmath561 and @xmath562 are small ( i.e. , @xmath563 ) the points of @xmath564 and @xmath565 are forming euclidean lines . all other sets of @xmath31 and @xmath349 meet either @xmath394 or @xmath402 , and all other slopes are small ( at most @xmath566 ) so all other members of @xmath352 are forming euclidean lines , too . finally , lemma [ le:28 ] completes the proof that @xmath1 should be at most @xmath82 . if @xmath16 is a prime ( and @xmath567 ) , then @xmath554 is a linear hypergraph . * proof.*well - known and easy . @xmath568 implies that there exist @xmath569 @xmath570 implying @xmath571 a contradiction . [ le:33 ] if @xmath16 is a prime , @xmath572 , @xmath515 , then the whole @xmath554 with @xmath573 is @xmath49-free . * proof.*suppose that the families @xmath574 and @xmath575 form a grid @xmath49 . apply minkowksi s theorem of simultaneous approximation ( lemma [ le : mink ] ) for the vector @xmath576 . there exists an @xmath577 such that for all @xmath146 $ ] we have @xmath578 and @xmath579 ( mod @xmath16 ) . the collections @xmath580 and @xmath581 , @xmath582 have the same intersection pattern , i.e. , they form a grid , too . then lemma [ le:31 ] implies that @xmath140 . we use induction on @xmath32 to show that @xmath583 holds for @xmath11 with appropriate @xmath584 . let @xmath16 be the largest prime @xmath585 . it is well - known @xcite that @xmath586 , where @xmath587 is an absolute constant . let @xmath588 be disjoint @xmath16-sets , and let @xmath538 be the grid - free hypergraph of size @xmath589 given by lemma [ le:33 ] . consider a grid - free , linear hypergraph @xmath54 on @xmath16 vertices . by induction hypothesis , there is such a hypergraph of size @xmath590 . put a copy of @xmath54 , @xmath591 , into each @xmath332 after randomly permuting their vertices , ( @xmath592 ) . the union of these hypergraphs , @xmath593 , is obviously linear . suppose that @xmath594 and @xmath595 are forming a grid in @xmath18 . since @xmath538 is grid - free there should be an edge , say @xmath596 such that @xmath597 . the vertices of @xmath598 are covered by the edges of @xmath349 , each meeting @xmath598 in a distinct singleton , so @xmath599 . if all @xmath600 then , since @xmath601 , we obtain that @xmath602 belongs to @xmath603 for each @xmath146 $ ] . call such a grid of _ type 0 _ and their number is denoted by @xmath604 . otherwise , there is an edge @xmath605 . we claim that this edge @xmath222 is unique . if @xmath222 and @xmath606 , then all @xmath607 meet @xmath332 in at least two vertices , hence all @xmath608 , hence @xmath31 and then @xmath349 are subfamilies of @xmath591 , but @xmath591 is grid - free . it follows that @xmath609 and @xmath610 are parts of @xmath538 and they are forming an @xmath611 grid ( on @xmath612 ) . call this @xmath352 a grid of _ type @xmath340 _ and denote the number by type @xmath340 grids by @xmath613 . we obtain a grid - free family of size at least @xmath614 if we leave out an edge from each grid in @xmath18 . next we estimate the expected size of @xmath615 when the vertices of each @xmath603 are permuted randomly and independently . concerning type 0 , there are at most @xmath616 choices of the vertex disjoint @xmath617 . given @xmath618 and @xmath340 the probability that @xmath619 indeed belongs to @xmath591 is exactly @xmath620 , which is at most @xmath621 . these events are independent , and the number of ways to choose @xmath622 is at most @xmath623 , so the expected number of type 0 grids @xmath624 given @xmath538 one can count type @xmath625 grids as follows ( the cases @xmath626 are similar ) . choose the edges @xmath627 , @xmath628 . each of @xmath629 intersect both of them . let @xmath630 , @xmath631 be the pair joining a vertex in @xmath632 and @xmath633 if @xmath221 intersects them in these two vertices . clearly , the edges @xmath630 form a matching between the vertices @xmath632 and @xmath634 and they determine @xmath629 . similarly , a matching between @xmath635 and @xmath636 determines @xmath637 . finally , choosing a vertex @xmath638 the number of possible choices so far is @xmath639 the vertices @xmath640 and @xmath641 are called @xmath642 and @xmath643 , resp . , @xmath644 . the probability that @xmath645 and @xmath646 both belong to @xmath647 is at most @xmath648 yielding @xmath649 . thus , for @xmath650 the expected number of grids in @xmath18 @xmath651 so there is a choice of permutations to make @xmath18 grid - free deleting only @xmath652 edges . this gives @xmath653 a short calculation shows that the right hand side is at least @xmath654 with some @xmath655 . since the turn function is monotone we have to consider only the case when @xmath1 divides @xmath32 , @xmath656 . let @xmath657 be an @xmath1-sum - free set of size @xmath658 provided by lemma [ le : behrend ] and let @xmath538 be the family defined by ( [ egyenes ] ) in subsection [ ss:31 ] . by lemma [ le:31 ] for @xmath11 @xmath554 is a linear hypergraph containing no grid . since the set of slopes @xmath318 is an @xmath1-sum - free set we can not have three lines with slopes @xmath659 forming a triangle , either . otherwise , we get @xmath660 for some @xmath661 ( mod @xmath16 , but it is not important here since all @xmath662 ) . finally , @xmath663 for some @xmath664 , as stated . * proof * of theorem [ th : main ] for @xmath67 . here we establish the lower bound ( [ eq : th_r=3 ] ) for @xmath665 when @xmath666 , and @xmath16 is a prime . first , recall that for @xmath67 @xmath554 may contain a grid , see ( [ eq:3 ] ) . therefore , to avoid grids and triangles at the same time we need to choose the slopes in @xmath318 more restrictively . for example , @xmath318 could be a set in @xmath667 which is @xmath300-free , @xmath304-free and @xmath316-free simultaneously . then lemma [ le:210 ] gives that @xmath538 is linear and grid - free . also the @xmath316-free property implies that @xmath554 has no triangles either . lemma [ le : a6 ] implies @xmath668 and then ( [ eq : sz ] ) completes the proof of the lower bound ( [ eq : th_r=3 ] ) . since @xmath554 contains neither grids nor triangles it is @xmath82-union free , too , by ( [ eq : conj9 ] ) . this implies @xmath669 . since @xmath554 is regular , ( and can be split into matchings ) , it is an optimal @xmath82-superimposed design . the bound @xmath670 exceeds the bound ( [ dyr ] ) for @xmath671 by dyachkov and rykov @xcite for @xmath67 . below we further improve both lower bounds with a different construction . define the hypergraphs @xmath672 and @xmath673 as follows on @xmath83 and @xmath674 vertices . + @xmath675 , + @xmath676 . + note that both are three - partite and the 3-partition of their vertices is unique . [ le:3-partite_union ] suppose that @xmath18 is a three - partite , linear hypergraph . it is @xmath82-union - free if and only if it avoids @xmath677 , @xmath672 , and @xmath673 . _ proof : _ we start like in the proof of corollary [ co:15 ] . suppose , that @xmath205 , @xmath678 , @xmath207 and @xmath679 form a linear @xmath82-uniform hypergraph . then @xmath209 . since @xmath210 , to cover the elements of @xmath211 the family @xmath212 must contain @xmath82 sets . we obtain @xmath680 . moreover , the sets @xmath681 meet @xmath211 in distinct elements . in the case of @xmath682 the latest property implies that every @xmath683 is covered by a unique @xmath48 , and every @xmath684 is covered by a unique @xmath47 , so @xmath685 is 2-regular , on 9 vertices , we obtain the grid @xmath677 . in the case of @xmath686 , say @xmath687 and @xmath688 we have that @xmath689 is a singleton and it must be the same element as @xmath690 . taking into the account that @xmath685 is 3-partite we obtain that it is isomorphic to @xmath673 when @xmath627 and @xmath691 are disjoint , and it is isomorphic to @xmath672 when @xmath627 and @xmath691 meet . finally , in the case @xmath692 , say @xmath688 , the other four sets meet @xmath691 in exactly one element , so there is a vertex @xmath108 of @xmath691 of degree at least @xmath82 . if @xmath693 then @xmath694 and @xmath695 covers @xmath696 and we could not finish because @xmath697 is 3-partite . similarly , if @xmath698 then @xmath699 must be covered by @xmath694 , a contradiction . so if the three configurations are avoided then @xmath18 is 3-union - free . * proof of proposition [ pr:10 ] . * the probabilistic lower bound from lemma [ le : erdos ] and the previous lemma imply the lower bound ( [ eq : u3 ] ) @xmath700 * proof of proposition [ pr:16 ] . * here we prove ( [ eq : k3 ] ) claiming @xmath701 . the proof is a refined version of the previous proof . let @xmath18 be the three - partite complete hypergraph with parts @xmath702 , @xmath703 and @xmath704 where @xmath705 as before . split it into @xmath706 perfect matchings @xmath707 where the second coordinates are taken modulo @xmath32 . we also call these _ parallel classes_. choose independently each of the @xmath706 matchings with probability @xmath153 , @xmath153 will be defined as @xmath708 . call the obtained random hypergraph @xmath54 . count the expected number of the arising configurations @xmath709 , @xmath672 , @xmath673 , and @xmath677 , those we want to avoid . here we have to be more careful , because although each edge belongs to @xmath54 with probability @xmath153 ( so its expected size is @xmath710 ) the choices of edges are not independent . more precisely , the probability that a subhypergraph @xmath711 appears in @xmath54 is exactly @xmath712 , where @xmath713 is the number of different parallel classes in @xmath711 . intersecting triples always belong to different classes , so they are independent , so the expected number of @xmath57 s is @xmath714 and the expected number of @xmath672 s is @xmath715 . the expected number of grids in @xmath54 with independent hyperedges is @xmath716 and the expected number of @xmath673 s with independent edges is at most @xmath717 . a configuration @xmath673 might have only one pair of dependent triples , namely the disjoint pair @xmath718 , @xmath719 . the number of these configurations is at most @xmath720 ( first we chose the triple corresponding to @xmath718 in @xmath721 ways , then its parallel edge at most @xmath32 ways , and finally the 7th vertex at most @xmath722 ways ) . so the expected number of these is at most @xmath723 . consider , finally , the grids @xmath218 containing parallel triples . note that the parallel classes of @xmath711 and @xmath724 are distinct , because every @xmath47 meets every @xmath48 . the number of these configurations is @xmath725 ( like before , first we chose a triple in @xmath721 ways , then its parallel edge at most @xmath32 ways , and finally the three remaining vertices at most @xmath721 ways ) . so if the number of independent classes @xmath726 , then the expected number of these grids is bounded by @xmath717 . if @xmath727 , then one of the three edges of the grid , say @xmath711 , consists of three parallel edges . the number of these configurations is at most @xmath728 , so we have an upper bound @xmath729 for the expected number of these whenever @xmath730 . finally , if @xmath731 , then both @xmath711 and @xmath724 consists of parallel edges . the number of those systems is at most @xmath732 so we have an upper bound @xmath733 . altogether the total expected number of configurations we want to avoid is bounded by a constant multiple of @xmath734 this is less than half of @xmath735 with our choice of @xmath153 . so we still have @xmath736 of the edges if we leave out an edge from each of the configurations we want to avoid , and thus we make the rest 3-union - free . moreover , conveniently , we can erase the unwanted edges together with all its parallels ( if @xmath711 is in @xmath18 then all of its shifted copy belongs to @xmath18 ) , so there is a random choice of @xmath18 where the remaining 3-union - free part is still large and union of matchings . dyachkov and rykov @xcite showed that there exists an optimal @xmath625-superimposed code ( i.e. , a regular graph ) with @xmath737 edges . it is easy to see , that there are @xmath240-regular @xmath32-vertex graphs for all @xmath738 ( for @xmath739 even ) . indeed , if @xmath32 is even , and @xmath740 then , e.g. , by baranyai s theorem @xcite the edge set of @xmath741 can be decomposed into @xmath742 perfect matchings . take @xmath240 of them . for odd @xmath32 ( and @xmath240 even ) @xmath741 can be decomposed into @xmath743 @xmath157-factors by a theorem of tutte . take @xmath744 of them . dyachkov and rykov @xcite showed that there exist optimal 2-superimposed designs ( i.e. , a @xmath157-union - free , @xmath240-regular graphs ) for @xmath745 . this can be improved to @xmath746 indeed , such an optimal design is just a @xmath240-regular , triangle and @xmath747-free graph . such a graph can be constructed using ( [ egyenes ] ) by taking a mod @xmath16 sidon set @xmath748 . the defined @xmath538 is the desired bipartite graph . we have an obvious upper bound @xmath749 . since @xmath750 by @xcite we have an upper bound @xmath751 giving the right order of magnitude . to determine the coefficient of the @xmath752 seems to be a very difficult question . erds and simonovits @xcite conjecture that @xmath753 they showed @xmath754 , i.e. , forbidding @xmath755 in magnitude is the same as forbidding all non - bipartite graphs . our main result is that the widely investigated transversal design @xmath538 ( see ( [ egyenes ] ) ) with @xmath573 is @xmath49-free for @xmath11 . if @xmath318 is @xmath1-sum - free then in addition @xmath538 has no triangles . it is natural to ask what other small substructures can be avoided this way . a. de bonis and u. vaccaro , optimal algorithms for two group testing problems and new bounds on generalized superimposed codes , _ ieee transactions on information theory , _ * 52 * ( 10 ) , 2006 , pp . 46734680 . w. g. brown , p. erds , and v. t. ss , some extremal problems on @xmath1-graphs , in : _ new directions in the theory of the graphs _ , proceedings of the third annual arbor conference on graph theory , academic press , new york , 1973 , pp . 5563 . p. erds , extremal problems in graph theory , in : m. fiedler ( ed . ) , _ theory of graphs and its applications _ , academic press , new york , 1964 , pp . p. erds , on extremal problems of graphs and generalized graphs . _ israel j. math . _ * 2 * ( 1964 ) , pp . 183-190 . p. erds , problems and results in combinatorial analysis . _ in : _ colloquio internazionale sulle teorie combinatorie ( rome , 1973 ) , tomo ii , 317 , atti dei convegni lincei , 17 , accad . lincei , rome , 1976 . p. erds and z. fredi , the greatest angle among @xmath32 points in the @xmath756dimensional euclidean space , _ annals of discrete math . _ * 17 * ( 1983 ) , _ combinatorial mathematics ( marseille - luminy , 1981 ) , north - holland math . stud . , 75 _ , pp . 275283 , north - holland , amsterdam - new york , 1983 . , on a problem of graph theory , _ studia sci hungar . _ * 1 * ( 1966 ) , 215235 . d. k. ray - chaudhuri and r. m. wilson , the existence of resolvable block designs , _ survey of combinatorial theory _ ( proc . , colorado state univ . , fort collins , colo . , 1971 ) , north - holland , amsterdam , 1973 , pp . 361375 . v. t. ss , an additive problem in different structures , _ proc . of the second int . conf . in graph theory , combinatorics , algorithms , and applications _ , san fra . california , july 1989 . siam , philadelphia , 1991 , pp .

we report observations from the _ hubble space telescope _ ( hst ) and the vla on the galactic environment , metallicity , and d / h in strong low - redshift ly absorption systems toward the bright bl lac object pks 2155 - 304 . these studies are intended to clarify the origin and chemical evolution of gas at large distances from galaxies . with ghrs / g160 m data at km s resolution , we detect a total of 14 ly absorbers , six of them clustered between and 18,500 km s . although _ orfeus _ studies claimed lyman continuum ( lyc ) absorption at with n(h i ) = , the ly data suggest a range , n(h i ) . even higher columns , needed for consistency with the _ orfeus _ lyc results , are possible if the ly line core at km s contains narrow h i components . we identify the ly cluster with a group of five h i galaxies offset by kpc from the sightline . the two strongest absorption features cover the same velocity range as the h i emission in the two galaxies closest to the line of sight . if the ly is associated with these galaxies , they must have huge halos ( kpc ) of highly turbulent , mostly ionized gas . the ly absorption could also arise from an extended sheet of intragroup gas , or from smaller primordial clouds and halos of dwarf galaxies . we see no absorption from si iii , c iv , or deuterium ly at the expected positions of the strongest ly absorbers . photoionization models yield ( ) limits of ( si / h ) , ( c / h ) , and ( d / h ) if n(h i ) has the _ orfeus _ value of . the limits increase to 0.023 solar metallicity and d / h if n(h i ) is only . these limits can be improved with further studies by hst / stis and measurements of the lyc and higher lyman series absorption by fuse . however , the current data suggest that the intergalactic gas in this group has not been enriched to the levels suggested by x - ray studies of intracluster gas . because of their low metallicity and large distance from galaxies , these absorbers could be primordial gas clouds . 470 pt = 10000

astro-ph9807246

the origin of the low - redshift ly@xmath0 absorption clouds discovered by the _ hubble space telescope _ ( bahcall et al . 1991 ; morris et al . 1991 ; stocke et al . 1995 ) remains a mystery , made more tantalizing by the possibility that they may contain significant mass . are they remnants of the high - redshift ly@xmath0 forest , or are they associated with past episodes of star formation , galactic outflows , and galaxy interactions ? based on their frequency , estimated size , and an ionization correction , they appear to contain a substantial amount ( @xmath20 ) of dark baryons ( shull et al . 1996 ; shull 1997 ) . there are also strong indications that some of these absorbers are associated with galaxies ( lanzetta et al . 1995 ; stocke et al . 1995 ; van gorkom et al . 1996 ) , although the distances to the nearest bright galaxies are often quite large ( morris et al . 1993 ; stocke et al . 1995 ; shull et al . 1996 ; grogin & geller 1998 ) . from numerical models , it now appears that the high-@xmath21 ly@xmath0 forest is part of a complicated gaseous structure , formed by the gravitational fluctuations of dark - matter potentials ( cen et al . 1994 ; hernquist et al . 1996 ; zhang et al . 1997 ) and at least partially `` polluted '' by heavy element nucleosynthesis during the epoch of galaxy formation . at high redshift , heavy elements ( c iv , si iv , c ii ) have been detected at @xmath22 times solar metallicity in 5075% of the ly@xmath0 forest clouds above @xmath23 @xmath6 column density ( cowie et al . 1995 ; tytler 1995 ) . from x - ray measurements of fe - line strengths in galaxy clusters , mushotzky & loewenstein ( 1997 ) suggest that intragroup gas might be enriched to levels of 10% solar metallicity at @xmath24 . however , extrapolating these ideas to low redshift or to all ly@xmath0 clouds is difficult . although a few metal - line absorption systems have been detected at moderate redshift , no low - redshift ly@xmath0-only absorber has yet been examined for the presence of heavy elements , which would indicate contamination by star formation . to investigate this possibility in the ly@xmath0 forest requires finding absorbers of sufficient column density to detect metals [ n(h i ) @xmath25 @xmath6 ] and located appropriately distant @xmath27 kpc ) from neighboring galaxies to suggest primordial gas . we now believe to have found an ideal target and absorbers , based on our newly analyzed ultraviolet spectra of pks 2155 - 304from the _ hubble space telescope _ ( hst ) and 21-cm emission images from the _ very large array _ ( vla ) . toward this bright , variable ( @xmath28 ) bl lac object , a strong ly@xmath0 line at @xmath29 km s@xmath2 was initially identified by the iue satellite ( maraschi et al . this and other ly@xmath0 lines were confirmed by low - resolution hst spectra with the faint object spectrograph ( fos / g130h ) ( allen et al . 1993 ) and with the goddard high resolution spectrograph ( ghrs / g140l ) ( bruhweiler et al . 1993 ) . the absorption was resolved by hst into at least two and possibly three systems . using fits to lyman - continuum ( lyc ) absorption near 960970 in low - resolution _ orfeus _ spectra , appenzeller et al . ( 1995 ) claimed that the 17,000 km s@xmath2 absorbers had a combined column n(h i ) @xmath30 @xmath6 . from our hst / ghrs measurements of the corresponding ly@xmath0 absorption , we estimate a range , n(h i ) @xmath7 @xmath6 . later in this paper , we will assess the accuracy of the h i column densities associated with the lyc absorption and discuss the conditions under which the _ orfeus _ and hst measurements could be reconciled . previously , our group observed the pks 2155 - 304 field with the vla ( van gorkom et al . 1996 ) and found evidence for three h i galaxies corresponding to ly@xmath0 absorbers at @xmath31 , 16,500 , and 17,000 km s@xmath2 . we associated the 17,000 km s@xmath2 absorbers with a small group of galaxies offset from the pks 2155 - 304 sightline by @xmath32 kpc for a hubble constant @xmath33 km s@xmath2 mpc@xmath34 . in this paper , we combine new hst and vla observations with theoretical interpretation of the pks 2155 - 304 sightline , which is unusual in the large number of strong ly@xmath0absorbers and their association with four large galaxies located within @xmath35 mpc . this number of galaxies would be high by chance , based upon the observed two - point correlation function . the average space density of galaxies in this local region , @xmath36 mpc@xmath37 , is @xmath38 times the large - scale average density of @xmath39 galaxies ( marzke , huchra & geller 1994 ) . this suggests the presence of a small group that has recently turned around from the hubble flow , much like our local group . these galaxies probably have not undergone significant mergers , but the presence of strong , broad ly@xmath0 absorption from extended gas may indicate some dynamical interaction . in 2.1 , we describe new hst / ghrs observations at 20 km s@xmath2 resolution between 1258 and 1293 ( @xmath40 km s@xmath2 ) . we detect 7 ly@xmath0 absorbers with equivalent widths ranging from 68 to 467 m . we also reanalyze an archival ghrs / g160 m spectrum at 1223 1258 ( @xmath41 km s@xmath2 ) and identify 7 definite ly@xmath0 absorbers ( 21201 m ) . in 2.2 , we present new vla studies of h i emission in the 2155 - 304 field from @xmath42 km s@xmath2 , a velocity range that includes four ly@xmath0 absorption systems . in 3 , we discuss the results of these studies . the new vla data provide accurate correspondence with the ly@xmath0 absorbers , suggesting that the absorption might arise from extended halos or intragroup gas over a region 1 mpc in diameter that could total @xmath43 in mass , if bound . the hst data also allow us to set limits on the metallicity of the strongest absorbers from the absence of si iii @xmath11 and c iv @xmath12 , and they provide a limit on d / h from residual ( d i ) ly@xmath0 absorption in the shortward wing of the strongest absorber . in 4 we summarize our conclusions and give suggestions for further study . somewhat contrary to the kinematic evidence , the low metallicities , [ si / h ] @xmath44 solar and [ c / h ] @xmath45 solar , would suggest that some primordial gas may still reside amidst the large - scale filaments of galaxies . the target , pks 2155 - 304 , lies at redshift @xmath46 or @xmath47 km s@xmath2(falomo , pesce & treves 1993 ) . it was observed by hst during cycle 6 on october 5 , 1996 , using the ghrs with the g160 m grating and post - costar optics . the continuum flux near 1280 was @xmath48 ergs @xmath6 s@xmath2 @xmath2 , about 70% of the median flux for this object over the past 15 years of observations in the iueagn database ( penton , shull & edelson 1998 ) . over the interval 12581293 , we obtained a resolution of 4.5 km s@xmath2 per 0.018 quarter - stepped pixel or @xmath1 km s@xmath2 per resolution element . the wavelength scale was determined by assuming that the galactic interstellar s ii absorption features at 1250.584 , 1253.811 , and 1259.519 lie at zero velocity in the local standard of rest ( lsr ) . based on the bell labs 21-cm survey ( stark et al . 1992 ) , the dominant h i absorption in this direction lies at @xmath49 , within the accuracy ( @xmath50 km s@xmath2 ) of the observations . this correction in the wavelength scale in the new data ( fig . 1 ) was @xmath51 , and that in the archival data ( fig . 2 ) was @xmath52 . our observations correspond to redshifted velocities @xmath40 km s@xmath2 ( @xmath53 ) in the ly@xmath0 line . according to previous convention , we quote heliocentric velocities , @xmath54 , and compute expected line positions from @xmath55 . we do not make relativistic corrections to the velocities . figure 1 shows our new ghrs / g160 m data ( 12581293 ) with s / n @xmath56 . figure 2 shows our reanalysis of an unpublished archival ghrs / g160 m spectrum with s / n @xmath57 ( 12231258 ) taken with pre - costar optics . the 17,000 km s@xmath2 ly@xmath0 absorbers are the strongest of the low - redshift ( @xmath58 ) absorbers found in hst searches other than associated absorbers with @xmath59 . because of the exceptional brightness of the background source ( @xmath60 mag ) and the high h i column densities ( @xmath61 @xmath6 ) , these absorbers are ideally suited for measuring heavy - element abundances in the low-@xmath21 forest . as of this writing , our group has studied @xmath62 low - redshift ly@xmath0 absorbers toward 11 bright quasars and seyfert galaxies ( stocke et al . 1995 ; shull et al . 1996 ; shull 1997 ) with more absorbers under analysis ( penton et al . our goal is to understand the physical structure of these clouds and their possible connections with galaxy halos , large - scale structure , and voids . our data reduction method was described in these earlier papers . complete results on the seven targets in our hst cycle 6 observations , together with calibration , are described in penton et al . ( 1998 ) . the ghrs spectra were taken through the @xmath63 large science aperture , using the standard quarter - diode sub - stepping pattern to yield pixels of 0.018 in fp - split mode . we recalibrated our spectra using iraf / stsdas / calhrs and the final ghrs reference files ( sherbert & hulbert 1997 ) with polynomial background subraction . one noteworthy point concerns our treatment of the hst / ghrs error vectors , which affect how we gauge the statistical significance of weak features . we found that the error vectors produced by the iraf / stsdas program * specalign * in the hst data handbook do not agree with those obtained by standard propagation of errors from individual subexposures . we have therefore chosen to perform our own error propagation and spectral coaddition using the idl software package . spectral coaddition was weighted by exposure time on a quarter - stepped , pixel - by - pixel basis , removing blemished pixels . this procedure gives some pixels less exposure time than others , but it prevents photocathode blemishes from being erronously flagged as absorption features . in some cases , this procedure results in no exposure time being available for some pixels ; straight lines between the last known good flux values are then used in figures 1 and 2 . the significance of the detected lines ( in @xmath64 ) is defined as the integrated s / n per resolution element of the fitted absorption feature . the status of `` definite absorber '' is given to features greater than @xmath13 significance ( penton et al . table 1 lists the identified absorption features and their equivalent widths . we see the expected galactic interstellar lines of s ii @xmath65 and si ii @xmath66 , as well as a high - velocity cloud in si ii at @xmath67 km s@xmath2 , which was also seen in c iv at @xmath68 km s@xmath2 by sembach et al . we also confirm the strong intergalactic ly@xmath0 absorbers near 17,000 km s@xmath2 . a blowup of the 12781292 region shows that many of the ly@xmath0 lines have velocity substructure , which we model as separate gaussian components . we included these additional components to reduce the @xmath69 per degree of freedom whenever the resulting components were above @xmath13 significance . in no case did we allow more than two components per blended absorption feature , although we occasionally tested for three . in addition to the definite ly@xmath0 absorbers , we found a number of features at the @xmath70 level that might be verified with future stis data . examples of these `` possible features '' occur at 1229.0 , 1243.2 , 1249.5 , 1264.8 , 1286.45 , and 1290.58 . the local continua are uncertain near these features , and some of the features are too broad ( @xmath71 km s@xmath2 ) to be discrete ly@xmath0 absorbers by our criteria . in these cases , we have decided to be conservative and not classify them as definite . we also analyzed an archival spectrum ( fig . 2 ) taken with the ghrs / g160 m using pre - costar optics . this spectrum was reduced and analyzed in a manner identical to that in figure 1 , with the wavelength scale zero point set to the lsr using the galactic interstellar s ii lines at 1250.584 and 1253.811 . between 1223 and 1258 , we detect 7 definite ly@xmath0 lines and three galactic interstellar lines : the two s ii resonance lines and the weak mg ii doublet . the strong ly@xmath0 feature at 1236 is easily detected and is resolved into two subcomponents separated by @xmath72 km s@xmath2 . this pair of lines , at @xmath73 and 5121 km s@xmath2 , lies at the same velocity as the h i detected galaxy at @xmath74 , @xmath75 just off the eastern edge of figure 3 and discussed in detail by van gorkom et al . this sub-@xmath39 galaxy is located @xmath76 kpc from the pks 2155 - 304 sightline . the absorption line at 1238.744 lies 0.077 blueward of the expected position of galactic n v @xmath77 . we identify this line as ly@xmath0 because of the slight wavelength offset and what would be an unusual strength for n v. based on the absence of the weaker line of the n v doublet at 1242.804 , we believe that n v @xmath77 contributes at most 30% ( at the @xmath13 level ) to the equivalent width of the ly@xmath0 feature at 1238.744 . the pks 2155 - 304 sightline appears to be unusual , both in the number of ly@xmath0 absorbers and in their strength . in total , from 1223 1293 , we identify 14 ly@xmath0 absorbers with significance greater than @xmath78 . with a correction for regions blocked by galactic s ii and si ii , these 14 ly@xmath0 absorbers correspond to an uncorrected frequency @xmath79 above 21 m ( n@xmath80 @xmath6 ) , for n@xmath81 @xmath6 . this frequency is somewhat higher than the mean value , to similar absorption strength , in sightlines studied with the ghrs during cycles 2 and 4 ( shull et al . 1996 ) . probably the most unusual aspect of this sightline , however , is the large number of strong ly@xmath0 absorbers near 17,000 km s@xmath2 . the strong correspondence in recession velocity between these ly@xmath0 absorbers and the surrounding galaxies argues that these clouds are not ejected from the bl lac object . rather , they appear to be intervening clouds at distances given by the hubble law , @xmath82 , as we assume throughout this paper . our previous results ( van gorkom et al . 1996 ) showed that two of the strongest ly@xmath0 absorbers found toward pks 2155 - 304 at 17,100 km s@xmath2 are located within a loose group of galaxies . three galaxies were detected in h i , two at the velocities of the absorbers , but no individual galaxy could be identified as being _ associated _ with the absorber . the high column densities observed in the absorbers suggested that there may be an extended component of intergalactic neutral gas . column densities n(h i ) @xmath83 @xmath6 are rarely detected in emission ( van gorkom 1993 ) , and common lore suggests that it may be ionized by the intergalactic uv background . the ly@xmath0 results suggested that this group may be the ideal place to look for diffuse h i at very low column densities . we therefore reobserved the group in hope of detecting a low surface brightness diffuse h i component . the observations were made in 1996 may with the vla in the 1 km ( d ) array with an extended north arm ( 3 km ) to compensate for the low declination of the source . the total integration time was 40 hrs spread over 8 different runs . all observations were centered at the radio position of pks 2155 - 304 , @xmath84 , @xmath85 ( b1950 ) . we used a total bandwidth of 6.25 mhz centered at 16,880 km s@xmath2 ; the usable velocity range is about 1300 km s@xmath2 . on - line hanning smoothing was employed , after which every other channel was discarded , leaving a set of 31 independent channels and resulting in a velocity resolution of 46 km s@xmath2 . the bl lac object is a radio continuum source with a variable flux density . we measured a flux density of 0.45 jy at 1.4 ghz . extreme care was taken to properly calibrate the bandpass . every 25 minutes a bandpass calibration was done , and the bandpass solution for these individual scans was interpolated to do the correction . as a result , our observations are limited by noise rather than by spectral dynamic range , which is better than 3000:1 . the u - v data for each of the 8 days were calibrated independently and inspected for interference and residual calibration errors . subsequently , the u - v data of all runs were combined . the continuum was subtracted by making a linear fit in frequency to the calibrated complex visibilities of the line free channels ( 2 - 5 and 23 - 25 ) . the resulting data were clipped at a level of 0.7 jy to remove man - made and solar interference . images were made using a taper that compromised between optimal surface brightness sensitivity and maximal sidelobe suppression ( using the task imagr in aips , with robustness factor 1 ) , resulting in a synthesized beam of @xmath86 . the rms noise in the channel images is 0.15 mjy beam@xmath2 . the instrumental parameters are summarized in table 2 . to search for h i , we imaged the entire primary beam ( @xmath87 ) . to calculate h i masses , we use the luminosity distance ( sandage 1975 ) assuming that the group is at @xmath88 and using @xmath89 km s@xmath2 mpc@xmath2 and @xmath90 . our 6 @xmath64 h i mass limit in the center of the field is @xmath91 . at full resolution , the column density sensitivity is @xmath92 @xmath6 ( @xmath93 over 57 kpc @xmath94 34 kpc @xmath94 46 km s@xmath2 ) . the data were smoothed spatially and in velocity down to a resolution @xmath95 , resulting in a detection limit of @xmath96 @xmath6 ( 5 @xmath64 over 126 kpc @xmath94 189 kpc @xmath94 92 km s@xmath2 ) . owing to a lack of short spacings , the observations are much less sensitive to h i emission that is completely smooth on scales larger than @xmath97 in a single 46 km s@xmath2 velocity channel . these values are valid for the center of the field . farther from the center , they have to be corrected for the change in primary beam response . the primary beam pattern is roughly gaussian with a fwhm of @xmath98 , and the limits are a factor two worse @xmath97 from the field center . the properties of the five galaxies detected in h i are summarized in table 3 and shown in figure 3 . perhaps the most interesting result of these observations is what we do not find . we do not detect a faint intergalactic component in h i , nor do we detect any diffuse extended h i associated with the galaxies . compared to van gorkom et al . ( 1996 ) , our new data are of much better quality and only show more clearly that the h i emission is confined to the galaxies at column densities well above @xmath99 @xmath6 . in addition to the 3 galaxies detected in the previous observations , we detect to the southwest an s0 galaxy cataloged in the apm catalog ( loveday 1996 ) with a velocity range partly outside our band . we see h i to the northwest at 16,375 km s@xmath2 , moving toward the center of the galaxy at lower velocities . the lowest velocity where we have usable , though very noisy data , is at 16237 km s@xmath2 . since this emission peaks only slightly to the northwest of the center , we infer that the systemic velocity of the galaxy must be about 16200 km s@xmath2 . this galaxy was not detected in our previous observations , because in those data that velocity range was seriously affected by interference . the other new detection is a tiny dwarf galaxy to the east at 17,156 km s@xmath2 . in the current data , the galaxy only shows up in one channel , but at the 8 @xmath64 level . going back to our previous data , we found the dwarf just above the noise in our oldest data set ( and at the edge of the band there ) . those data were taken with 11 km s@xmath2 velocity resolution , and the emission seems to cover about 60 km s@xmath2 . an overlay of just the dwarf on the digitized poss is shown in figure 4 . a hint of some faint light ( @xmath100 ) can be seen on the blue poss image , but it is clear from the image that similar dwarfs could easily be missed optically . in h i the dwarf is just above our detection limit . although somewhat brighter galaxies with larger h i masses could have been detected closer to the sightline , in fact none was seen ( fig . are we getting any closer to identifying galaxies associated with the ly@xmath0 absorption ? there are 3 galaxies with h i emission in the range 16,900 17,100 km s@xmath2 , the velocity of the deepest ly@xmath0 absorption . thus , in the first instance , it is not obvious that the absorption would be _ associated _ with any one of the galaxies . in our previous work ( van gorkom et al . 1996 ) we noticed a curious phenomenon : in all cases , the ly@xmath0 absorption was at the systemic velocity of the closest galaxy . in the current , much improved , ly@xmath0 data we notice something even more curious : the two strong features at 17,100 km s@xmath2and at 16,200 km s@xmath2 cover exactly the same velocity range as the h i emission in the two nearest galaxies . in figure 3 we show the total h i emission contours overlaid on an optical image ( greyscale ) , and we also indicate for each galaxy the sense of rotation and velocity extent of the h i emission . as in our previous data , there is no evidence for ly@xmath0corotating with the h i disks . the broad widths of the ly@xmath0 absorbers in the current data is intriguing . it seems unlikely that the two strong absorbers are associated with their nearest galaxies only . if the width of the ly@xmath0 lines were to reflect the potential of the nearest galaxies , each would have a halo characterized by a flat rotation curve extending out to 400 and 500 kpc respectively . perhaps more likely , the widths could indicate that the clouds probe a group potential . in that scenario we may be viewing two groups along the line of sight , one centered at 17,100 km s@xmath2 , the other at 16,200 km s@xmath2 . the detection of 3 galaxies with systemic velocities close to 17,100 km s@xmath2 makes it plausible that there is indeed a loose group at this velocity . our velocity coverage does not extend to sufficiently low velocities to make any statements about the presence of a group at 16,200 km s@xmath2 . the combined hst and vla data suggest that the 17,000 km s@xmath2 ly@xmath0 absorption lines arise in a group environment . between 16,100 and 18,100 km s@xmath2 , we observe 6 ly@xmath0 absorbers and four large ( vla ) galaxies at similar velocities . the velocity centroid and radial velocity dispersion of the four galaxies are @xmath101 km s@xmath2 and @xmath102 km s@xmath2 . the equivalent statistics for the 6 ly@xmath0 absorbers are @xmath103 km s@xmath2 and @xmath104 km s@xmath2 . the fact that the strongest ly@xmath0 absorbers ( 1284.484 and 1285.097 ) lie close to @xmath105 suggests that the center of the gravitational potential well may be centered at @xmath106 km s@xmath2 . before we interpret the larger velocity dispersion of ly@xmath0 lines , it is worth noting that this galaxy group was observed in h i emission with the vla over a more limited velocity range ( 16,283 17,571 km s@xmath2 ) than the ly@xmath0 lines . the simplest interpretation of the data is that the absorbing gas arises in intragroup material stripped or blown out of the gas - rich galaxies . however , there is no direct evidence for this interpretation in the h i emission data . nor are galaxy h i halos expected to be this dense at such enormous radii ( dove & shull 1994 ) . combining the 740 km s@xmath2 velocity dispersion of the 6 definite ly@xmath0 absorbers with the mean projected separation , @xmath107 mpc between the four galaxies , we obtain a virial mass estimate , @xmath108 the estimated velocity dispersion and size / mass scale are those of a modest group , less than the size of the virgo cluster . the mean distance of these galaxies at @xmath109 km s@xmath2 is @xmath110 mpc or @xmath111 . the break in the standard luminosity function is @xmath112 or @xmath113 ( marzke et al . 1994 ) in the zwicky ( blue ) magnitude system . thus , an @xmath39 galaxy in this group would lie at @xmath114 . using two slightly brighter galaxies in the field for calibration of the src - j plate magnitudes ( lauberts & valentijn 1989 ) , we have estimated b@xmath115 magnitudes for the five galaxies around pks 2155 - 304 . in terms of @xmath39 , they are : north ( @xmath116 ) , east ( @xmath117 ) , south ( @xmath118 ) , southwest ( @xmath119 ) , and dwarf ( @xmath120 ) . thus , three of the h i galaxies have luminosities essentially at or above @xmath39 . although drawing statistical inferences from only four galaxies is risky , we have attempted to estimate the possible number of fainter galaxies , both observationally and theoretically . in a cfht image ( wurtz et al . 1997 ) centered on pks 2155 - 304 , of radius @xmath121 kpc at @xmath122 km s@xmath2 , we found five galaxies in excess of the background to a limit of @xmath123 , about 3.5 magnitudes below @xmath39 . over the larger field of @xmath124 kpc radius , defined by the four h i galaxies , this corresponds to @xmath125 excess galaxies . however , this photometry does not constrain the redshift of these excess galaxies ; most , if not all , could lie at the redshift , @xmath126 , of pks 2155 - 304 ( falomo et al . 1993 ) . since bl lac objects are routinely found in poor clusters ( wurtz et al . 1997 ) , these five excess galaxies are likely to be associated with pks 2155 - 304 . if we integrate a schechter luminosity function , @xmath127 , down to @xmath128 ( @xmath123 ) , we would expect between 3 and 10 times more galaxies , compared to those above @xmath129 and @xmath117 , respectively . scaling from the observed h i galaxies above those limits , we would expect to find between 10 and 30 galaxies associated with this group , down to @xmath128 . therefore our preliminary optical survey is consistent with these expectations . however , since our radio h i survey detected only one dwarf galaxy in the field , gas - rich dwarfs may be scarce in this group . we are conducting a spectroscopic survey of the pks 2155 - 304 field down to @xmath130 to test our hypothesis that the large vla galaxies are accompanied by smaller galaxies , some closer to the pks 2155 - 304 sightline . we also investigated the possibility that hot gas associated with the strong absorbers at 17,100 km s@xmath2 might be responsible for the 600 ev absorption feature seem by the objective grating spectrometer on the _ einstein observatory _ ( canizares & kruper 1984 ) and by the bbxrt spectrograph ( madejski et al . this possibility would only work if the x - ray absorption arose from an o v k - edge ( 626 ev rest frame , 592 ev observed ) or an o viii k@xmath0 line ( 653 ev rest frame , 618 ev observed ) . we can rule out the o v hypothesis ( @xmath131 k ) because it would require an unreasonably large hydrogen column , n@xmath132 @xmath133 , for oxygen metallicity @xmath134 , in order to produce the observed optical depth , @xmath135 . the resulting gas densities would be quite high , across a 300 kpc slab , and the cooling time would be less than @xmath136 yrs . line absorption by o viii would be more feasible , requiring a column density n@xmath137 . however , neither of these cluster hot - gas models would explain the observed broad ( @xmath138 km s@xmath2 ) absorption widths . we conclude that the 600 ev absorption must be produced elsewhere , probably within pks 2155 - 304 . because the four bright galaxies are h i - rich spirals , it is also possible that these galaxies are not bound , based upon recent spectroscopic work on similar galaxy groups ( mulchaey & zabludoff 1998 ; zabludoff & mulchaey 1998 ) . even if they are bound , these galaxies may not have virialized or closely interacted . if this is the case , the ly@xmath0 clouds may also be non - virialized , and our estimate of @xmath139 would be too large . this interpretation suggests that these clouds are portions of the gaseous filament out of which these spiral galaxies formed . if these clouds and galaxies are portions of a bound group , then we would expect the clouds to have metallicities of 0.1 0.3 solar , similar to the stripped and virialized gas in rich clusters and elliptical - rich groups ( mushotzky & loewenstein 1997 ) . in the unbound case , not only would x - ray emission be unlikely , but the metallicity of the gas would be substantially lower than 10% solar ( see 3.2 ) . in our earlier papers on the environments of the low-@xmath21 absorbers ( shull et al . 1996 ; van gorkom et al . 1996 ) , we estimated a space density , @xmath140 , of low-@xmath21 ly@xmath0 clouds with n(h i ) @xmath141 @xmath6 and characteristic size scale @xmath142 , in order to explain their frequency per unit redshift . this space density is @xmath143 times larger than that of @xmath39 galaxies , but comparable to that of dwarf galaxies with @xmath144 , suggesting a possible connection . however , the h i absorption cross sections of these dwarfs are uncertain , and it is unclear whether the extended gas was stripped out , blown out , or existed primordially . the main results of our comparison between the ly@xmath0 absorption and the 21-cm emission can be summarized as follows : ( 1 ) 21-cm emission has been detected in this region , spread over 1 mpc of sky and 1300 km s@xmath2 of velocity ; ( 2 ) the large distances to the nearest bright galaxies ( 400 - 800@xmath145 kpc ) ; and ( 3 ) the agreement in velocity ( within @xmath146 km s@xmath2 ) between several of the ly@xmath0 absorbers and the h i galaxies to the south , southwest , and north . the proper interpretation of the ly@xmath0 absorbers toward pks 2155 - 304 hinges on several geometrical issues . does the ly@xmath0 absorption occur in a smoothly distributed layer , 200 - 600 kpc in depth , or in smaller , denser clumps ? some evidence for the latter interpretation comes from the fact that the ly@xmath0 absorption occurs in discrete systems in velocity space . less certain is whether the absorbers are kinematically associated with the large galaxies . although a single sight line through the region is not necessarily typical , the vla observations suggest that the absorbing gas is distributed across a region @xmath35 mpc in diameter . as a first approximation , we will model the absorption as a homogeneous slab . we then explore more complex distributions , either in intergalactic clouds or in dwarf - galaxy halos . in the homogeneous approximation , the total mass of gas in the vicinity of the cluster of galaxies around pks 2155 - 304 can be estimated by assuming a slab of radius 500 kpc , depth @xmath147 kpc , mean column density n(h i ) @xmath148 @xmath6 , exposed to an ionizing radiation field with specific intensity @xmath149 ergs @xmath6 s@xmath2 hz@xmath2 sr@xmath2 at 1 ryd and spectral slope @xmath150 , as described in 3.2 . ( the bl lac , pks 2155 - 304 can not dominate the ionizing flux incident on these clouds , unless its actual redshift is much less than the value , @xmath46 , from falomo et al . if @xmath151 and @xmath152 are constant , then n(h i ) @xmath153 , so that total density , @xmath154 , and gas mass , @xmath155 , scale as [ n(h i)]@xmath156 . for the parameters above , the mean densities are @xmath157 cm@xmath37 and @xmath159 cm@xmath37 , and the total gaseous mass , including helium , is @xmath160 ^ 2 \left [ \frac { \rm n(h~i ) } { 2 \times 10^{16}~{\rm cm}^{-2 } } \right ] ^{1/2 } \ ; .\ ] ] the gas mass could therefore approach 1% of the @xmath161 required to bind the group , a relatively small gas fraction compared to the gas found in clusters and elliptical - rich groups . likewise , the luminous mass in these galaxies is small . if the gas is clumped into denser parcels with the same total covering factor , the neutral fraction would increase and the total gas mass would decrease inversely with the characteristic depth of the absorbers . let us now consider inhomogeneous models for the gas distribution around pks 2155 - 304 , involving dwarf galaxies or primordial gas filaments . let the total projected area of the gas be @xmath162 with depth @xmath152 , where @xmath163 mpc and @xmath164 kpc . assume that this volume is filled by an ensemble of @xmath165 clouds , each of radius @xmath166 kpc , with a velocity dispersion 750 km s@xmath2 . the cloud filling factors in area and volume are , @xmath167 and @xmath168 , where @xmath169 could exceed 1 if clouds overlapped in projected area . using the given parameters , their ratio is @xmath170 . since we detected many absorbers along the sightline , it is likely that @xmath171 , which suggests that @xmath172 . a consequence of this model is that these `` clouds '' should collide with one other on a crossing time @xmath173= ( 0.6 mpc)/(750 km s@xmath2 ) @xmath174 yr . therefore , if this group is bound , we would expect considerable gas stripping , with the possibility of shocks , hot gas ( @xmath175 k ) , and tidal plumes . there is no evidence for these effects , although there has been no search for extended soft x - ray emission or o vi absorption from this region . if the group of galaxies and clouds is not bound , then this cloud ensemble could either be primordial gas which never participated in galaxy formation or gaseous halos of luminous or dwarf galaxies . based upon an extensive survey of quasar fields , for which ultraviolet spectra were obtained by the hst key project team , lanzetta et al . ( 1995 , hereafter l95 ) identified @xmath176 of large equivalent width ( @xmath177 ) ly@xmath0 absorption lines with bright galaxies at impact parameters @xmath178 kpc away from the sightline . the three strongest close pairs of ly@xmath0 clouds in table 1 would not be resolved into multiple components at the 1 resolution of the key project spectra . while all three cloud complexes are easily strong enough to be included in the l95 study , none is as close to a bright galaxy as those in the l95 survey : @xmath179 kpc for the 5100 km s@xmath2 absorber , @xmath180 kpc for the 16,300 km s@xmath2 absorber , and @xmath181 kpc for the 17,000 km s@xmath2 absorber . our h i detection survey in this field ( van gorkom et al . 1996 and 2.2 herein ) reaches depths at least comparable to the l95 survey ( 0.10.3 @xmath39 ) . thus , the bright galaxy halos in this field may be anomalously larger than those found by l95 in other fields , and also substantially larger than expected theoretically ( dove & shull 1994 ) . however , we strongly suspect that bright galaxy halos can not account for these clouds . are there undetected dwarf galaxies at the ly@xmath0 cloud redshifts closer to the pks 2155 - 304 sightline ? if so , then dwarf - galaxy halos could be responsible for these absorptions as extrapolated by l95 and others . chen et al . ( 1998 ) found a best - fit ly@xmath0 halo size that scales with galaxy luminosity as @xmath182 , a somewhat stronger dependence than the @xmath183 found for mg ii absorbing halos ( steidel 1995 ) . with the chen et al . scaling relation , the two weaker absorption complexes ( 5100 and 16,200 km s@xmath2 ) would be consistent with the l95 and chen et al . observations if any fainter galaxies ( @xmath184 ) were found closer to pks 2155 - 304 at those redshifts . however , the 17,000 km s@xmath2 absorber is so strong that a dwarf galaxy would need to be very close to the sightline to satisfy the chen et al . halo - size relationship . specifically , we would require impact parameters @xmath185 kpc for @xmath186 and @xmath187 kpc for @xmath188 . the only bright galaxy sufficiently near to pks 2155 - 304 to satisfy the first condition is galaxy g4 of falomo et al . ( 1993 ) , which has @xmath189 and @xmath190 kpc . however , g4 has @xmath191 , the same as pks 2155 - 304 . while there are some closer galaxies , their numbers are modest : 1 3 excess galaxies to @xmath192 based on the independent data from wurtz et al . ( 1997 ) and falomo et al . the only two galaxies with spectroscopy in hand are galaxies 1 and 2 of falomo et al . ( 1993 ) at @xmath191 , located 4 arcsec e and 25 arcsec se of the bl lac , respectively . at this point , there is no dwarf galaxy candidate for any of these three strong absorption complexes . thus , although unlikely , the dwarf - galaxy halo hypothesis remains possible . we detected one faint dwarf ( @xmath193 ) at @xmath194 from the pks 2155 - 304 sightline , and we may find other faint galaxies at the redshifts of the three strong absorption complexes . however , there is no direct evidence in favor of the dwarf - halo , and the primordial cloud hypothesis must be taken seriously for all three ly@xmath0 absorbers . we can also use our spectra to set limits on the metallicity of the strongest ly@xmath0 absorbers . as seen in figure 1 , metal lines should be searched for in three strong absorption systems , corresponding to ly@xmath0 lines at @xmath195 ( @xmath196 ) , @xmath197 ( @xmath198 ) , and @xmath199 ( @xmath200 ) . in the limited wavelength range of our ghrs / g160 m spectrum , the only expected strong metal line is the si iii @xmath201 resonance line . no si iii absorption is present at the 1271.73 location of system 1 . although we see a hint of si iii absorption at the expected positions [ 1274.79 and 1275.40 , see fig . 1 ] corresponding to systems 2 and 3 , we treat this absorption as _ upper limits_. using the unsmoothed data to detect weak , unresolved features , we obtain a formal @xmath78 error on equivalent width of 23 m ( rest - frame 22 m ) . to convert the observed si iii / h i to limits on abundances ( si / h ) , we must make an ionization correction based on photoionization conditions in the absorbers . appenzeller et al . ( 1995 ) claimed to detect weak lyc absorption with the far - uv spectrograph aboard _ orfeus_. the ly@xmath0 clouds at @xmath202 were estimated to have a combined column density n(h i ) = @xmath5 @xmath6 . we have derived the expected column densities of metal ions by modeling the strong absorbers as slabs with n(h i ) = @xmath17 @xmath6 and total depth ranging from 5 to 800 kpc , comparable to or less than the offset distance from the nearest galaxies . the slabs are assumed to be illuminated on both sides by an ionizing spectrum with specific intensity at 1 ryd of @xmath149 ergs @xmath6 s@xmath2 hz@xmath2 sr@xmath2 , consistent with a recent calculation by shull et al . ( 1998 ) that gives @xmath203 , based on low-@xmath21 seyfert galaxies and a new igm opacity model . the spectral index of the ionizing background is taken as @xmath204 ( zheng et al . 1997 ) . as a `` standard model '' , we assume a homogeneous slab , of depth 400 kpc and 0.003 solar metallicity . the photoionization equilibrium is computed with the model cloudy version 90.03 ( ferland 1996 ) . the results are given in table 4 . in general , the c iii @xmath205 and c iv @xmath12 lines are the best tracers of metals , while si iii @xmath11 is a factor of 510 weaker , assuming relative solar abundances of ( si / h)@xmath206 and ( c / h)@xmath207 ( grevesse & anders 1989 ) . however , the c iii line lies shortward of the hst band , and its observation must await the launch of the fuse satellite in early 1999 . for unsaturated absorption , the predicted equivalent widths can be written : @xmath208 \left [ \frac { \rm [ c / h ] } { 0.003 } \right ] \ ; , \\ \nonumber \\ w_{\lambda}(\rm si~iii ) = ( 22~{\rm m\aa } ) \left [ \frac { \rm n(h~i ) } { 2 \times 10^{16}~{\rm cm}^{-2 } } \right ] \left [ \frac { \rm [ si / h ] } { 0.003 } \right ] \ ; . \end{aligned}\ ] ] following the convention of our earlier ly@xmath0 work , we treat `` definite absorbers '' as those with @xmath78 significance . the observed 23 m ( @xmath78 ) limit on si iii @xmath11 ( rest equivalent width 22 m ) corresponds to n(si iii ) @xmath209 @xmath6 . if we scale the column density of the strong absorption components to n(h i)@xmath210 @xmath6 , the ( @xmath78 ) upper limit on metallicity can be written , @xmath211 \leq ( 0.003 ) \left [ \frac { 2 \times 10^{16}~{\rm cm}^{-2 } } { \rm n(h~i ) } \right ] \left ( \frac { \rm si } { \rm h } \right)_{\odot } \ ; .\ ] ] another limit on metallicity comes from c iv @xmath212 . from their low - resolution ( ghrs / g140l ) spectrum , bruhweiler et al . ( 1993 ) quote a ( @xmath213 ) upper limit of 110 m for c iv @xmath12 . we recalibrated the pre - costar ghrs / g140l spectrum using iraf / stsdas / calhrs with the final calibration files . as before , subexposure coaddition was performed using our own idl routines , which merge the subexposures by exposure time with photocathode blemishes removed ( fig . we find a @xmath78 limit of @xmath214 m ( 127 m rest frame ) or n(c iv ) @xmath215 @xmath6 for a linear curve of growth . this yields a metallicity limit , @xmath216 \leq ( 0.005 ) \left [ \frac { 2 \times 10^{16}~{\rm cm}^{-2 } } { \rm n(h~i ) } \right ] \left ( \frac { \rm c } { \rm h } \right)_{\odot } \ ; .\ ] ] figures 6 and 7 illustrate how these metallicities depend on the two key parameters of the photoionization models : the mean hydrogen density @xmath217 and the spectral slope @xmath218 of the ionizing background . the observed column density , n(h i ) , and the assumed cloud depth , @xmath152 , determine the mean neutral density , @xmath219 \left [ \frac { 400~{\rm kpc } } { d } \right ] \ ; .\ ] ] the neutral fraction , @xmath220 , is set by photoionization equilibrium and is proportional to the ionization parameter , @xmath221 , as are the other metal - ion column densities . the precise behavior of the ratios ( si iii / h i ) and ( c iv / h i ) depend on the range of @xmath222 where these particular ions peak in fractional abundance ( donahue & shull 1991 ) . the spectral slope , @xmath218 , affects the ionization rates above the relevant ionization thresholds ( 16.34 , 33.46 , 45.13 ev for si ii , iii , iv ; 24.38 , 47.87 , 64.48 ev for c ii , iii , iv ) . for a fixed radiation intensity , @xmath223 , decreasing the assumed cloud depth @xmath152 results in a higher @xmath217 , smaller ionization parameter @xmath222 , larger hydrogen neutral fraction , and lower n(c iv ) and n(si iii ) . for @xmath224 and @xmath225 kpc , the predicted column densities are : @xmath226 } { 0.003 } \right ) \\ \nonumber \\ { \rm n(c~iv ) } & = & ( 1.8^{+0.7}_{-0.5 } \times 10^{13}~{\rm cm}^{-2 } ) \left ( \frac { \rm [ c / h ] } { 0.003 } \right ) \ ; . \end{aligned}\ ] ] the topology of c iv and si iii differ , as seen in the excluded regions of figures 6 and 7 . for denser absorbers , the ionization equilibrium shifts to lower ionization parameter ( @xmath222 ) in which case c iv / h i decreases ( fig . 7 ) while si iii / h i remains about the same ( fig . 5 ) . for si iii , the ionization solutions can be double - valued , since si iii changes more strongly with @xmath222 than h i ( donahue & shull 1991 ) . at sufficiently low and sufficiently high ionization parameter , silicon lies in other states ( si ii or si iv ) . as a result , the h i absorption toward pks 2155 - 304 could still contain metals at 0.01 solar and arising either in a homogeneous slab , of depth @xmath227200 - 400 kpc , or in denser parcels such as 1040 kpc halos of dwarf galaxies or intergalactic clouds . thus , it is easier to hide metals from detection in the c iv lines than in si iii . the observed limit on si iii absorption with n(h i ) = @xmath17 @xmath6 and 0.003 metallicity formally allows solutions with @xmath228 kpc or @xmath229 kpc . limits on c iv absorption allow @xmath230 kpc . the combined c iv and si iii limits , together with the numerous discrete ly@xmath0 absorbers in velocity space , suggest that the gas may be inhomogeneous on scales of tens of kpc . however , it is puzzling how clumped absorbers could maintain a high density contrast when these clumps would be expected to collide on a billion - year time scale . such collisions would also shock - heat the gas possibly making c iv lines more detectable . filamentary sheets , with large aspect ratios , provide a more plausible scenario consistent with these constraints . before concluding this section , it is worth examining the accuracy of the measurements of n(h i ) . our assumed value , n(h i ) @xmath148 @xmath6 , comes from the claimed detection of a lyc depression in _ orfeus _ data ( appenzeller et al . these data suggest a flux discontinuity shortward of 970 , corresponding to the @xmath231 ly@xmath0 absorbers , but the lyman break is not definitive , owing to the moderate ( 0.5 ) spectral resolution , low signal - to - noise ( s / n = 15 ) , and uncertainties in defining and extrapolating the true continuum between 920 and 1100 . for similar reasons , these data are not sufficiently accurate to determine n(h i ) from a curve - of - growth analysis of the higher lyman series , ly@xmath232 ly@xmath233 . because of ly@xmath0 line saturation in our hst / ghrs data , we can not confirm the large h i columns implied by the lyc absorption . however , we can provide both a firm lower limit and give reasonable estimates from the strongest ly@xmath0 features at 12841285 . the optical depth of the ly@xmath0 line per unit velocity is @xmath234 , where @xmath235 is the h i column density per unit velocity . we may integrate this , for @xmath236 and @xmath237 , to obtain , @xmath238 where @xmath239 is measured in km s@xmath2 . in our post - costar data , with s / n @xmath56 , this formula provides only a minimum column density , since we can not distinguish optical depths @xmath240 . flux calibration and background subtraction are difficult with the ghrs one - dimensional detectors . an integration of the data ( fig . 8) yields values @xmath241 ranging from @xmath242 @xmath6 . values of @xmath17 @xmath6 needed for consistency with the _ orfeus _ lyc measurement would require unresolved narrow components in the line core . although we have some skepticism about adopting the ( @xmath17 @xmath6 ) column density from _ orfeus _ , it is worth remembering that line saturation and multiple velocity components make column density measurement extremely difficult from just one line . a prime example of this difficulty comes from the ly@xmath0 absorbers seen toward 3c 273 at @xmath243 km s@xmath2 and @xmath244 km s@xmath2 ( weymann et al . recent _ orfeus _ measurements of these components ( hurwitz et al . 1998 ) find ly@xmath232 absorption equivalent widths larger by factors of 1.5 and 2.4 than values predicted by ly@xmath0 voigt profile fits of hst / ghrs data ( @xmath245 and @xmath246 km s@xmath2 for the 1000 km s@xmath2 cloud , and @xmath247 , @xmath248 km s@xmath2 for the 1600 km s@xmath2 cloud ) . by analyzing both ly@xmath0 and ly@xmath232 , hurwitz et al . ( 1998 ) derive a best - fit of @xmath249 and @xmath250 km s@xmath2 . they conclude that , if the h i absorption arises in a single cloud , its column density is higher by at least a factor of 4 compared to the value of weymann et al . ( 1995 ) . thus , it could be that the strong , saturated ly@xmath0 lines towards pks 2155 - 304 have columns well above @xmath251 @xmath6 and even as high as @xmath17 @xmath6 . this could arise if narrow h i components are hidden within the line core at @xmath8 km s@xmath2 . for example , 20 components , each with @xmath252 km s@xmath2and n(h i ) @xmath253 @xmath6 , spread stochastically over @xmath38 km s@xmath2 would reproduce the ly@xmath0 and lyc data . each component would have central optical depth @xmath254 and an ( isolated ) equivalent width of 240 m . saturation and velocity overlap would allow their accumulated ly@xmath0 absorption to give the observed @xmath255 , while the lyc optical depth could reach values @xmath256 . without better data on the lyc absorption edges or the higher lyman series lines , we can not verify the ly@xmath0 absorption ( @xmath17 @xmath6 ) in the strong absorber at @xmath257 km s@xmath2 . confirmation of the lyman limit and higher lyman series will await our planned studies of pks 2155 - 304 following the scheduled feb . 1999 launch of the far ultraviolet spectroscopic explorer ( fuse ) . with fuse , if n(h i ) is as large as @xmath17 @xmath6 , the lyman edge at 963.36 will have @xmath258 , and many higher lyman lines will be detectable . even if we adopt the minimum values , n(h i ) @xmath259 @xmath6 and @xmath260 in the line core , we should be able to detect higher lyman lines with @xmath261 , @xmath262 , and @xmath263 . it would also be helpful to obtain a better ly@xmath0 spectrum of the 1270 - 1290 region with hst / stis , whose two - dimensional array detectors will improve the background subtraction and flux calibration . the metal - line searches can also be improved with hst , using stis and cos ( see 3.4 ) . if n(h i ) is smaller than the value ( @xmath17 @xmath6 ) assumed in figures 5 and 6 , our limits on metallicity increase . for homogeneous clouds , of constant density , several quantities follow from simple scaling relations of n(h i ) . first , our inferred value of d / h scales inversely with n(h i ) , and would therefore go up . second , one would find a tradeoff of cloud depth ( @xmath152 ) and metallicity ( [ c / h ] or [ si / h ] ) . if n(h i ) were decreased by a factor 10 , the metallicity limits would rise by the same factor . if @xmath152 is held constant , a simple scaling of the metallicity with n(h i ) is not possible . as we discussed earlier , the relative ionization fractions of h i , c iv , and si iii all vary with ionization parameter , @xmath222 . if we retain our estimate of the radiation background and assume @xmath147 kpc , reducing n(h i ) by a factor of 10 changes the metallicity limits to c / h@xmath264 and si / h@xmath265 times solar . for the range in ionization parameter @xmath266 appropriate to these assumptions about cloud density and radiation field , the greater sensitivity to @xmath222 of the si iii fraction causes the si iii constraints on metallicity to weaken much more rapidly with decreasing n(h i ) . regardless of the exact value of n(h i ) , the most sensitive metal lines remain upper limits . therefore , one must take seiously the possibility that these clouds are primordial . a similar analysis can be done using the strong absorbers to search for weak deuterium ly@xmath0 ( 1215.339 ) features in the shortward wings of h i ly@xmath0 ( 1215.667 ) . no deuterium feature is seen in the wings of the h i system 1 ( h i at 1281.393 , d i at 1281.044 ) to a level of @xmath267 m . in system 2 ( h i at 1284.484 , d i at 1284.133 ) the limit is @xmath268 m . for a linear curve of growth , these equivalent widths translate to n(d i ) @xmath269 @xmath6 and @xmath270 @xmath6 , respectively . if we attribute column densities n(h i ) of @xmath271 @xmath6 and @xmath17 @xmath6 to systems 1 and 2 , respectively , these limits give a deuterium abundance limits of , @xmath272 \ ; \ ; { \rm ( system~1 ) } \\ \nonumber \\ \left ( \frac { \rm d}{\rm h } \right ) \leq ( 2.8 \times 10^{-4 } ) \left [ \frac { 2 \times 10^{16}~{\rm cm}^{-2 } } { \rm n(h~i ) } \right ] \ ; \ ; { \rm ( system~2 ) } \end{aligned}\ ] ] these limits can be improved significantly with higher - precision data on the d i line ( with hst / stis ) and with better h i column densities from higher lyman - series lines ( with the fuse spectrograph ) . at the current level of accuracy , neither limit provides a strong constraint on big bang nucleosynthesis ( schramm & turner 1998 ) . the interstellar medium value has been reported in the range d / h @xmath273 ( linsky et al . 1993 , 1995 ) . however , a controversy still exists over the high - redshift d / h observed in qso absorption systems . a recent determination of d / h in two qso lyman - limit systems ( burles & tytler 1997 ) gives a `` low value '' , d / h @xmath274 , while songaila and cowie ( cf . songaila 1997 ) quote `` high values '' with a range d / h @xmath275 . the high values of d / h would obviously require a large destruction rate of deuterium through star formation . thus , it would be helpful to obtain a detection of deuterium in the strong ly@xmath0 systems at @xmath202 toward pks 2155 - 304 . here , the measured metallicity is much less than solar values , implying little astration of d. we expect that future hst and fuse observations can make a factor of 3 improvement in sensitivity , which could set a limit on d / h @xmath276 . our photoionization models demonstrate that , for clouds with @xmath277 solar metallicity , it should be possible , with better hst / stis data , to detect the 1548.2 and 1550.8 resonance lines of c iv . we may also be able to detect c ii @xmath278 , n v @xmath279 , si iv @xmath280 , and si iii @xmath11 . with the fuse satellite , we intend to search for c iii @xmath205 , c ii @xmath281 , and o vi @xmath282 . the c iii line should be detectable down to well below 0.001 solar metallicity . the c ii lines might be present if the gas is at higher density ( lower @xmath222 ) . the o vi line is weaker in the standard model ( @xmath147 kpc , @xmath150 ) , but it might be enhanced by a number of effects : a hard photoionizing radiation field above 114 ev , a collisionally ionized component due to hot gas , or oxygen abundance enhancement by massive - star nucleosynthesis . we probably will not detect o i @xmath283 , which is weak , even at 3% solar metallicity , because the o i abundance is tied by charge exchange to the ratios ( h ii / h i ) and ( o ii / o ) . because of the o vi ionization correction , the strength of the @xmath284 absorption line depends both on the shape of the ionizing spectrum and on the cloud depth . for example , if @xmath285 ( @xmath147 kpc ) the o vi column density increases by over a factor of 4 , while if @xmath150 ( @xmath286 kpc ) it increases by a factor of 2 . the _ total _ h i column in the absorbers could be higher than the assumed value , @xmath17 @xmath6 . in addition , the si and o abundances could be enhanced ( songaila & cowie 1996 ; giroux & shull 1997 ) relative to c in regions dominated by `` prompt nucleosynthesis '' from massive - star supernovae . thus , if si / c @xmath287 and o / c @xmath288 , we may be able to detect lines of o vi , si ii , si iii , and si iv . as shown by donahue & shull ( 1991 ) , the ratios of column densities from multiple ion states of the same element ( e.g. , si ii / iii / iv and c ii / iv ) can be modeled to provide diagnostics of the intensity and spectrum of the 14 ryd ionizing radiation field . detections of these ions would allow us to discriminate between an extragalactic radiation field due to agn or due to o - stars from starburst galaxies . the sightline to pks 2155 - 304 is unique among the agn studied thus far for low - redshift ly@xmath0 absorbers . it has the highest frequency of absorbers , and they are identified with a concentration of bright galaxies . the ly@xmath0 absorbers at @xmath29 km s@xmath2 may be an extreme example of the previously known association of ly@xmath0 clouds with the extended halos of galaxies ( lanzetta et al . 1995 ; stocke et al . 1995 ) . of greater interest is the enormous inferred gaseous extent ; the nearest bright galaxies lie @xmath9 off the sightline . these offsets are so large that they are unlikely to represent equilibrium gaseous disks ; the orbital times at such distances would be enormous . we have interpreted these absorbers as large sheets of intragroup gas , or as smaller primordial clouds and halos of dwarf galaxies . the clumpiness of the ly@xmath0 absorption in velocity space suggests that some spatial structure is present . however , as discussed in 3.1 , a medium that is clumpy in both space and velocity would be subject to disruptive collisions . * the metallicity limits for the ly@xmath0 absorbers give ( si / h ) and ( c / h ) less than 0.003 solar if n(h i ) = @xmath289 @xmath6 ( the value from orfeus ) . the greatest uncertainty in these metallicities is the precise value of n(h i ) . we will use the fuse satellite to measure the h i columns through lyman limit absorption and higher lyman series lines . we hope to see discrete absorption edges at 961.04 , 963.36 , and 963.82 . * our metallicity limit , [ si / h ] @xmath290 solar contradicts the hypothesis that these clouds are metal - enriched intragroup gas . mushotzky & loewenstein ( 1997 ) suggest that intracluster gas might be enriched early ( at @xmath24 ) to levels of 10% solar metallicity . evidently , the intergalactic gas around pks 2155 - 304 has not been enriched to the levels observed in x - ray emitting intracluster gas . * if the `` small group '' hypothesis is correct , a search for x - rays with axaf imagers would test whether any hot gas has been produced by tidal effects or stripping . * future spectral observations with hst can provide even better limits on d / h , si iii , and c iv than those reported here . with fuse , we will be able to search for the expected strong c iii @xmath291 line , the weak o vi @xmath282 , as well as absorption lines of s ii , c ii , s ii , and n ii which may be present if the absorption occurs in denser halos . * the low metallicity limits and the large projected distances from the nearest galaxies suggest that these clouds could be primordial . this work was based on observations with the nasa / esa _ hubble space telescope _ obtained at the space telescope science institute , which is operated by aura , inc . under nasa contract nas5 - 26555 and on observations made with nrao s very large array . the nrao is operated by associated universities , inc . under a cooperative agreement with the national science foundation . we thank michael fall and richard mushotzky for helpful discussions on the evolution of intracluster gas . this work was supported by hst guest observer grant go-06593.01 - 95a and by the astrophysical theory program ( nasa grant nagw-766 and nsf grant ast96 - 17073 ) to the university of colorado and by an nsf grant ( ast96 - 17177 ) to columbia university . @xmath292 column densities are computed for cloud with n@xmath293 @xmath6 , @xmath147 kpc , and 0.003 solar metallicity , irradiated by qso ionizing spectrum with @xmath150 and @xmath149 ergs @xmath6 s@xmath2 hz@xmath2 sr@xmath2 . observed wavelengths correspond to ly@xmath0 components at 1281.393 , 1284.484 , and 1285.097 , at redshifts @xmath294 , @xmath198 , and @xmath200 . equivalent widths are quoted in the rest frame .

let be a connected oriented graph . a set is convex in if , for every pair of vertices , the vertex set of every-geodesic , ( shortest directed path ) and every-geodesic in is contained in . the convexity number , , of a non - trivial oriented graph , , is the maximum cardinality of a proper convex set of . the strong convexity spectrum of the graph , , is the set . in this paper we prove that the problem of determining the convexity number of an oriented graph is-complete , even for bipartite oriented graphs of arbitrary large girth , extending previous known results for graphs . we also determine , for every pair of integers . convexity number , convex set , spectrum , oriented graph , grid 05c 12 , 05c20

1703.02654

graphs considered in the paper are finite , without loops or multiple edges . in a graph @xmath13 , @xmath14 and @xmath15 ( @xmath16 and @xmath17 ) denote the vertex set and the edge set of @xmath7 , respectively . for undefined concepts and notation we refer the reader to @xcite . for two vertices @xmath18 and @xmath19 in a graph @xmath7 , a @xmath20-geodesic is a shortest path between @xmath18 and @xmath19 . a set @xmath5 of vertices of @xmath7 is convex if the vertices of every @xmath20-geodesic is contained in @xmath5 for every @xmath21 . according to duchet , convexity in graphs has been studied since the early seventies , when abstract convexity was studied in different contexts ( @xcite is an outdated , but very nice , survey on the subject ) . convexity in graphs has taken many different directions , and different related parameters have been defined and widely studied , e.g. , the hull number @xcite , the geodetic number @xcite , and the convexity number @xcite of a graph . recent papers on this subjects include @xcite , where the decision problem associated with these three parameters are shown to be @xmath10-complete , even when restricted to bipartite graphs , and in the case of the geodetic number , even when restricted to bipartite chordal graphs . chartrand , fink and zhang generalized the concept of convexity to oriented graphs , and defined the convexity number for an oriented graph ; oriented analogues of the hull number and geodetic number are defined in @xcite . we focus on the convexity number of oriented graphs ; although this generalization was introduced in 2002 , and the proof given by gimbel in @xcite on the @xmath10-completeness of determining the convexity number of an arbitrary graph is one of the shortest and neatest @xmath10-completeness proofs ever done , the problem of determining the convexity number of an oriented graph was not known to be @xmath10-complete until now . we prove that determining the convexity number of an oriented graph is @xmath10-complete even when restricted to bipartite graphs of girth @xmath22 , with @xmath23 . an _ oriented graph _ is an orientation of some graph . in an oriented graph @xmath24 , @xmath14 and @xmath15 ( @xmath25 and @xmath26 ) denote the vertex set and the edge set of @xmath0 , respectively . an _ oriented subgraph _ @xmath27 of an oriented graph @xmath24 is an oriented graph with @xmath28 and @xmath29 . an oriented graph is _ connected _ if its underlying graph is connected . a _ directed path _ is a sequence @xmath30 of vertices of an oriented graph @xmath0 such that @xmath31 are distinct and @xmath32 for @xmath33 . an oriented graph is _ strongly connected _ ( or _ strong _ ) if for every pair of distinct vertices @xmath18 and @xmath19 , there exists a directed path from @xmath18 to @xmath19 . the _ girth _ of an oriented graph is the length of a shortest directed cycle . a @xmath20-_geodesic _ in a digraph @xmath0 is a shortest @xmath20-directed path and its length is @xmath34 . a nonempty subset , @xmath5 , of the vertex set of a digraph , @xmath0 , is called a _ convex set _ of @xmath0 if , for every @xmath21 , every vertex lying on a @xmath20- or @xmath35-geodesic belongs to @xmath5 . for a nonempty subset , @xmath36 , of @xmath25 , the _ convex hull _ , @xmath37 $ ] , is the minimal convex set containing @xmath36 . thus @xmath38 = s$ ] if and only if @xmath5 is convex in @xmath0 . the _ convexity number _ , @xmath6 , of a digraph @xmath0 is the maximum cardinality of a proper convex set of @xmath0 . a _ maximum convex set _ @xmath5 , of a digraph @xmath0 , is a convex set with cardinality @xmath6 . since every singleton vertex set is convex in a connected oriented graph @xmath0 , @xmath39 . the _ degree _ , @xmath40 , of a vertex @xmath19 in an oriented graph is the sum of its in - degree and out - degree ; this is , @xmath41 . a vertex , @xmath19 , is an _ end - vertex _ if @xmath42 . a _ source _ is a vertex having positive out - degree and in - degree @xmath43 , while a _ sink _ is a vertex having positive in - degree and out - degree @xmath43 . for a vertex @xmath19 of @xmath0 , the in - neighborhood of @xmath19 , @xmath44 , is the set @xmath45 and the out - neighborhood of @xmath19 , @xmath46 , is the set @xmath47 . a vertex @xmath19 of @xmath0 is a _ transitive vertex _ if @xmath48 , @xmath49 and , for every @xmath50 and @xmath51 , @xmath52 . for graphs @xmath7 and @xmath53 , their _ cartesian product _ , @xmath54 , is the graph with vertex set @xmath55 , and such that two vertices @xmath56 and @xmath57 are adjacent in @xmath54 if either @xmath58 and @xmath59 is an edge in @xmath53 , or @xmath60 and @xmath61 is an edge in @xmath7 . for a vertex @xmath22 of @xmath7 , the subgraph of @xmath54 induced by the set @xmath62 is called an @xmath53-fiber and is denoted by @xmath63 . similarly , for @xmath64 , the @xmath7-fiber , @xmath65 , is the subgraph induced by @xmath66 . we will have occasion to use the fiber notation @xmath65 and @xmath63 to refer instead to the set of vertices in these subgraphs ; the meaning will be clear from the context . it is clear that all @xmath7-fibers are isomorphic to @xmath7 and all @xmath53-fibers are isomorphic to @xmath53 . as mentioned , the concept of convexity number of an oriented graph was first introduced by chartrand , fink and zhang in @xcite , where they proved the following pair of theorems . let @xmath0 be a connected oriented graph of order @xmath67 . then @xmath68 if and only if @xmath0 contains a source , a sink or a transitive vertex . [ no2 ] there is no connected graph of order at least @xmath69 with convexity number @xmath70 . taking an interesting direction for the subject of convexity in oriented graphs , in @xcite , tong , yen and farrugia introduced the concepts of convexity spectrum and strong convexity spectrum of a graph . for a nontrivial connected graph @xmath7 , we define the _ convexity spectrum _ , @xmath71 , of a graph @xmath7 , as the set of convexity numbers of all orientations of @xmath7 , and the _ strong convexity spectrum _ , @xmath72 , of a graph @xmath7 as the set of convexity numbers of all strongly connected orientations of @xmath7 . if @xmath7 has no strongly connected orientation , then @xmath72 is empty . the _ lower orientable convexity number _ , @xmath73 , of @xmath7 is defined to be @xmath74 and the _ upper orientable convexity number _ , @xmath75 , is defined to be @xmath76 . hence , for every nontrivial connected graph @xmath7 of order @xmath77 , @xmath78 . tong , yen and farrugia calculated the convexity and strong convexity spectra of complete graphs and also constructed , for every @xmath79 , a graph @xmath7 with convexity spectrum @xmath80 . it is not very surprising that the strong convexity spectra of @xmath81 for @xmath82 is a `` large '' set , @xmath83 , missing only @xmath84 and @xmath85 . nonetheless , we find very surprising that , in one hand , tong and yen proved in @xcite that @xmath86 for every pair of integers @xmath87 , and , in the other hand , we prove that the strong convexity spectrum of an @xmath88 grid , for any pair of integers @xmath89 , only lacks the set of integers @xmath90 . so , an interesting question arises from the previous observation : what property in a graph determines a large strong convexity spectrum ? we can discard regularity and high degrees ; grids are not regular graphs , and have both small maximum and minimum degree . although we did not find what is so special about grids in terms of convexity , we managed to calculate the strong convexity spectra of all grids . the rest of this paper is ordered as follows . section [ snp ] is devoted to prove the @xmath10-completeness of the problem of determining the convexity number of a given oriented graph ; the problem remains @xmath10-complete even when restricted to bipartite oriented graphs of arbitrarily large girth . in section [ sgrids ] , we prove some basic results on the convexity number of general oriented graphs , and also introduce a concept of main importance to this work : the whirlpool orientation of a grid . using this concept , we prove that @xmath91 for any grid @xmath7 . we finish the section with a result excluding some values from the strong convexity spectra of certain grids . section [ scssg ] is devoted to calculate the strong convexity spectra of @xmath92 and @xmath93 grids for every integer @xmath67 . in section [ smain ] , the strong convexity spectra of @xmath88 grids for every pair of integers @xmath94 is calculated . we define the problem oriented convexity number as follows . given an ordered pair @xmath95 , consisting of an oriented graph @xmath0 and a positive integer @xmath96 , determine whether @xmath0 has a convex set of size at least @xmath96 . this first section is devoted to prove the @xmath10-completeness of oriented convexity number . [ redu ] oriented convexity number restricted to bipartite oriented graphs of girth @xmath97 is @xmath10-complete . let @xmath0 be an oriented graph . given a subset @xmath98 , it can be verified in polynomial time whether @xmath99 is a convex set . hence , oriented convexity number is in @xmath10 . in order to prove @xmath10-hardness ( and hence @xmath10-completness ) , we reduce an instance @xmath100 of the well - known @xmath10-complete problem clique to an instance @xmath101 of oriented convexity number such that @xmath102 if and only if @xmath103 , the encoding length of @xmath101 is polinomially bounded in terms of the encoding length of @xmath100 , and @xmath0 is bipartite . let @xmath100 be an instance of clique . let us assume that @xmath104 and @xmath7 is connected ; we construct @xmath0 as follows . for every vertex @xmath18 of @xmath7 we create a directed hexagon , @xmath105 , with two antipodal distinguished vertices @xmath106 and @xmath107 . for every edge @xmath108 we add the arcs @xmath109 and @xmath110 to @xmath0 . we create four additional vertices @xmath111 with arcs @xmath112 for @xmath113 and arcs @xmath114 for every @xmath115 . of theorem [ redu ] when @xmath7 is @xmath116 . ] clearly , @xmath117 and @xmath118 . it is direct to observe that the digraph @xmath0 is bipartite and strongly connected . in figure [ redufig ] the classes of the bipartition are given by the vertices of the same color ( black and white ) . we also claim the following statements to hold . [ c1 ] let @xmath18 be a vertex in @xmath7 and let @xmath99 be a convex set of @xmath0 with @xmath119 . if @xmath120 , then @xmath121 . [ c2 ] let @xmath99 be a convex set of @xmath0 with @xmath119 . if @xmath122 for @xmath123 , then @xmath124 . [ c3 ] let @xmath125 be such that @xmath126 . if @xmath99 is a convex set of @xmath0 such that @xmath127 , then @xmath128 . [ c4 ] if @xmath5 is a clique of @xmath7 , then @xmath129 is a convex set of @xmath0 . it follows from claims [ c1]-[c4 ] that @xmath99 is a proper convex set of @xmath0 with @xmath119 , if and only if there exists a clique @xmath5 in @xmath7 such that @xmath129 . considering a maximum clique of @xmath7 , and a maximum proper convex set of @xmath0 , we obtain @xmath130 . hence , if @xmath131 , then @xmath7 contains a clique of size at least @xmath96 if and only if @xmath0 contains a convex set of size at least @xmath132 . since the encoding length of @xmath101 is linearly bounded in terms of the encoding length of @xmath100 , oriented convexity number is @xmath10-complete . we will consider two cases . first , assume that @xmath133 . let @xmath134 and @xmath135 be vertices in @xmath136 . clearly , @xmath137 is a @xmath138-geodesic in @xmath0 and @xmath139 is a @xmath140-geodesic in @xmath0 . hence , @xmath121 . for the second case suppose that @xmath141 and @xmath142 . therefore , every @xmath138-directed path in @xmath0 uses the vertex @xmath106 , and every @xmath140-directed path in @xmath0 uses the vertex @xmath107 . thus , @xmath133 and we are back to the first case . suppose first that @xmath143 with @xmath144 . assume without loss of generality that @xmath145 . every @xmath146-directed path in @xmath0 uses the vertices @xmath147 and @xmath148 , thus @xmath149 . for every @xmath115 , @xmath150 is a @xmath151-geodesic in @xmath0 . we conclude from claim [ c1 ] that @xmath121 for every @xmath115 , and it follows that @xmath152 . if @xmath122 for @xmath123 and @xmath153 for some @xmath154 , then every @xmath155-directed path in @xmath0 uses the vertex @xmath147 and every @xmath156-directed path in @xmath0 uses the vertex @xmath148 . thus , @xmath149 and we are done . it follows from claim [ c1 ] that @xmath157 . since @xmath158 , it is easy to observe that @xmath159 is an @xmath160-geodesic in @xmath0 . hence , @xmath161 and claim [ c2 ] guarantees @xmath128 . it is direct to verify that @xmath162 is a convex set of @xmath0 for every @xmath115 . let @xmath163 and @xmath164 . if @xmath141 and @xmath165 , then every @xmath166-directed path in @xmath0 uses the vertices @xmath106 and @xmath167 ; moreover , it must contain an @xmath168-directed path . our claim follows from noting that @xmath169 . it is easy to observe that the directed @xmath97-cycles in the previous construction can be replaced by directed @xmath170-cycles , and the directed path @xmath171 can be replaced by a directed path of length @xmath77 , in order to get the result for an oriented bipartite graph of arbitrary large girth . we begin this section with two straightforward lemmas regarding convex sets in strong oriented graphs . the first lemma is a direct observation , so the proof will be omitted . [ girthcon ] let @xmath0 be a strongly connected oriented graph . if @xmath98 is a convex set such that @xmath119 , then @xmath99 induces a strong subdigraph of @xmath0 therefore , @xmath172 , where @xmath173 stands for the girth of @xmath7 . [ comcon ] let @xmath0 be an oriented graph . if @xmath98 is a maximal convex set , then @xmath174 is a connected subdigraph of @xmath0 . otherwise , let @xmath175 be the connected components of @xmath176 . since @xmath99 is a convex set of @xmath0 , @xmath177 is a convex set of @xmath0 properly containing @xmath99 , a contradiction . the following observation is simple , but also very useful while searching for adequate orientations to realize specific convex numbers . the proof is straightforward and thus will be omitted . [ ai - ao ] let @xmath7 be a triangle - free graph and let @xmath0 be an orientation of @xmath7 with a convex set @xmath99 . let @xmath178 be adjacent to a vertex @xmath179 . * if @xmath180 , then @xmath181 . * if @xmath182 , then @xmath183 . let @xmath77 be an integer . we denote by @xmath184 the path on @xmath77 vertices , and we will assume without loss of generality that @xmath185 . the @xmath186-_grid _ is the cartesian product @xmath187 . hence , if @xmath188 , then @xmath189 . although it is not standard , and it can be impractical in a different context , for the sake of simplicity we will denote the ordered pair @xmath190 as @xmath191 . also , we will use the canonical embedding of the grid @xmath188 in the plane to define orientations of @xmath7 , and directed paths in an orientation @xmath0 of @xmath7 . to achieve this goal , we will use directions to `` move '' on the grid , denoted as a sequence of movements using the symbols @xmath192 , which stand for up , down , left and right . as an example , consider a directed cycle denoted in the usual way , _ i.e. _ , @xmath193 ; with our notation we have the sequence @xmath194 , starting at vertex @xmath191 . in the following paragraph , there is another example of the orientations that can be defined in this way . let @xmath195 be integers and @xmath7 be the canonical plane embedding of the grid @xmath187 . let @xmath196 be a connected subgraph of the interior dual of @xmath7 and let @xmath53 be the subgraph of @xmath7 induced by the faces in @xmath197 ( hence , the interior dual of @xmath53 is @xmath196 ) . we define a _ whirlpool _ to be an oriented graph obtained from @xmath53 by the following orientation of its edges . @xmath198 an example of a whirlpool is depicted in figure [ whirlpool ] , where the graph @xmath196 has the gray faces of the grid as its vertex set . in the rest of the figures , the gray squares will always correspond to directed cycles . as the following proposition shows , whirlpools have a very important property related to convexity . [ lcon1grids ] if @xmath0 is a whirlpool , then @xmath199 . let @xmath0 be a whirlpool . hence , there exist a pair of integers @xmath195 and a subgraph @xmath53 of the grid @xmath187 such that @xmath0 is obtained from @xmath53 by the aforementioned orientation of its edges . we affirm that every @xmath69-cycle in @xmath53 is an oriented @xmath69-cycle in @xmath0 . if @xmath99 is a @xmath69-cycle in @xmath53 , then @xmath200 for some @xmath201 . it is not difficult to observe that if @xmath202 , then @xmath99 is an oriented cycle in @xmath0 . else , @xmath203 ( the cycle @xmath99 in reverse order ) is an oriented cycle in @xmath0 . hence , @xmath0 is strongly connected . since every edge of @xmath53 belongs to a @xmath69-cycle , it follows that every arc of @xmath0 belongs to a directed @xmath69-cycle . recalling that @xmath0 is triangle free , it is clear that @xmath204 implies @xmath205 . hence , if @xmath204 , then the vertex set of every @xmath69-cycle containing the arc @xmath206 is contained in the convex hull of @xmath207 . but @xmath0 is strongly connected , so using the fact that the interior dual of @xmath53 is connected , it can be shown inductively that for every pair @xmath208 of vertices of @xmath0 , the convex hull of @xmath209 is @xmath25 . therefore , @xmath199 . [ con=1 ] for every grid @xmath7 , @xmath210 . observe that if an oriented grid @xmath0 contains a whirlpool @xmath211 as a subdigraph , then every convex set containing at least two adjacent vertices of @xmath211 must contain @xmath212 . also , if @xmath211 is a whirlpool , then the digraph @xmath213 obtained by the reversal of every arc of @xmath211 has the same properties as @xmath211 ; we will call such a digraph an _ anti - whirlpool_. for a given integer @xmath96 and an oriented graph @xmath0 , it is easier to prove @xmath214 than proving @xmath215 . the following result excludes some values from the strong convexity spectra of grids . although simple , the complete proof of the lemma is to long to be included here . the proof of the first item of the lemma is complete , as well as the cases @xmath216 and @xmath217 of the second item . the proofs for the cases @xmath218 and @xmath219 can be obtained with similar arguments . [ forbid ] let @xmath195 be integers . if @xmath188 , then : * @xmath220 . * for every @xmath221 , if @xmath222 , then @xmath223 . by theorem [ no2 ] we have @xmath224 . observing that every connected subdigraph of @xmath7 with @xmath225 or @xmath226 vertices has at least one vertex of degree @xmath227 , and thus does not admit a strong orientation , it follows from lemma [ girthcon ] that @xmath228 . any strong orientation of @xmath7 has neither sinks nor sources . also , @xmath229 and hence the orientations of @xmath7 can not have transitive vertices . it follows that @xmath230 . let @xmath231 be integers such that @xmath232 and suppose that a strong orientation @xmath0 of @xmath7 has a convex set @xmath99 of cardinality @xmath233 . let @xmath234 be the set @xmath235 and assume without loss of generality that @xmath180 . since @xmath0 is strong , @xmath236 and @xmath237 . it follows from observation [ ai - ao ] that @xmath238 and @xmath239 . let us denote by @xmath240 the vertices above , below , to the left and to the right of @xmath241 , respectively , in g. since @xmath232 , we can assume without loss of generality that either @xmath242 or @xmath243 is a directed path in @xmath0 , and hence a geodesic . but this contradicts that @xmath99 is a convex set . hence , a convex set of cardinality @xmath233 can not exist . let @xmath231 be integers such that @xmath244 and suppose that a strong orientation @xmath0 of @xmath7 has a convex set @xmath99 of cardinality @xmath245 . let @xmath246 be the set @xmath235 . let us assume without loss of generality that @xmath180 and @xmath247 . we have two cases . first consider @xmath248 . again , we have two cases . our first subcase is @xmath249 . as in the previous argument , either @xmath243 or @xmath250 is a geodesic in @xmath0 , contradicting that @xmath99 is a convex set . our second subcase is @xmath251 . we will assume without loss of generality that @xmath252 . hence , either @xmath253 or @xmath254 is a geodesic in @xmath0 , contradicting that @xmath99 is a convex set . as a second case , consider @xmath255 , with two subcases . our first subcase is @xmath249 . either @xmath256 or @xmath257 is a geodesic in @xmath0 , a contradiction . the second subcase is @xmath251 . hence , at least one of the following directed paths is present in @xmath0 , and it is a geodesic : @xmath258 . but every case results in a contradiction . hence , a convex set of cardnality @xmath245 can not exist . the following pair of results deal with the convexity spectra of @xmath92 grids . [ 2*n - part1 ] let @xmath67 be an integer and let @xmath7 be the grid @xmath259 . if @xmath260 is an integer such that @xmath261 and @xmath262 , then @xmath263 . let @xmath264 be the orientation of @xmath265 as a whirlpool . we will consider two cases . first , suppose that @xmath266 . let @xmath267 be the orientation of @xmath268 as a whirlpool , if @xmath260 is odd , or as anti - whirlpool if @xmath260 is even . in either case , the orientation of the edges @xmath269 and @xmath270 result in parallel arcs , _ i.e. _ , we have either the arcs @xmath271 and @xmath272 or the arcs @xmath273 and @xmath274 . we will assume that @xmath260 is odd , the remaining case can be dealt similarly . if @xmath0 is the digraph obtained by orienting the two remaining edges as @xmath275 and @xmath276 , then it is clear that @xmath277 and @xmath278 . but @xmath279 , and hence , @xmath280 is a convex set of cardinality @xmath281 . if @xmath99 is a convex set of @xmath259 such that @xmath282 , then @xmath283 . from the previous observations about @xmath284 and @xmath285 , we conclude that @xmath286 . since @xmath264 and @xmath267 are whirlpools , we obtain @xmath287 . therefore , @xmath280 is a maximum convex set of @xmath0 . if @xmath288 , then we will assume that @xmath77 is odd , otherwise we are back in the previous case . let @xmath267 be the orientation of @xmath289 $ ] as a whirlpool , if @xmath260 is odd , or as anti - whirlpool if @xmath260 is even . again , we will assume that @xmath260 is odd . orient the remaining edges in the following way : @xmath290 and @xmath291 . an argument similar to the one used in the previous case shows that @xmath292 and @xmath293 are convex sets of @xmath0 , and clearly @xmath294 . if @xmath99 is a convex set of @xmath259 such that @xmath282 , then @xmath295 . since @xmath296 and @xmath297 , then @xmath298 . but @xmath299 and @xmath300 are @xmath301-geodesics in @xmath0 . since @xmath302 and @xmath303 are whirlpools in @xmath0 , we obtain @xmath124 . hence , @xmath292 is a maximum convex set of @xmath0 . if @xmath67 is an integer and @xmath7 is the grid @xmath259 , then @xmath304 let @xmath0 be a strong orientation of @xmath7 and @xmath99 a maximum convex set of @xmath0 . it follows from lemma [ comcon ] and the fact that @xmath305 $ ] is strong that , for some @xmath306 , either @xmath307 , or @xmath308 . hence , there are not odd integers greater than @xmath227 in @xmath8 . assume without loss of generality that @xmath309 for some @xmath306 . also assume without loss of generality that @xmath310 . we will consider two cases . consider for the first case @xmath311 . since @xmath0 is strong , either @xmath312 or @xmath313 . in the former case @xmath314 is a @xmath315-geodesic in @xmath0 , a contradiction . in the latter case , it is direct to verify that @xmath316 is also a convex set ; since @xmath99 is maximum , we get @xmath266 . for the second case consider @xmath317 . again , since @xmath0 is strong , either @xmath312 or @xmath313 . in the former case it is easy to check that @xmath316 is a convex set of @xmath0 , hence , @xmath266 . in the latter case , if @xmath318 , then @xmath319 is a convex set of @xmath0 , a contradiction . hence @xmath320 and we have two further cases . if @xmath321 , then @xmath316 is a convex set of @xmath0 , thus @xmath266 . if @xmath322 , then @xmath323 is a convex set of @xmath0 , and hence @xmath324 . although a bit more complex , similar arguments can be used for the @xmath93 grids . [ 3*n - c=3k ] let @xmath67 be an integer and let @xmath7 be the grid @xmath325 . if @xmath260 is an integer such that @xmath306 , then @xmath326 . with @xmath327 and @xmath328 . ] we consider two cases . if @xmath329 , then let @xmath264 be the orientation of @xmath330 as a whirlpool . suppose that @xmath77 is even , the remaining case is analogous . orient the remaining edges as @xmath331 . it is straightforward to verify that this orientation @xmath0 of @xmath7 is strong and @xmath280 is a maximum convex set of @xmath0 . for @xmath332 , let @xmath264 be the orientation of @xmath333 as a whirlpool ( note that @xmath264 is empty for @xmath334 ) . let @xmath267 be the orientation of @xmath335 as a whirlpool . orient @xmath336 as a directed cycle . also , orient @xmath337 as @xmath338 if @xmath260 is even , and reverse each of these arcs if @xmath260 is odd . finally , orient the remaining arcs as @xmath339 . let @xmath0 be the resulting orientation of @xmath7 , and let @xmath340 be the induced subdigraph @xmath341 $ ] of @xmath0 . we will assume that @xmath260 is even , the remaining case can be dealt similarly . observe that @xmath342 , and @xmath343 . since @xmath344 , @xmath345 , @xmath346 , and @xmath347 , it is clear that @xmath348 is a convex set of @xmath0 with @xmath349 . let @xmath99 be a convex set of @xmath0 such that @xmath350 . since every convex set in a strong digraph induces a strong subdigraph , we observe that @xmath351 . but @xmath267 is a whirlpool , and hence @xmath352 . note that @xmath353 and also that @xmath354 is a directed path in @xmath0 . thus , @xmath355 . recall that @xmath346 and consider the directed path @xmath356 to conclude @xmath357 . since @xmath260 is even , @xmath358 is a directed cycle in @xmath0 . this implies @xmath359 , but @xmath264 is a whirlpool , and hence @xmath360 and @xmath128 . therefore @xmath361 . [ 3*n - c=3k+2 ] let @xmath67 be an integer and let @xmath7 be the grid @xmath325 . if @xmath260 is an integer such that @xmath332 , then @xmath362 . with @xmath327 and @xmath363 . ] let @xmath264 be the orientation of @xmath364 as a whirlpool . we will consider two cases . for the first case consider @xmath365 , and let @xmath267 be the orientation of @xmath366 as a whirlpool , if @xmath260 is even , or as an anti - whirlpool , if @xmath260 is odd . if @xmath260 is even , orient the remaining edges as @xmath367 to obtain the orientation @xmath0 of @xmath7 . it is direct to verify that @xmath0 is strong . observing that @xmath368 and @xmath369 it is easy to verify that @xmath280 is a convex set of @xmath0 . if @xmath99 is a convex set of @xmath0 such that @xmath370 , then @xmath371 . but @xmath372 , and hence @xmath351 . recalling that @xmath267 is a whirlpool and observing that @xmath373 is a @xmath374-geodesic in @xmath0 , we conclude that @xmath124 . hence @xmath375 . if @xmath260 is odd , orient the remaining edges as @xmath376 to obtain the orientation @xmath0 of @xmath7 . this orientation is , locally , the dual orientation of the case when @xmath260 is even , so analogous arguments show that @xmath280 is a convex set and @xmath375 . as a second case , assume that @xmath377 or @xmath378 . when @xmath260 is odd , orient @xmath379 as a directed path , and orient the arcs @xmath380 . if @xmath381 , let @xmath267 be the orientation of @xmath382 as an anti - whirlpool and orient the remaining edges of @xmath7 as @xmath383 to obtain @xmath0 . clearly @xmath0 is strong . also , it is direct to verify that @xmath280 is a convex set of @xmath0 with @xmath384 vertices . let @xmath99 be a convex set of @xmath0 such that @xmath370 . if @xmath351 , then @xmath352 . but @xmath385 is a @xmath386-geodesic in @xmath0 , and thus , @xmath387 for @xmath113 . also , @xmath388 is a @xmath389-geodesic in @xmath0 . this implies @xmath128 , because @xmath359 . otherwise , and because @xmath390 , @xmath360 and @xmath391 for some @xmath392 . in any case , @xmath379 is the union of a @xmath393-geodesic and a @xmath394-geodesic in @xmath0 ( and the case @xmath395 is finished ) . since @xmath396 is a @xmath397-geodesic in @xmath0 , we have @xmath351 and @xmath128 . when @xmath260 is even , as in the previous case , we can orient the remaining edges of @xmath7 to obtain , locally , an orientation that is dual to the orientation when @xmath260 is odd . hence , analogous arguments can be followed to prove that @xmath280 is a maximum convex set of @xmath0 . therefore , @xmath398 . [ 3*n - c=3k+1 ] let @xmath67 be an integer and let @xmath7 be the grid @xmath325 . if @xmath260 is an integer such that @xmath399 , then @xmath400 . for @xmath327 and @xmath401 ] we will assume that @xmath260 is odd , the remaining case can be dealt similarly . orient @xmath402 as a whirlpool to obtain @xmath403 . we will consider two cases . for the first case , suppose that @xmath378 . let @xmath264 be the digraph obtained from @xmath404 by orienting @xmath405 as @xmath403 , @xmath406 as a directed path , and the remaining edge of @xmath407 $ ] as @xmath408 . also , orient @xmath409 , and @xmath410 as directed paths . orient the remaining edge as @xmath411 to obtain the digraph @xmath0 . it is immediate to verify that @xmath0 is a strong digraph . note that @xmath412 and @xmath413 . observing that @xmath414 , @xmath415 , @xmath416 , and @xmath417 , it is easy to conclude that @xmath280 is a convex set of @xmath0 with @xmath384 vertices . let @xmath99 be a convex set of @xmath0 such that @xmath418 . since @xmath395 , then @xmath419 and hence @xmath420 . also , there is at least one vertex @xmath421 . regardless of the choice of @xmath19 , the directed path starting at @xmath422 and defined by the sequence @xmath423 results from the union of a @xmath424-geodesic and a @xmath425-geodesic . hence , @xmath426 and therefore @xmath427 as a second case , assume that @xmath428 . let @xmath264 and @xmath267 be the digraphs obtained by orienting both @xmath404 and @xmath429 as whirlpools . orient @xmath430 and @xmath431 as directed paths . if @xmath432 , orient @xmath433 , and orient the same edge as @xmath434 otherwise . finally , orient the remaining edges as @xmath411 to obtain the digraph @xmath0 . it is direct to verify that @xmath0 is strong . observe that @xmath435 and @xmath436 . noting that @xmath437 , @xmath438 and @xmath439 , it is not hard to verify that @xmath280 is a convex set of @xmath0 . let @xmath99 be a convex set of @xmath0 such that @xmath418 . if @xmath440 , then @xmath441 , and @xmath360 . if @xmath442 and @xmath351 , then @xmath352 . but @xmath443 is a @xmath444-geodesic in @xmath0 , which implies @xmath445 . the directed path with initial vertex @xmath446 and defined by the sequence @xmath447 is a @xmath448-geodesic in @xmath0 . from here we observe that @xmath441 and thus @xmath360 . if @xmath442 and @xmath449 , then @xmath441 , and @xmath360 . hence , in every case @xmath360 . since there is at least one vertex from @xmath450 in @xmath99 , necessarily @xmath451 . the directed path starting at @xmath452 and defined by the sequence @xmath453 is a @xmath454-geodesic , and hence @xmath352 . but if @xmath455 , it is easy to verify that @xmath124 . hence , @xmath427 . so far , we have every integer of the convexity spectrum of @xmath325 , except for @xmath69 . our next theorem deals with the remaining case . [ n*3-c=4 ] if @xmath456 is an integer and @xmath7 is the grid @xmath325 , then @xmath457 for @xmath458 . ] consider the standard plane embedding of @xmath7 and color the interior faces gray and white with a checkerboard - like pattern , coloring the square on the bottom left corner with gray . we will define an orientation of the arcs of @xmath7 using this coloring , an example can be seen on figure [ n*3-c=4fig ] . there are two rows of squares . enumerate the gray squares in each row from left to right . orient the bottom left corner square as a whirlpool and , from here , orient all the gray squares in its row alternating whirlpool and anti - whirlpool orientations . orient all the gray squares in the upper row following the same principle , but start orienting as an anti - whirlpool the first gray square . at this point , every arc dividing two interior faces of @xmath7 has received an orientation . every remaining unoriented edge @xmath459 of @xmath7 divides a white square from the exterior face of @xmath7 . thus , the edge @xmath459 lies in exactly one square of @xmath7 , and has one parallel arc @xmath460 in the square . it @xmath459 is not an edge of a corner square , orient it in the same direction as @xmath460 . there are four edges belonging to the white corner squares that remain unoriented . orient the remaining edges as @xmath70-paths in such way that there are not white oriented squares . let @xmath0 be the digraph obtained by this orientation . clearly , @xmath0 is strong and the vertices of each gray square conform a convex set . let @xmath99 be a convex set of @xmath0 such that @xmath461 . there must be two gray squares @xmath462 and @xmath463 such that @xmath464 and @xmath465 . since @xmath19 is an interior vertex , it belongs to two white squares . let @xmath18 be a vertex in the opposite corner in one of these white squares @xmath466 . assume without loss of generality that @xmath19 is the lower left corner of @xmath462 , the upper right corner of @xmath463 , and the lower right corner of @xmath466 ; the remaining cases can be dealt similarly . if @xmath18 is the middle vertex of a @xmath467-path or a @xmath468-path of length two , then @xmath469 , and hence @xmath470 . otherwise , let @xmath241 and @xmath471 be the upper right and lower left corners of @xmath466 , respectively . hence , @xmath472 and either the sequence @xmath473 starting from @xmath241 determines an @xmath3-directed path of length @xmath69 , or the sequence @xmath474 determines a @xmath4-directed path of length @xmath69 . in either case , @xmath470 . it can be verified inductively that @xmath124 , and hence , @xmath475 . if @xmath456 is an integer and @xmath7 is the grid @xmath325 , then @xmath476 \setminus \ { 2 , 3 , 5 , 7\}.\ ] ] by virtue of theorem [ no2 ] and lemmas [ forbid ] , [ 3*n - c=3k ] , [ 3*n - c=3k+2 ] , [ 3*n - c=3k+1 ] and [ n*3-c=4 ] , it remains to prove that @xmath477 . let @xmath0 be a strong orientation of @xmath7 and @xmath99 a convex set of @xmath0 . lemmas [ girthcon ] and [ comcon ] imply that @xmath99 induces a strong subdigraph of @xmath0 and that @xmath316 induces a connected subdigraph of @xmath0 , respectively . but every connected subgraph of @xmath7 with @xmath478 vertices has either a vertex of degree @xmath227 , and thus does not admit a strong orientation ; or does not have a connected complement . hence , @xmath477 . the following lemma is the cornerstone of the vast majority of the arguments we will use in this section . it is easy to check that the vertices on the boundary of each of the gray filled squares conform a convex set . we affirm that any convex set in @xmath53 has at most @xmath69 vertices . let @xmath99 be a maximum convex set in @xmath53 and suppose that @xmath461 . observe that the intersection of @xmath99 with the vertices of each gray filled square is either empty , or it has one vertex , or it has four vertices . hence , since @xmath480 , the vertices of at least two squares are contained in @xmath99 . if the vertices of two gray filled squares different from the central one are contained in @xmath99 , then there are at least two vertices of the central square in @xmath99 ; thus , the vertices of the central square are contained in @xmath99 . so , by symmetry , we need only to consider two cases . the first case is when the square on the lower left corner and the central square are contained in @xmath99 . assume that the vertex in the lower left corner of @xmath53 is @xmath481 . it is easy to check that @xmath482 and @xmath483 are @xmath484- and @xmath485-geodesics , respectively . from here , @xmath486 , and @xmath487 is a @xmath488-geodesic in @xmath53 . therefore , there are at least two vertices of each gray filled square in @xmath99 and we can conclude that @xmath489 , a contradiction . in the second case , we have the upper left corner and the central square contained in @xmath99 . now , @xmath490 is a @xmath491-geodesic in @xmath53 . hence , there are at least two vertices of the lower left corner square in @xmath99 . so , the lower left corner square is contained in @xmath99 and we have the condition of the first case . as the reader would expect , the main part of the argument in the next lemma s proof is the construction of the orientation . the following lemmas will use similar orientations , so the descriptions will be very detailed in the first ones , and will loose detail as the lemmas progress . consider the standard plane embedding of @xmath7 and color the interior faces gray and white with a checkerboard - like pattern , assigning gray to the square on the bottom left corner ( like in figure [ 4 * 4-c=4 ] ) . we will define an orientation of the arcs of @xmath7 using this coloring . enumerate the rows of squares from bottom to top . enumerate the gray squares in each row from left to right . orient the bottom left corner square as a whirlpool and , from here , orient all the gray squares in the first column and first row alternating whirlpool and anti - whirlpool orientations . now , the first square or every odd row is oriented , so we can orient all the gray squares in the odd rows alternating whirlpool and anti - whirlpool orientations . a similar idea can be used to orient all the gray squares in even rows , but start orienting as an anti - whirlpool the first gray square on the second row . at this point , every arc dividing two interior faces of @xmath7 has received an orientation . we will consider two cases . first , suppose that @xmath77 and @xmath496 are even integers , hence every corner square of @xmath7 is gray , and every remaining unoriented edge @xmath459 of @xmath7 divides a white square from the exterior face of @xmath7 . thus , the edge @xmath459 lies in exactly one square of @xmath7 , and has one parallel arc @xmath460 in the square . orient @xmath459 in the same direction as @xmath460 . all the edges of @xmath7 are now oriented ; let @xmath0 be the resulting oriented graph . figure [ 4 * 4-c=4 ] is an example of this orientation . it is easy to verify that @xmath0 is strongly connected , and the vertices of every gray square conform a convex set of @xmath0 . if @xmath99 is a convex set of @xmath0 such that @xmath461 , then @xmath99 intersects the vertices on at least two different gray squares @xmath462 and @xmath463 in odd columns and rows . since @xmath99 induces a strong subdigraph of @xmath0 , we may assume without loss of generality that @xmath462 and @xmath463 are gray squares in the same row and adjacent odd columns . if @xmath462 and @xmath463 are in row @xmath497 and columns @xmath260 and @xmath498 , then the vertices of @xmath462 and @xmath463 , together with the vertices of the squares @xmath466 and @xmath499 in row @xmath500 and columns @xmath260 and @xmath498 , induce a subdigraph of @xmath0 isomorphic to the digraph @xmath53 of figure [ 4 * 4-c=4 ] . therefore @xmath501 . we can repeat this argument using squares @xmath463 and @xmath499 and the squares in column @xmath502 and rows @xmath497 and @xmath500 . iterating this process we obtain @xmath124 . hence , @xmath503 . for the second case , assume that @xmath77 or @xmath496 is an odd integer . by virtue of lemma [ n*3-c=4 ] , we assume that @xmath94 . observe that there are exactly two white corner squares in @xmath7 . except for the edges in the white corner squares , orient the remaining edges of @xmath7 as in the previous case . for each of the white corner squares we have the two cases depicted in figure [ n*m - c=4fig ] ( the squares we are interested in are the bottom right corners ) , and two isomorphic cases obtained by reversing all the arcs of the previous ones . we will consider two cases . if @xmath504 , complete the orientation @xmath0 of @xmath7 as in figure [ n*m - c=4fig ] . again , it is direct to verify that @xmath0 is strong and the vertices of every gray square conform a convex set of @xmath0 . let @xmath99 be a convex set of @xmath0 such that @xmath461 . as in the previous case , it suffices to show that there are two consecutive gray squares in the same row or the same column that intersect @xmath99 . since @xmath461 , there @xmath99 intersects at least two different gray squares hence , the desired condition is held , unless the gray squares are precisely those adjacent to one of the white corners . we will assume without loss of generality that one white corner square is the one in the bottom right , as in figure [ n*m - c=4fig ] . let @xmath462 and @xmath463 be gray squares to the left and above the white corner , respectively . let @xmath505 be the vertex in the upper left corner of @xmath506 , @xmath507 . consider first the situation depicted by the digraph on the left in figure [ n*m - c=4fig ] . it is clear that @xmath508 , and also that , starting from @xmath509 , the sequence @xmath510 determines a @xmath511-directed path of length @xmath69 . hence , @xmath99 intersects two consecutive gray squares in the same row . for the situation depicted by the digraph on the right in figure [ n*m - c=4fig ] , it is clear that @xmath512 . it is also clear that , starting from @xmath509 , the sequence @xmath513 determines a @xmath511-directed path of length @xmath97 . since @xmath99 is convex , it intersects two consecutive gray squares in the same row . as a final case , assume without loss of generality that @xmath514 . since @xmath77 is an odd integer , the two white corners are those on the right side of @xmath7 . in the situation depicted by the digraph on the left in figure [ n*m - c=4fig ] , use precisely that orientation and the same argument as in the previous case . in the remaining case , use the orientation of figure [ n*m - c=4fig ] for the bottom right corner , and orient the upper right corner also as a directed path of length @xmath70 ( assume that it goes up and left ) . this orientation is strong , the vertices of each gray square conform a convex set , and the same argument as the previous case shows that we can find two consecutive gray squares in the same column that intersect @xmath99 , and hence @xmath124 . the idea of this proof is to generalize the orientation used in lemma [ n*m - c=4 ] , but using grids of size @xmath520 instead of squares in the odd - numbered rows and columns . an example of this orientation is depicted in figure [ n*m - c = abfig ] . first , suppose that @xmath521 and @xmath522 . enumerate the rows and columns of squares of @xmath7 from down to up and from left to right , respectively . color with gray and white the squares of @xmath7 in a checkerboard - like pattern , but considering rectangles of squares instead of single squares , in the following way . color the squares in the first @xmath523 columns and @xmath524 rows , and the square in the @xmath460-th column and @xmath525-th row wih gray . color the squares in the @xmath460-th column and the first @xmath524 rows , and the squares in the @xmath525-th column and the first @xmath523 rows in white . color the rest of @xmath7 with a tiling of this coloring . enumerate the rows and columns of gray rectangles in the pattern from left to right and from down to up . clearly , the gray rectangles in the even numbered rows ( columns ) are squares . we can assume that @xmath526 , hence , the gray rectangles in the odd numbered rows ( columns ) are proper rectangles . orient the gray squares in the even rows as in the proof of lemma [ n*m - c=4 ] ( in particular , the first gray square of the second row is an anti - whirlpool ) . orient the gray rectangles in the odd rows as whirlpools or anti - whirpools in such way that every gray square ( possibly except the last gray square in every row or every column ) in a even row , together with the four gray squares sharing a vertex with it , induce a digraph isomorphic to @xmath53 ( figure [ 4 * 4-c=4 ] ) . of course , to achieve this end , we also need to orient four additional arcs dividing either two white squares of @xmath7 , or a white square and the exterior face of @xmath7 . now , every unoriented edge of @xmath7 divides two white squares , or a white square from the exterior face in an even row or column of gray squares . orient every edge dividing two white squares in the same direction as the closest arc in a gray square in the same row or column . there are unoriented edges parallel to the arcs oriented in the previous step ; orient those edges in the same direction as the arcs they are parallel to . the remaining unoriented edges form paths on the exterior face of @xmath7 joining pairs of gray rectangles . orient those paths as directed paths in such way that , if any , the corner vertices of @xmath7 in a white square have in - degree and out - degree equal to one . figure [ n*m - c = abfig ] shows an example of this orientation . first , suppose that @xmath539 . orient @xmath540 as a whirlpool . now , re - orient the corresponding arcs in the fiber @xmath541 to obtain a directed path in the same direction ( left or right ) as the arc between @xmath542 and @xmath543 . also , orient the fiber @xmath544 as a directed path in the same direction as the arc between @xmath542 and @xmath543 . assume without loss of generality that @xmath544 is oriented right . finally orient down every edge in @xmath545 except for @xmath546 , which is oriented up . if @xmath0 is the resulting oriented graph , then it is easy to verify that @xmath547 is a convex set of @xmath0 , and @xmath548 . if @xmath522 , then orient @xmath549 as a whirlpool and @xmath550 as a whirlpool or as an anti - whirlpool in such way that the fibers @xmath551 and @xmath552 are isomorphic . orient the remaining edges arbitrarily , as long as there is one arc going up and one arc going down , to obtain @xmath0 . we will show that @xmath292 is a convex set of @xmath0 . let @xmath553 be a @xmath554-path in @xmath0 such that every intermediate vertex of @xmath553 belongs to @xmath293 . then @xmath553 can be codified as @xmath555 . let @xmath556 be the sequence such that @xmath557 from the way we oriented @xmath302 and @xmath303 we conclude that @xmath558 defines a @xmath554-path in @xmath0 , strictly shorter than @xmath553 and such that @xmath559 . hence , @xmath292 is a convex set of @xmath0 . recalling that @xmath560 , and using the fact that every convex set with more than @xmath561 vertices has at least two vertices in @xmath292 and at least two vertices in @xmath293 , we conclude that @xmath548 . consider a coloring of the squares of @xmath7 similar to the coloring used in the proof of lemma [ n*m - c = ab ] , with the following differences . in the lower left corner of the aforementioned orientation of @xmath7 we have a rectangle , @xmath569 , of @xmath570 gray squares . consider the subgraph @xmath302 of @xmath7 obtained by deleting the first @xmath571 columns and the first @xmath572 rows of vertices of @xmath7 . color @xmath302 as in the proof of lemma [ n*m - c=4 ] . finally , complete the checkerboard - like pattern with gray rectangles of size @xmath573 squares in the first row , and with gray rectangles of size @xmath574 squares in the first column . we want our largest convex set to be a segment of @xmath569 , the gray rectangle in the lower left corner . color white the squares in the first @xmath96 columns from the top row of @xmath569 . also color white the squares in the first @xmath580 rows of the rightmost column of @xmath569 to obtain the gray region @xmath581 . now , orient @xmath7 as in lemma [ n*m - c = ab ] , orienting @xmath581 as a whirlpool . orient all the remaining edges to the right and down to obtain @xmath0 . an example of this orientation is depicted in figure [ n*m - c = ab - k - lfig ] . now , observe that the top right corner of @xmath581 coincides with the top right corner of @xmath569 . hence , this orientation has the same properties as the orientation of lemma [ n*m - c = ab ] . we can assume without loss of generality that @xmath582 . otherwise , assuming that @xmath583 , we have that @xmath584 for some pair of integerers @xmath585 such that @xmath586 , @xmath587 . hence , @xmath588 . from here , it is easy to observe that the only proper convex sets of @xmath0 are the vertices in each of the gray regions of @xmath0 . since the largest one is @xmath581 , which has @xmath589 vertices , we conclude @xmath590 . first , consider the case @xmath594 and @xmath514 . if @xmath595 , then the result follows from lemma [ n*m - c = ab - k - l ] . the orientations for @xmath596 are shown in figure [ 5 * 4-c = nm - kfig ] . it can be easily verified that the set of white vertices is a largest convex set for this orientation . now suppose that @xmath597 . the case @xmath598 follows from lemma [ n*m - c = ab - k - l ] . the orientation for @xmath599 is depicted in figure [ 5 * 4-c = nm - kfig ] . again , it is not hard to verify that the sets of white vertices are largest convex sets for each of the orientations . finally , suppose that @xmath600 . if @xmath601 , then the result follows from lemma [ n*m - c = ab - k - l ] . the orientations for @xmath602 are depicted in figure [ 5 * 4-c = nm - kfig ] . as in the previous cases , the sets of white vertices are largest convex sets for the corresponding orientation . observe the orientations of the @xmath609 and @xmath610 grids in figure [ 5 * 4-c = nm - kfig ] for @xmath599 . clearly , the orientation of the latter can be obtained from the orientation of the former by adding an additional row of squares at the bottom of the grid , and orienting this new row as a whirlpool . naturally , three arcs of the former orientation should change its direction , in this case , these are the arcs @xmath611 and @xmath612 of the former oriented graph . we have a similar situation with the orientation of the @xmath613 and @xmath610 grids for @xmath599 and @xmath614 , respectively . again , the latter can be obtained from the former by adding one row and one column of squares , orienting the new row as an anti - whirlpool and the new column following certain pattern . the idea of this proof is to observe that the three orientations of the @xmath610 grids , shown in figure [ 5 * 4-c = nm - kfig ] , can be naturally extended to obtain the desired orientations . first , observe that , independently of the value of @xmath77 , we can always add a new row of squares at the bottom of the grid . notice that we are extending the largest convex set of the grid , but the complement of such set remains unchanged . hence , if we consider the aforementioned orientations of the @xmath610 grid , we can conclude that @xmath616 for every integer @xmath608 . to add a new column of squares we have two different behaviors . first , consider the orientations for @xmath599 and @xmath598 . in this case , to add a new column , we need to change the direction of one arc , as in the @xmath609 and @xmath610 grids for the @xmath599 argument . but , to add further columns we will not need to change the direction of any arc , just add a column of squares oriented as a whirlpool . following this procedure we will obtain an orientation of the grid @xmath187 for every pair of integers @xmath617 , for @xmath599 and @xmath598 , respectively . in the other hand , when @xmath614 , adding a new column will not preserve the complement of our largest convex set , the complement will grow larger . if we add a new column , following the pattern as in the example in figure [ n*m - c = nm - kfig ] , we will obtain an orientation of the grid @xmath187 for every pair of integers @xmath89 , for @xmath618 . we can generalize the orientations used for @xmath599 and @xmath598 , respectively , to obtain orientations of the grid @xmath187 for @xmath619 and @xmath620 . the idea is to extend the pattern horizontally , we will explain how it is done for the case @xmath621 , the remaining case is analogous . we want to preserve the @xmath622 directed path defined by the sequence @xmath623 , and the whirlpool ( or anti - whirlpool ) of @xmath624 squares in the upper right corner of the grid . the square of the upper left corner of the grid is oriented as a whirlpool and the rest of the grid is oriented as a whirlpool or anti - whirlpool , except for some alternating squares adjacent to the aforementioned directed path . the remaining unoriented edges are oriented as @xmath625 and , if it is still unoriented , @xmath626 . following a similar idea , orientations for every grid @xmath187 for any pair of integers @xmath627 and any @xmath628 can be obtained . we depict the orientations for @xmath629 for @xmath630 in figure [ n*m - c = nm - kfig ] . since all these orientations follow a similar pattern , it can be easily proved that each one has convexity number @xmath631 , as desired . now , consider the case @xmath514 . if @xmath634 ( mod 4 ) , then lemma [ n*m - c = nb ] gives the desired orientation . else , we give basic orientations that can be enlarged adding columns of whirlpools , either to the largest convex set , or to its complement . the orientation for @xmath599 is the orientation used for the @xmath613 grid ( figure [ 5 * 4-c = nm - kfig ] ) . clearly , additional rows of whirlpools can be added in the bottom of the grid . we give the orientations for @xmath635 in figure [ n*m - c = nm - kfig2 ] . again , the set of black vertices is the complement of the largest set of the oriented graph . since it is easy to observe that the given orientations have the desired properties , and when we extend them the arguments remain the same , this concludes the proof ot the lemma . as a final remark , the reader might have noticed that the proofs are lengthy and technical , which makes them hard to follow . it would be a good problem to find a short proof for the main theorem of this article . p. duchet , convexity in combinatorial structures , proceedings of the 14th winter school on abstract analysis . circolo matematico di palermo , palermo , 1987 . rendiconti del circolo matematico di palermo , serie ii , supplemento no .

we explicitly show that the new polynomial invariants for knots , upto nine crossings , agree with the ooguri - vafa conjecture relating chern - simons gauge theory to topological string theory on the resolution of the conifold . tifr / th/00 - 53 + september , 2000 * on link invariants and topological string amplitudes * + p. ramadevi + _ physics department , + indian institute of technology bombay , + mumbai 400 076 , india + _ tapobrata sarkar + _ department of theoretical physics , + tata institute of fundamental research , + homi bhabha road , mumbai 400 005 , india _ +

hep-th0009188

recently , there has been a lot of progress in understanding @xmath0 chern - simons theory on @xmath1 and its related invariants , the knot invariants @xcite , from a topological string theory point of view @xcite . in a striking conjecture , ooguri and vafa @xcite have proposed a formulation of knot invariants on @xmath1 from considerations of topological string theory amplitudes on the resolved conifold , based on earlier work by gopakumar and vafa@xcite that is similar in spirit to the celebrated ads / cft correspondence . this conjecture , which was tested for the simplest knot , called the unknot ( a circle in @xmath1 ) in @xcite and for the trefoil knot ( toral knot of type ( 2,3 ) ) by labastida and marino @xcite , provides an extremely interesting `` physical '' interpretation for the coefficients in the polynomial invariants . the purpose of this paper is to critically examine this conjecture for a wide class of knots using a simpler method of directly computing invariants for any knot or link carrying arbitrary representations of @xmath0 presented in refs . @xcite . in @xcite , the structure of the knot invariants was deduced from topological string amplitudes in the presence of d-2 branes ending on d-4 branes , and the result was expressed in terms of certain integers representing the number of d-2 branes . this constitutes a physical interpretation of the integer coefficients appearing in two - variable homfly polynomials for knots . in @xcite , this prediction was tested on the chern - simons field theory side for toral knots ( with explicit results for the right handed trefoil knot ) using the formulation of the toral knot operators introduced in @xcite . in particular , the authors in @xcite developed a group theory argument in relating the vacuum expectation values ( vevs ) of operators on the topological string theory side to invariants of knots carrying arbitrary @xmath0 representation in chern - simons theory . their explicit calculations for the right handed trefoil knot agreed with ooguri - vafa conjecture on the polynomials appearing in the vev of such operators@xcite . the computation becomes very tedious for knots with higher number of crossings within the toral knot operator formalism . we will use a simpler method @xcite for directly obtaining the invariants for any knot and link ( includes torus knots ) carrying arbitrary @xmath0 representations . then , incorporating the group theoretic results presented in ref . @xcite , we will compute the new polynomial invariants for a wide class of knots confirming ooguri - vafa conjecture . we can evaluate multi - component link invariants within chern - simons field theoretic framework but their relations to the observables on the topological closed string theory needs to be determined to check the conjecture . the organisation of the paper is as follows . in the first part of section 2 , we briefly recapitulate the salient features of the conjecture in @xcite . next , we review the procedure of @xcite in computing new polynomial invariants from invariants of knots carrying arbitrary @xmath0 representation . in section 3 , we briefly present the method of directly calculating knot invariants for arbitrary representations of @xmath0 @xcite . in section 4 , we present our results on knot and link invariants for arbitrary representations , upto 9 crossings . in section 5 , we conclude with discussions and comments on some open problems . in this section , we begin by briefly recapitulating the essential details of the argument of @xcite which proposes an equivalence between @xmath0 chern - simons gauge theory on @xmath1 and closed a - twisted topological string theory on the @xmath2 resolved conifold geometry . for a large @xmath3 gauge theory , in thooft s double line notation , denoting by @xmath4 the genus of the ( triangulated ) reimann surface formed by the double lines , and by @xmath5 the number of faces in the triangulation of this surface the free energy can be written as the summation @xmath6 where @xmath7 is the thooft coupling related to the perturbative gauge coupling constant @xmath8 by @xmath9 . for the case of chern - simons theory on @xmath1 , the coefficients @xmath10 were shown to be equal to the partition function of the a - twisted open topological string theory with @xmath4 handles and @xmath5 boundaries with the target space as the cotangent space of @xmath1 @xcite . in the context of d - branes , this implies that the chern - simons theory is equivalent to the open topological string theory obtained by wrapping @xmath3 d - branes on the base @xmath1 of the cotangent space . on the other hand , we can consider the cotangent space @xmath11 to be the deformed conifold that can , via a conifold transition , be resolved into the total space of the line bundle @xmath12 over @xmath2 . in @xcite , it was conjectured that the closed topological string theory on this resolved conifold geometry is dual to the @xmath0 chern - simons theory on @xmath1 with the khler parameter @xmath13 of the blown up @xmath2 being related by @xmath14 to the coupling constant @xmath15 of the chern - simons theory . elegant computations were done in @xcite , ( following earlier work done by these authors in @xcite ) in order to check this conjecture at the level of the free energy ( the vacuum amplitude @xmath16 of the gauge theory being related to the free energy as @xmath17 ) , and their results strongly supported the said conjecture . in @xcite , this conjecture was further extended to the level of observables of the chern - simons theory , which are the gauge invariant and metric independent wilson loop operators @xmath18 associated with a knot @xmath19 , ( which is an embedding of a circle @xmath20 on a three manifold ) , and carrying an irreducible representation @xmath21 of the gauge group , in our case the gauge group being @xmath0 . these operators are defined in the usual way as trace of the holonomy @xmath22 : @xmath23\end{aligned}\ ] ] with @xmath24 being the trace over the representation @xmath21 and p denotes path ordering . the vacuum expectation value ( vev ) of the wilson loop operator @xmath25 gives the knot invariant @xmath26 $ ] for knot @xmath19 embedded in @xmath1 : @xmath27~=~\langle w_r(c ) \rangle ~=~ { \int [ { \cal d}a_{\mu } ] w_r(c ) \exp ( is ) \over \int [ { \cal d}a_{\mu } ] \exp ( is ) } ~. \label { 2five}\ ] ] here @xmath28 denotes the chern - simons field theory action : @xmath29 where @xmath30 is a one form valued in the lie algebra of @xmath0 gauge group and @xmath15 is the coupling constant which takes integer values . besides the wilson loop associated with a single curve @xmath19 ( knot ) , we could consider product of wilson loops for a link @xmath31 made up of component knots @xmath32 carrying representation @xmath33 as observables . the invariant for such a multi - component link is given by @xmath34~=~ \langle w_{r_1 , r_2 , \ldots , r_s } ( l ) \rangle ~=~ { \int [ { \cal d}a_{\mu } ] \prod_i w_{r_i}(c_i ) \exp ( is)~\over \int [ { \cal d}a_{\mu } ] \exp ( is ) } . \label { 2six}\ ] ] for @xmath35 s chosen to be defining representation of @xmath0 , the invariant gives two - variable homfly polynomial @xcite ( upto unknot normalisation ) which reduces to the celebrated jones polynomials for @xmath36 gauge group . we will present the essential ingredients for directly evaluating these knot and link invariants in the next section . the observables in the topological string theory on a deformed confiold , are constructed using the following basic idea @xcite : define a lagrangian submanifold ( a 3-cycle on which the symplectic form vanishes ) corresponding to a given knot @xmath19 in @xmath1 such that @xmath19 represents the intersection of the 3-cycle with @xmath1 . wrapping @xmath37 d - branes on the 3-cycle now gives a chern - simons theory on the 3-cycle in addition to the @xmath0 chern - simons theory on @xmath1 , and quantizing the action for the complex scalar field on @xmath19 gives rise to the effective operator @xmath38 \label { oper}\ ] ] where @xmath39 and @xmath40 are the holonomies transforming under @xmath0 and @xmath41 groups respectively and the trace is over the defining representation . clearly , these operators involve traces of powers of holonomy besides the wilson loops and products of wilson loops which has not been considered within chern - simons field theoretic framework . we will see later in this section that the vev of these new operators can be related to chern - simons invariants for knots carrying arbitrary @xmath0 representation . the form of the vev of @xmath42 was conjectured by invoking duality of this theory with closed topological string theory on the resolved conifold . this was done using a method analogous to the schwinger computation of the free energy of a particle in a background field . it is known @xcite that in type iia compactification on calabi - yau , the topological string amplitudes at genus @xmath4 compute corrections to the low energy theory of the form @xmath43 ( with @xmath44 and @xmath45 being the self dual field strengths of the curvature and the graviphoton respectively ) . these can be computed via a one - loop calculation where the relevant d - brane states , that are d-2 branes wrapped on cycles bound to d-0 branes , are integrated out . the lagrangian submanifold on which d - branes are wrapped in the chern - simons calculation is identified on the resolved conifold geometry , and the schwinger computation is done with @xmath37 d-4 branes wrapping the lagrangian 3-cycle and extending along an @xmath46 in non - compact space time , on which the d-2 branes might end . the @xmath47 gauge field on the non - compact @xmath46 gives rise to a magnetic two form field on the d-4 brane , and each d - brane state that enters in the ( two dimensional ) schwinger calculation is characterised by its magnetic charge @xmath21 ( with @xmath21 transforming in some representation of the @xmath48 gauge theory on the d-4 brane ) , bulk d-2 brane charge @xmath49 and spin @xmath50 . denoting the number of such states by @xmath51 , the closed topological string amplitude is @xcite @xmath52 where is the variable @xmath13 defined in @xcite and equal to @xmath53 of @xcite . ] @xmath54 $ ] ; @xmath55 . in order to obtain the above conjectured form ( [ lmeq ] ) from chern - simons theory , we need to device methods of computing vevs of trace of powers of holonomy . an elegant group theoretic approach @xcite proves useful in arriving at the form ( [ lmeq ] ) and also obtaining the new polynomial invariant @xmath56 in terms of invariants of knots carrying arbitrary @xmath0 representation in chern - simons theory . we will see in section 4 that these @xmath56 for a wide class of knots indeed obey eqn . ( [ lmaeq ] ) . the basic idea involved in the group theoretic method is to use the frobenius formula ( see , for eg . @xcite ) to compute product of traces of arbitrary powers of the holonomy . introducing a class function @xmath57 , this formula reads as @xmath58 where @xmath59 denotes the conjugacy class determined by the sequence @xmath60 ( i.e there are @xmath61 1-cycles , @xmath62 2-cycles etc ) in the permutation group @xmath63 ( @xmath64 ) , @xmath65 is the young tableau with @xmath66 number of boxes , @xmath67 is the @xmath0 representation corresponding to the young tableau @xmath68 . @xmath69 gives the character of the conjugacy class @xmath59 in the representation @xmath68 of the permutation group @xmath63 . rewriting the exponent in the eqn . ( [ oper ] ) and @xmath70 in the eqn . ( [ lmeq ] ) in terms of the class functions , the following equality can be obtained @xcite : @xmath71 where @xmath72 means `` @xmath73 divides @xmath74 '' and @xmath75 . for example , @xmath76 for the vector @xmath77 and the `` connected '' vev @xmath78 is defined through the relation @xmath79=\sum_{{\bf \vec k}}\frac{\vert c({\bf \vec k } ) \vert}{l ! } g_{{\bf k}}^{(c)}\prod_jx^{k_j } \label { recurs}\ ] ] where @xmath80 , @xmath81 is the number of elements in the conjugacy class @xmath59 . from eqn . ( [ compa ] ) , we deduce @xmath82 using eqns . ( [ lmbasis ] , [ recurs ] , [ masterlm ] ) , @xmath83 can be expressed in terms of the vev of @xmath84 and certain lower order terms . we refer the reader for more details to @xcite , but for the moment , we list the formal expressions for @xmath83 for @xmath85 0.5 cm for the single box representation of @xmath0 , the invariant @xmath86 is simply given by the expectation value of @xmath39 in the single box representation , i.e @xmath87 the r.h.s of the above equation is the unnormalised homfly polynomial , as we will come to in the next section . similarly , for the horizontal and vertical double box representations of @xmath0 , we get from eq . ( [ masterlm ] ) , @xmath88 and similarly , for representations of @xmath0 with three boxes in the young tableau , we get the equations @xmath89 the above equations can be used to find the values of @xmath90 in various representations in a recursive manner . for example , the quantity @xmath91 is , as we have mentioned , the unnormalised homfly polynomial . using this and the value of @xmath92 ( the method of calculation will be presented in the next section ) we can evaluate @xmath93 from eq . ( [ twoboxes ] ) . we can then use this value of @xmath93 in eq . ( [ threeboxes ] ) to calculate @xmath94 using the expression for @xmath95 . in ref . @xcite , the conjecture ( [ lmeq]),([lmaeq ] ) was verified for the simplest curve called unknot : @xmath96 implying @xmath97 for representations other than the defining representation . for a class of toral knots ( explicit results for right - handed trefoil ) , @xmath98 was shown to agree with the conjecture ( [ lmaeq ] ) @xcite . in the next section , we will present our direct method of computing @xmath99 for a wide class of knots and links ( which includes torus knots ) carrying arbitrary representation of @xmath0 which will be used in section 4 to evaluate @xmath100 and hence proving ooguri - vafa conjecture for such knots . # 1#2 plus 2pt minus 1pt # 1 # 1#2 # 1 # 1|#1 # 1| # 1#2|#2 # 1#2#3#4 # 1#2#3#4 in this section , we present the methods of obtaining invariants of knots and links carrying arbitrary @xmath0 representations in chern - simons theory @xcite . for representations other than defining representations placed on knots / links , we will refer to the invariants as generalised knot / link invariants . explicit computations are done for some knots and links obtained from braids which includes toral knots of type @xmath101 . these invariants will be useful in obtaining the new polynomial invariants which are related to amplitudes in topological string theory . in order to directly evaluate the knot and link invariants , we require the following two ingredients : * the functional integral over the space of matrix valued one forms @xmath30 is evaluated by exploiting the connection between the chern - simons theory in three dimensional space with boundary and the @xmath102 wess - zumino conformal field theory on the two dimensional boundary @xcite . * any knot / link can be represented as a closure or plat or capping of braids @xcite . the computation of these invariants has been considered in detail in refs . @xcite . in order to illustrate the technique of direct evaluation of the invariant , we will concentrate on knots and links @xmath103 obtained as a closure of a two - strand braid with @xmath104-crossings . @xmath103 represents toral knots of type ( 2,m ) for @xmath104 odd and two - component links for @xmath104 even . the details we present for this class of knots and links are general enough for deriving invariants of knots and links obtained from any @xmath73-strand braid by closure or plat or capping of braids . consider the three - manifold @xmath1 containing @xmath103 carrying representation @xmath21 as shown in fig . 1 . in the two component knots of link @xmath105 . ] let us slice the three - manifold into two three dimensional balls as shown in fig.2(a ) and ( b ) . the two dimensional @xmath2 boundaries of the three - balls are oppositely oriented and have four points of intersections with the braid , which we refer to as four punctures . now , exploiting the connection between chern - simons theory and wess - zumino conformal field theory , the functional integrals of these three - balls correspond to states in the space of four point correlator conformal blocks of the wess - zumino conformal field theory @xcite . the dimensionality of this space is dependent on the representation of @xmath0 placed on the strands and the number of punctures on the boundary . these states can be written in a suitable basis . two such choices of bases ( @xmath106 , and @xmath107 ) are pictorially depicted in fig.3(a),(b ) . here @xmath108 and @xmath109 as allowed by the @xmath102 wess - zumino conformal field theory fusion rules . the basis @xmath110 is chosen when the braiding is done in the side two parallel strands . in other words , it is the eigen basis corresponding to the braiding generators @xmath111 and @xmath112 : |_s^side= b_1 |_s^side= b_3 @xmath113 for right - handed half - twists between two parallel strands carrying different representation @xmath114 being : _ s^(+)(r_1,r_2 ) = _ r_1r_2^s q^c_r_1 + c_r_2 + c_r_1 - c_r_2 /2- c_s/2 . [ pevalue ] where @xmath115 and @xmath116 are the quadratic casimir in the representations @xmath117 and @xmath50 respectively . in terms of the highest weight @xmath118 for representation @xmath21 , the quadratic casimir is given by @xmath119 where @xmath120 is the weyl vector equal to the sum of the fundamental weights of @xmath0 . similarly for braiding in the middle two anti - parallel strands @xmath121 , we choose the basis @xmath122 : b_2 |_t^cent= ( _ t^(- ) ( r,|r ) anti - parallel strands being : _ t^(-)(r_1 , |r_2)= _ r_1 |r_2^t q^-c_r_1 - c_r_2 /2 + c_t/2 where @xmath123 . these two bases are related by a duality matrix : |_s^side = a_st ( ) @xmath0 quantum racah coefficients which are known for some special representations @xcite . for example , the duality matrix for @xmath21 taken to be defining representation of @xmath0 ) is a @xmath124 matrix : ( ) where the number in square bracket refers to the quantum number defined as = q^n 2 - q^-n 2 q^1 2 - q^-1 2 [ quan ] with @xmath125 ( also called deformation parameter in quantum algebra @xmath126 ) related to the coupling constant @xmath15 as @xmath127 . we will see that the invariants are polynomials in @xmath128 . now , let us determine the states corresponding to figs.2(a ) and ( b ) . since the braiding is in the side two parallel strands , it is preferable to use @xmath129 as basis states . let @xmath130 be the state corresponding to fig . clearly , we can write the state for fig . 2(b ) as |_2 = b_1^m |_1 this state should be in the dual space as its @xmath2 boundary is oppositely oriented compared to the boundary in fig . . then the link invariant is given by v_r[l_m ] = _ 1 |b_1^m |_1 [ inv ] for determining the polynomial , we will have to express the states as linear combination of the basis states @xmath110 . the coefficients in the linear combination are chosen such that @xmath131 \right ) = ( { \rm dim}_q r)^2,\ ] ] where the quantum dimension for any irreducible representation @xmath21 with highest weight @xmath132 is given by dim_q r = _ > 0 [ . ( + ) ] , [ qdim ] where @xmath133 s are the positive roots and @xmath120 is equal to the sum of the fundamental weights of the lie group @xmath0 . the square bracket refers to usual definition of quantum number ( [ quan ] ) . the above mentioned restrictions determine the state @xmath134 ( see @xcite ) as : |_1 = _ s r r |_s^side substituting it in eqn . ( [ inv ] ) and using the braiding eigenvalue ( [ pevalue ] ) , we obtain @xmath135~=~ \sum_{s \in r \otimes r } { \rm dim}_q r_s ( \lambda_{s}^{(+)}(r , r))^m~. \label { knot}\ ] ] for a two - component link with different representations , the multi - coloured link invariant will be @xcite @xmath136~=~ \sum_{s \in r_1 \otimes r_2 } { \rm dim}_q r_s ( \lambda_{s}^{(+)}(r_1,r_2))^{2m}~. \label { link}\ ] ] for @xmath137 denoting @xmath0 defining representation , we get the homfly polynomial @xmath138 $ ] @xcite ( upto unknot normalisation ) @xmath139 = { ( \lambda^{1\over 2 } - \lambda^{-1 \over 2 } ) \over ( q^{1 \over 2 } - q^{-1\over 2 } ) } p_{l_m}[l = i\lambda , m = i ( q^{1/2}- q^{-1/2})]~.\ ] ] so far , we have elaborated for a class of knots and links obtained from closure of two - strand braids with @xmath104 crossings and directly obtained the invariant ( [ knot ] , [ link ] ) . the essential ingredients presented for this class is sufficient to derive the invariant for any knot / link obtained from @xmath73-strand braids . let us now determine explicitly the irreducible representations and braiding values for some @xmath0 representations . for symmetric representation placed on component knots @xmath140 the irreducible representations @xmath141 in @xmath102 wess - zumino conformal field theory in the large @xmath15 limit will be @xmath142 the quantum dimension ( [ qdim ] ) for @xmath143 and parallel braiding eigenvalues ( [ pevalue ] ) for strands carrying symmetric representation @xmath144 can be rewritten in a neat form : @xmath145 [ n+1 ] \ldots [ n+n+\ell-1 ] ) ( [ n-1][n ] \ldots [ n+n-\ell-2 ] ) \over [ 2 \ell]!~ [ n- \ell ] ! ~ ( [ n+\ell+1 ] [ n+\ell ] \ldots [ 2 \ell + 2])}~,\nonumber\\ \l^{(+)}_{\rho_{\ell } } ( r_n , r_n ) & = & ( -)^{n-\ell}~q^{(n-1)n \over 2 } q^{\frac{n(n+1)}{2 } - \frac{\ell(\ell + 1)}{2 } } \end{aligned}\ ] ] similarly , for antisymmetric representations placed on the component knots , the irreducible representations @xmath146 will be .5 cm @xmath147 we can again rewrite the quantum dimension ( [ qdim ] ) for @xmath146 and eigenvalues in parallel strands ( [ pevalue ] ) carrying @xmath148 representations as follows : @xmath149 [ n-1 ] \ldots [ n-2n+\ell+1 ] ) ( [ n+1][n ] \ldots [ n-\ell+2 ] ) \over [ 2n-2\ell]!~ [ \ell]!~ ( [ 2n-2 \ell+2 ] [ 2n-2 \ell+3 ] \ldots [ 2n-\ell+1 ] ) } \nonumber\\ \l^{+}_{{\hat { \rho}}_{\ell } } ( \hat r_n , \hat r_n ) & = & ( -)^{n-\ell } q^{(n-1)n \over 2 } ~q^{n- \ell ( n+1)+ \frac{\ell(\ell + 1)}{2 } } ~.\end{aligned}\ ] ] even though we have explicitly given the braiding eigenvalues for symmetric and antisymmetric representations , the general procedure presented in this section can be used to obtain eigenvalues for strands carrying mixed symmetry representations as well . for example @xmath150 , the irreducible representations in the tensor - product will be @xmath151 we will write the quantum dimensions for the above representations and the braiding eigenvalues for parallel strands which will be useful for computation of new polynomial invariants in topological string theory . @xmath152[n-2][n-3 ] \over ~[6][2]^2 } & ; & \lambda_{\rho_1 } ^{(+)}(r , r ) = a q^{-3 ( n^3+n^2 - 8n-12 ) \over2 n ( n+2)}\label{rone}\\ { \rm dim}_q \rho_2= c { [ n][n+2][n-2]\over [ 3 ] [ 5 ] } & ; & \lambda_{\rho_2}^{(+)}(r , r)=a q^{- 3 ( n^2 - 6 ) \over 2n}\label{rtwo}\\ { \rm dim}_q \rho_3= c { [ n][n+2][n-2]\over [ 3 ] [ 5 ] } & ; & \lambda_{\rho_3}^{(+)}(r , r)=-a q^{- 3 ( n^2 - 6 ) \over 2n}\label{rthree}\\ { \rm dim}_q \rho_4 = c { [ n][n+2][n+3 ] [ 3]\over [ 2]^2 [ 5 ] [ 4 ] } & ; & \lambda_{\rho_4}^{(+ ) } ( r , r)= a q^{-(3n^2 + 5n-18 ) \over 2n}\label{rfour}\\ { \rm dim}_q \rho_5 = c { [ n][n+1][n+2 ] \over [ 2]^2 [ 4 ] [ 3 ] } & ; & \lambda_{\rho_5}^{(+ ) } ( r , r ) = -a q^{-3 ( n+3 ) ( n-2 ) \over 2n}\label{rfive}\\ { \rm dim}_q \rho_6 = c { [ n][n-1][n-2 ] \over [ 2]^2 [ 3 ] [ 4 ] } & ; & \lambda_{\rho_6}^{(+ ) } ( r , r)= a q^{-3 ( n-3 ) ( n+2 ) \over 2n}\label{rsix}\\ { \rm dim}_q \rho_7 = c { [ n+2][n+3][n-2 ] \over [ 6 ] [ 2]^2 } & ; & \lambda_{\rho_7}^{(+)}= -a q^{-3 ( n^4 + 6n^3 + 5n^2 - 24n-36 ) \over 2n ( n+2 ) ( n+3)}\label{rseven}\\ { \rm dim}_q \rho_8= c { [ 3][n][n-2][n-3 ] \over [ 2]^2 [ 4 ] [ 5 ] } & ; & \lambda_{\rho_8}^{(+)}= -a q^{-(3n^2 - 5n -18 ) \over 2n } \label{reight}\end{aligned}\ ] ] where @xmath153[n+1][n-1]/ [ 3]$ ] and @xmath154 . the @xmath155 signs in the braiding eigenvalues are deduced from topological equivalence @xmath156 . for the anti - parallelly oriented strands carrying symmetric representation @xmath144 , the braiding eigenvalue ( [ aeval ] ) is given by ^(- ) _ _ ( r_n , |r_n ) = ( -)^ q^(n-1)/2 q^^2/2 ; = 0,1 , , n . here these eigenvalues for @xmath157 correspond respectively to the irreducible representations in the product @xmath158 : @xmath159 here boxes with dot represents a column of length @xmath160 . the quantum dimension for @xmath161 ( [ qdim ] ) is : @xmath162[n+1 ] \ldots [ n+2\ell-1 ] ) \prod_{a=1}^{n-2 } ( [ n - a ] [ n - a+1 ] \ldots [ n - a+\ell-1 ] ) \over [ \ell ] ! ( [ n+2 \ell-2 ] [ n+2 \ell-3 ] \ldots [ n+\ell-1 ] ) \prod_{a=1}^{n-2 } ( [ a][a+1 ] \ldots [ a + \ell-1])}\ ] ] similarly , for antisymmetric representations @xmath163 , we have the following eigenvalues in anti - parallely oriented strands : @xmath164 where the representations @xmath165 appear in the following tensor product : .5 cm @xmath166 .8 cm and the quantum dimension for @xmath165 is @xmath167[n-1 ] \ldots [ \ell+1 ] ) ( [ n+1 ] [ n ] \ldots [ n-\ell+2 ] ) \over [ n-2 \ell ] ! [ \ell ] ! ( [ n-2\ell+2 ] [ n-2\ell+3 ] \ldots [ n-\ell+1])}\ ] ] we will list the knot invariant formula for some knots ( carrying arbitrary @xmath0 representation @xmath21 ) as shown in fig . 4 . in this subsection , we go beyond torus knots and compute the invariants for the knots @xmath192 and @xmath193 using the methods described in the last section . we will do so for representations containing one and two boxes in the young tableau . for the knot @xmath192 , we obtain from eq . ( [ 3.33 ] ) following @xmath194\ ] ] for representations consisting of two boxes in the young diagram , we get the following expressions from eqs . ( [ 3.33 ] ) and ( [ twoboxes ] ) @xmath195\nonumber\\ f_{\twover}\left(4_1\right)&=&\denomneg \left[q^{-\frac{5}{2}}\lambda^{-3}{\left(\lambda-1\right)}^2 \left(\lambda - q^2\right)\right.\nonumber\\ & ~&\left . \times\left(\lambda^2-q^2\right)\left(\lambda q-1\right)\right ] \end{aligned}\ ] ] similar expressions can be calculated for the knot @xmath193 . we obtain , in this case , from eqs . ( [ 3.35 ] ) and ( [ twoboxes ] ) , the following result for @xmath180 @xmath196 \end{aligned}\ ] ] and the representations @xmath175 and @xmath176 , we obtain , in a similar fashion as before , @xmath197 \nonumber\\ f_{\twover}\left(6_1\right)&=&\denomneg \left[q^{-\frac{9}{2}}\lambda^{-3}{\left(\lambda-1\right)}^2 \left(\lambda - q\right)\left(\lambda q-1\right ) \right.\nonumber\\&~&\left . \left\{-q^5+\lambda^2q^2\left(1 - 2q^2\right ) + \lambda^4\left(1+q^2\right)- \lambda^3q\left(q^3 - 1\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda q^3\left(1 - 2q - q^2\right)\right\}\right ] \end{aligned}\ ] ] these results are again in full agreement with the conjecture of ooguri and vafa @xcite thus providing further strong evidence in support of their arguments . we will now calculate the new invariants obtained from knots that are connected sums . the knots @xmath198 where @xmath104 and @xmath73 are odd integers are of this form . we will explicitly evaluate the new invariants for the knots @xmath199 ( as drawn in figure 4 . ) for representations upto two boxes in the young diagram , although from our methods , these can be easliy extended to general @xmath104 and @xmath73 . let us start with the knot @xmath200 . for the defining representation , the invariant is given by eq . ( [ 3.38 ] ) @xmath201\ ] ] for the representation @xmath202 , we obtain from eqs . ( [ 3.38 ] ) and ( [ twoboxes ] ) , @xmath203\end{aligned}\ ] ] and similarly for @xmath204 , we obtain @xmath205\end{aligned}\ ] ] next , let us consider the knot @xmath206 . using eqs . ( [ 3.39 ] ) and ( [ twoboxes ] ) , we get the following results : for @xmath207 , the invariant polynomial is given by @xmath208\ ] ] for @xmath202 , we get @xmath209\end{aligned}\ ] ] and , for @xmath204 , we obtain @xmath210 \end{aligned}\ ] ] in this subsection , we will consider invariants obtained for two and three component links carrying defining representations . we have seen that @xmath211 for knots are exactly the knot polynomials in the defining representation ( [ onebox ] ) . however , naively extending this equality to links does not match with ooguri - vafa conjecture . for the @xmath212-component link denoted by @xmath213 ( also called the hopf link in the knot theory literature ) we have from eq . ( [ 3.41 ] ) the following link invariant in defining representation : @xmath214 \label{pa1}\ ] ] similarly , from the methods presented in the last section , we find that for the @xmath212-component links @xmath215 ( [ 3.42 ] ) and @xmath216 ( [ 3.43 ] ) , the link invariants in defining representations are respectively , @xmath217 \label{pa2}\\ v_{\one,\one}(6_1 ^ 2)&= & \frac{\lambda^{1\over 2}-\lambda^{-{1\over 2 } } } { \left(q^{1\over 2}-q^{-{1\over 2}}\right)^2 } \left[q^{-3}\lambda^{5\over 2 } \left\{-1+q - q^2+q^3-q^4+q^5-q^6+\right.\right.\nonumber\\ & ~&\left.\left.\lambda\left(q - q^2 + q^3-q^4+q^5\right)\right\}\right ] \label{pa3}\end{aligned}\ ] ] similarly , for three component links @xmath218 , the invariant in defining representation @xcite and @xmath104 , defined in @xcite are related to @xmath125 and @xmath219 as @xmath220 @xmath221 \nonumber\\ v_{\one,\one,\one}(6_3 ^ 3)&= & \frac{\lambda^{1\over 2}-\lambda^{-{1\over 2 } } } { \left(q^{1\over 2}-q^{-{1\over 2}}\right)^3 } \left(q^{-2}\lambda^2\right ) \left[q\left(1-q+q^2\right)- \lambda^{-1}\left(1-q+^2\right ) \left(1+q^2\right)\right.\nonumber\\&~&\left . + \lambda^{-2}q\left(2 - 3q+2q^2\right)\right ] \label{pa5}\end{aligned}\ ] ] from these link invariants in defining representation , it is clear that they can not be directly equated to new polynomial invariant @xmath222 as there are extra powers of @xmath223 in the denominator , violating the ooguri - vafa conjecture . recently labastida et al @xcite ( see note added ) have derived new polynomial invariants for links involving link invariants in chern - simons theory . hence using their results and the link invariants in defining representations , @xmath222 can be shown to obey the conjecture . in this paper , we have performed non - trivial tests of the conjecture in @xcite relating topological string amplitudes on the resolved conifold geometry to chern - simons gauge theory on @xmath1 . the elegant group theoretic procedure developed in ref . @xcite to determine topological string amplitudes from vevs of wilson loops of chern - simons theory enables us to verify the conjecture for a number of knots upto nine crossings . the methods developed in @xcite established a simple and powerful way of computing generalised knot and link invariants in @xmath0 chern - simons theory . in particular , we have gone much beyond the results presented in @xcite using our tools , and have found beautiful agreement with the conjecture of @xcite for new polynomial invariants for these knots . we believe that our results firmly establishes the said conjecture . further , this is the first time the integer coefficients of these ( new ) polynomial invariants in knot theory has meaning in terms of certain d - brane computations in closed topological string theory . we have the machinery to obtain knot and link invariants in chern - simons theory . however , we have been able relate only knot invariants to closed topological string amplitudes . it would be extremely interesting to generalise the topological string theory computation to objects that are @xmath73-component links in the dual theory , and to check the new polynomial invariants that arise in this case using the methods developed in this paper . another interesting direction will be to explore the relationship of perturbative chern - simons invariants - namely , vassiliev invariants to objects in the dual string side . we leave the study of such issues to a future publication . .5 cm * acknowledgments * : p.r would like to thank csir for the grant . .5 cm * note added * : after this work appeared in the bulletin board , the very important paper @xcite by labastida , marino and vafa appeared , which has treated the case of multi - component links observables in large n chern - simons theory and dual topological string picture , in great details . in this appendix , we present the explicit results for the values of @xmath83 for @xmath21 containing three boxes in its young tableau . we will do so for the @xmath185 knot , i.e a two stranded braid with seven crossings . the results are @xmath224 \\ f_{\threever}(7_1)&=&\denomneg \left[q^{-\frac{63}{2}}\lambda^{15\over 2}\left(1-q+q^2\right ) \left(1-q^{1\over 2}+q - q^{\frac{3}{2}}+q^2 -q^{\frac{5}{2}}+q^3\right)\right.\nonumber\\&~&\left . \left(1+q^{1\over 2}+q+q^{\frac{3}{2}}+q^2 + q^{\frac{5}{2}}+q^3 \right ) \left(1+q^4\right ) \left\{q^{\frac{17}{2}}+q^{\frac{21}{2 } } + q^{\frac{23}{2 } } + q^{\frac{25}{2}}+q^{\frac{27}{2 } } \right.\right . \nonumber\\&~&\left.\left . + 3q^{\frac{29}{2 } } + 2q^{\frac{31}{2}}+3q^{\frac{33}{2}}+3q^{\frac{35}{2 } } + 2q^{\frac{37}{2}}+3q^{\frac{39}{2}}+3q^{\frac{41}{2 } } + 2q^{\frac{43}{2}}+2q^{\frac{45}{2}}+q^{\frac{47}{2 } } \right.\right.\nonumber\\&~&\left.\left . + \lambda\left(-q^{\frac{13}{2}}-2q^{\frac{15}{2 } } -3q^{\frac{17}{2}}-4q^{\frac{19}{2}}- 5q^{\frac{21}{2}}-6q^{\frac{23}{2}}- 8q^{\frac{25}{2}}-10q^{\frac{27}{2}}- 13q^{\frac{29}{2 } } \right.\right.\right . \nonumber\\&~&\left.\left.\left . -15q^{\frac{31}{2 } } -17q^{\frac{33}{2}}-18q^{\frac{35}{2 } } - 18q^{\frac{37}{2}}-18q^{\frac{39}{2 } } - 16q^{\frac{41}{2 } } \right.\right.\right.\nonumber\\&~&\left.\left.\left . -13q^{\frac{43}{2 } } - 10q^{\frac{45}{2}}-6q^{\frac{47}{2 } } -3q^{\frac{49}{2}}-q^{\frac{51}{2}}\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda^2\left(2q^{\frac{11}{2}}+3q^{\frac{13}{2}}+ 7q^{\frac{15}{2}}+8q^{\frac{17}{2}}+ 12q^{\frac{19}{2}}+ 13q^{\frac{21}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 18q^{\frac{23}{2}}+21q^{\frac{25}{2}}+ 28q^{\frac{27}{2}}+32q^{\frac{29}{2}}+ 39q^{\frac{31}{2}}+43q^{\frac{33}{2}}+ 48q^{\frac{35}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 48q^{\frac{37}{2}}+ 46q^{\frac{39}{2}}+40q^{\frac{41}{2}}+ 32q^{\frac{43}{2}}+24q^{\frac{45}{2}}+ 16q^{\frac{47}{2}}+9q^{\frac{49}{2 } } \right.\right.\right.\nonumber\\&~&\left.\left.\left . + 4q^{\frac{51}{2 } } + q^{\frac{53}{2}}\right ) + \lambda^3\left(-q^{\frac{9}{2}}-4q^{\frac{11}{2 } } - 6q^{\frac{13}{2}}-10q^{\frac{15}{2}}- 14q^{\frac{17}{2}}-16q^{\frac{19}{2}}- \right.\right.\right.\nonumber\\&~&\left.\left.\left . 20q^{\frac{21}{2}}-26q^{\frac{23}{2}}- 30q^{\frac{25}{2}}-38q^{\frac{27}{2}}- 46q^{\frac{29}{2}}-52q^{\frac{31}{2}}- 60q^{\frac{33}{2 } } \right.\right.\right.\nonumber\\&~&\left.\left.\left . -66q^{\frac{35}{2}}- 65q^{\frac{37}{2}}-62q^{\frac{39}{2}}- 54q^{\frac{41}{2}}-42q^{\frac{43}{2}}- 32q^{\frac{45}{2}}-22q^{\frac{47}{2}}- \right.\right.\right.\nonumber\\&~&\left.\left.\left . 12q^{\frac{49}{2}}-6q^{\frac{51}{2 } } - 2q^{\frac{53}{2}}\right)+\lambda^4\left(2q^{\frac{9}{2 } } + 3q^{\frac{11}{2}}+7q^{\frac{13}{2}}+ 8q^{\frac{15}{2}}+12q^{\frac{17}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 13q^{\frac{19}{2}}+17q^{\frac{21}{2}}+ 19q^{\frac{23}{2}}+25q^{\frac{25}{2}}+ 29q^{\frac{27}{2}}+36q^{\frac{29}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 40q^{\frac{31}{2}}+47q^{\frac{33}{2}}+ 49q^{\frac{35}{2}}+50q^{\frac{37}{2}}+ 46q^{\frac{39}{2}}+40q^{\frac{41}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 32q^{\frac{43}{2}}+24q^{\frac{45}{2}}+ 16q^{\frac{47}{2}}+9q^{\frac{49}{2}}+ 4q^{\frac{51}{2}}+q^{\frac{53}{2}}\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda^5\left(-q^{\frac{9}{2}}-2q^{\frac{11}{2 } } - 3q^{\frac{13}{2}}-4q^{\frac{15}{2}}- 5q^{\frac{17}{2}}-6q^{\frac{19}{2}}- 7q^{\frac{21}{2}}-8q^{\frac{23}{2}}- 10q^{\frac{25}{2 } } \right.\right.\right.\nonumber\\&~&\left.\left.\left . -12q^{\frac{27}{2}}- 15q^{\frac{29}{2}}-17q^{\frac{31}{2}}- 19q^{\frac{33}{2}}-20q^{\frac{35}{2}}- 19q^{\frac{37}{2}}-18q^{\frac{39}{2}}- 16q^{\frac{41}{2 } } \right.\right.\right.\nonumber\\&~&\left.\left.\left . -13q^{\frac{43}{2}}- 10q^{\frac{45}{2}}-6q^{\frac{47}{2 } } - 3q^{\frac{49}{2}}-q^{\frac{51}{2}}\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda^6\left(q^{\frac{11}{2}}+q^{\frac{15}{2}}+ q^{\frac{17}{2}}+q^{\frac{19}{2}}+ q^{\frac{21}{2}}+2q^{\frac{23}{2}}+ q^{\frac{25}{2}}+2q^{\frac{27}{2}}+ 3q^{\frac{29}{2}}+3q^{\frac{31}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 3q^{\frac{33}{2}}+4q^{\frac{35}{2}}+ 2q^{\frac{37}{2}}+3q^{\frac{39}{2}}+ 3q^{\frac{41}{2}}+2q^{\frac{43}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 2q^{\frac{45}{2}}+q^{\frac{47}{2}}\right)\right\ } \right ] \\ f_{\mixed}(7_1)&=&\denom \left[q^{-\frac{61}{2}}\lambda^{15\over 2}\left(1-q+q^2\right ) \left(1-q^{1\over 2}+q - q^{\frac{3}{2}}+q^2-q^{\frac{5}{2}}+ q^3\right ) \left(1+q^{1\over 2 } \right . \nonumber\\ & ~&\left . + q+q^{\frac{3}{2}}+q^2 + q^{\frac{5}{2}}+q^3\right)\left(1+q^4\right ) \left\{q^{\frac{39}{2}}+2q^{\frac{41}{2}}+3q^{\frac{43}{2 } } + 3q^{\frac{45}{2}}+4q^{\frac{47}{2 } } \right.\right.\nonumber\\&~&\left.\left . + 3q^{\frac{49}{2}}+ 4q^{\frac{51}{2}}+3q^{\frac{53}{2}}+3q^{\frac{55}{2}}+ 2q^{\frac{57}{2}}+q^{\frac{59}{2}}+\lambda\left(-q^{\frac{35}{2 } } -3q^{\frac{37}{2}}-7q^{\frac{39}{2}}- 12q^{\frac{41}{2 } } \right.\right.\right.\nonumber\\&~&\left.\left.\left . -16q^{\frac{43}{2}}- 20q^{\frac{45}{2}}-23q^{\frac{47}{2}}- 23q^{\frac{49}{2}}-23q^{\frac{51}{2}}- 20q^{\frac{53}{2}}-16q^{\frac{55}{2}}- 12q^{\frac{57}{2}}- \right.\right.\right.\nonumber\\&~&\left.\left.\left . 7q^{\frac{59}{2}}- 3q^{\frac{61}{2}}- q^{\frac{63}{2}}\right ) + \lambda^2\left(q^{\frac{33}{2}}+4q^{\frac{35}{2}}+ 11q^{\frac{37}{2}}+19q^{\frac{39}{2 } } + 31q^{\frac{41}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 40q^{\frac{43}{2}}+ 52q^{\frac{45}{2}}+58q^{\frac{47}{2}}+ 62q^{\frac{49}{2}}+58q^{\frac{51}{2}}+ 52q^{\frac{53}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 40q^{\frac{55}{2}}+31q^{\frac{57}{2 } } + 19q^{\frac{59}{2}}+ 11q^{\frac{61}{2}}+ 4q^{\frac{63}{2 } } + q^{\frac{65}{2}}\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda^3\left(-2q^{\frac{33}{2}}- 7q^{\frac{35}{2}}-16q^{\frac{37}{2}}- 28q^{\frac{39}{2}}-42q^{\frac{41}{2}}- 56q^{\frac{43}{2}}-70q^{\frac{45}{2 } } - \right.\right.\right.\nonumber\\&~&\left.\left.\left . 80q^{\frac{47}{2}}-84q^{\frac{49}{2}}- 80q^{\frac{51}{2}}-70q^{\frac{53}{2}}- 56q^{\frac{55}{2}}- \right.\right.\right.\nonumber\\&~&\left.\left.\left . 42q^{\frac{57}{2}}- 28q^{\frac{59}{2}}-16q^{\frac{61}{2}}- 7q^{\frac{63}{2}}-2q^{\frac{65}{2}}\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda^4\left(q^{\frac{33}{2}}+6q^{\frac{35}{2}}+ 12q^{\frac{37}{2}}+23q^{\frac{39}{2}}+ 32q^{\frac{41}{2}}+44q^{\frac{43}{2}}+ 53q^{\frac{45}{2}}+62q^{\frac{47}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 63q^{\frac{49}{2}}+62q^{\frac{51}{2}}+ 53q^{\frac{53}{2}}+44q^{\frac{55}{2}}+ 32q^{\frac{57}{2}}+23q^{\frac{59}{2 } } + 12q^{\frac{61}{2}}+6q^{\frac{63}{2 } } + q^{\frac{65}{2}}\right ) \right.\right.\nonumber\\&~&\left.\left . + \lambda^5\left(-2q^{\frac{35}{2}}- 5q^{\frac{37}{2}}-9q^{\frac{39}{2}}- 14q^{\frac{41}{2}}-18q^{\frac{43}{2}}- 22q^{\frac{45}{2}}-25q^{\frac{47}{2}}- \right.\right.\right.\nonumber\\&~&\left.\left.\left . 25q^{\frac{49}{2}}-25q^{\frac{51}{2}}- 22q^{\frac{53}{2}}-18q^{\frac{55}{2}}- 14q^{\frac{57}{2}}-9q^{\frac{59}{2}}- \right.\right.\right.\nonumber\\&~&\left.\left.\left . 5q^{\frac{61}{2}}-2q^{\frac{63}{2}}\right ) + \lambda^6\left(q^{\frac{37}{2}}+q^{\frac{39}{2}}+ 3q^{\frac{41}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 3q^{\frac{43}{2}}+ 4q^{\frac{45}{2}}+4q^{\frac{47}{2}}+ 4q^{\frac{49}{2}}+4q^{\frac{51}{2}}+ 4q^{\frac{53}{2}}+ \right.\right.\right.\nonumber\\&~&\left.\left.\left . 3q^{\frac{55}{2}}+ 3q^{\frac{57}{2}}+ q^{\frac{59}{2}}+ q^{\frac{61}{2}}\right)\right\}\right ] \end{aligned}\ ] ] 99 e. witten , `` quantum field theory and the jones polynomial , '' commun . * 121 * ( 1989 ) , 351 . l. kauffman , _ knots and physics _ , world scientific ( 1993 ) , + w. b. r. lickorish , _ an introduction to knot theory _ , springer - verlag ( 1998 ) + e. guadagnini , _ the link invariants of chern - simons field theory _ , walter de gruyter ( 1993 ) . r. gopakumar , c. vafa , `` m - theory and topological strings , i , '' hep - th/9809187 . r. gopakumar , c. vafa , `` on the gauge theory/ geometry correspondence , '' hep - th/9811131 . h. ooguri , c. vafa , `` knot invariants and topological strings , '' nucl . phys . * b577 * , 419 , ( 2000 ) , hep - th/9912123 . j. m. f. labastida , m. marino , `` polynomial invariants for torus knots and topological strings , '' hep - th/0004196 . p. ramadevi , t.r . govindarajan , r. k. kaul , `` three - dimensional chern - simons theory as a theory of knots and links ( iii ) . compact semisimple group , '' nucl . * b 402 * ( 1993 ) 548 , hep - th/9212110 . p. ramadevi , `` chern - simons theory as a theory of knots and links , '' , ph.d . j. m. f. labastida , p. m. llatas , a. v. ramallo , `` knot operators in chern - simons theory , '' nucl * b348 * ( 1991 ) 651 . e. witten , `` chern - simons gauge theory as a string theory , '' hep - th/9207094 p. freyd , d. yetter , j. hoste , w.b.r . lickorish , k. millett , and a. ocneanu , `` a new polynomial invariant of knots and links , '' bull . a.m.s * 12 * ( 1985 ) 239 . i. antoniadis , e. gava , k. s. narain , t. r. taylor , `` topological amplitudes in string theory , '' nucl . phys . * b413 * ( 1994 ) 162 , hep - th/9307158 . w. fulton , j. harris , _ representation theory , a first course _ , springer - verlag , 1991 . j. birman , `` braids , links and mapping class groups , '' , annals of mathematics studies * 82 * , princeton university press . w.b.r . lickorish and k.c . millett , `` a polynomial invariant of oriented links , '' topology * 26 * ( 1987 ) 107 . j. m. f labastida , m. marino , c. vafa , `` knots , links and branes at large n , '' hep - th/0010102

in the region of sgr b2 there are several condensations heated externally by nearby hot stars . therefore ho far ir lines are expected to probe only an external low density and high temperature section of these condensations , whereas millimeter - wave lines can penetrate deeper into them where the density is higher and t lower . we have conducted a study combining ho lines in both spectral regions . first , observations of several ho thermal lines seen in absorption toward sgr b2(m ) at a spectral resolution of km s have been analyzed . second , an _ iram_30 m telescope map of the para ho line at 183.31 ghz , seen in emission , has also been obtained and analyzed . the ho lines seen in absorption are optically thick and are formed in the outermost gas of the condensations in front of the far ir continuum sources . they probe a maximum visual extinction of to 10 mag . radiative transfer models indicate that these lines are quite insensitive to temperature and gas density , and that ir photons from the dust play a dominant role in the excitation of the involved ho rotational levels . in order to get the physical conditions of the absorbing gas we have also analyzed the co emission toward sgr b2(m ) . we conclude , based on the observed co=76 line at 806.65 ghz with the _ caltech submillimeter observatory _ , and the lack of emission from the far ir co lines , that the gas density has to be lower than . using the values obtained for the kinetic temperature and gas density from oh , co , and other molecular species , we derive a water column density of ( 9) in the absorbing gas . hence , the water vapor abundance in this region ,(ho ) , is(1 - 2) . the relatively low ho / oh abundance ratio in the region , - 4 , is a signature of uv photon dominated surface layers traced by far ir observations . as a consequence the temperature of the absorbing gas is high , t - 500 k , which allows very efficient neutral neutral reactions producing ho and oh . on the other hand , the 183.31 ghz data provide a much better spatial and spectral resolution than the far - ir iso data . this maser line allows to trace water deeper into the cloud , i.e. , the inner , denser ( ) and colder ( t k ) gas . the emission is very strong toward the cores . the derived water vapor abundance for this component is a few . there is also moderate extended emission around sgr b2 main condensations , a fact that supports the water vapor abundance derived from far ir ho lines for the outer gas .

astro-ph0601336

the determination of water abundance in space is a long standing problem in astronomy . theoretical models predict that water can be the most abundant species in _ warm _ molecular clouds after h@xmath0 and co @xcite . therefore , the determination of its spatial distribution and abundance contributes to a better knowledge of the chemical and physical processes that take place in the interstellar medium ( ism ) . unfortunately , water is an abundant molecule in our atmosphere making particularly difficult the observation of its rotational lines and vibrational bands from earth . even so , some observations of water lines have been performed from ground based and airborne telescopes : the @xmath21 at 22 ghz @xcite , the @xmath22 at 325 ghz @xcite , the @xmath23 at 321 ghz @xcite and the @xmath5 at 183.31 ghz ( watson et al . 1980 ; cernicharo et al . 1990 , 1994 , 1996 , 1999a ; gonzlez - alfonso et al . , 1998 ) . due to the maser nature of these lines , the analysis and interpretation of the spectra is not obvious . among these lines , only two have been used to map the extended emission of water vapor in orion : the @xmath5 -hereafter the 183 ghz line- @xcite and the @xmath22 at 325 ghz @xcite . however , although we know from iso and swas observations that water is extended ( @xmath325@xmath24@xmath1025@xmath24 ) in sgr b2 ( cernicharo et al . 1997 , neufeld et al . 2003 , goicoechea et al . 2004 ) , little is known about its excitation conditions and its detailed spatial distribution . an alternative to indirectly estimate the water abundance in the galactic center ( gc ) is to use related species such as hdo @xcite or h@xmath25o@xmath26 @xcite . in none of these cases , the determination of @xmath13(h@xmath0o ) is straightforward . the iso mission @xcite , and specially , the _ long wavelength spectrometer _ , lws , @xcite , and the _ short wavelength spectrometer _ , sws , @xcite , have provided a unique opportunity to observe several h@xmath0o lines in a great variety of astronomical environments . nevertheless , the majority of these observations were performed at the low spectral resolution of the grating mode ( @xmath31000 km s@xmath4 ) , which produces a critically strong dilution in the search for molecular features in most ism sources . nevertheless , the sgr b2 cloud has been analyzed and studied in detail with the lws/_fabry perot _ , which provided a velocity resolution of @xmath335 km s@xmath4 ( goicoechea et al . 2004 , hereafter g04 ) . opposite to what is found toward other star forming regions such as orion @xcite , the observations of the @xmath27 line at @xmath3179.5 @xmath28 m in sgr b2 show that the line appears in absorption rather than in emission @xcite . afterward , the launch of the _ submillimeter wave astronomy satellite _ , swas , @xcite , and odin @xcite allowed the observation of the @xmath29 fundamental transition of both h@xmath30o at 557 and h@xmath31o at 548 ghz , first detected by the _ kuiper airborne observatory _ , kao , @xcite . although the velocity resolution is @xmath31 km s@xmath4 , the large beam of swas ( @xmath34@xmath24 ) makes these observations more sensitive to the cold and less dense gas . swas observations have provided a reliable estimate of the water vapor abundance in the low excitation clouds located in the line of sight toward sgr b2 @xcite . however , the fact that only the ground - state absorption line is detected makes difficult a detailed study of water vapor excitation mechanisms in sgr b2 itself . this is the most massive cloud in the galaxy , with @xmath310@xmath32 , @xcite , and a paradigmatic object in the gc region as its geometrical properties , physical conditions and chemical characteristics make it a miniature galactic nucleus with @xmath315@xmath24 extent ( g04 and references therein ) . the main star - forming regions in sgr b2 are located within three dust condensations , labeled as ( n ) , ( m ) and ( s ) . these condensations are embedded in a @xmath310 pc moderate density ( @xmath33 @xmath8 ) cloud @xcite . in addition , these structures are surrounded by lower density components of warm ( t@xmath1@xmath34100 k ) gas , hereafter the sgr b2 warm envelope . these conditions have been derived mainly from absorption observations of nh@xmath25 metastable lines @xcite , oh lines ( goicoechea & cernicharo , 2001 ; 2002 ) , and h@xmath0co lines @xcite in the radio domain . nevertheless , the warm and low density gas is poorly traced by radio observations of other molecular species ( generally excited by collisions in the denser regions and thus observed in emission ) . however , the warm envelope represents the strongest contribution to the absorption features produced by many light hydrides in the far - ir spectrum of sgr b2 @xcite . the origins of the observed rich chemistry and the heating mechanisms in the sgr b2 warm envelope are a subject of intense debate . the matter is complicated due to the different observational signatures to be integrated in the same picture : high temperatures derived from nh@xmath25 absorption lines @xcite , fine structure emission from the photo ionized and photo dissociated gas @xcite , sio and x ray distribution @xcite , etc . in all possible scenarios , water plays a significant role . several mechanisms allow its formation and survival in the warm envelope . dissociative recombination of h@xmath25o@xmath26 leads to the production of h@xmath0o and oh . this processes depend on the specific @xmath35 branching ratio for the h@xmath0o formation channel . unfortunately , the determination of @xmath35 with different experimental procedures has also yielded different values , from @xmath35=0.05 @xcite to @xmath35=0.25 @xcite , while most chemical models have used @xmath35@xmath30.35 . in addition , water could also be produced in the gas phase by the endothermic reaction : @xmath36 however , the gas temperature must exceed @xmath3300 k to overcome the activation barrier @xcite . at these temperatures , the reaction : @xmath37 also contributes to the formation of oh . therefore , h@xmath0o and oh column densities can be used to determine the role of the neutral - neutral reactions in their formation / destruction routes . still , the exact h@xmath0o / oh ratio will be determined by @xmath35 , the temperature and by photodissociation processes if uv radiation is present . as an example , neufeld et al . @xcite studied a diffuse cloud ( @xmath38@xmath31 , @xmath39=100 @xmath8 ) toward w51 and showed that the presence of a warm gas component ( t@xmath1@xmath40400 k ) could explain the observed variations of the h@xmath0o / oh ratio respect to other diffuse clouds . finally , high oxygen depletion onto water ice mantles in dust grains could have taken place during the evolution of the cool gas in sgr b2 . photodesorption and/or evaporation for dust temperatures above @xmath390 k , could release some water back into the gas phase enhancing the h@xmath0o abundance expected from pure gas phase formation @xcite . however , gas and dust are thermally decoupled in the outer layers of sgr b2 , where the dust temperatures are significantly lower , t@xmath41@xmath142030 k @xcite , than gas temperatures , t@xmath1@xmath14300 k @xcite . thus , t@xmath41 seems too low to produce significant evaporation of water ice mantles . therefore , a detailed study of the far ir h@xmath0o lines and of the 183.31 ghz extended emission is needed to constrain the water abundance and the physical characteristics of the absorbing / emitting region . in this work we present and analyze the far - ir observations of several thermal lines of water vapor toward sgr b2(m ) and the first map of the 183.31 ghz maser emission of para h@xmath0o around sgr b2 main condensations . the layout of the paper is as follows : in section [ sct : observ ] we summarize the far ir , submm and mm observations and data reduction . the spectra and maps are presented in sec.[sct : results ] . section [ sct : analysis ] is devoted to the analysis of co observations ( [ ssct : co ] ) and water vapor observations ( [ ssct : h2o ] ) with different radiative transfer methods . the main implications of our work are discussed in sec . [ sct : discussion ] , where photochemistry models for h@xmath0o and oh are also presented . a summary is given in sec . [ sct : summary ] . most pure rotational lines of water vapor that play a role in the radiative heating and cooling of the dense ism appear in the terahertz domain . thus , airborne or satellite observations are needed to avoid the earth s atmosphere blocking . in particular , many pure rotational lines of h@xmath0o appear in the far ir coverage of the lws spectrometer @xcite on board iso @xcite . we have used the lws fabry - perot ( fp ) instrument to search for h@xmath0o and h@xmath42o lines toward sgr b2(m ) . the lws / fp spectral resolution is @xmath43 7000 - 1000 and has a circular aperture of about 80@xmath44 in diameter . the majority of detected lines have been observed in the time awarded to our iso proposals . however , an extensive inspection of the public iso data base has been carried out in order to examine and average all available water lines . most of the water lines were present in the astronomical observation template ( aot ) lws04 observations , which give a large spectral sampling , wavelength precision and s / n ratios . the number of scans in this mode was @xmath3412 depending on the expected absorption produced by the different species . these lws products have been processed and compared through off - line - processing ( olp ) pipeline 6.0 and 10.1 versions . there are no major differences except that recent pipelines produce @xmath4510% less absorption in some lines due to continuum level differences from one olp to another . the data were analyzed using the iso spectrometers data reduction package isap . the mean fp continuum flux of each line deviates by @xmath4520% and this can be taken as the flux calibration error @xcite . after checking the continuum level of the aot lws01 observations , a polynomial baseline was fitted to each spectra and adopted as fp continuum level . [ t ] the water line at 183.31 ghz can be observed from high mountain - top sites under very dry conditions @xcite . the observations in the direction of sgr b2 were achieved with the iram-30 m telescope , with a half power beam width ( hpbw ) of @xmath313@xmath44 , in december 1999 and january 2004 . the source transits at 24 degrees elevation at pico veleta , but even so the average atmospheric opacity over a bandwidth of 0.5 ghz centered at 183.3 ghz was 1.7 , i.e. an atmospheric water column of 0.9 mm along the line of sight . we used and sis receiver designed to cover the band 130 - 184 ghz . in its higher end , the receiver temperature is about 160 k and the sideband rejection is at least 18 db . the backend used was a 512 two - pole filter with half - power widths and spacings equal to 1.0 mhz . with these conditions , and in this configuration , system temperatures were typically around 3000 k. the pointing was checked using the already known strong emission of w49n @xcite . once it was verified that the sgr b2 emission was quite compact , the map was carried out in wobbler - switching mode in order to obtain very flat baselines . observations were also tried in january 2001 , 2002 and 2004 . in the 2001 run the weather was poor while in the last run it was possible to have a 30 minutes window of observing time with good atmospheric transmission . these runs allowed us to check that the line profile in the position peak of sgr b2(m ) was the same of december 1999 and to confirm that the extent of the water emission toward this condensation was larger than the beam size . in the january 2004 observing run we also had very good atmospheric transmission to perform observations around sgr b2(n ) . however , the measured line fluxes were lower than in previous observations by a constant factor . as the line profiles did not change , we scaled the 1999 observations to those the reference taken in the ( 0,+5 ) position in 2004 ( see fig . [ mapa_183 ] ) . the co 7 - 6 ( 806 ghz ) observations were performed on march 10 , 2002 with the 10.4 m _ caltech submillimeter observatory _ ( cso ) located at the summit of mauna kea ( hawaii ) . the receiver is a helium - cooled sis mixer operating in double - sideband mode ( dsb ) providing an instantaneous bandwidth of 0.95 ghz and designed to fully cover the 780 - 910 ghz atmospheric window ( see kooi et al . the pointing was checked using a strong co point - like source in the nearby w28 molecular cloud , and was kept within 3@xmath44 accuracy for a hpbw of @xmath310@xmath44 . two different acousto - optic spectrometer ( aos ) backends were used with 2048 and 1024 channels respectively for a total spectral coverage of 1.5 and 0.5 ghz in each case . the zenith atmospheric water vapor column was @xmath30.5 mm during the observations , resulting on system temperatures ranging from 6000 to 10000 k depending on the receiver and the elevation . co emission is very extended around sgr b2 , so we had to perform position switched scans setting the off position 1 degree away in azimuth . an additional 5% increase had to be applied to t@xmath46 in order to second - order correct the standard chopper - wheel calibration method at high frequencies with large atmospheric opacities ( pardo et al . the far ir spectrum of sgr b2(m ) is dominated by the absorption produced by nh@xmath25 , oh and h@xmath0o rotational lines @xcite . the detected far - ir lines of water vapor are shown in fig . [ obs_iso_agua ] . spectroscopical and observational data are tabulated in table [ tab - water_lines ] . except the @xmath27 ground state line of @xmath47-h@xmath0o at @xmath3179.5 @xmath28 m ( @xmath31669.9 ghz ) , all water lines have a similar profile and are centered at sgr b2(m ) velocities . the @xmath3179.5 @xmath28 m line is saturated and absorbs from 150 to + 100 km s@xmath4 , which therefore includes the water vapor located in the foreground gas toward the gc and the warm gas around sgr b2(m ) . the widespread absorption produced by the @xmath27 line has been previously presented in cernicharo et al . ( 1997 ) and enlarged in g04 ( 9@xmath24@xmath1027@xmath24 ) , while the @xmath29 absorption has been mapped ( 26@xmath24@xmath1019@xmath24 ) by swas ( neufeld et al . 2003 ) . these observations probe that low excitation h@xmath0o is present in the clouds intersecting the line of sight toward large areas of sgr b2 . the average velocity of all h@xmath0o lines observed with the iso / lws fp is @xmath4860@xmath95 km s@xmath4 , in agreement with the velocity of other related oxygen species such as h@xmath25o@xmath26 or oh ( goicoechea & cernicharo , 2001 ; 2002 ) . taking into account the wavelength calibration error of the lws / fp instrument , this velocity is compatible with the expected @xmath4865 km s@xmath4 cloud seen in radio observations @xcite . we note , however , that velocities close to @xmath4860 km s@xmath4 are also associated with gas surrounding most sgr b2 continuum sources , so that the bulk of the h@xmath0o absorption can arise from them . possible overlapping with other molecular species occurs at some wavelengths . in particular , the o 4@xmath494@xmath50 line is blended with hf @xmath6=21 at 121.697 @xmath28 m @xcite and the o 2@xmath511@xmath52 line at 181.053 @xmath28 m has a small contribution from h@xmath25o@xmath26 q(1,1 ) @xcite . observations of the 183.31 ghz h@xmath0o line ( e@xmath53@xmath14205 k ) toward sgr b2 are presented in fig . [ mapa_183 ] . the emission appears at the lsr velocities of sgr b2 with no contribution from the line of sight clouds . the emission appears at least in the 50 - 75 km s@xmath4 range . note the different line shapes and intensities of the 183.31 ghz emission for positions in front and around the main condensations ( here the line appears much wider ) . finally , fig . [ co_7_6_cso ] shows the co @xmath6=76 line ( e@xmath53@xmath14150 k ) observed toward sgr b2(m ) . this co line shows a strong self absorption at @xmath5470 km s@xmath4 ( the velocity of the 183.31 ghz h@xmath0o line peak at his position ) so that the profile peaks at @xmath3 @xmath4855 and @xmath4885 km s@xmath4 . similar patterns are shown by lower - j co lines . the warm gas present in the outer layers of sgr b2(m ) might be expected to radiate in high@xmath6 co lines . the co @xmath6=1413 transition at 185.999 @xmath28 m is the one with the lowest energy level ( e@xmath53@xmath14581 k ) within the range of iso / lws detectors . however , we have not detected any emission / absorption from co ( 3@xmath55 limits are @xmath452@xmath1010@xmath56 w @xmath12 ) in the iso / lws spectra toward the sgr b2 region @xcite , with both grating and fp . co spectroscopical data and line flux upper limits are tabulated in table [ tab - co_lines ] . nevertheless , recent studies of the large scale co @xmath6=76 emission toward the gc with the _ antarctic submillimeter telescope and remote observatory _ ( ast / ro ) have shown that the emission is concentrated toward sgr b and sgr a complexes @xcite . [ co_7_6_cso ] shows higher spectral / angular resolution observations of this line taken with the cso telescope toward the sgr b2(m ) position . hence , it seems that at a given @xmath57 level , the rotational co line emission disappears from the sgr b2 spectrum . = 76 line at 806 ghz toward sgr b2(m).,width=302 ] we have performed nonlocal radiative transfer calculations to try to reproduce the lack of high@xmath6 co lines in the far ir spectrum of sgr b2(m ) , and to help constraining the physical parameters needed to model the water vapor absorption . in fig . [ co_models ] we show the predictions of a nonlocal model for several high@xmath6 transitions of co. the nonlocal model used here is an adaptation of the radiative transfer code developed by gonzlez alfonso & cernicharo @xcite with the inclusion of dust in the transfer @xcite . we have implemented it for high@xmath6 levels of co ( see sec . [ sssct : nonlocal ] for further details on the specific model developed for sgr b2(m ) ) . the collisional rates have been taken from flower ( 2001 ) . the dust continuum emission has been modeled following goicoechea and cernicharo ( 2002 ) and g04 ( see section 4.2.1 ) . we have run an array of models for co abundances of 10@xmath58 and 10@xmath15 with t@xmath1 varied from 100 to 500 k and @xmath17 from 5.0@xmath1010@xmath59 to 6.4@xmath1010@xmath60 @xmath8 . from these results it is clear that in order to suppress the far ir co emission , and to match the co @xmath6=76 line emission , low h@xmath0 density is required . in particular , if the co @xmath6=76 line arises from a layer of gas at t@xmath1@xmath3100 k the model with @xmath17=2@xmath1010@xmath7 @xmath8 correctly reproduces the absence of far ir lines . however , if it arises from the same outer layer of warm oh ( t@xmath1@xmath3300 k ) detected in the far ir @xcite , the limit for the density will be @xmath4510@xmath7 @xmath8 . in any case , the lack of high@xmath6 co lines is an evidence of the low density gas located in front of the far ir continuum source . in the following section we will constrain the physical parameters of this layer by studying the available water vapor lines . the main observational result of this work is that the far ir water lines toward sgr b2(m ) appear in absorption , while the 183.31 ghz line is seen in emission around and in the main condensations . the fact that the continuum emission in the far ir is optically thick @xcite indicates that the h@xmath0o absorption lines arise from regions where the excitation temperatures ( t@xmath61 ) are smaller than the dust temperatures inferred from the continuum emission . however , the problem of the h@xmath0o line excitation toward sgr b2 is not straightforward . apart from self absorption and the possible excitation by collisions with molecules ( e.g. h@xmath0 ) , atoms ( e.g. he ) and @xmath62 ( if the ionization fraction is significant ) , the level population can be primarily determined by the thermal emission of dust , the role of which is essential in the excitation of molecules such as h@xmath0o or oh which have many rotational lines in the far and mid ir . this represents a major difference with respect to the excitation treatment of molecules observed in the radio domain where , generally , one can neglect the excitation by dust photons . if the excitation is dominated by ir photons , the two transitions arising from the ground levels of ortho - water , 2@xmath511@xmath52 and 1@xmath631@xmath52 , will determine how the higher energy levels will be populated . even if collisions are important , the presence of an optically thick far ir continuum will strongly affect the t@xmath61 of water and , thus , it must be carefully taken into account in the models . in the case of sgr b2 , this means that the external dust layers of the cloud will absorb the possible water line emission from the inner regions . knowledge of the geometry of the region to be modeled and the relative filling factors of the dust and gas in the beam of the lws instrument , are also important for the models . for this reason , high angular resolution ground - based observations of the 183.31 ghz water line are particularly important to model the water vapor radiative transfer in sgr b2 . in this section we analyze the h@xmath0o observations with a _ multi - molecule large velocity gradient _ ( lvg ) model . for sgr b2(m ) we have adopted a spherical geometry with two components ( see fig . [ fig_h2o_geometry ] ) : a uniform continuum core with a diameter of @xmath323@xmath44 ( @xmath31 pc for a distance of 8.5 kpc ) and a shell of variable thickness and distance to the core . the presence of an external shell of molecular gas ( not resolved by the iso / lws beam ) surrounding a central condensation is indicated by the analysis of the far ir oh @xcite and nh@xmath25 lines @xcite and it is also supported by h@xmath0o 183.31 ghz line observations ( fig . [ mapa_183 ] ) . in particular goicoechea & cernicharo ( 2002 ) found an angular size of @xmath342@xmath44 for the oh envelope . the bulk of the h@xmath0o absorption lines must arise from the oh layers , or inside them . following the oh model geometry and figs . [ mapa_183 ] and [ co_7_6_cso ] , a total size of @xmath330@xmath44 ( @xmath31.2 pc ) for the core@xmath48shell cloud has been adopted . the core is considered as a gray body with an opacity at 80 @xmath28 m of 2.5 , with a dust opacity law given by @xmath64 $ ] , and a dust temperature of 30 k. these values are consistent with the color temperatures and dust emissivities derived from the analysis of the iso / lws continuum observations at the same wavelengths of the detected far ir h@xmath0o lines @xcite . although dust temperatures can be slightly larger ( t@xmath41@xmath36080 k ) inside the cloud ( from the analysis of the millimeter continuum emission ) or slightly lower ( t@xmath41@xmath320 k ) in the outer and colder diffuse layers ( from the analysis of extended iras continuum emission ; gordon et al . 1993 ) , we judged t@xmath41@xmath1430 k as the most representative value for the dust grains coexisting with the molecular species detected in the far ir . the bulk of the continuum emission detected by iso can be hardly fitted with dust temperatures below 30 k. the emission of higher temperature dust arising from the innermost regions of sgr b2 is hidden in the fir by the huge amount of foregroung gas and colder dust . for these models we have considered an ortho h@xmath0o column density of 1.8@xmath1010@xmath11 @xmath12 and a para h@xmath0o column density of 0.6@xmath1010@xmath11 @xmath12 . these are lower limits suggested by the nonlocal radiative transfer models ( see below ) . the lvg model computes the statistical equilibrium population of the rotational levels for ortho h@xmath0o and para h@xmath0o independently . the collisional rates were scaled from those of h@xmath0o he collisions @xcite . several lvg computations for some selected ortho and para h@xmath0o transitions in the thz domain are shown in fig . [ lvg_mod ] . the excitation temperature t@xmath61 of each transition is shown in each panel as a function of t@xmath1 and @xmath17 . the thick contour corresponds to the equivalent temperature of the continuum core . therefore , to the left of this contour , water lines appear in absorption . from the lvg models it is clear that high density and temperature are required to observe the far ir water lines in emission . for the range of densities implied by the co observations ( @xmath4510@xmath7 @xmath8 ) , the h@xmath0o lines observed by the iso / lws are correctly predicted in absorption . the temperature of the absorbing layer is , however , more difficult to estimate because the possible solutions for a given ( low ) density model are not very sensitive to temperature variations , as indicated by the smooth change of t@xmath61 for constant density as the kinetic temperature goes from 500 to 100 k. lvg models predict that , in addition to dust photons , collisions play a role in the excitation of the lowest h@xmath0o rotational levels . the higher energy levels are pumped from the lowest levels by absorption of far ir photons . in the case of the ortho 1@xmath631@xmath52 line observed by _ swas _ ( neufeld et al . 2000 ) , only moderate densities ( @xmath655@xmath1010@xmath7 @xmath8 ) are needed to observe the line in emission . this is the case of the extended emission in the 1@xmath631@xmath52 line observed in the _ 180 pc molecular ring _ around the gc between @xmath66@xmath3@xmath4880 and 120 km s@xmath4 ( neufeld et al . 2003 ) . for the density conditions derived for the sgr b2 velocity range and for the line of sight clouds , the 557 ghz line is correctly predicted in absorption . a similar behavior is expected for the para 1@xmath671@xmath68 line at @xmath3269.3 @xmath28 m ( @xmath31113 ghz ) that will be observed by future heterodyne instruments such as hifi / herschel . among all the moderate excitation para h@xmath0o lines ( e@xmath69@xmath45450 k ) , only the 3@xmath702@xmath71 appears in the mm domain . contrary to other h@xmath0o maser lines accessible from ground based telescopes , relatively low t@xmath1 and density are required for the 3@xmath702@xmath71 line inversion . these conditions allowed the first detection of extended water emission in orion ( cernicharo et al . 1994 ) , and a strong dependence of the emission with t@xmath1 was revealed . hence , the 183.31 ghz line could be an excellent tracer of the warm gas in molecular clouds . we have used the same lvg model to analyze the 183.31 ghz line inversion mechanism that produces extended emission in sgr b2 ( fig . [ mapa_183 ] ) . [ lvg_mod_183 ] shows different excitation models for the 3@xmath702@xmath71 transition . since a fraction of the 183.31 ghz line emission may arise from the warm envelope in front of the far ir continuum source , models with and without a central continuum source ( the same described at the beginning of the section ) have been considered . for a given t@xmath1 ( from 20 to 500 k ) , each panel shows computations of different para h@xmath0o column densities as a function of t@xmath72 ( in k ) and h@xmath0 volume density . for low temperatures ( t@xmath1@xmath4540 k ) and moderate densities ( @xmath4510@xmath60 @xmath8 ) , the line will be observable for large @xmath73(p h@xmath0o ) values only if a continuum source is present . due to the minor role played by collisions in the pumping of the rotational levels at these temperatures , the line intensity is almost independent of the density . nevertheless , as the temperature increases , collisions start to be significant , and the expected intensity of the 183.31 ghz line becomes less sensitive to the models with or without the continuum source . the 183.31 ghz emission is produced by the population inversion of the 3@xmath70 and 2@xmath71 levels . this is due to the different rates ( far from the thermalization densities ) at which both levels can be populated . the 3@xmath70 level is radiatively connected with the 2@xmath71 and 2@xmath74 lower energy levels . the a@xmath75 coefficient for the 3@xmath702@xmath74 at @xmath3138.5 @xmath28 m is @xmath310@xmath7 times larger than the a@xmath75 of the 3@xmath702@xmath71 transition . the @xmath3138.5 @xmath28 m line is seen in strong absorption in the iso / lws spectra ( see fig . [ obs_iso_agua ] ) . the 2@xmath71 level is radiatively connected with the 2@xmath67 and 1@xmath67 lower energy levels . the 2@xmath711@xmath67 at @xmath3100.9 @xmath28 m produces the strongest absorption of para h@xmath0o in the far ir ( see fig . [ obs_iso_agua ] ) . on the other hand , the 3@xmath70 level is radiatively connected with the 3@xmath76 , 4@xmath76 and 4@xmath77 higher energy levels , while the 2@xmath71 level is connected with the 3@xmath78 ( see the @xmath367.1 @xmath28 m line in fig . [ obs_iso_agua ] ) and with the 3@xmath70 higher energy levels . the different radiative pathways and rates ( for @xmath79@xmath80@xmath81 ) at which the 3@xmath70 and 2@xmath71 levels can be populated produce the inversion . lvg models ( see fig . [ lvg_mod_183 ] ) show that this mechanism can be efficient to produce 183.31 ghz line emission in regions of relatively low density . the observed extended emission at 183.31 ghz has a brightness temperature of 10 k in average ( see fig . [ mapa_183 ] ) . assuming that it arises from the low density regions in which the the far ir continuum sources are embedded , with t@xmath1@xmath82300 - 500 k and @xmath17@xmath310@xmath7 @xmath8 , we derive a value for @xmath73(p h@xmath0o ) of @xmath405@xmath1010@xmath83 @xmath12 . this value is higher than the one derived from the far - ir water lines ( see sect . [ sssct : nonlocal ] ) possibly because the 183.31 ghz line penetrates deeper into this dusty environment . toward the main condensation , the bulk of the emission seems to arise from the cold ( t@xmath8440 k ) and dense gas . under these conditions , a brightness temperature of the line of @xmath3100 k would translate into a p h@xmath0o column density of @xmath310@xmath85 @xmath12 . however , an important contribution could come from the embedded high temperature and high density core condensations where the column density should also be much larger . to take into account the radiative coupling between regions with different physical and/or excitation conditions , the radiative transfer has to be treated with nonlocal techniques , more sophisticated than the lvg approximation . we have adapted the radiative transfer used in sec . 5.1 for co , to the ground vibrational states of ortho and para h@xmath0o respectively . the model includes all the water rotational levels with transitions between 40 @xmath28 m and 183.31 ghz . the level population is computed in statistical equilibrium considering collisional excitation and radiative excitation by line and continuum photons . this is computed consistently assuming that the water molecules and the dust grains are coexistent . the geometry , core@xmath48shell dimensions , are the same of that considered in the lvg models . the shell was divided in 41 layers . the central dust condensation has been modeled with identical parameters than in section 4.2.1 . we have considered 14 rotational levels ( e@xmath86 @xmath45 608 k ) of ortho - water for model with t@xmath1 below 100 k , and up to 30 rotational levels ( e@xmath87 @xmath45 1290 k ) for the higher temperature models . we have adopted a turbulence velocity of 8 km s@xmath4 . the continuum radiation field has been treated as in gonzlez - alfonso & cernicharo @xcite by considering a spectral range of 70 km s@xmath4 centered on each water transition . collisional rates have been taken from green et al . obviously all the water transitions in the continuum core are thermalized to the dust temperature due to the large opacity in the far ir . the computed line profiles are a result of convolving the brightness temperature with the angular resolution of the lws detectors ( @xmath380@xmath44 ) . the resulting spectral resolution of the synthetic water lines is 1 km s@xmath4 , as no convolution with the spectral resolution has been performed . to test the sensitivity of the model to the physical parameters , models were computed with @xmath73(h@xmath0o ) and t@xmath1 of 1.8@xmath1010@xmath11 , 9@xmath1010@xmath11 and 4.5@xmath1010@xmath83 cm @xmath88 , and 40 , 100 , 200 , 300 and 500 k respectively , while densities were increased from 5@xmath1010@xmath89 to 6.4@xmath1010@xmath90 cm @xmath91 multiplying by 2 in each step . the different nonlocal radiative transfer models for the first two h@xmath0o column densities are shown in figs . [ mod_1_h2o ] and [ mod_2_h2o ] . models for other column densities have been also ran . however , the observed absorption depth is not reproduced for column densities below 10@xmath11 cm @xmath88 . in the @xmath73(h@xmath0o)=1.8@xmath1010@xmath83 cm @xmath88 case too many lines would be in emission , contrary to the observations . the main problem to interpret the h@xmath0o absorption toward sgr b2 arises from the large opacities of the far ir water lines , @xmath310@xmath59 , and even @xmath310@xmath7 for the ortho 2@xmath511@xmath52 line at @xmath3179.5 @xmath28 m . under these conditions , many weak water lines have to be detected to constrain the physical conditions and the column density . therefore , we strength that it is difficult to obtain physical parameters from the observation of few far ir water rotational lines . in addition , as the radiative excitation dominates the population of the far ir levels , lines with a weak dependence on the dust excitation should be investigated . taking into account the 14 water lines detected in the far ir , the nonlocal results imply that models are not very sensitive to the temperature and that the only indication about the column density has to be searched in weak far ir h@xmath0o lines or in the 183.31 ghz line ( see sec . even so , the far ir absorption arises in the low density external layers of gas , while it is very likely that the 183.31 ghz line could have an important contribution from inner and denser regions . another complication arises from the fact that for models of low @xmath73(h@xmath0o ) values and t@xmath1@xmath92200 k , it is also difficult to distinguish between different h@xmath0 densities ( see fig . [ mod_1_h2o ] ) . limits to t@xmath1 and @xmath79(h@xmath0 ) have to be searched in weak lines below 70 @xmath28 m . the models for high water column densities ( fig . [ mod_2_h2o ] ) predict absorption lines in the lws range at @xmath356.3 , @xmath357.6 , @xmath358.7 , @xmath378.7 , @xmath399.5 , @xmath3125.4 and @xmath3136.5 @xmath28 m . at the spectral resolution and sensitivity of the lws / fp , none of these lines has been detected . this implies that @xmath73(h@xmath0o)@xmath934.5@xmath9410@xmath83 @xmath12 toward the warm envelope of sgr b2 . the @xmath3136.5 @xmath28 m line ( also predicted by the models with large column densities ) is contaminated by the absorption produced by the c@xmath25 @xmath95(8 ) rovibrational line , which is also predicted by the models of tri atomic carbon @xcite . some of these lines are predicted ( even in emission ) by the models with large @xmath73(h@xmath0o ) . models with @xmath73(h@xmath0o)@xmath821.8@xmath1010@xmath11 @xmath12 are consistent with iso detections and upper limits to other iso lines . only the @xmath367.1 @xmath28 m line is weaker than the observations . therefore , models shown in fig . [ mod_1_h2o ] give a lower limit to the water vapor column density in the outer and warm ( 300 - 500 k ) envelope . taking into account the difficulties implied by the h@xmath0o modeling in the far ir , we found that @xmath96 @xmath12 is the best h@xmath0o column density to fit the iso observations ( see figure [ mod_2_h2o ] ) . the co analysis ( sec . 5 ) , and the studies in the far ir of oh @xcite and the ammonia lines @xcite and our co data , also support that the water absorption lines arise in the warm and low density ( @xmath9210@xmath7 @xmath8 ) layer in front of sgr b2(m ) . the h@xmath0o column density derived from iso observations is below the lower limit of @xmath73(h@xmath0o ) estimated for the 183.31 ghz line for this component ( see sec . [ sssct : lvg ] ) as it is likely that an important fraction of the 183.31 ghz emission arises from the inner and denser regions closer to sgr b2 main cores or even from them , as pointed out above . this component , with a mean h@xmath0 density of @xmath6510@xmath19 @xmath8 , has been traced by the nh@xmath25 non metastable emission lines , t@xmath1@xmath14100 k , @xcite and by the emission of hc@xmath25n rotational lines , t@xmath1@xmath142040 k , @xcite . the determination of the temperature from mm emission lines is also complicated in these regions completely obscured to iso observations . therefore , the observed differences in the 3@xmath702@xmath74 line intensity and shape can be a combination of @xmath73(h@xmath0o ) and/or t@xmath1 variations across the region . the results presented in this section have been compared to those obtained from another radiative transfer based on a different approach ( asensio ramos & trujillo bueno 2003 ; asensio ramos & trujillo bueno 2006 ) . this code is a generalization to spherical geometry of the very fast iterative multilevel gauss - seidel ( muga ) and multilevel successive overrelaxation ( musor ) methods developed by trujillo bueno & fabiani bendicho ( 1995 ) for the case of a two - level atom and generalized by fabiani bendicho , trujillo bueno & auer ( 1997 ) to multilevel atoms in cartesian geometries . the code also allows the application of the standard multilevel accelerated @xmath97-iteration ( mali ) ( olson , auer & buchler 1986 ) . the angular information for the calculation of the mean intensity is obtained by solving the radiative transfer equation along its characteristic curves ( straight trajectories with constant impact parameter ) with the aid of the short - characteristics formal solver with parabolic precision ( kunasz & auer 1988 ) . the statistical equilibrium equations are linearized with the aid of the preconditioning scheme developed by rybicki & hummer ( 1991 , 1992 ) with the introduction of an approximate @xmath97 operator that can be efficiently obtained in the framework of the short - characteristics technique . the convergence rate of the muga and musor schemes is equivalent to that obtained with the introduction of a nonlinear @xmath97 operator , with the advantage of not being necessary neither to calculate nor to invert such nonlinear operator . interestingly , the time per iteration is similar to that obtained for the standard @xmath97-iteration or the mali method . the computing time for the muga scheme is reduced by a factor 4 with respect to mali , while the musor scheme leads to an order of magnitude of improvement in the total computing time with respect to mali . the calculations have been performed with the same geometry , the same molecular data ( collisional and radiative transitions ) , and the same physical conditions of the previously described nonlocal code . the emerging line profiles from both codes are very similar , with differences below 2 - 3 % . both codes predict lines in absorption / emission for the same physical conditions , with identical spectral shapes in the cases where re - emission is found in the line wings . this test of consistency allows us to be very confident in the results presented in figs . [ mod_1_h2o ] and [ mod_2_h2o ] since they were obtained with numerical methods based on completely different approaches . the h@xmath0o ( oh ) column densities derived from iso observations in the warm envelope are within an order of magnitude of the shock model predictions for the same low density region @xcite . nevertheless , @xmath98shocks have been invoked to explain the heating of large amounts of warm gas in the gc and also in the warm envelope @xcite . in addition , the shock models of @xcite correctly reproduce the observed column densities of @xmath73bearing species such as nh@xmath0 and nh@xmath25 that could be ultimately related with the dust grain chemistry @xcite . a qualitative explanation came from the observation of large oh column densities ( h@xmath0o / oh@xmath32 - 4 ) in the warm envelope . according to goicoechea & cernicharo ( 2002 ) , if a far uv radiation field illuminates the outer regions of the cloud , water molecules could be photodissociated producing an enhancement of the oh abundance expected only from its formation through h@xmath25o@xmath26 dissociative recombination . the presence of such an extended far uv radiation field with @xmath38@xmath310@xmath89 - 10@xmath99 is inferred from the [ oiii ] , [ niii ] , [ nii ] , [ cii ] , and [ oi ] extended line emission @xcite . therefore , an important contribution from h@xmath0o photodissociation may explain the observed h@xmath0o / oh abundance ratio in the external layers of the envelope . + to investigate in more detail the o chemistry in the envelope , we have run several photochemistry calculations that include depth dependent photodissociation , h@xmath25o@xmath26 dissociative recombination and neutral neutral reactions . we have used the latest version of the public and available pdr model of le bourlot et al . @xcite . the model does not include oxygen grain surface chemistry , but it is consistent with the low densities and high temperatures found in sgr b2 envelope . following @xcite , we assume a @xmath38=5@xmath1010@xmath89 radiation field and a @xmath100=@xmath79(h)+2@xmath79(h@xmath0)=5@xmath1010@xmath89 @xmath8 density . assuming that the outer gas layers of sgr b2 envelope are directly illuminated by such a radiation field , the model solves the uv transfer and the chemistry up to a@xmath101=20 mag . as a reference model we solve the thermal balance explicitly for an initial gas temperature of 500 k. the resulting h@xmath0o and oh column densities and the h@xmath0o / oh ratio are shown in fig . [ mods_pdr ] as a function of the visual extinction through the cloud . this @xmath102 pdr can only maintain the temperature above @xmath3100 k in the first 2 magnitudes of the cloud , where the uv radiation efficiently photodissociates h@xmath0o to form oh . this results in a low h@xmath0o / oh ratio . inside the cloud , the h@xmath0o and oh production is rapidly dominated by h@xmath25o@xmath26 dissociative recombination and the h@xmath0o / oh ratio tends to a constant value that basically depends on the assumed branching ratio @xmath35 ( we have taken @xmath35=0.25 ) . at least two different scenarios can produce large column densities of gas at high t@xmath1 . the first one is the presence of several clumpy pdrs within iso s beam . in this scenario , far ir h@xmath0o and oh observations will only trace the pdr clump surfaces . the second one is the presence of shocks . it is very likely that widespread low velocity shocks can locally heat the gas within the envelope and preserve a fraction of the molecular gas with temperatures up to @xmath3500 k @xcite , but the same is possible in the clumpy pdr scenario . with these temperatures neutral - neutral reactions play an important role in the chemistry . to simulate this situation , we have run several iso thermal models with temperatures ranging from 500 to 50 k. some selected results are also shown in fig . [ mods_pdr ] . from these models it is clear that both @xmath73(h@xmath0o ) and @xmath73(oh ) are clearly enhanced by neutral - neutral reactions ( fig . [ mods_pdr ] , upper panel ) . if the temperature is high , these models reproduce the observational values much better . hence , a significant fraction of warm gas seems to be needed to reproduce the large h@xmath0o and oh column densities observed toward sgr b2 . this conclusion agrees with the large temperatures derived from oh observations in the far ir @xcite . the gas temperature determines the contribution of neutral - neutral reactions ( 1 ) and ( 2 ) . as the temperature of the gas increases , reaction ( 1 ) contributes to a enhancement of the h@xmath0o / oh ratio ( fig . [ mods_pdr ] , lower panel ) . note that a moderate increase of the radiation field ( to @xmath38=10@xmath7 ) shifts the h@xmath0o / oh curve to larger a@xmath101 inside the cloud until photodissociation becomes less important . in the shielded regions of the cloud , the predicted column densities are similar to those with @xmath38=5@xmath1010@xmath59 . the h@xmath0 column density ( or a@xmath101 ) responsible of the warm gas observed from far ir absorption lines ( the location of the species in optical depth ) has been traditionally difficult to establish . a comparison between the observational h@xmath0o / oh ratio and photochemistry models shows that the bulk of the h@xmath0o / oh absorption must arise from the surface of sgr b2 ( a maximum a@xmath101 of @xmath35 to 10 mag ) , in agreement with the large opacities derived from the radiative transfer models . even assuming an homogeneous surface cloud , most of the water vapor will arise from a@xmath101@xmath4510 mag if t@xmath1@xmath3500 k while @xmath4050@xmath103 of water can be at a@xmath101@xmath6510 mag if t@xmath1@xmath3300 k. therefore , an accurate description of the thermal structure of the cloud ( with far ir diagnostics tracing the same gas ) will be needed to establish more detailed conclusions . still , in a inhomogeneous medium , several pdr - like clump surfaces locally heated to t@xmath1@xmath3300500 k by low velocity shocks could be entirely responsible for the far ir h@xmath0o / oh absorption . taking into account the uncertainties implied in the determination of column densities in the envelope , we take @xmath13(h@xmath0o)@xmath14(1 - 2)@xmath1010@xmath15 as a lower limit . there is a determination of hd column density toward sgr b2 of @xmath310@xmath104 @xmath12 , which translates into @xmath10510@xmath106 @xmath12 ( polehampton et al . however , taking into account the huge absorption by dust at 112 @xmath28 m ( @xmath1074 ) , it is unlikely that in this hd column density the cold gas , where high column densities are expected , is accounted for . we derive , see @xcite , that the h@xmath0 column density in the warm gas is @xmath310@xmath108 @xmath12 . besides , the 183.31 ghz observations imply that the water abundance is at least an order of magnitude lower in the core regions .. however , we underline the importance of further detection of weak h@xmath0o lines and/or lines with a lesser dependence from the dust emission to refine the models and better constrain the physical parameters of the region ( e.g. the temperature ) , that also determine much of the chemistry . specially important will be the input of the herschel / hifi observations for water lines below @xmath32 thz . + another difficult problem is to place the origin of the fuv radiation field revealed by the fine structure line observations in the region @xcite . in principle , fuv photons could arise from the massive stars near the sgr b2(m ) core and/or from another stellar population within the envelope itself ( not resolved yet by observations ) . the permeating effect of the radiation field will be determined by the clumpyness and inhomogeneity of the medium surrounding the stars , and by the energy of the stellar photons . in addition , x rays observations could complement this scenario of widespread low velocity shocks and uv radiation . energetic euv and/or x ray photons can penetrate deeper in the neutral cloud than fuv photons , and thus , could also play a role in the chemistry , as they can also induce many photo ionization and photo dissociation processes . the correlation found in the sgr b complex between the 6.4 kev fe@xmath109 line and the sio emission could also indicate that the x ray sources also drive the shocks in the region @xcite . gas temperatures in xdrs can easily reach t@xmath1@xmath3300500 k because of the more efficient gas heating by x ray induced photoelectrons from the gas ( and not from dust grains as in pdrs ) . under these conditions , neutral reactions dominate the chemistry and especially oh reaches large abundances . however , xdr models predict low h@xmath0o / oh@xmath450.1 abundance ratios @xcite , and thus , they can not be the dominant scenario explaining the far ir water and oh lines toward sgr b2 @xcite . sgr b2 shows diffuse emission in the k@xmath110 line of fe@xmath109 at 6.4 kev @xcite . a dozen of compact x ray sources have been detected within sgr b2 cloud , and they may explain the whole x ray emission found in the region . many of the detected x ray sources are not associated with radio ( hii@xmath111 regions created by massive stars ) nor with ir sources . according to these observations , the intrinsic x ray luminosity toward sgr b2(m ) is @xmath112@xmath9210@xmath113 erg s@xmath4 , which translates into a x ray flux incident on molecular gas of @xmath114@xmath920.10.001 erg @xmath12 s@xmath4 ( assuming @xmath30.11 pc from the x ray source ) . hence , the x ray field within sgr b2(m ) is in the low tail , or even weaker , than those studied by maloney et al . ( 1996 ) . as noted by g04 , the low [ oi]63 @xmath28m/[cii]158 @xmath28 m and ( [ oi]63+[cii]158)/fir intensity ratios observed in the region favor a dominant pdr origin for these lines . both shock @xcite and xdr @xcite models predict larger ratios . we have carried out far ir observations of several thermal absorption lines of water vapor toward sgr b2(m ) and have mapped the 183.31 ghz water line around the main dust condensations of the complex . the main conclusions of this work are the following : 1 . the detected water absorption lines are very opaque and arise from the warm envelope around sgr b2(m ) . the observation of the co @xmath6=76 line and the lack of far ir co lines at iso s sensitivities ( 3@xmath55 limits below @xmath32@xmath1010@xmath56 w @xmath12 ) imply that the density of such layer is @xmath79(h@xmath0)@xmath310@xmath7 @xmath8 . 3@xmath702@xmath74 line at 183.31 ghz shows @xmath340@xmath44@xmath1040@xmath44 extended emission around sgr b2(m ) and @xmath340@xmath44@xmath1020@xmath44 around sgr b2(n ) . this is the first observation of that line in a gc source and represents further evidence that water vapor is extended in warm molecular clouds . lvg and nonlocal radiative transfer calculations have been carried out to extract the water vapor abundance and to constrain the physical parameters of the absorbing / emitting regions . because of the radiative excitation by dust photons , the far ir water lines are not very sensitive to t@xmath1 . taking into account the analysis of the related species oh @xcite , the water absorption must arise from warm gas at similar temperatures , i.e. , from 300 to 500 k. for this warm envelope , we found @xmath73(h@xmath0o)=@xmath96 @xmath12 . an important fraction of the 183.31 ghz emission arises from the inner , denser and colder gas located closer to the main cores . we estimate a water abundance of a few@xmath1010@xmath20 in the denser regions . photochemistry models show that a component of warm gas , @xmath3300 - 500 k , is needed to activate the neutral - neutral reactions and reproduce the large h@xmath0o and oh column densities observed in the envelope . we show that oh and h@xmath0o far ir observations toward sgr b2 are surface tracers of the cloud ( a maximum a@xmath101 of 5 to 10 mag ) . we found @xmath13(h@xmath0o)@xmath14(1 - 2)@xmath1010@xmath15 in these regions . although irradiated by fuv , and possibly more energetic photons , affecting the h@xmath0o / oh ratio in the outermost layers , a clumpy structure for the pdr is needed . alternatively , low velocity shocks could maintain the gas heating through the envelope . due to the complexity of sgr b2 ( and also of the gc ism as a whole ) a multiple scenario is needed to explain the modest angular resolution far ir observations . the input of chemical models , and higher sensitivity and larger spatial resolution far ir observations will lead to a better understanding of the gc environment . we thank spanish dges and pnie for funding support under grants panaya2000 - 1784 , esp2001 - 4516 , aya2002 - 10113-e , esp2002 - 01627 , aya2003 - 02785-e and aya2004 - 05792 . cso observations were supported by nsf grant ast-9980846 . jrg was also supported by the french _ direction de la recherche _ in the latest stages of the work . we thank j. le bourlot for useful suggestions and for his help with the pdr model . zmuidzinas , j. , blake , g. a. , carlstrom , j. , keene , j. , miller , d. , schilke , p. & ugras , n. g. 1995 , in astronomical society of the pacific , airborne astronomy symposium on the galactic ecosystem : from gas to stars to dust , volume 73 , 33 - 40 . cccccccccc p h@xmath0o & @xmath5 & 1635.439 & 183.3 & 205 & 3.53e@xmath11506 & 1120@xmath960 k km s@xmath4 & & 30@xmath93 + o h@xmath31o & @xmath27 & 181.053 & 1655.8 & 79 & 5.45e@xmath11502 & @xmath115(2.68@xmath90.26)e@xmath11518 & 0.73@xmath90.02 & 52@xmath92 + o h@xmath0o & @xmath116 & 180.488 & 1661.0 & 160 & 2.99e@xmath11502 & @xmath115(3.44@xmath90.46)e@xmath11518 & 0.74@xmath90.01 & 64@xmath93 + o h@xmath0o & @xmath27 & 179.527 & 1669.9 & 80 & 5.47e@xmath11502 & @xmath115(6.76@xmath90.29)e@xmath11514 & 0.06@xmath90.03 & 204@xmath923 + o h@xmath0o & @xmath117 & 174.626 & 1716.7 & 163 & 4.94e@xmath11502 & @xmath115(1.86@xmath90.46)e@xmath11518 & 0.82@xmath90.02 & 44@xmath92 + p h@xmath0o & @xmath118 & 138.527 & 2164.1 & 205 & 1.22e@xmath11501 & @xmath115(1.27@xmath90.03)e@xmath11518 & 0.89@xmath90.02 & 41@xmath92 + o h@xmath0o & @xmath119 & 121.719 & 2462.9 & 516 & 1.20e@xmath11501 & @xmath115(5.40@xmath90.29)e@xmath11518 & 0.96@xmath90.01 & 62@xmath99 + o h@xmath0o & @xmath120 & 108.073 & 2773.9 & 160 & 2.52e@xmath11501 & @xmath115(6.07@xmath90.05)e@xmath11518 & 0.64@xmath90.02 & 53@xmath92 + p h@xmath31o & @xmath121 & 102.008 & 2938.9 & 194 & 2.52e@xmath11501 & @xmath115(9.38@xmath92.62)e@xmath11519 & 0.96@xmath90.01 & 72@xmath98 + p h@xmath0o & @xmath121 & 100.983 & 2968.7 & 196 & 2.55e@xmath11501 & @xmath115(6.02@xmath90.11)e@xmath11518 & 0.66@xmath90.01 & 55@xmath99 + p h@xmath0o & @xmath122 & 89.988 & 3331.4 & 297 & 3.45e@xmath11501 & @xmath115(1.99@xmath90.14)e@xmath11518 & 0.90@xmath90.01 & 67@xmath92 + o h@xmath0o & @xmath123 & 75.380 & 3977.0 & 271 & 3.25e@xmath11501 & @xmath115(4.32@xmath90.07)e@xmath11518 & 0.69@xmath90.02 & 51@xmath92 + p h@xmath0o & @xmath124 & 67.089 & 4468.6 & 410 & 1.20e@xmath4800 & @xmath115(1.36@xmath90.08)e@xmath11518 & 0.93@xmath90.03 & 76@xmath98 + o h@xmath0o & @xmath125 & 66.437 & 4512.4 & 376 & 1.22e@xmath4800 & @xmath115(8.39@xmath90.14)e@xmath11519 & 0.92@xmath90.02 & 46@xmath93 + ccccccccc co & @xmath126 & 185.9 & 1611.8 & 581 & 2.95e@xmath11504 & @xmath452.28e@xmath11518 & @xmath451.007 + co & @xmath127 & 173.6 & 1726.6 & 663 & 3.64e@xmath11504 & @xmath451.02e@xmath11518 & @xmath451.003 + co & @xmath128 & 162.8 & 1841.3 & 752 & 4.42e@xmath11504 & @xmath451.14e@xmath11518 & @xmath451.003 + co & @xmath129 & 153.3 & 1956.0 & 846 & 5.31e@xmath11504 & @xmath451.49e@xmath11518 & @xmath451.003 +

for five - dimensional braneworlds with an orbifold topology for the extra dimension , we discuss the validity of recent claims that a gauge exists where the two boundary branes lie at fixed positions and the metric satisfies where labels the transverse dimensions . we focus on models where the bulk is empty apart from a negative cosmological constant , which , in the case of cosmological symmetry , implies the existence of a static frame with schwarzschild - ads geometry . considering the background case with the branes moving apart after a collision , we show that such a gauge can be constructed perturbatively , expanding in either the time after collision or the brane velocity . finally we examine how cosmological perturbations can be accommodated in such a gauge . damtp-2006 - 25

hep-th0604053

the large number of recent publications on braneworld cosmology is a testament to the richness and difficulty of this comparatively new field ( see , for example , @xcite ) . despite the large gap in complexity between possible string / m - theory realisations and the greatly simplified models currently being investigated , even the simplest models are extremely rich in new physics . many such five - dimensional toy models @xcite are motivated by the dimensional reduction of hoava - witten m - theory @xcite to five dimensions but miss out most of the resulting large number of bulk degrees of freedom.the extra dimension is assumed to have the topology of an @xmath0 orbifold , with two 3-branes residing at the boundaries of the spacetime at the two fixed points . the increase in complexity from a purely four - dimensional universe to one confined to a brane embedded in a five - dimensional bulk is considerable . we shall therefore mainly focus on the randall - sundrum model @xcite , where the bulk is empty apart from a cosmological constant . we assume that the branes contain no matter and carry equal and opposite tensions finely tuned with respect to the bulk cosmological constant such that their induced cosmological constants vanish , although the conclusions of this paper will not depend sensitively on these assumptions . in the background case , where the transverse dimensions are assumed to possess cosmological symmetry , the lack of non - gravitational degrees of freedom in the bulk then implies the existence of a _ birkhoff _ frame in which the bulk is manifestly static with anti - de - sitter(ads)-schwarzschild geometry , parameterised by @xmath4 ( related to the black hole mass ) and the lengthscale @xmath5 . the velocities of the branes are then determined entirely by their positions , and the system can be solved exactly even in the presence of matter on the branes ; the only deviation from four - dimensional einstein gravity is the presence in the friedmann equation of quadratic terms in the brane stress - energy and a dark radiation term proportional to @xmath4 , which contains all of the gravitational effects of the bulk . when one comes to include cosmological perturbations @xcite , however , the problem rapidly becomes intractable , largely due to the considerable increase in number of the gravitational degrees of freedom in going from four to five dimensions . to achieve any sort of analytic result on the propagation of braneworld perturbations one needs to resort to approximations . the simplest is to work to lowest order in the brane velocities , assuming also that the energy content of the branes is much less than their tensions . the result is a four - dimensional effective scalar - tensor theory @xcite , valid at low energies , formulated for example in terms of the induced metric of one of the branes . the scalar degree of freedom , known as the radion , corresponds to the distance between the two branes . attempting to go beyond the very limiting low - energy approximation requires a method of solving the five - dimensional einstein equations . while a great deal of progress has been made recently @xcite , the problem is still very difficult . the five - dimensional analysis is rendered simpler by choosing a gauge in which the branes are located at fixed positions . the further assumptions that the coordinate @xmath1 parameterising the extra dimension is chosen such that both the cross terms @xmath6 ( where @xmath7 labels the transverse time and space dimensions ) vanish and @xmath8 result in a particularly useful gauge , henceforth referred to as a _ simple fixed - brane _ ( sfb ) gauge . several publications @xcite have already assumed , without proof , the existence of sfb gauges to derive effective theories with different , more useful regimes of validity than the low - energy scalar - tensor theory . in particular they focused on the small - radion limit pertinent to recent attempts to model the big bang as a brane collision @xcite . this paper is organised as follows . in [ 5d formalism ] we discuss the general problem of constructing sfb gauges , specialising in [ background ] to the special case of background spacetimes with cosmological symmetry . for the case of two branes moving apart after a collision , initially considered in the birkhoff frame , we construct in [ t expansion ] the required coordinate transformation as a power series in the sfb time @xmath9 , equivalent to an expansion in the radion . we show that such a gauge can be found correct to arbitrary orders in @xmath9 , although there is in general no analytic expression for the terms in the series and one must resort to numerical methods . we also comment on the convergence of this series and its relation to the horizon structure of the bulk . a second method is derived in [ c expansion ] , namely an expansion in @xmath10 , which is equivalent to the velocity of the branes . this turns out to be simpler , with analytic expressions available at all orders . finally we discuss cosmological perturbations in [ perturbations ] , and show the existence of the gauge transformations which will accommodate general perturbations into an sfb gauge . we take our spacetime to be an @xmath0 orbifold with two boundary branes at the fixed points . we assume first of all the existence of a gaussian - normal coordinate system relative to the positive tension brane at @xmath11 , with metric [ gn metric ] s^2=^2 + h_(,)^^.throughout this paper greek indices will run over @xmath12 and denote the transverse spatial @xmath13 and temporal @xmath14 coordinates , and @xmath15 etc . are five - dimensional indices running over @xmath12 and @xmath16 for the orbifold dimension . the location of the positive - tension brane at @xmath17 is not crucial for the discussion , of more importance is the form of the metric with @xmath18 and @xmath19 for the sake of simplicity in the following argument . though the coordinate system ( [ gn metric ] ) is particularly simple , the second brane will , in general , lie on a spacetime - dependent trajectory @xmath20 . implementing the israel junction conditions on a moving brane is a complicated procedure @xcite ; far simpler is to use a coordinate system in which the branes lie at fixed positions , at the cost of losing the simple form of the metric ( [ gn metric ] ) . a general coordinate transformation to bring the branes to fixed positions will introduce off - diagonal terms @xmath21 and a non - trivial @xmath22 to the metric . however , the aim of this paper is to show the existence of a ` simplest possible ' sfb gauge where both the non - diagonal metric terms vanish and @xmath22 has no dependence on the extra dimension @xmath23 [ fixed gauge ] s^2=d(x)^2y^2+g_(x , y)x^x^,in which the branes lie at the fixed positions @xmath24 and @xmath25 . the metric coefficient @xmath26 can then be viewed as a four - dimensional scalar field which plays the rle of the radion , giving the proper distance between the branes on curves of constant @xmath27 . note that , as discussed in @xcite , this definition of the radion is frame dependent . as we shall see in [ background dynamics ] , this function is expected to be monotonically increasing ( for the homogeneous cosmological background ) for the case of two branes moving apart after a collision , and vanishes only at the surface of collision . defining the diffeomorphism [ f , g 5d ] ^=f^(x^,y),=g(x^,y),we must then find @xmath28 functions @xmath29 and @xmath30 such that [ diag ] g+h_(f^,g)f^&=&0 + [ k ] ( ) ^2+h_(f^,g)&=&d(x^)^2>0.the new metric is then of the form ( [ fixed gauge ] ) . @xmath31 must then satisfy boundary conditions related to the positions @xmath32 of the branes in the original frame ( [ gn metric ] ) [ g bc ] g(x,0)=_+(f(x,0)),g(x,1)=_-(f(x,1))in order to bring the positions of the branes in the new frame to @xmath24 for the positive - tension brane and @xmath25 for the negative - tension brane . note that there is a residual gauge freedom of @xmath33-reparameterisation which leaves both the form of the metric and the brane positions invariant . this system of non - linear , partial differential equations is extremely difficult to tackle in general . however , since we are ultimately interested in cosmological braneworlds , we will at first only need to consider these equations for the homogeneous case , effectively reducing the dimensionality of the problem from five to two . furthermore , this paper is only concerned with randall - sundrum style braneworlds where the bulk contains only a ( negative ) cosmological constant ; staticity of the bulk then follows in the homogeneous case from birkhoff s theorem . this static _ birkhoff frame _ is of the form ( [ gn metric ] ) and we investigate its properties in the next section . we will describe two methods of constructing the coordinate transformation satisfying ( [ diag ] ) and ( [ k ] ) for the background , and then turn our attention to cosmological perturbations . in this section we re - derive some standard results on the propagation of branes in the static birkhoff frame background @xcite , with metric [ birkhoff ] s^2=y^2-n^2(y)t^2+a^2(y)^2 where [ a ] a^2(y)&=&e^-2 y / l+ e^2 y / l + [ n ] n^2(y)&=&a^2-a(y)=-n(y)/l the bulk geometry is schwarzschild - ads , with the ads lengthscale @xmath5 related to the cosmological constant @xmath34 by @xmath35 and the parameter @xmath4 proportional to the mass of the bulk black hole . we assume that the brane tensions are fine - tuned to the usual values @xmath36 where the five - dimensional gravitational constant has been set to unity for convenience . the spatial curvature is assumed to vanish for simplicity . in this frame the branes will not be static , but follow trajectories @xmath37 . assuming for the moment that the branes are empty , the israel junction conditions are simply [ junction ] k^a _ b=-^a_bwhere @xmath38 are the extrinsic curvatures of the branes . the space - space components of this equation then give the first - order equations of motion @xcite [ dydt ] ( ) ^2=cwhere @xmath39 etc . since we are interested in brane collisions we assume that @xmath40 are of opposite signs , specifically that @xmath41 and @xmath42 . the proper time @xmath43 on the branes can then be determined , giving the world velocities and accelerations as [ world velocity ] = & , & = + [ acceleration ] = -&,&=-the above equations imply that , for empty , fine - tuned , spatially - flat branes , one can not achieve a consistent embedding into the negative mass schwarzschild - ads spacetime with @xmath44 . it also implies that , under the same conditions , the @xmath45 solution is the trivial case where the branes remain static and the frame reduces to the usual ads warp - factor profile , although from ( [ acceleration ] ) we see that the branes are still accelerating . we shall therefore henceforth only consider the case of _ positive _ spacetime mass @xmath46 . in this case there is a horizon in the bulk where @xmath47 vanishes , given by [ horizon ] y_=.the coordinate system ( [ birkhoff ] ) will only be valid in a particular coordinate patch bounded by the horizon . one would therefore expect any results in this paper derived from this coordinate system only to be valid up to the point where one of the branes hits the horizon ( in other words , for the physical slice of spacetime bounded by the two branes to develop a horizon ) . since @xmath48-translation is a symmetry of the metric ( [ birkhoff ] ) we can assume that the brane collision occurs at @xmath49 , @xmath50 . 1 shows the branes trajectories in the birkhoff frame with @xmath51 . note that a brane crossing the horizon will do so in finite proper time but will require an infinite @xmath52 , since from ( [ dydt ] ) we see that the brane velocity with respect to the birkhoff time @xmath48 tends to zero as the horizon is approached . the corresponding picture as a penrose diagram is illustrated in fig . 2 , showing the region of the extended schwarzschild - ads spacetime corresponding to the coordinate patch ( [ birkhoff ] ) . in this picture , the horizon at @xmath53 can be interpreted as the event horizon of the bulk black hole . note that the physical spacetime is just the slice bounded by the two branes ( or , rather , two copies of it due to the orbifold nature of the complete spacetime ) . importantly , it follows from ( [ dydt ] ) and ( [ world velocity ] ) that the friedmann equation on the brane is simply @xmath54 , i.e. the background geometry is finite at the collision @xcite . the event is , however , singular from a five - dimensional perspective , in that the fifth dimension momentarily vanishes . in general perturbations will diverge due to the blue - shifting of their energies as the radius of the fifth dimension vanishes , although several authors recently have claimed that this need not be an obstacle to the propagation either of cosmological perturbations @xcite or , more fundamentally , string excitations @xcite across the bounce . . the shaded area denotes the region of extended schwarzschild - ads covered by the coordinate system ( [ birkhoff ] ) . the trajectory of the positive- tension brane is denoted by the solid ( dashed ) line . the physical region of spacetime is that bounded by the two branes . the diagonal lines bounding the coordinate patch represent the horizon @xmath53 . region i therefore represents the portion of the future of the collision which the @xmath55 coordinates cover.,width=340 ] . the shaded area denotes the region of extended schwarzschild - ads covered by the coordinate system ( [ birkhoff ] ) . the trajectory of the positive- tension brane is denoted by the solid ( dashed ) line . the physical region of spacetime is that bounded by the two branes . the diagonal lines bounding the coordinate patch represent the horizon @xmath53 . region i therefore represents the portion of the future of the collision which the @xmath55 coordinates cover.,width=302 ] we now simplify the analysis of [ 5d formalism ] to the homogeneous , static background of [ background dynamics ] . the metric components ( [ birkhoff ] ) and the brane trajectories ( [ dydt ] ) do not , by homogeneity , depend on the spatial coordinates @xmath13 . we can therefore just leave these coordinates alone , , and focus on the two - dimensional @xmath55 portion of the metric . as before , we consider a diffeomorphism [ fg 2d ] t = f(y , t ) , y = g(y , t).the conditions ( [ diag ] ) and ( [ k ] ) on the form of the new metric become [ eqn 1 ] g_y g_t&=&n(g)^2 f_y f_t + [ eqn 2 ] g_y^2&=&n(g)^2 f_y^2+d(t)^2,giving [ fixed metric ] s^2=d(t)^2y^2 - b(y , t)^2t^2 + a(y , t)^2 ^2where @xmath56 it is at this point that the assumption that the metric ( [ birkhoff ] ) is static ( i.e. that we are concerned only with randall - sundrum - style braneworlds , with no non - gravitational degrees of freedom in the bulk ) comes in . this implies that @xmath57 is only a function of @xmath30 , rather than @xmath58 as would be expected more generally . one can then eliminate @xmath59 entirely using ( [ eqn 1 ] ) and ( [ eqn 2 ] ) , giving ( up to an arbitrary sign ) [ fy ] f_y&= & + [ ft ] f_t&=&the jacobian of this coordinate transformation must not vanish , which implies from ( [ fy ] ) and ( [ ft ] ) that @xmath60 we can therefore conclude that @xmath61 away from the collision which , by convention , we shall assume occurs at @xmath50 . in the @xmath62 plane , the collision corresponds to the vanishing of the metric component @xmath63 , and also to the curve @xmath64 . this curve has zero proper length , lying at constant @xmath9 , and wlog we can take the collision to correspond to the line @xmath65 on which @xmath59 vanishes and @xmath30 takes the constant value @xmath66 . + as we saw in [ background dynamics ] , the distance between the two branes is a monotonically - increasing function after the collision . this implies that one can exploit the @xmath9-reparameterisation of the metric to fix [ t gauge fix ] d(t)=t.it turns out that this gauge choice ensures the existence of a perturbative expansion of the functions @xmath59 and @xmath30 in powers of @xmath9 , as well as corresponding to the linear approximation for close branes used in @xcite . + continuity implies that @xmath67 , which from ( [ fy ] ) and ( [ ft ] ) reduces to a second - order partial differential equation for @xmath68 : [ g ] g_t g_yy+t-+(g_tg_y^2-t^2 g_t)=0writing ( [ dydt ] ) as @xmath69 we find [ gt0 ] .g_t|_y=0&=&y_+(f(0,t)=-.v(g)f_t|_y=0 + [ gt1 ] .g_t|_y=1&=&y_-(f(1,t)=+.v(g)f_t|_y=1substituting ( [ ft ] ) for @xmath70 we find , for example , .g_t|_y=0&=&-.|_y=0 + .g_y^2|_y=0&=&.|_y=0using ( [ n ] ) and ( [ dydt ] ) then gives the boundary conditions [ gy bc ] . g_y|_y=0,1=t.|_y=0,1together with @xmath71 note that the sign of @xmath72 in ( [ gy bc ] ) is specified by the choice @xmath73 , @xmath74 , i.e. that the orientation of the new coordinate system is the same as the old one . the fact that the branes are moving in opposite directions , not immediately apparent from the boundary conditions since the @xmath75 sign in ( [ gt0 ] ) has been lost , follows from the choice of gauge ( [ t gauge fix ] ) . we now proceed to solve ( [ g ] ) subject to the boundary conditions ( [ gy bc ] ) as a perturbative expansion in powers of @xmath9 . specifically , we write [ power ] g(y , t)=y_0+_n=1^g_n(y ) t^nwhere the boundary conditions ( [ gy bc ] ) will specify the derivatives of @xmath76 at @xmath77 in terms of the values there of the lower - order @xmath78 @xmath79 . ( [ g ] ) becomes a set of ordinary differential equations for the @xmath76 ; the aim therefore is to identify the correct initial value @xmath80 for which the two boundary conditions are simultaneously satisfied . once @xmath81 has been determined to the required order , @xmath82 can in principle be recovered from ( [ ft ] ) with boundary condition @xmath83 . the @xmath84 contribution to ( [ g ] ) gives [ g1 ] g_1 g_1-g_1 ^ 2 + 1=0with ( [ gy bc ] ) implying [ g1bc ] g_1(0)=g_1(1)=where @xmath85 etc . we then obtain the result [ g1 ] g_1(y)=((2y-1)),=the corresponding result for @xmath59 is f_1(y)=((2y-1)).this is just the relevant coordinate transformation in the approximation of constant brane velocities , as detailed in @xcite . before examining the existence of @xmath76 for arbitrary @xmath57 , we consider the simpler case of @xmath86 . [ g2 ] g_1 g_2 - 2g_1 g_2 + 2g_2 g_1+(g_1 g^2_1-g_1)=0expanding out [ boundary series ] k(g(y , t))=k(y_0)+tg_1(y)k(y_0)+o(t^2)for an arbitrary analytic function @xmath87 , we identify the boundary conditions as @xmath88 giving [ g2bc ] g_2(1)=-g_2(0)=.there is no simple analytic general solution to this equation , so one must proceed numerically . the task is to find the value of @xmath89 such that , after integrating ( [ g2 ] ) subject to the initial conditions at @xmath24 , @xmath90 is given by ( [ g2bc ] ) , for example using a bisection procedure . the existence of @xmath91 is discussed below . for tuned tensions , i.e. the boundary conditions ( [ gy bc ] ) , the @xmath76 all have definite parity @xmath92 about @xmath93 . to show this , we substitute the power series ( [ power ] ) into ( [ g ] ) and consider the coefficient of @xmath94 , giving [ gn ] g_1 g_n - 2 g_1 g_n+ n g_1 g_n = p_n(y)where @xmath95 is composed of a finite sum of products of the @xmath78 and their first or second derivatives , but only containing @xmath78 for @xmath96 ( c.f . ( [ g2 ] ) ) . examining the contributions from the various terms in ( [ g ] ) individually , we write symbolically ( i.e. ignoring numerical constants ) @xmath97 where the symbol @xmath98 denotes the terms in the coefficient of @xmath94 which do not contain @xmath76 , i.e. the contribution to the rhs of ( [ gn ] ) . working under the inductive hypothesis that @xmath99 has parity @xmath100 for @xmath101 ( which is obviously true for @xmath102 and @xmath103 ) , we see that each summand has the parity @xmath104 . similarly , @xmath105 makes a contribution @xmath106 again , the parity of these terms is @xmath104 by the inductive hypothesis . next , we examine @xmath107 , which is multiplied by a function of @xmath30 in ( [ g ] ) , so we will need all orders up to @xmath57 in its expansion . we find that g_t g_y^2&~&_k=2^t^k ( _ m=1^k-1 _ p=1^k - m g_m g_p g_k+1-m - p ) + [ term 3 ] & ~ & _ k=2^n t^k h_k(y)+o(t^n+1)where @xmath108 is a function of parity @xmath109 involving only @xmath99 for @xmath110 ; there is no contribution at order @xmath94 or below from @xmath76 . @xmath111 is obviously also of this form , so we just need to look at the @xmath112 term . any function @xmath113 with a power series expansion can be written symbolically in the form [ function expansion ] k(g)~k(y_0)+_k=1^t^k(_m=1^k _ i_j>0 , i_j = kg_i_1 ... g_i_k),which can be re - expressed as @xmath114 where @xmath115 is a sum of products of @xmath78 with @xmath116 , with total parity @xmath117 . putting this all together , we see that the rhs of ( [ gn ] ) can be written as p_n&~ & _ m=2^n-1g_m g_n+1-m + _ m=2^n-1 g_mg_n+1-m + & & + _ m=0^n-2g_m(y ) h_n - m(y).each summand in each term has the same parity , @xmath104 . therefore , looking back to ( [ gn ] ) , we see that the whole equation is invariant under parity if @xmath76 has parity @xmath92 . whether or not this is the case depends on the boundary conditions ; in the fine - tuned case , the boundary conditions are just that @xmath118 where @xmath119 and @xmath120 note that this equation implies that @xmath121 at the boundaries is specified entirely in terms of the values there of the @xmath99 for @xmath101 . from ( [ function expansion ] ) , @xmath122 has parity @xmath123 . hence [ boundary parity ] g_n(1)=(-1)^n-1 g_n(0)and the inductive hypothesis is proven . one can now show the existence of @xmath76 for @xmath57 odd - since the function will be odd , it and its second derivative must vanish at @xmath93 . the equation of motion will then prescribe the value of @xmath124 which , together with @xmath125 , will give a unique solution . for even @xmath57 , one must use a more general argument . we are searching for the value of @xmath80 which gives a particular value for @xmath126 . taking @xmath127 for @xmath128 very large , but @xmath129 to its finite , prescribed value , the lhs of ( [ gn ] ) will dominate over the rhs apart from at the finite number of points where either @xmath76 or its derivatives vanish . therefore , [ gn asymptotic ] g_n(y)n h_n(y ) n = g_n(0)+where @xmath130 is the solution to the initial value problem @xmath131 provided that @xmath132 ( which can be shown numerically for specific values of @xmath57 ) , we then have that @xmath133 as @xmath134 , @xmath135 as @xmath136 . hence , by continuity , there will be ( at least one ) intermediate value of @xmath80 where @xmath126 takes the required value . hence we can always find a @xmath76 satisfying both ( [ gn ] ) and the two boundary conditions for @xmath121 at @xmath77 following from ( [ gy bc ] ) . the above argument is not sensitive to the precise nature of the boundary conditions ; in other words , in the case of general brane tensions and brane matter content , the boundary conditions ( [ gy bc ] ) would be modified , but one would still expect to be able to find suitable functions @xmath137 to perform the coordinate transformation to the sfb frame . however , the entire analysis is dependent on the existence of a static bulk frame such as ( [ birkhoff ] ) ( though the precise nature of the functions @xmath138 and @xmath57 will , again , not affect the argument ) . it is not clear if this result could be generalised to more complicated braneworlds where the bulk is not empty , such as dilatonic or supergravity - inspired models with bulk scalar fields @xcite . to @xmath139 plotted for @xmath140 . note the alternating parities.,width=340 ] up to @xmath141 with the exact result . the dotted curves correspond to the exact trajectories as in fig . 1 , showing highly accurate agreement with the numerical results . for comparison the corresponding trajectories computed working only to first order in @xmath9 are also shown for the same range of @xmath9 , with clear differences.,width=340 ] the horizon structure of the bulk has ramifications for the regions of validity of the coordinate system and the radius of convergence of the power series ( [ power ] ) . for @xmath46 there is a horizon , whilst for @xmath44 there is a naked singularity and no consistent embedding of flat , empty , finely - tuned branes . the @xmath45 case , for fine - tuned branes , represents the trivial evolution where the branes remain fixed at their original positions , and the birkhoff frame automatically satisfies the required gauge conditions ( [ fixed gauge ] ) . we therefore only need consider the case @xmath46 . the first six non - trivial @xmath76 have been computed numerically using a bisection procedure to search for @xmath80 in the specific case of @xmath142 , @xmath143 . the resulting @xmath91 to @xmath139 are plotted in fig . 3 . fig . 4 gives the calculated brane trajectories back in the @xmath144 frame ; for example , @xmath145 gives @xmath146 , and integrating ( [ ft ] ) with @xmath147 gives @xmath148 , hence one can plot the trajectories @xmath149 and compare them with the exact solutions . the accuracy of the agreement is high , as can be seen from the fig . also plotted are the trajectories calculated using only the first - order expression for @xmath81 , with the same range of @xmath9 ; these clearly differ wildly from the higher - order results as the horizon is approached . depending on the initial conditions , either the positive- or negative - tension brane will hit this horizon at a finite value of @xmath150 . for the initial conditions chosen above , this corresponds to @xmath151 . at this point , @xmath152 will vanish at the relevant boundary , and so ( [ g ] ) will break down and one would not expect the coordinate system ( [ fixed metric ] ) to be valid beyond this value of @xmath9 . in particular , one would expect this value of @xmath9 to correspond to the radius of convergence of the series ( [ power ] ) . it is clear from fig . 3 that the boundary values of @xmath137 are increasing in modulus as @xmath57 increases , which is a manifestation of the lack of convergence of the series for large @xmath9 in this case . in the fine - tuned , empty - brane case with boundary conditions ( [ gy bc ] ) , the symmetry of the problem ( i.e. the manifest reflection symmetry under @xmath153 in the trajectory plots ) implies that the same coordinate system will also be valid before the collision , for @xmath154 . the advantage of an expansion in the sfb time @xmath9 is that it is ideally adapted to working in the limit of close branes , a regime of considerable interest recently in the literature @xcite . in effect , an expansion in @xmath9 is equivalent to an expansion in the distance between the two branes . working even just to first order in @xmath9 or @xmath155 whilst making no other approximations , one can obtain interesting , non - perturbative results unattainable using the moduli space approximation @xcite . whether or not the full series ( [ power ] ) converges is then not important if one simply wants to proceed perturbatively in @xmath9 . an alternative method of finding the required gauge ( [ fixed gauge ] ) is to consider an expansion in the velocity of the branes ( [ dydt ] ) . specifically , we define the parameter [ h ] h = and expand in this rather than in @xmath9 . this amounts to an expansion about the @xmath45 ads spacetime where the ( finely - tuned ) branes are at fixed positions in the birkhoff frame ( [ birkhoff ] ) . this method turns out to be conceptually and analytically simpler than the expansion in @xmath9 , though it is of less direct use for the near - brane regime , being an expansion about a static configuration . in this sense , working to linear order in @xmath156 will reproduce the results of the moduli space approximation . as discussed below , we will in fact expand about the state where the branes are static and coincident , in order to simplify the analysis somewhat and to tie in with the rest of the paper assuming that the branes are moving apart from an initial collision . it would not , however , be difficult to generalise the results to a different initial configuration . firstly we define a new coordinate @xmath157 , in terms of which the birkhoff frame metric can be written [ h birkhoff ] s^2&=&z^2 - n(z)^2t^2 + a(z)^2 ^2 + a(z)^2&=&z^2 + + n(z)^2&=&a(z)^2-for @xmath158 the branes are static , at @xmath159 and @xmath160 . since we are interested in modelling the aftermath of a collision we shall take @xmath161 . whilst simplifying the resulting analysis considerably one must then remember that , for @xmath158 , the metric will be singular . therefore we shall expect @xmath162 in the sfb frame ( [ fixed metric ] ) to vanish for @xmath158 . bearing this in mind , we then perform a coordinate transformation as before , keeping the @xmath9-gauge general for the time being ( as opposed to the choice @xmath147 of the previous section ) . expanding in @xmath156 , we write [ z ] z(y , t)&=&z_0(1+_n=1^h^n z_n(y , t ) ) + [ t ] t(y , t)&=&t+_n=1^h^n t_n(y , t),where for @xmath158 we can set @xmath163 since the branes are already at fixed positions . we then solve @xmath164 order by order to determine the functions @xmath165 and @xmath166 , together with the boundary conditions ( [ dydt ] ) , [ z bc ] = z _ in the @xmath62 frame , the metric becomes [ yt metric ] s^2&=&(z_y^2-n(z)^2t_y^2)y^2 + & & + 2(z_y z_t - n(z)^2t_y t_t)yt + & & + ( z_t^2-n(z)^2t_t^2)t^2+a(z)^2 ^ 2and so , using ( [ z ] ) and ( [ t ] ) , we obtain [ dygyy1 ] _ y g_yy&=&2(l^2 _ 1 z_1-z_0 ^ 2 _ 1 t_1 ) h^2+o(h^3 ) + [ gyt1 ] g_yt&=&-z_0 ^ 2 _ 1h+o(h^2).next we examine the boundary conditions ( [ z bc ] ) , & = & = z_0 _ 1(0,t ) h+o(h^2 ) + & = & z_+=+o(h^2)giving [ z1bc1 ] _ 1(0,t)=and , similarly , [ z1bc2 ] _ 1(1,t)=-.([gyt1 ] ) gives the simple condition that @xmath167 and , choosing the collision to occur at @xmath65 , implies [ t1 ] t_1(y , t)=0.the condition ( [ dygyy1 ] ) on @xmath162 reduces to @xmath168 which , together with the boundary conditions ( [ z1bc1 ] ) , ( [ z1bc2 ] ) and @xmath169 , gives [ z1 ] z_1(y , t)=(1 - 2y)therefore , to @xmath170 , the required coordinate transformation is [ order h result ] z(y , t)&=&z_0+(1 - 2y)+o(h^2 ) + t(y , t)&=&t+o(h^2).the metric ( [ yt metric ] ) is then given by [ order h metric ] s^2&=&(4+o(h^3))y^2 + o(h^2)y t + & & + ( z_0 ^ 2 + 2(1 - 2y)h+o(h^2))(-t^2+^2)which , to @xmath170 , is of the required form ( [ fixed gauge ] ) . as a check , we can calculate the proper times on the two branes , [ proper times ] t_=(z_0+o(h^2))tgiving the hubble parameters as [ hubble ] h_=+o(h^2).the function @xmath63 , giving the proper distance between the branes along lines of constant @xmath23 , can be read off from ( [ order h metric ] ) as [ d(t ) order h ] d(t)=h+o(h^2)which gives [ check ] h_=d(t)+o(h^2),in agreement with the low - energy result of @xcite . the analysis at higher orders is straightforward . the constraint @xmath171 reduces to an equation of the form [ z general n ] z_n(y , t)=_i , j^ c^(n)_ij t^i y^jwhere the superscript means that the sum is finite , i.e. a binomial in @xmath23 and @xmath9 . this can easily be integrated to obtain @xmath165 up to two arbitrary functions of @xmath9 , which are fixed by the boundary conditions ( [ z bc ] ) and the requirement that @xmath172 . similarly , the other constraint @xmath173 is of the form [ t general n ] t_n(y , t)=_i , j^ d^(n)_ij t^i y^jgiving @xmath166 up to an arbitrary function of @xmath9 . this is a manifestation of the @xmath9-reparameterisation invariance , and can be fixed for example by setting @xmath174 . in this particular gauge , the next two sets of functions are [ z2 ] z_2(y , t)&=&y(y-1 ) + [ t2 ] t_2(y , t)&=&-y(1-y ) + [ z3 ] z_3(y , t)&=&\{4(l^2+t^2 z_0 ^ 2)(- ) . + & & . + y(l^2-t^2 z_0 ^ 2)-l^2 } + [ t3 ] t_3(y , t)&=&(9 - 12y+8y^2 ) . this transformation satisfies all the required boundary conditions , as well as the conditions on the metric ( [ fixed gauge ] ) , up to @xmath175 : @xmath176 in order to prove the relations ( [ z general n ] ) and ( [ t general n ] ) , we use induction . specifically , we aim to prove that the general form of @xmath177 and @xmath178 is a binomial function of @xmath9 and @xmath23 of the form @xmath179 from ( [ z1 ] ) and ( [ z2]-[t3 ] ) , we see that this is indeed the case for @xmath180 . we now assume that we have found the @xmath181 and @xmath182 for @xmath101 , satisfying the required boundary conditions . we then look at the constraint that @xmath162 is independent of @xmath23 , which becomes [ dygyy ] z z_y z_yy = z_y^3+(n(z)n(z ) z^3 z_y t_y^2+n(z)z^3t_y t_yy).we aim for an equation of the form ( [ z general n ] ) and , examining the lhs of this equation , we see that the lowest order in @xmath156 at which @xmath183 appears is at @xmath184 , z z_y z_yy & & ( z_0 ^ 3 z_1z_n+t^2 b(y , t ) ) h^n+1 + [ zn term ] & = & ( -z_0 ^ 2 z_n+t^2 b(y , t))h^n+1where the symbolic notation @xmath185 represents a binomial function in @xmath9 and @xmath23 depending only on the binomial functions @xmath181 for @xmath101 . the factor of @xmath186 multiplying this function is displayed explicitly since we need to keep track of powers of @xmath9 to ensure the rhs of ( [ z general n ] ) is indeed a binomial with no negative powers of @xmath9 . now starting on the rhs of ( [ dygyy ] ) , we note that @xmath187 and hence , at @xmath184 , does not contain @xmath188 , only the lower @xmath181 . hence we can write symbolically @xmath189 both @xmath190 and @xmath191 have a factor @xmath192 in front , and can be written with the inductive hypothesis as [ tt ] t_y^2 , t_yt_yy ~ t^2 h^4 b(y , t)+ t^2 o(h^n+2).furthermore , as in ( [ function expansion ] ) , a general analytic function of @xmath193 can be written [ f(z ) ] f(z)=f(z_0)+_k=1^h^k(_m=1^k_i_j>0,=k z_i_1 ... z_i_k)combining ( [ tt ] ) and ( [ f(z ) ] ) , we see that the term with coefficient @xmath194 on the rhs of ( [ dygyy ] ) contributes , at @xmath184 , a term of the form @xmath195 . hence , putting all of the above together , we have shown that @xmath177 is given in terms of the @xmath181 for @xmath101 by an equation of the form ( [ z general n ] ) with an additional pre - factor of @xmath9 . this equation can always be integrated to give [ expression for z_n ] z_n(y , t)=t b_n(y , t)+a_n(t ) y+b_n(t)where @xmath196 is a binomial expression depending on the @xmath181 for @xmath101 , and @xmath197 and @xmath198 are arbitrary functions , which are then uniquely fixed by the boundary conditions . unlike the expansion in @xmath9 , we can in principle calculate all the @xmath177 analytically , though this is not a terribly helpful procedure . what is important is that the boundary conditions are trivial to implement - ( [ z bc ] ) will prescribe values for @xmath188 at @xmath24 and @xmath25 , which , together with @xmath172 , fix @xmath197 and @xmath198 uniquely from ( [ expression for z_n ] ) . to show the existence of the @xmath178 , we need to examine the equation [ t equation ] t_yt_t = z_y z_twhich follows from the metric ( [ yt metric ] ) . @xmath178 enters the lhs at @xmath199 , @xmath200 where again @xmath185 stands for binomial terms composed of already - known functions , the @xmath182 for @xmath101 in this case . in fact the rhs of ( [ t equation ] ) does not contain @xmath177 at @xmath199 , since both @xmath201 and @xmath202 are @xmath170 . therefore , the equation for @xmath178 is of the form ( [ t general n ] ) , specifically @xmath203 with the factor of @xmath9 coming from the @xmath201 . this can be integrated to [ expression for t_n ] t_n(y , t)=yt _ n(y , t)+d_n(t)where once again @xmath204 is a binomial function of @xmath23 and @xmath9 . the arbitrary function @xmath205 encodes the @xmath9 reparameterisation invariance of the gauge . because of the pre - factor @xmath23 in ( [ expression for t_n ] ) , the gauge @xmath206 corresponds to setting @xmath207 . we have therefore shown that there exist simple , binomial expressions for @xmath166 and @xmath165 for all @xmath57 , with an extremely simple if tedious procedure for finding them explicitly . it should be emphasised , as in [ t expansion ] , that the precise nature of the boundary conditions are not important . hence , although the explicit solution constructed in this section corresponds to empty , fine - tuned branes , it would be simple in principle to repeat the process for more general situations with both de - tuned branes and matter at the background level . we now consider more general spacetimes , where the spatial homogeneity is broken by cosmological perturbations . our starting point is to consider a general background spacetime , taking the metric to be of the form [ gn background ] s_0 ^ 2=y^2-n(y , t)^2t^2+a(y , t)^2 _ ijx^i x^jwhere @xmath208 is the relevant maximally - symmetric spatial metric of constant curvature . in this frame the branes are taken to have loci @xmath37 . we then assume that a coordinate transformation [ y and t ] t = f(y , t),y = g(y , t)can be constructed as described previously , satisfying [ cond 1 ] g_y g_t - n(f , g)^2 f_y f_t = 0,[cond 2 ] g_y^2 - n(f , g)^2 f_y^2 = d(t)^2,with boundary conditions @xmath209 @xmath210 bringing the metric into the standard form [ fixed background ] s_0 ^ 2=d(t)^2 y^2 - b(y , t)^2 t^2 + a(y , t)^2 _ ij x^i x^jwith the branes fixed at @xmath24 and @xmath25 ( the precise forms of @xmath211 and @xmath212 are not important ) . note that , in this section , we shall not fix the @xmath9-reparameterisation invariance as in ( [ t gauge fix ] ) , which will allow us to consider more general situations than the branes moving apart after a collision , such as perturbations about static branes . we now allow the presence of cosmological perturbations , both to the background metric ( [ gn background ] ) and , in the case of scalar perturbations , to the positions of the branes themselves . we then attempt to find an infinitesimal gauge transformation which will not only retain the properties [ requirement ] g_5=g_55=0of the metric , but bring the brane positions back to @xmath24 and @xmath25 if necessary . firstly , we shall perform the coordinate transformation ( [ y and t ] ) to bring the background into the form ( [ fixed background ] ) , before constructing the gauge transformation required to accommodate both the brane positions and the bulk metric perturbations into the sfb gauge at first order . note that , in this section , we are still not considering any non - gravitational degrees of the freedom in the bulk , which would in principle also need perturbing , such as a bulk scalar field . firstly , we consider the simplest case of tensor perturbations . in five dimensions , the most general infinitesimal tensor perturbation @xcite to ( [ gn background ] ) only affects the transverse spatial components , i.e. [ tensor ] g_ija(y , t)^2(_ij+h_ij)x^i x^ j.this will therefore not affect the properties ( [ requirement ] ) of the metric in the new frame ( [ fixed background ] ) . furthermore , the brane trajectories are scalars , and are therefore not affected by a purely tensor perturbation . hence tensor perturbations can trivially be accommodated into the gauge ( [ fixed background ] ) . there are six vector degrees of freedom for five - dimensional perturbations about the background ( [ gn background ] ) , encoded in three divergenceless vectors : [ vector ] s^2&=&s_0 ^ 2 + q^(t)_i x^i t + q^(y)_i x^i y + & & + a(y , t)^2 f_(i|j)x^i x^jwhere @xmath213 denotes the covariant derivative with respect to the metric @xmath208 and @xmath214 . performing the background coordinate transformation ( [ y and t ] ) , we obtain the perturbed version of ( [ fixed background ] ) s^2&= & d(t)^2 y^2 -b(y , t)^2t^2 + ( q^(t)_i f_y + q^(y)_i g_y ) y x^i + & & + ( q^(t)_i f_t + q^(y)_i g_t ) t x^i + & & + a(y , t)^2(_ij+2 f_(i|j))x^i x^jtherefore , in order to satisfy the gauge conditions ( [ requirement ] ) , we require the linear combination [ vector combination ] q^(t)_i f_y + q^(y)_i g_yto vanish . under the most general infinitesimal vector gauge transformation @xmath215 the three vector perturbations transform as q^(y)_i&=&-a(y , t)^2 _ y _ i + q^(t)_i&=&-a(y , t)^2 _ t _ i + f_i&=&-_iwhere @xmath216 . there are therefore two gauge - invariant vector perturbations ( i.e. four degrees of freedom ) corresponding to @xmath217 and @xmath218 , and any one vector perturbation can be set to zero . in particular , we can remove the linear combination ( [ vector combination ] ) with suitable @xmath219 satisfying @xmath220 again there is no vector perturbation to the brane trajectories , hence these still lie at @xmath24 and @xmath25 . scalar perturbations require a more detailed treatment , both because there are more of them in general and because one must also assume that the brane positions are perturbed . in five dimensions there are seven scalar degrees of freedom , [ scalar ] s^2&=&s_0 ^ 2+_55y^2 + 2_05y t - n^2(y , t)_00t^2 + & & + q^(y)_,i y x^i + q^(t)_,it x^i + & & + 2a(y , t)^2(2+e_,ij)x^i x^jthree of these can be removed at this stage by a scalar gauge transformation ; effectively , this is just the statement that one can always choose a gaussian - normal gauge for the bulk where the off - diagonal terms @xmath221 , @xmath222 and the @xmath22 perturbation vanish : [ gn scalar ] s_gn^2&=&s_0 ^ 2-n^2(y , t)t^2 + q^(t)_,it x^i + & & + 2a(y , t)^2(2+e_,ij)x^i x^jwhere @xmath223 . performing the coordinate transformation ( [ y and t ] ) , we find s^2&=&y^2 - 2n^2 f_y f_t y t + q_,i f_y x^i y + & & -(b^2+n^2 f_t^2)t^2+q_,if_t x^i t + [ fixed scalar metric 1 ] & & + a^2x^i x^jwhere all the problematic terms have been written out in the first line , and @xmath224 etc . the background brane positions are fixed in this frame , but in the presence of scalar perturbations will be displaced to [ brane positions ] y_+(,t)=y_+(,t),y_-(,t)=1+y_-(,t)in five dimensions there are four gauge - invariant scalar degrees of freedom , hence we can not eliminate any more of these perturbation variables with gauge transformations . we can , however , perform another gauge transformation to rearrange them in such a way as to satisfy the conditions ( [ requirement ] ) , whilst simultaneously bringing the two branes back to the fixed positions @xmath24 and @xmath25 . under the most general infinitesimal scalar coordinate transformation [ scalar transformation ] x^ax^a+^a,^a=(,,^i),the metric perturbations transform as g_55&=&-2 d^2 -2d + g_50&=&-d^2 + b^2 + g_5i&=&-_i where @xmath225 and @xmath226 . hence , from ( [ fixed scalar metric 1 ] ) , we find the three conditions necessary to fix the required gauge : [ xi ] & & 2d^2+2d+(n^2f_y^2)=0 , + [ zeta ] & & +d^2 - 2-q f_y=0 + [ chi ] & & b^2-d^2-n^2f_yf_t=0,together with the boundary conditions [ xi bc ] ( y=0)=-y_+(,t),(y=1)=-y_-(,t)which , from ( [ brane positions ] ) , will bring the brane positions to @xmath24 and @xmath25 . @xmath227 and @xmath228 have no boundary conditions to satisfy ; once one has found the required @xmath229 , they follow immediately from ( [ zeta ] ) and ( [ chi ] ) respectively . eliminating @xmath230 from ( [ xi ] ) with ( [ xi ] ) , we obtain the equation of motion for @xmath229 , @xmath231 interestingly only one of the original seven possible five - dimensional perturbations appears in this equation . in principle some information about the values of @xmath232 and its derivatives at the boundaries can be inferred from the isral junction conditions , though in practice this is not relevant . the perturbation is absorbed into a ` forcing ' term @xmath233 , with ( [ xi eqn ] ) taking the form of a generalised diffusion equation . for static randall - sundrum branes ( e.g. @xmath45 for empty branes ) , the original background configuration ( [ gn background ] ) is already in an sfb gauge with @xmath234 . in this simple case , the equation of motion ( [ xi eqn ] ) reduces to [ static xi equation ] = f(y , t,)the general solution of this equation is [ static xi solution ] ( y , t,)=^(y , t,)+(1-y)a(,t ) + yb(,t),where @xmath235 is a particular solution to ( [ static xi equation ] ) , and the linear term represents the most general homogeneous solution . @xmath236 and @xmath237 are arbitrary functions which can be set so that @xmath229 satisfies arbitrary boundary conditions at @xmath24 and @xmath25 . in particular , one can always solve the boundary conditions ( [ xi bc ] ) with the choice [ a and b ] a(,t)&=&-(y_+(,t)+^(0,t , ) ) + b(,t)&=&-(y_-(,t)+^(1,t,))hence any cosmological perturbations about a static , fixed - brane background can always be accommodated into an sfb gauge . for non - trivial @xmath238 the process of constructing a solution to ( [ xi eqn ] ) for arbitrary boundary conditions ( [ xi bc ] ) is far more complicated and one must , once again , resort to a series solution . the boundary conditions ( [ xi bc ] ) can be dealt with by a simple redefinition of the forcing function @xmath233 , by setting [ remove bc ] ( y , t,)=w(y , t,)-(1-y)_+(,t)-y_-(,t).the new function @xmath239 has the simpler boundary conditions that @xmath240 at @xmath24 and @xmath25 , and satisfies the equation [ new equation ] w+k(y , t)=g(y , t,)where @xmath241 we now expand the functions @xmath242 , @xmath239 and @xmath243 in powers of @xmath9 . considering the case once again where the branes are moving apart after a collision which occurred at @xmath65 , we see from its definition in ( [ xi eqn ] ) that @xmath244 , hence there is no @xmath245 term in its expansion . we therefore write k(y , t)&=&_t=1^t^n k_n(y ) + g(y , t,)&=&_t=0^t^n g_n(y , ) + w(y , t,)&=&_t=0^t^n w_n(y,),which , upon substitution into ( [ new equation ] ) , gives [ w0 ] w_0&=&g_0 + [ wn ] w_n+n k_1 w_n&=&g_n-_m=1^n-1 m w_m k_n+1-m_ntogether with the boundary conditions @xmath246 ( [ wn ] ) is just a simple , linear sturm - liouville problem which can always be solved with an eigenfunction expansion subject to certain conditions . specifically , we consider the eigenfunction equation [ eigenvalue equation ] _ m(y)+_m k_1(y)_m(y)=0,_m(0)=_m(1)=0.the eigenfunctions @xmath247 for @xmath248 will form a complete , orthogonal basis , and so we can expand _ t,)&=&k_1(y)_m g_nm(,t)_m(y ) , + w_n(y , t,)&=&_m w_nm(,t)_m(y)and these sums converge absolutely . substituting these expressions back into ( [ wn ] ) , we find [ coefficients ] w_nm = and hence the required solution to ( [ xi eqn ] ) subject to the boundary conditions ( [ xi bc ] ) has been constructed . from ( [ coefficients ] ) , we see that this solution only exists if the eigenvalues of ( [ eigenvalue equation ] ) are either never integer - valued , or , if @xmath249 for some integer @xmath128 , then the corresponding @xmath250 vanishes . note that this conclusion is consistent with @xmath9-reparameterisation . for example , if we reparameterise @xmath251 then @xmath252 where @xmath128 is the minimal positive integer such that @xmath253 . the eigenvalues of ( [ eigenvalue equation ] ) will therefore all be multiplied by @xmath128 , and @xmath254 is only non - zero for @xmath57 a multiple of @xmath128 . hence the conclusion remains the same . this condition on the functions @xmath238 and @xmath255 can be regarded as necessary and sufficient for scalar perturbations to be accommodated within the sfb gauge . note that if @xmath256 , the above argument will simplify considerably since each equation for @xmath257 will just be of the form w_n=_nk_1=0which can be integrated immediately , with two arbitrary constants to fix the boundary conditions . hence , in this case , it will always be possible to find the required gauge . however , for branes moving apart after a collision , from the above argument @xmath258 will be proportional always to its value @xmath259 in the gauge @xmath147 , which will not vanish . hence one will have to check the eigenspectrum of the problem to verify the existence of the required sfb gauge . the precise nature of the function @xmath260 encoding both the bulk perturbation @xmath232 and the perturbations @xmath261 to the brane positions is not important for the above discussion . hence one would expect , subject to the above caveats , to be able to include the effects of matter perturbations on the brane if desired . since the perturbative construction of the background coordinate transformation in [ t expansion ] and [ c expansion ] were not sensitive to the precise nature of the boundary conditions either , one can expect to include matter at the background level as well . hence we have presented in this paper a method for constructing sfb gauges very generally for randall - sundrum cosmologies . the aim of this paper was to demonstrate the existence of gauges for randall - sundrum two - brane braneworlds in which the metric takes the particularly simple form ( [ fixed gauge ] ) whilst , at the same time , the branes are located at fixed positions . such gauges have been assumed to exist already in the literature @xcite , without proof , since they simplify considerably the analysis of the five - dimensional einstein equations necessary to derive analytic results for effective theories . we first examined the dynamics of the system in the background , deriving from the trajectories of the branes the necessary boundary conditions for the coordinate transformation functions . these were then found using two different power series expansions , expanding either in time or in the parameter @xmath10 related to the mass of the bulk black hole . the former approach corresponds to an expansion in the radion , the proper distance between the two branes , and is adapted ideally to the case of two branes moving apart soon after a collision . this regime is of interest in recent attempts to model the big bang as a collision between two branes @xcite . however , the non - linear equations which one needs to solve order by order in @xmath9 do not admit an analytic solution . one can , however , demonstrate the existence of all terms in the series , satisfying the required boundary conditions , corresponding to being able to perform the necessary coordinate transformation correct to @xmath262 for arbitrary @xmath128 . whilst no analytic information about the convergence of the series is possible , numerical simulations show good agreement with the exact results , and we conjectured that the radius of convergence of the series would be equal to @xmath263 , the time taken for the first brane to hit the horizon . a second method , namely an expansion in @xmath10 , corresponds to an expansion in the velocity of the branes . it is possible to obtain analytic expressions for the coordinate transformation functions , and the resulting sfb metric , correct to arbitrary order in @xmath156 . this is because the equations at each order are simple , linear differential equations , involving just binomial functions which are trivial to integrate . results on cosmological backgrounds are of limited interest ; the important question is whether or not one can find an infinitesimal gauge transformation to accommodate arbitrary cosmological perturbations , at first order , into an sfb gauge . we addressed this question last , and showed that this is trivially the case for tensor and vector perturbations . scalar perturbations require considerably more work because one must also take into account a perturbation of the brane trajectories themselves , and hence impose boundary conditions on the infinitesimal coordinate transformation @xmath264 . the second - order partial differential equation for @xmath229 can be solved by an eigenvalue expansion , giving a well - defined constraint on the eigenvalues of ( [ eigenvalue equation ] ) for the sfb gauge to exist . whilst the arguments in this paper do not apply to braneworlds with a ` non - empty ' bulk ( i.e. with non - gravitational bulk degrees of freedom ) , they are not sensitive to the precise nature of the israel junction conditions ; one can include both detuned tensions and brane matter without altering the conclusions .

it has recently been argued that the rate per unit volume of baryon number violation ( topological transitions ) in the hot , symmetric phase of electroweak theory is of the form in the weak - coupling limit , where is a non - perturbative numerical coefficient . over the past several years , there have been attempts to extract the rate of baryon number violation from real - time simulations of classical thermal field theory on a spatial lattice . unfortunately , the coefficient will not be the same for classical lattice theories and the real quantum theory . however , by analyzing the appropriate effective theory on the lattice using the method of hard thermal loops , i show that the _ only _ obstruction to precisely relating the rates in the real and lattice theories is the fact that the long - distance physics on the lattice is not rotationally invariant . ( this is unlike euclidean - time measurements , where rotational invariance is always recovered in the continuum limit . ) i then propose how this violation of rotational invariance can be eliminated and the real b violation rate measured by choosing an appropriate lattice hamiltonian . i also propose a rough measure of the systematic error to be expected from using simpler , unimproved hamiltonians . as a byproduct of my investigation , the plasma frequency and debye mass are computed for classical thermal field theory on the lattice . * v * -20pt

hep-ph9701393

the violation of baryon number ( b ) in the hot , symmetric phase of electroweak theory plays a crucial role in scenarios for electroweak baryogenesis . the rate of b violation is tied , through the electroweak anomaly , to the the rate of topological transitions of the electroweak gauge fields . for the symmetric phase , this rate is not calculable by any perturbative method . in this paper , i address whether the topological transition rate can , in principle , be extracted from lattice simulations . the discussion will reveal some surprising features of real - time thermal field theory on the lattice . it has long been appreciated that , at finite temperature , topological transitions in real time are not directly related to topological transitions in euclidean time @xcite . as a result , there is no apparent way to measure the real - time thermal transition rate in a standard , euclidean - time , lattice simulation of quantum field theory . several years ago , ambjrn and krasnitz @xcite cleverly implemented the observation @xcite that a full simulation of _ quantum _ field theory is not actually required . topological transitions occur in the symmetric phase through large configurations which are essentially classical . indeed , _ all _ long - distance bosonic physics in a hot , ultrarelativistic plasma is effectively classical because of bose enhancement . the number of quanta per mode in low - energy modes is given by the bose distribution @xmath2 and , by the correspondence principle , this is the classical regime . unlike quantum field theory , real - time simulations of classical field theory are tractable : just evolve the classical equations of motion . classical thermal field theory has an ultraviolet catastrophe , historically famous in the context of black - body radiation . because every mode has energy @xmath3 by the classical equipartition theorem , and because there are an infinite number of modes per unit volume in continuum field theory , the energy density is infinite . in quantum field theory , in contrast , the ultraviolet contribution is cut off at momenta of order @xmath4 . ambjrn and krasnitz reasoned that the details of short - distance physics should not affect the long - distance physics of topological transitions . in their simulations , they put their classical system on a spatial lattice and then evolved it in continuous real time . the ultraviolet catastrophe was cut off by the lattice spacing , which they progressively made smaller and smaller . son , yaffe , and i @xcite have recently pointed out that short - distance effects do not decouple as cleanly as ambjrn and krasnitz hoped . in particular , the short - distance modes cause damping of the long - distance dynamics , and this damping affects the transition rate . we showed that damping reduces the transition rate by a factor of @xmath5 , where @xmath6 is the weak fine structure constant . because the damping is caused by short - distance physics , it is not universal , and so a comparison of theories with different short - distance physics becomes non - trivial . in the classical lattice theory used by ambjrn and krasnitz , for example , damping reduces the rate instead by a factor of @xmath7 where @xmath8 is the lattice spacing . as i shall discuss , there is an even more serious problem : because damping is dominated by short - distance physics , it knows about the anisotropies of the lattice . as we shall see explicitly , and as was first noted by bodeker _ et al . _ @xcite , the effective long - distance physics of the classical lattice theory is not even rotationally invariant . this is in striking contrast to the familiar situation of euclidean - time simulations , where rotational invariance is always recovered in the continuum limit . i shall assume that our analysis of damping in ref . @xcite is correct and will eventually be borne out by numerical simulations at sufficiently weak coupling . in this paper , i focus on the natural follow - up question : given that the real quantum theory and the classical lattice theory have different long - distance physics , is there any way to measure the real topological transition rate ? the answer is yes in principle but an exact calculation requires a careful choice of lattice action and the numerical extraction of a somewhat awkward limit . but i shall also propose a rough numerical measure of the suitability of generic lattice actions and argue that even results from simple , canonical actions should be in the right ballpark if properly interpreted . i shall not address at all the complicated problem of how one measures topological transitions on the lattice in the first place , which has a long , difficult history for euclidean - time quantum simulations and a shorter but still confusing one for real - time classical simulations . i simply focus on dynamics and assume the measurement problem will eventually be solved , using cooling or some other technique . section 2 of this paper outlines my method for taking a measurement of the topological transition rate in a theory with one ultraviolet cut - off ( _ e.g. _ a classical spatial lattice theory ) and using it to predict the rate in the same theory with a different ultraviolet cut - off ( _ e.g. _ the real , continuum , quantum field theory ) . the procedure will require , however , that the long - distance dynamics be rotationally invariant in both cases , and this poses a problem for the lattice that will eventually be dealt with . section 2 rests on a very rough and schematic discussion of the effective long - distance dynamics , and i return in section 3 to do a better job of reviewing the correct long - distance theory . i review the derivation of `` hard thermal loop '' effective theory but add a small twist . the usual discussions in the literature are based on the assumption that hard ( _ i.e. _ high - momentum ) excitations in the plasma move at the speed of light . this is not true for lattice theories , and i show how the usual results easily generalize . in section 4 , as a warm - up example of a rotational - invariant classical theory , i consider continuum classical thermal field theory with the ultraviolet regulated by higher - derivative interactions . in section 5 , i turn to the canonical definition of the classical theory on a simple cubic spatial lattice . by explicit computation , the damping at long distances is shown to be anisotropic and to have an interesting structure of cusp and logarithmic singularities in its angular distribution . i discuss the origin of these singularities . then i argue that measurements of the topological transition rate in simple lattice theories can still ( if properly interpreted ) be used to estimate the real transition rate , and i propose a rough measure of the systematic error arising from the anisotropy of the lattice . in section 6 , i propose that the real transition rate could in principle be measured arbitrarily well from lattice simulations by implementing a lattice version of the higher - derivative continuum theory discussed in section 4 . in section 7 , i briefly discuss some possibilities of alternative lattice theories that may be more rotational - invariant than the canonical one , yet easier to implement than my proposal of section 6 . section 8 explores two quantities important to thermal field theory but slightly tangential to the main thrust of this paper : the plasma frequency and the debye mass . i compute both for classical , thermal gauge theory on the lattice . the results provide one test of whether lattice simulations of a given size are indeed in the small coupling limit , and i comment on the application of this test to the simulations of ref . section 9 offers my conclusions and a summary of what remains to be done . the origin of the basic scales associated with topological transitions is reviewed in the introduction of ref . @xcite , and i shall simply quote here the standard result : topological transitions proceed through non - perturbative gauge configurations of spatial size @xmath9 . in ref . @xcite , we then showed that the time scale associated with such transitions is @xmath10 and that the transition occurs through slowly varying magnetic configurations . ( more technically , the relevant , low - frequency gauge fields are transversally rather than longitudinally polarized . ) my current endeavor can be summarized as follows . consider the effective long - distance and long - time theory corresponding to frequencies and spatial momenta of order @xmath11 . what is the correspondence of those theories between ( a ) the real quantum theory , and ( b ) a cut - off classical theory ? the important characteristics of the long - distance effective theory for topological transitions , as discussed in refs . @xcite , can be roughly introduced as follows : the behavior of the long - distance modes is analogous to a pendulum moving in hot molasses , where the molasses represents the short - distance modes , and non - perturbative physics such as the topological transition rate corresponds to large - amplitude fluctuations of the pendulum . a slightly non - linear pendulum is described by an equation of the form @xmath12 a pendulum in ( idealized ) hot molasses is described instead by a langevin equation of the form @xmath13 there is a damping term with coefficient @xmath14 due to the viscous interaction with the molasses . there is also a random force @xmath15 due to the buffeting of the pendulum by random thermal fluctuations of the molasses . the two are related by the fluctuation - dissipation theorem . ignoring the non - linearity of the pendulum , the random force in ( [ eq : pendulum ] ) must have a white - noise spectrum with normalization @xmath16 : @xmath17 the frequency response of such a pendulum is @xmath18 with a power spectrum @xmath19 which includes small - amplitude , high - frequency fluctuations and larger - amplitude , low - frequency ones . if the pendulum is strongly damped , @xmath20 , then the characteristic inverse time for the largest amplitude fluctuations is @xmath21 in this case , the @xmath22 term is ignorable in ( [ eq : response ] ) , and the large - amplitude fluctuations of the system can be described by replacing ( [ eq : pendulum ] ) with the simpler equation @xmath23 in the case of gauge theory , the damping coefficient with the plasmon damping rate : they refer to different kinematic regimes . @xmath14 describes the damping of nearly - static magnetic fields in the plasma , for which @xmath24 and @xmath25 is @xmath26 . in this case , @xmath25 is dominated by hard contributions . the plasmon damping rate , on the other hand , describes the damping of propagating plasma waves , for which @xmath27 and @xmath25 is @xmath28 . in this case , the dominant contribution to @xmath25 depends on soft as well as hard physics . ] @xmath14 is determined from the imaginary part of the ( retarded ) thermal self - energy @xmath29 . the effective equation of the long - distance modes is , in frequency and momentum space , @xmath30 where @xmath31 . the dominant contribution to @xmath32 for @xmath33 is well - known @xcite and is generated by interactions with short - distance ( momenta @xmath34 ) degrees of freedom . in the frequency region of interest ( @xmath35 ) , and for transverse ( _ i.e. _ magnetic ) fluctuations , it is given in weak coupling by @xmath36 where @xmath37 is the ( leading order ) debye mass . for the case of pure gauge theory , @xmath38 where @xmath39 is the adjoint casimir and @xmath40 for su(@xmath41 ) . so , for each mode @xmath42 , the damping coefficient @xmath14 of ( [ eq : pendulum ] ) is just @xmath43 for the spatial momentum scale of topological fluctuations , @xmath44 , ( [ eq : omega ] ) confirms that @xmath45 and shows that the effective theory can be described by a simplified langevin equation of the form ( [ eq : pendulum simplified ] ) . so far , i have been very schematic . most significantly , i have blithely ignored all the terms in the above equations labeled `` higher order in @xmath46 . '' since i am ultimately interested in non - perturbative phenomena , i need to do a better job . i will review the full , unadulterated equations of motion of the effective theory in the next section . however , the schematic equations presented so far are adequate to explain the basic idea of how to compare theories with different ultraviolet behavior . ignoring the higher - order terms altogether for now , consider two theories with different ultraviolet cut - offs say , the real quantum theory and some cut - off classical theory . the self - energy @xmath32 will be different in the two theories , and so the damping constant @xmath14 will be different . the long - distance effective equations of motion are then also superficially different : [ eq:2 eqs]@xmath47 but , as i ve written them , these two equations are trivially related by a rescaling of time : @xmath48 [ that this transformation maps @xmath49 can be verified from ( [ eq : noise 0 ] ) . ] that means that the topological transition rates @xmath50 of the two theories are related by a simple rescaling of time : @xmath51 the prescription for measuring the real topological rate is then : ( a ) to measure it in some uv cut - off classical theory , ( b ) to perturbatively compute the self - energy and hence the values of @xmath14 in both the real theory and the classical theory , and then ( c ) to convert the measured rate by ( [ eq : convert 1 ] ) . the fly in the ointment is that generically , in lattice theories , @xmath14 is not rotationally invariant but depends on the direction of @xmath42 with respect to the axis of the lattice . i shall demonstrate this later by explicit calculation . this means that there is no single rescaling of time that can relate the equations ( [ eq:2 eqs ] ) for all the long - distance modes of the theories . so there is another step required in the prescription : ( step 0 ) find a classical lattice theory where the ultraviolet cut - off is as rotationally invariant as possible . i should emphasize that , unlike the more familiar case of euclidean - time simulations , this last step is a theoretical necessity and not merely a numerical convenience . in euclidean simulations , improving the rotational invariance of the action improves the rate of approach to the continuum limit as the lattice spacing is decreased , but ( in principle ) any lattice action will do as long as the lattice spacing is small enough . here , that is not the case . the chink in the armor of rotation invariance can already be seen in the result ( [ eq : pi ] ) for the self - energy in the real , rotational - invariant theory . in euclidean time , consider effective interactions of the long - distance degrees of freedom that are induced by the short - distance degrees of freedom . such interactions can be taylor expanded in the small , long - distance momenta @xmath52 , @xmath53 , and @xmath54 . the cubic symmetry of the lattice is enough to guarantee that any interaction up to two derivatives @xmath55 is in fact rotationally invariant as well . interactions involving four derivatives need not be rotationally invariant , but such interactions are irrelevant and decouple from the long - distance physics . the crucial assumption of this argument is the analyticity of the interactions at @xmath56 . it has long been known that real - time thermal interactions generated by short - distance physics _ fail _ this analyticity criterion , as can be seen explicitly in ( [ eq : pi ] ) . the standard mechanism by which rotational invariance is recovered in euclidean time is therefore no longer operative . before proceeding to flesh out the proposals made in this section , i should fix a point of nomenclature . often in this paper i shall refer to the `` real '' rate @xmath57 of topological transitions in continuum , quantum , pure non - abelian gauge theory at high temperature and weak coupling . if one s interest is in actual electroweak theory in the same limit , with its accompanying higgs and fermion fields , it is easy to convert if one ignores the small effects of the weinberg mixing angle . the debye mass in electroweak theory is @xcite @xmath58 for a single doublet higgs and @xmath59 families . by the conversion procedure ( [ eq : convert 1 ] ) outlined above and by ( [ eq : gamma 0 ] ) , the electroweak rate @xmath60 is related to the pure su(2 ) gauge theory rate `` @xmath57 '' by @xmath61 eq . ( [ eq : ew ] ) and the analysis of this paper applies whenever the temperature is sufficiently high that the infrared dynamics of the higgs is irrelevant at lengths of @xmath62 . this is the case either ( 1 ) far above the electroweak phase transition or crossover , or ( 2 ) in the symmetric phase at the transition in cases where there is a 1st - order transition and the transition is not exceedingly weak . other cases could be handled by including the infrared dynamics of the higgs into the discussion and into simulations . it is now well - established how to write a set of non - local equations describing the dominant interactions of long - distance modes in hot , real - time , non - abelian gauge theory when the random force term @xmath63 discussed earlier is ignored . these interactions go by the name of `` hard thermal loops , '' since they are generated by integrating out the hard , short - distance modes of the theory . though there is a great deal of literature on this subject , it is almost universally geared to hard particles that travel at the speed of light . in a lattice theory , in contrast , the hard particles have a more complicated dispersion relationship than @xmath64 . it will therefore be useful to briefly review the derivation of the hard thermal loop equations so that i can motivate the ( simple ) modifications required for applying them to lattice theories . hard thermal loops were originally derived through diagrammatic analysis by braaten and pisarski . for my purpose , the quickest way to get at them is by an alternate method formalized by blaizot and iancu @xcite , which i shall present heuristically . ( see also ref . the method is a generalization of the description of qed plasmas via vlasov equations . the vlasov equations are an effective description where soft photons are represented by gauge fields @xmath65 in the usual way , but all the hard , charged particles are represented by a locally defined flow density @xmath66 , which is the density of particles at time @xmath67 and position @xmath68 with ( hard ) momentum @xmath69 . the first vlasov eq . follows from applying liouville s theorem to @xmath70 and using the equations of motion : ) can be ignored in the application to topological transitions , see ref . @xmath71 where @xmath72 is the 3-velocity and where @xmath73 . ( note that @xmath74 does not transform as a lorentz 4-vector . ) in the real world of ultrarelativistic plasmas , hard excitations move at the speed of light and @xmath75 is just @xmath76 . here enters my simple generalization of the standard discussion in the literature : for more general underlying field theories , @xmath72 should be identified with the _ group _ velocity of the hard excitations , since it is the group velocity which describes the physical rate of motion of charge . if the dispersion relationship for the hard modes is @xmath77 , then the group velocity is @xmath78 ( for those who find this style of argument overly heuristic , an explicit check of final results will be made in appendix [ apndx : diagrammatic ] using a diagrammatic approach . ) the second vlasov equation , which is for the gauge field , is simply @xmath79 where i have added a flavor index @xmath80 to make explicit that there may be different types of particles with different charges in the plasma . i have also introduced the short - hand notation @xmath81 now linearize the vlasov equations in the gauge fields and in deviations @xmath82 of @xmath70 from its equilibrium distribution @xmath83 . @xmath83 depends only on the energy @xmath84 of excitations and not on @xmath85 or @xmath67 . using @xmath86 the linearized vlasov equations are @xmath87 these equations can be solved formally for @xmath82 , yielding @xmath88 @xmath89 is an infinitesimal that has been inserted to pick out the retarded solutions to the equations . for qed with a single massless fermion , @xmath90 where the factor of @xmath91 counts the two polarizations each of the fermion and anti - fermion . it is easy to guess how the forgoing generalizes to non - abelian gauge theories . from ( [ eq : linearized a ] ) it is clear that the @xmath82 of interest must carry an adjoint color index . ( rather than a simple color - diagonal number operator , @xmath82 now represents a correlation between the colors that the gauge bosons convert between . ) but then the convective derivative @xmath92 in ( [ eq : linearized n ] ) must be generalized to a covariant convective derivative @xmath93 , where @xmath94 is understood to act in the adjoint representation . similarly , the derivative in ( [ eq : linearized a ] ) should be covariant . the resulting equation for the soft modes is @xmath95 where @xmath96 has been replaced by @xmath97 and the factor of 2 counts the two helicity states of the hard gauge bosons . readers desiring a much more detailed justification should refer to ref . @xcite . for our current application to topological transitions , where frequencies are small compared to spatial momenta and the fluctuations are magnetic , we are only interested in the small frequency limit of ( [ eq : blaizot ] ) . working in @xmath98 gauge , this limit gives @xmath99 where @xmath100 . as long as the energy @xmath84 respects parity , the integral over @xmath101 would vanish if not for the @xmath89 factor . using @xmath102 where @xmath103 stands for principal part , the effective equation for the frequencies and momenta of interest may be rewritten as @xmath104 where @xmath105 this is indeed of the form of the `` non - linear system in molasses '' ( [ eq : pendulum ] ) discussed schematically in section [ sec : basics ] , except that it is missing the random force term . the underlying diagrammatic origin of the result ( [ eq : gamma ] ) for the damping factor can be roughly understood by considering the process of fig . [ fig : absorb ] , which shows a soft excitation being absorbed by the small - angle scattering of a hard excitation . the factor @xmath106 in ( [ eq : gamma ] ) comes from the interaction vertex , and the second factor @xmath107 is because the amplitude of fig . [ fig : absorb ] should be squared to get a rate . the @xmath108 is just the @xmath109 limit of the energy - conserving delta function @xmath110 associated with this diagram . finally , @xmath111 for a bose or fermi distribution is just @xmath112 times the product @xmath113 of the initial - state probability distribution and the final - state bose enhancement or fermi blocking factor . more details are given in appendix [ apndx : diagrammatic ] . the final result ( [ eq : basic 1 ] ) of the previous section is in fact sufficient for the comparison of different theories , but , for the sake of completeness , i will review the random force term as well . the vlasov equations of the previous section describe the relaxation or propagation of deviations from equilibrium ; they do not , as formulated above , describe the fluctuations that occur in equilibrium itself . however , fluctuations are related to dissipation through the fluctuation - dissipation theorem . introduce a random force term by writing @xmath114 if we momentarily ignore the non - linearity of ( [ eq : basic ] ) , which arises from the non - abelian nature of the theory , then the characteristics of the random force can be taken from ( [ eq : noise 0 ] ) : @xmath115 where @xmath8 and @xmath116 are adjoint color indices . huet and son @xcite have shown how to modify this equation to include the non - linearity and restore gauge invariance . the result is to simply replace @xmath117 above by a straight , adjoint - charge wilson line connecting @xmath85 to @xmath118 : @xmath119 where @xmath120 @xmath121 runs on a straight path from @xmath85 to @xmath118 , @xmath122 denotes path ordering , and the @xmath123 are the ( real , anti - hermitian ) adjoint representation generators . the basic equations for the effective theory are then ( [ eq : basic ] ) and ( [ eq : noise ] ) . the damping coefficient @xmath14 of ( [ eq : gamma ] ) can be rewritten as @xmath124 where @xmath125 depends only on the direction @xmath126 . note that @xmath127 is transverse : @xmath128 in a theory that is fundamentally rotationally invariant , it must then have the form @xmath129 where @xmath130 is a number independent of @xmath55 , obtained by averaging @xmath131 over @xmath126 : the analog of the @xmath37 in ( [ eq : pi ] ) . however , @xmath132 is not actually the debye mass in theories where @xmath133 . see section [ sec : plasma freq ] . ] @xmath134 by examining the effective equations ( [ eq : basic ] ) and ( [ eq : noise ] ) , it is then easy to check that the time rescaling suggested in section [ sec : basics ] will indeed relate any two rotational - invariant theories : @xmath135 if @xmath84 is monotonically increasing , @xmath130 can be rewritten in the simpler form @xmath136 the result ( [ eq : md 0 ] ) for the real quantum theory can now be reproduced by using the massless bose distribution @xmath137 for @xmath138 . before i move on to lattice theories and the difficulties with rotational invariance , it is useful to first make the preceding discussion concrete with a simple , rotational - invariant example of a classical thermal field theory . consider a classical continuum description of non - abelian gauge theory where i regulate the ultraviolet by higher - derivative terms . the hamiltonian density is @xmath139 \ , , \label { eq : high d}\]]where @xmath80 has an expansion @xmath140 and @xmath141 is ( minus ) the conjugate momentum to @xmath142 . at momenta low compared to @xmath143 , this is just normal gauge theory . however , provided @xmath144 grows sufficiently rapidly at large @xmath145 , the higher - derivative interactions will cut off the ultraviolet catastrophe of classical thermal field theory at a momentum scale of order @xmath143 . to compare this cut - off classical theory to the real quantum theory , i will compute the damping coefficient @xmath130 of ( [ eq : m ] ) . for any classical theory , the equilibrium distribution @xmath146 is simply @xmath147 one quick way to see this is to start with the bose distribution of the quantum theory and , for the first and last time in this paper , put in the explicit factor of @xmath148 : @xmath149 this just recapitulates the discussion ( [ eq : classical regime ] ) of the introduction . the remaining element we need is the frequency @xmath150 as a function of momentum @xmath101 for perturbative excitations . for the theory ( [ eq : high d ] ) , it is @xmath151 the coefficient @xmath130 of ( [ eq : m ] ) is then ( for monotonically increasing @xmath80 ) @xmath152 this converges provided @xmath144 grows faster than @xmath145 for large @xmath145 . in other words , a four - derivative interaction is not enough to cut off the ultraviolet , but six derivatives do the job . by ( [ eq : convert ] ) and ( [ eq : md 0 ] ) , the topological transition rate @xmath153 in this classical , higher - derivative theory is related ( in weak coupling ) to the rate @xmath57 in the real ( pure gauge ) quantum theory by @xmath154 it will be useful later on to have the specific result for @xmath80 of the form @xmath155 , corresponding to an isolated six - derivative cut - off . in this case , @xmath156 where @xmath157 is the usual euler gamma function . we now have almost everything we need to compute the damping coefficient @xmath158 , or equivalently @xmath159 , for lattice theories ; we just need the dispersion relationship @xmath84 . let s start by studying the simple , classical , kogut - susskind hamiltonian used by ambrn and krasnitz @xcite for simulations on a simple cubic ( sc ) , spatial lattice . the tree - level dispersion relationship for propagating excitations on the lattice is @xmath160 \ , , \label{eq : sc omega}\ ] ] where @xmath8 is the lattice spacing . for thermal field theory on the lattice , the parameters @xmath161 , @xmath8 , and @xmath4 always appear in the combination @xmath162 for dimensionless quantities . and @xmath8 can be scaled out of the hamiltonian @xmath163 so that the distribution @xmath164 becomes @xmath165 . ] i shall henceforth work in lattice units by setting @xmath8 = @xmath4 = 1 . the conversion back to physical ones for su(@xmath41 ) rather than @xmath166 . ] is @xmath167 , @xmath168 , @xmath169 , and @xmath170 . the group velocity @xmath171 is @xmath172 plugging into ( [ eq : gamma ] ) with the classical distribution ( [ eq : n0 classical ] ) then yields in a simple cubic lattice theory was first written down for qed coupled to massless scalars in eq . ( b11 ) of ref . my ( [ eq : sc gamma ] ) is that equation s @xmath24 limit , once the overall normalization in ref . @xcite is changed to be appropriate for pure , non - abelian gauge theory . ] [ eq : sc gamma]@xmath173 where the integrals are over the brillouin zone @xmath174 . the delta function can be used to eliminate one of the three integrations , and i have done the remaining two numerically . ( the detailed expression is given in appendix [ apndx : numerical ] . ) [ fig : impi]b shows the resulting angular distribution of the trace @xmath175 normalized by its angular average @xmath176 . not only is the result not rotationally invariant , but it has an interesting structure of singularities as well . there are cusps and , in certain special directions , @xmath158 is infinite ! as we shall see , these singularities are mild and integrable . also , they are true singularities only in the weak - coupling limit : they are rounded off by higher - order corrections but become sharper and sharper as @xmath177 . for any fixed direction @xmath126 of the soft momentum , @xmath158 as given by ( [ eq : gamma ] ) can be rewritten @xmath178 where the integral is over the surface @xmath179 in @xmath101 space on which @xmath180 is orthogonal to @xmath55 . if there is a point on this surface where the vector @xmath181 vanishes , there will be a singularity in @xmath14 . if there is a entire line of points where the vector vanishes , the singularity will be more severe . this is analogous to van howe singularities in the density of states . and has singularities ( typically cusps ) when @xmath182 vanishes for some @xmath101 . ] naive counting of constraints suggests the existence of singularities is to be expected : 5 degrees of freedom in @xmath101 and @xmath126 versus 4 constraints in @xmath183 and @xmath184 . to summarize , singularities in @xmath14 will be caused by momenta @xmath101 such that there is a null vector @xmath126 , orthogonal to @xmath185 , of the matrix @xmath186 . in the case of the dispersion relationship ( [ eq : sc omega ] ) , @xmath187 at @xmath188 and its lattice reflections , this matrix is proportional to @xmath189 and so every direction of @xmath126 orthogonal to @xmath190 is associated with a singularity of @xmath14 . this is the origin of the cusp singularities in fig . [ fig : impi]b , which are along the plane @xmath191 and its three lattice reflections . for @xmath192 and general @xmath193 , the matrix ( [ eq : sc matrix ] ) has ( 1,-1,0 ) as a null vector , which is orthogonal to @xmath194 . therefore , @xmath126 in the ( 1,-1,0 ) direction or its lattice reflections is associated with an entire line of singularities in @xmath101 space . this is the origin of the infinities in these @xmath126 directions in fig . [ fig : impi]b . ( see section [ sec : nospike ] for a more general discussion . ) for these values of @xmath126 , @xmath195 vanishes linearly as one approaches the @xmath101-space line of singularities . the surface integral in ( [ eq : surface int ] ) is therefore only logarithmically singular . physically , there are not , in fact , infinite spikes in the angular distribution of @xmath196 in the lattice theory . recall that @xmath197 for the long - distance physics of interest . the infinite size of the spikes occurs only in the formal limit @xmath198 that i applied to study the theory in weak coupling . the taming of the singularity can be seen from the vlasov equations by returning to the hard - thermal loop equation ( [ eq : blaizot ] ) before extracting the small @xmath199 limit . the imaginary part of the self - energy , for spatial polarizations , is @xmath200 where @xmath201 is the four - vector @xmath202 . the relevant difference from the earlier derivation of the damping constant @xmath158 of ( [ eq : gamma ] ) is simply that @xmath108 has been replaced by . this difference is sufficient to cut off the logarithmic infinities in @xmath158 from order @xmath203 to order @xmath204 . it is beyond my present purpose to explore the cut off of the logarithmic singularities in detail , but i should mention that the previous paragraph is misleadingly incomplete . based on the vlasov equations , i ve shown how the logarithms are cut off by effects sub - leading in the limit @xmath198 . at sub - leading order , however , the vlasov equations themselves are inadequate . one can see this from a more fundamental , diagrammatic analysis of the self - energy , discussed in appendix [ apndx : diagrammatic ] . the @xmath205 discussed above is nothing more than the @xmath206 limit of the energy - conserving delta function @xmath110 associated with fig . [ fig : absorb ] and discussed at the end of section [ sec : review ] . for @xmath207 of order @xmath208 in coupling , however , the correct sub - leading version is @xmath209 the @xmath210 term is the same order as the @xmath199 term . physically , it represents dispersion of the wave packets of the hard excitations an effect not built into the vlasov equations implicit treatment of hard excitations as a collection of classical particles . in any case , the divergences in the small - coupling limit are only logarithmic and so are mild enough not to cause any real problem with the effective theory specified by ( [ eq : sc m ] ) . topological transitions occur through configurations with spatial size @xmath211 and not through pure , infinite - extent , plane waves with momenta @xmath55 precisely in one of the divergent directions . since logarithmic singularities are integrable in @xmath126 , they will give only a finite damping effect to the evolution of any spatially localized configuration . if there were no infinite spikes , fig . [ fig : impi]b would look like at least a rough approximation to the rotational - invariant sphere of fig . [ fig : impi]a , and a procedure for estimating the real , quantum transition rate @xmath57 suggests itself : replace @xmath14 by its appropriate angular average @xmath212 , and use this value to relate the lattice measurement of @xmath50 to @xmath57 as in ( [ eq : convert 1 ] ) and section [ sec : app rot ] . what would be the expected systematic error of this procedure ? suppose @xmath213 and @xmath214 were the maximum and the minimum of the angular distribution of @xmath215 . ( i sha nt worry about the indices @xmath216 on @xmath158 for the moment . ) since the effect of damping decreases the transition rate , it seems reasonable that the measured @xmath50 is bigger than it would have been if @xmath215 had been isotropically equal to @xmath217 in all directions but smaller than if @xmath215 had been isotropically equal to @xmath218 in all directions . so the ratio of the extreme values @xmath218 and @xmath217 to @xmath212 would give a conservative estimate of the relative systematic error due to the anisotropy of the damping . i shall discuss later whether one might find lattice theories for which @xmath215 indeed has no infinite spikes . for the moment , focus on the case at hand . since the spikes are integrable , we can still replace @xmath14 by its average to obtain an estimate of @xmath57 from the lattice measurement of @xmath50 . @xmath212 is the damping felt by isotropic gauge configurations . the problem that remains is to find a plausible estimate of the systematic error , so that there is a measure for deciding whether one lattice theory with spikes is better than another . one obvious candidate is the root - mean - square deviation @xmath219 given by @xmath220 , where the overline indicates angular averaging . i have suppressed spatial indices on @xmath14 above and will now be more concrete . by comparison with the rotational - invariant case , where @xmath221 has the form @xmath222 due to its transversality , define @xmath223 for the lattice case , where @xmath224 denotes averaging over the direction of @xmath55 . the variation is @xmath225 ^ 2 \,.\ ] ] using the general result ( [ eq : gamma ] ) for @xmath158 in any theory , we have @xmath226 and @xmath227 for the kogut - susskind hamiltonian on a simple cubic lattice , numerical integration gives @xmath228 and @xmath229 using ( [ eq : convert ] ) and ( [ eq : gamma 0 ] ) , one can then estimate the real transition rate ( in the weak coupling limit ) from a lattice measurement as follows . the b violation rate @xmath50 is expected to scale as @xmath230 in weak coupling @xcite . in lattice units , measure the proportionality constant @xmath231 then the estimate is @xmath232 the fact that the rough indicator @xmath233 of the systematic error due to using the lattice is about @xmath234 suggests that the estimate ( [ eq : estimate ] ) should at least be in the right ballpark of the true answer . ( also , the minimum value of @xmath235 in fig . [ fig : impi]b is just 30% below the average . ) unfortunately , @xmath236 has not yet been measured , should not be confused with the coefficient @xmath237 presented in ref . @xcite , which was an attempt to extract the coefficient of a presumed @xmath238 scaling law for weak coupling a scaling law which ignores the effects of damping @xcite . ] and even the weak - coupling scaling law @xmath239 has yet to be verified on the lattice . to eliminate the fundamental systematic uncertainty caused by the anisotropy of the lattice , one needs to find an effectively rotational - invariant lattice theory . ( i re - emphasize that the search for an alternative theory is a theoretical necessity : just taking the small lattice spacing limit is by itself inadequate . ) as it turns out , an effectively rotational - invariant theory is , in principle , easy to achieve . the idea is simply to start with a rotational - invariant continuum theory with a higher derivative cut - off , such as discussed in section [ sec : higher derivatives ] , and then put it on a lattice that is fine compared to the continuum cut - off scale @xmath240 . a numerical extraction of @xmath50 would then require a careful double limit : @xmath241 where the order of limits is crucial . to be a little more specific about how the cut - off @xmath143 might be achieved on a spatial lattice , consider the tree - level dispersion relationship for propagating gluons with momentum small compared to the lattice spacing . for a given lattice hamiltonian ( with cubic symmetry ) , it will have an expansion in momentum of the form @xmath242 + o(p^6)\ ] ] in lattice units , where the coefficient @xmath243 is just a reminder that the units of time can be normalized arbitrarily . by including next - to - nearest - neighbor couplings , and perhaps next - to - next - to - nearest neighbor couplings , and so forth , the parameters of the interactions can be tuned to make any finite set of the coefficients @xmath244 of the expansion take on whatever values desired . in particular , there is some local hamiltonian @xmath245 for which ( 1 ) @xmath116 is a small number , which i ll call @xmath246 , ( 2 ) @xmath247 is @xmath248 , ( 3 ) @xmath249 , and ( 4 ) all other coefficients are @xmath250 : @xmath251 choosing units of time where @xmath252 , @xmath253 this describes a theory with a rotational - invariant four - derivative `` cut off '' at momentum @xmath254 , plus anisotropic terms that are suppressed by @xmath246 at that scale . unfortunately , the four - derivative `` cut - off '' in ( [ eq:4 deriv 2 ] ) is inadequate . as discussed in section [ sec : higher derivatives ] , a six - derivative interaction is required to cut off the uv contribution to damping . so , even for small @xmath89 , a generic theory with dispersion relationship ( [ eq:4 deriv ] ) will still produce significant anisotropy . to the damping rate is actually @xmath255 and so , technically , is small in the limit that @xmath256 was very large . but this is certainly an impractical limit for any numerical simulation , keeping in mind that , as discussed below , the @xmath257 and @xmath258 limits do not commute . ] we can fix the cut - off by going to six derivatives , requiring improved hamiltonians @xmath259 where ( 1 ) @xmath116 is a small number , which i ll now call @xmath260 , ( 2 ) the fourth - order coefficients @xmath247 and @xmath261 vanish , ( 3 ) the coefficient of @xmath262 is @xmath248 , ( 4 ) the sixth - order coefficients that break rotational invariance vanish , and ( 5 ) other coefficients are @xmath250 : @xmath263 choosing units of time where @xmath264 , @xmath265 this describes the desired rotational - invariant six - derivative cut - off at momentum @xmath254 plus anisotropies that are suppressed by @xmath246 at that scale . the relative size of the anisotropies in the damping coefficient @xmath14 will therefore be @xmath266 and disappear in the limit @xmath257 . i will not attempt in the present work to explicitly construct the improved lattice hamiltonian @xmath259 which gives this behavior . a very nice discussion of improving lattice hamiltonians is given by moore in ref . @xcite , where he constructs a classical yang - mills hamiltonian with vanishing @xmath267 terms . constructing a hamiltonian that satisfies the other constraints for @xmath259 is presumably straightforward in principle but tedious in practice . in everything above , i have referred to conditions on the tree - level dispersion relationship . the real dispersion relationship for hard excitations ( @xmath268 ) will be modified by perturbative corrections , whose strength is parametrized by @xmath269 . this is the reason that taking the @xmath258 limit before the @xmath270 limit would be crucial to extracting results from numerical simulations . for an improved hamiltonian giving the desired low - momentum dispersion relation ( [ eq:6 deriv ] ) , the conversion of lattice measurements to the real transition rate would be achieved by first measuring the coefficient @xmath271 and then taking @xmath272 which follows from ( [ eq : hd6 ] ) . @xmath273 above continues to mean @xmath162 where @xmath161 is the continuum coupling extracted from the hamiltonian @xmath259 . the @xmath274 limit discussed in the last section may be difficult to extract numerically . so it s worthwhile to consider whether there are relatively simple lattice theories where the damping coefficient @xmath14 , though anisotropic , is closer to rotational invariance than for the simple lattice hamiltonian discussed in section [ sec : sc ] . i will briefly discuss here one possibility for improvement : replacing the simple cubic lattice by a face - centered cubic ( fcc ) lattice . lattice @xcite , which is the four - dimensional analog of an fcc lattice , and the body - centered hypercubic ( bch ) lattice @xcite . ] the fcc lattice can plausibly lead to more rotational - invariant results because sites have more nearest neighbors than for the simple cubic lattice ( eight instead of six ) . the shape of the first brillouin zone of an fcc lattice is slightly complicated ( a `` truncated octahedron '' ) , but all that is needed here is that the cube @xmath284 in momentum space covers the unique momentum states exactly twice . so , when integrating over states , i take @xmath285 the sum is a notational reminder to sum over the two dispersion relationships @xmath286 \ , . \label{eq : fcc omega}\ ] ] the corresponding group velocities are @xmath287 and its permutations . numerical evaluation of the general formula ( [ eq : gamma ] ) for @xmath14 ( see appendix [ apndx : numerical ] ) gives the angular distribution of @xmath288 shown in fig . [ fig : impi]c . the position of the logarithmic spikes is explained in the next section . numerical integration of ( [ eq : gamma bar ] ) and ( [ eq : sigma ] ) gives @xmath289 and@xmath290 . the 26% result for the angular deviation of damping on the fcc lattice is indeed better than the 31% result for the sc lattice , but not by a lot . the minimum of @xmath235 in fig . [ fig : impi]c is 20% below the average . the simplest actions on both the sc and fcc lattices have logarithmic spikes in the angular distribution of @xmath14 , shown in figs . [ fig : impi]b and c. in the proposal of section [ sec : proposal ] for reaching the rotational - invariant limit , it does nt matter whether the hardest modes @xmath291 of the theory give rise to such spikes , because the effect of the hardest modes , and hence the strength of the spikes , will vanish as @xmath257 . for the purposes of surveying simpler lattice theories that are only approximately rotationally invariant , however , it s interesting to consider whether it s possible to avoid the spikes altogether . so i will take a moment to explain the generic property of the simple sc and fcc actions that causes the spikes and then discuss whether it can be avoided . recall the conditions @xmath292 and @xmath293 of section [ sec : sing origin ] for generating some sort of singularity . consider the energy @xmath84 of excitations for small but fixed @xmath193 in a lattice theory with cubic symmetry . [ fig : spike]a shows a qualitative contour plot of energy @xmath84 vs. @xmath294 and @xmath295 for some such lattice theory [ _ e.g. _ the simple cubic theory of ( [ eq : sc omega ] ) ] . there is a minimum at @xmath296 , which there must be if we are plotting the branch of @xmath84 that approaches the continuum limit @xmath297 for small @xmath101 . suppose that there is also a maximum somewhere along the line @xmath298 , as shown in fig . [ fig : spike]a . then one can argue as follows that there must be a singularity associated with the direction @xmath299 and with @xmath101 somewhere on the line @xmath298 ( @xmath193 still fixed ) . by reflection symmetry through the plane @xmath298 , the group velocity @xmath171 on the line must be perpendicular to @xmath126 , and the curvature @xmath300 on the line must have the block - diagonal form @xmath301 in the basis @xmath302 . near the minimum in the fixed @xmath193 plane , @xmath303 must be negative . near the maximum that by assumption lies on the line @xmath298 , it must be positive . therefore , there is a point inbetween where @xmath304 and so @xmath305 . this point then satisfies both conditions for generating a singularity associated with the direction @xmath299 . but now note that this argument did not depend on the value of @xmath193 . as long as the assumption that both a minimum and a maximum in @xmath306 simultaneously lie on the line @xmath298 holds for some _ range _ of @xmath193 , then there will be an entire curve of points in @xmath101 space that contributes to the singularity in the @xmath299 direction . as discussed earlier , such degeneracy gives rise to a logarithmic singularity in @xmath14 . a similar qualitative situation , for some different lattice theory [ _ e.g. _ the fcc theory of ( [ eq : fcc omega- ] ) ] , is shown in fig . [ fig : spike]b . the same argument now goes through with the direction @xmath307 and the pane of symmetry @xmath308 . this is the origin of the spikes in the fcc case of fig . [ fig : impi]c . to avoid spikes , look for lattice theories such that @xmath84 does _ not _ have lines of local maxima and minima in @xmath306 that simultaneously lie on the same plane of symmetry . to show that this is possible , i will give an example from massless scalar qed on a simple cubic lattice ( where momenta are @xmath309 ) . for scalar qed , the problem can be reduced to searching through the space of @xmath310 that are finite linear combinations of @xmath311 for integer vectors @xmath312 . this is because one can generate any such @xmath310 , @xmath313 by a gauged version of the local scalar interaction @xmath314 where @xmath32 is conjugate to @xmath315 . so i will therefore first construct an @xmath84 that works and then leave it as an exercise to the interested reader to construct the corresponding hamiltonian . the space of linear combinations of @xmath311 is equivalent , by trigonometric identities , to the space of parity - even polynomials in @xmath316 and @xmath317 . a simple example of an @xmath84 that avoids spikes is @xmath318 ^ 2 = 4 - 2 \prod_i ( 1 + \cos p_i ) \ , . \label{eq : spikeless 1}\ ] ] this example is a bit degenerate : the boundary of the brillouin zone @xmath309 is a constant energy surface , @xmath319 . a contour plot of @xmath84 for fixed @xmath193 is shown in fig . [ fig : nospike]a . there is a maximum along planes of symmetry , but it is a degenerate maximum . the nominal conditions for a singularity in the damping rate are met but only at the boundary , and at the boundary the group velocity @xmath75 corresponding to ( [ eq : spikeless 1 ] ) vanishes . the explicit factors of @xmath320 and @xmath321 in the formula ( [ eq : gamma ] ) for @xmath14 then compensate the divergence of the @xmath108 , and there are no singularities . here s an example that s degenerate in a different way : @xmath322 ^ 2 = \left\ { ( 1 - \cos p_x ) \ , ( \cos p_y - \cos p_z)^2 \right\ } \times \left\ { \hbox{cyclic permutations } \right\ } \ , . \label{eq : spikeless 2}\ ] ] this example has been constructed to give @xmath278 along all the planes of symmetry , and a fixed @xmath193 slice is shown in fig . [ fig : nospike]b . this example by itself does nt have the right @xmath323 behavior for the continuum theory , but we can now obtain an interesting non - degenerate example by an appropriate superposition of the two degenerate ones . for example , a @xmath193 slice of @xmath324 ^ 2 + 5\ , [ \omegap^{(2)}]^2 \label { eq : spikeless 3}\ ] ] is plotted in fig . [ fig : nospike]c . since all the machinery was put in place in section [ sec : review ] , it s interesting to look at some other quantities , characteristic of classical thermal gauge theories , that are not as directly related to the topological transition rate as the damping coefficient @xmath14 . in this section , i shall examine the plasma frequency and debye screening length at leading order in weak coupling . the plasma frequency @xmath325 is the frequency of small - amplitude , propagating , colored waves in the long wavelength limit . ( in quantum theories , it is also known as the plasmon mass . ) it can be extracted from the general hard - loop equation ( [ eq : blaizot ] ) by linearizing in @xmath326 and taking the @xmath327 limit for fixed @xmath199 . it s convenient to work in a covariant gauge , where this limit gives @xmath328 if the underlying theory has at least cubic symmetry , this can be rewritten as @xmath329 where @xmath330 for a classical theory , this is @xmath331 for the kogut - susskind hamiltonian on a simple cubic lattice , the dispersion relationship ( [ eq : sc omega ] ) gives @xmath332 in lattice units . the plasma frequency is in fact related to the physics of topological transitions . when the system crosses the potential energy barrier for such transitions , it oscillates many times back and forth across the barrier for each net transition @xcite . these oscillations are small amplitude , and their oscillation frequency is the plasma frequency . for comparison , a typical su(2 ) simulation by ambjrn and krasnitz @xcite has couplings given by @xmath333 , for which ( [ eq : plasmon ] ) predicts a plasma oscillation period of @xmath334 . their simulation has time steps of @xmath335 in lattice units . debye screening is the screening of static electric fields in a plasma . it can be seen from the small - amplitude , zero frequency behavior of ( [ eq : blaizot ] ) , which gives @xmath336 where @xmath337 for a classical theory , @xmath338 for the kogut - susskind hamiltonian on a simple cubic lattice , this gives @xmath339 in lattice units . in fact , there is an analytic result @xcite is @xmath340 where @xmath341 is given by eq . ( a.5 ) of ref . @xcite , except that equation has a typographic error : the elliptic function @xmath201 should be squared @xcite . ] for the integral , and ( [ eq : debye ] ) is more precisely @xmath342 ^ 2 \ ; \ca g^2 \,,\ ] ] where @xmath343 is the complete elliptic integral of the first kind . in the weak coupling limit , the inverse screening distance @xmath344 should be large compared to the lattice spacing . for comparison , the weakest coupling simulations in ref . @xcite have @xmath345 , for which ( [ eq : debye ] ) gives @xmath346 lattice spacings . so there is one physical quantity , at least , for which these simulations are only marginally in the weak - coupling limit . the implications for measuring the weak - coupling behavior of the topological transition rate are unclear . in this paper , i have presented a procedure for extracting the weak - coupling limit of the real topological transition rate from rates that could be measured in simulations . the concerns about rotational invariance originally raised by bodeker _ et al . _ @xcite are crucial , and a precise measurement of the rate requires a search for an effectively rotational - invariant lattice hamiltonian . unlike the rotational improvements familiar to euclidean studies , simply improving the rotational invariance of higher and higher derivative terms in the small momentum limit is not enough : one must also introduce a uv cut - off distance @xmath240 that is large compared to the lattice spacing , as discussed in section [ sec : proposal ] . there remain several interesting problems . one is to explicitly construct the improved hamiltonian of section [ sec : proposal ] . another is to survey other , simpler hamiltonians ( which would not require numerical extraction of the @xmath274 limit ) to find one with relatively small anisotropy as measured by @xmath347 . it would also be interesting to find pure gauge theories that avoid logarithmic spikes , like the scalar qed theories discussed in sec . [ sec : nospike ] . and , of course , there is the ongoing numerical problem of actually extracting lattice topological transition rates in the first place and verifying that they scale as @xmath348 at sufficiently small coupling . in this paper , i have focused on extracting the real topological rate from lattice simulations where the only degrees of freedom are standard gauge link variables . an interesting alternative suggested by hu and mller is to instead introduce additional particle degrees of freedom that are described by continuum positions @xmath349 and classical non - abelian charges and which interact with the lattice fields . the purpose of these particles is to mimic the hard degrees of freedom of the real quantum theory and induce , in the lattice theory , the correct long - distance physics of the real theory . i would just like to point out that any proposal of this type will require very careful attention to orders of limits , because there are in fact _ two _ contributions to the effective long - distance theory : that of the new particles , and that of the short - distance modes of the lattice gauge fields . the latter gets large as the lattice spacing is made small and will be anisotropic . so two non - commuting limits are required of any such scheme : the limit of small lattice spacing , and a limit where the coupling to the additional particles is somehow made large so that their contribution to long - distance physics dominates . i am particularly indebted to dam son for explaining to me blaizot and iancu s formulation of hard thermal loops and for helping me understand how to generalize it . i also thank steve sharpe , rajamani narayan , greg moore , larry mclerran , misha shaposhnikov , and especially larry yaffe for a variety of useful conversations . this work was supported by the u.s . department of energy , grant de - fg03 - 96er40956 . in this appendix , i will verify the formula ( [ eq : gamma ] ) for @xmath221and in particular its interpretation of @xmath75 as the group velocity by applying diagrammatic methods in a wide class of theories . specifically , i will discuss the calculation of the imaginary part @xmath25 of the gauge boson self - energy starting from one - loop euclidean diagrams of the form of fig . [ fig : pi]a , where the line in the loop represents hard excitations , be they gauge particles , scalars , or whatever . ( diagrams of the form of fig . [ fig : pi]b will not give a contribution to @xmath25 because they can not be cut . ) i will assume that tree - level propagators of the theory are of the form @xmath350 . color indices , which are just responsible for the overall factor of @xmath39 in the final result ( [ eq : gamma ] ) for @xmath14 , will be ignored . my first assertion is that the thermal contribution to @xmath351 is @xmath352 where @xmath353 is the distribution function for bosons ; @xmath354 is the integrated , squared amplitude for absorption of a gauge boson by the thermal bath ; and @xmath355 is the vertex shown in fig . [ fig : absorb ] . the sum above is over different species and polarizations @xmath356 of the hard particles , and the capital letters @xmath201 and @xmath103 will henceforth stand for the `` four - vectors '' @xmath202 and @xmath357 . @xmath358 represents the final state bose or fermi enhancement and should be replaced by just @xmath359 for a classical theory . ( [ eq : start ] ) is a simple consequence of the thermal optical theorem @xcite . rather than plodding through the one - loop case step by step here , i will give a quick pictorial argument for readers who are unfamiliar with it . by the contour trick for doing euclidean finite - temperature sums , or by simply starting directly with a real - time formulation of thermal perturbation theory , the thermal contribution to the self - energy diagram of fig . [ fig : pi]a is given by the diagrams of fig . [ fig : pi2 ] , which represent forward scattering of the soft excitation off of other excitations in the thermal bath . the imaginary part corresponds to cutting these diagrams and putting the cut line on shell . for the retarded self - energy , where @xmath199 is treated as @xmath360 , this cutting yields the net decay rate ( actually , the rate times @xmath361 ) of excitations : @xmath362 is the _ difference _ @xmath363 shown in fig . [ fig : impi2 ] of the square amplitudes @xmath364 for destroying soft quanta and @xmath365 for creating them . by the principle of detailed balance ( which is indeed obeyed diagrammatically ) , these amplitudes must satisfy @xmath366 from which one then has @xmath367 so return to ( [ eq : start ] ) . taking the limit that both @xmath199 and @xmath55 are small and the limit that @xmath24 gives @xmath368 the sum over flavors is now restricted to pairs of flavors that have degenerate dispersion relationships . the result ( [ eq : gamma ] ) , or its generalization to other theories , will now follow if @xmath369 can be replaced by @xmath370 : that is , if soft magnetic gauge bosons indeed couple to group velocity . i ll argue this by surveying a few instructive examples . as the first class of theories to check , start with a continuum scalar qed theory with a lagrangian with scalar interactions of the form @xmath371 where @xmath8 and @xmath116 are flavor indices . for the one flavor case , everything is trivial . the dispersion relationship is @xmath372 , and the trilinear coupling with the photon is @xmath373 ; so @xmath374 has the desired form . the case of multiple flavors is slightly less trivial and is worth understanding because it is analogous to the case of multiple branches in lattice non - abelian gauge theory . let @xmath375 be the unitary matrix that diagonalizes @xmath376 in flavor space , so that @xmath377 is diagonal . it s diagonal elements @xmath378 are just the @xmath310 for the different branches . in this energy eigenbasis , the trilinear coupling is not in general diagonal : @xmath379 \,,\ ] ] so that @xmath380 but the degeneracy condition on the flavor sum in ( [ eq : almost there ] ) eliminates the second term above , and so one obtains @xmath381 this is indeed the trivial generalization of ( [ eq : gamma ] ) to include a sum over branches . now consider scalar qed on a spatial lattice , with the gradient energy given by some sum of local terms of the form @xmath382 where @xmath383 is the product of u(1 ) link matrices along some path from @xmath85 to @xmath384 . ( the canonical case would be that @xmath85 and @xmath384 are nearest neighbors and the path is the single link between them . ) i am only interested in the @xmath385 coupling in perturbation theory and only in the limit of soft momentum for @xmath46 . in this limit , @xmath386 and the coupling is minimal . that is , the coupling has the form ( [ eq : scalar toy ] ) in momentum space in the soft photon limit , and the previous discussion applies . similar considerations hold for non - abelian lattice gauge theories . schematically , write @xmath46 = @xmath387 where @xmath388 are the soft - momentum modes and @xmath389 the hard momentum modes . under _ soft _ gauge transformations @xmath390 , we have @xmath391 and @xmath392 . gauge invariance under soft gauge transformations then implies that the @xmath393 coupling must be minimal in the soft limit . that is , it again has the form of ( [ eq : scalar toy ] ) in the soft limit with @xmath315 representing @xmath389 and @xmath394 where @xmath123 are the ( anti - hermitian ) adjoint - representation generators . as a check , it is easy to explicitly verify that the soft - hard - hard coupling of the kogut - susskind hamiltonian on a simple cubic lattice indeed has @xmath395 , where @xmath8 and @xmath116 are the polarizations of the hard gluons . for the sake of completeness , i will explicitly write down the two - dimensional integrals corresponding to ( [ eq : gamma ] ) that i did numerically to produce figs . [ fig : impi]b and c. where the @xmath400 sum is over the two dispersion relations ( [ eq : fcc omega ] ) ; @xmath280 and @xmath401 are given by ( [ eq : fcc epsilon ] ) and ( [ eq : fcc v ] ) with @xmath402 ; the @xmath193 sum is over two values @xmath397 with and @xmath404 \ , , \\ b & \equiv & \hat k_z \left [ \cos\left(p_x\over2\right ) + \cos\left(p_y\over2\right ) \right ] \ c & \equiv & - \hat k_x \sin\left(p_x\over2\right ) \cos\left(p_y\over2\right ) - \hat k_y \sin\left(p_y\over2\right ) \cos\left(p_x\over2\right ) \,.\end{aligned}\ ] ] the @xmath405 sign in ( [ eq : fcc pz ] ) is unrelated to that in ( [ eq : fcc omega ] ) .

in the restframe ultra - violet ( uv ) , two of the parameters that best characterize the range of emission - line properties in quasar broad emission - line regions are the equivalent width and the blueshift of the line relative to the quasar rest frame . we explore the connection between these emission - line properties and the uv through x - ray spectral energy distribution ( sed ) for radio - quiet ( rq ) quasars . our sample consists of a heterogeneous compilation of 406 quasars from the sloan digital sky survey ( at ) and palomar - green survey ( at ) that have well - measured emission - line and x - ray properties ( including 164 objects with measured ) . we find that rq quasars with both strong emission and small blueshifts can be classified as `` hard - spectrum '' sources that are ( relatively ) strong in the x - ray as compared to the uv . on the other hand , rq quasars with both weak emission and large blueshifts are instead `` soft - spectrum '' sources that are ( relatively ) weak in the x - ray as compared to the uv . this work helps to further bridge optical / soft x - ray `` eigenvector 1 '' relationships to the uv and hard x - ray . based on these findings , we argue that future work should consider systematic errors in bolometric corrections ( and thus accretion rates ) that are derived from a single mean sed . detailed analysis of the emission line may allow for sed - dependent corrections to these quantities .

1109.1515

while the broad emission - line region ( belr ) in active galactic nuclei ( agn ) remains an enigma , in recent years there has been considerable focus on radiation line driven and magneto hydrodynamic driven winds from an accretion disk ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) as a key component . the existence of such winds has been demonstrated for accretion disks in cataclysmic variable star systems and hydrodynamic simulations show that the phenomenon scales to the much larger quasar black hole masses ( e.g. , * ? ? ? * ) ; however , the detailed gas and kinematic structure of the wind remains uncertain . this is an acute problem , which is important to our understanding of black hole growth and its influence on the evolution of the host galaxy , especially if such winds are a significant source of agn `` feedback '' ( e.g. , * ? ? ? * ; * ? ? ? perhaps the most important / most common wind diagnostic is the emission - line blueshift @xcite , which we have studied for a large sample of quasars @xcite and in the context of the narrow - line seyfert 1 ( nls1 ) class @xcite . specifically , high - ionization emission lines , like @xmath01549 are found to peak at wavelengths blueward of their expected laboratory wavelength when referenced to low ionization lines ( which are more indicative of the quasar systemic redshift ; e.g. , @xcite ) . as far back as the early 1980s , this `` blueshift '' had been suggested as the signature of an outflow @xcite . while the exact physical origin of this blueshift is not agreed upon @xcite , one possibility is that it is related to the strength of the radiation pressure compared with the black hole mass and is sensitive to the eddington accretion rate @xcite . it may be that nls1s are the low - luminosity , local analogues of large civ blueshift quasars . if the typically steep hard - band photon indices of nls1s ( @xmath4 ) are signatures of high eddington accretion rates @xcite , then the same might be the case for large civ blueshift quasars ( e.g. , * ? ? ? * ) since a high accretion rate for a given mass is expected to be more conducive to wind driving through radiation pressure on spectral lines @xcite . indeed , the common link between large blueshift quasars , broad absorption line quasars ( balqsos ; @xcite ) , and traditionally defined nls1s @xcite is likely a strong radiation - pressure driven wind as all three categories exhibit wind features in emission and/or absorption @xcite and may tend toward high @xmath5 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the premise of a radiation line driven wind ( e.g. , * ? ? ? * ; * ? ? ? * ) is fairly simple and has applications to quasars ( e.g. , * ? ? ? * ; * ? ? ? the integrated opacity in ultra - violet ( uv ) bound - bound transitions accelerates gas as photons at specific wavelengths are scattered by ions in the gas . as the gas velocity increases , the doppler shift makes photons in new wavelength ranges able to contribute to the acceleration . this process allows a large amount of momentum to be extracted from continuum and emission - line photons , thus driving an outflow from the accretion disk . for this process to work , the quasar must reduce its x - ray continuum such that the ions which absorb in the uv are not destroyed by over - ionization . to some extent , we know that outflows driven at least in part by radiation pressure must be occurring in balqsos because we see large equivalent width absorption troughs where both energy and momentum have been removed from the continuum @xcite . in this paper , we follow - up on the work of @xcite , @xcite , and @xcite using sdss quasars at @xmath1 where the presence of emission in the optical provides a powerful diagnostic of accretion disk physics . in particular , @xcite showed that a number of different emission - line features were consistent with a two - component disk+wind model of the belr ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . emission - line features suggestive of a weaker than average ionizing spectrum ( relative to the uv ) were found in quasars with weak , highly blueshifted emission . on the other hand , quasars with emission - line features that indicate a strong ionizing spectrum were found to have very strong emission lines at the systemic redshift . these extrema can be crudely categorized respectively as being dominated by a `` wind '' component and by a `` disk '' component of the belr , although it seems that both components may be present in all objects . while some of the differences seen in the belr are likely derived from intrinsic differences in the spectral energy distribution ( sed ; e.g. , a large _ intrinsic _ uv to x - ray flux ratio is needed to produce a strong radiation line driven wind ; @xcite ) , processing of the ionizing continuum _ through _ the wind can have a significant affect on the emission - line features that arise within the disk component @xcite . here we explore the x - ray properties of quasars ( such as the optical / uv to x - ray flux ratio , @xmath6 ) over the full range of properties to see if the evidence that the extrema in emission - line properties traces extrema in the ionizing continuum is consistent with the x - ray properties of quasars . while our work focuses specifically on the emission line , our results should be viewed in the context of the broader `` eigenvector 1 ( ev1 ) '' literature that stemmed from the ground - breaking work of @xcite . it is beyond the scope of this paper to provide a full review of this literature , but a newcomer to the field would be well served by a review of @xcite and @xcite and references therein ; table 2 in the latter nicely summarizes a broad range of correlated emission and continuum properties of quasars . while much of the ev1 work is founded in low - redshift quasars and investigations of optical emission lines , investigations that bridge the gap to high - redshift quasars and uv emission lines include @xcite , @xcite , and @xcite . more recent work focused on includes @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . lastly , @xcite and @xcite are among those notable for making connections between optical / uv emission lines and x - ray properties of quasars . this paper is organized as follows . in section [ sec : data ] we describe the data used in our analysis . in section [ sec : analysis ] we investigate the range of x - ray properties as a function of emission - line parameters . in section [ sec : discussion ] we discuss the resulting seds . finally , our conclusions are presented in section [ sec : conclusions ] . equivalent widths are given in units of rest - frame , and velocities in kms@xmath7 with a sign convention such that positive velocities represent outflows in the quasar frame . spectral indices are given according to @xmath8 throughout , such that @xmath9 represents the local slope of the sed in a log - log plot . our investigation is based on archival data described in section [ sec : archive ] and new _ data described in section [ sec : newdata ] . many of the selection criteria used are common to all samples and are summarized here . in general , we have limited ourselves to quasars included in the @xcite quasar catalog from the 7th data release ( dr7 ; @xcite ) of the sloan digital sky survey ( sdss ; @xcite ) such that we can make use of the sdss spectroscopy of these sources . the exception is for some pg / bqs @xcite quasars where there exists space - based uv spectra of the region . we use the rest - frame eqw and blueshift as our main diagnostics . we have chosen not to include the commonly - used fwhm measurement in our analysis as that parameter is difficult to interpret in the disk+wind model of the emission region that we have adopted in @xcite . this concern is supported by the work of @xcite , who found that has contributions from both an outflowing and a gravitationally bound component ( the relative contributions of which may be dependent on the eddington ratio ) . @xcite showed that fwhm provides a biased estimator of the black hole mass , but argued that this bias can be calibrated out in the ensemble average . except for the pg quasars , we matched the objects to the database that we used in @xcite and extracted our own values for the line parameters and corrected sdss redshifts as given by @xcite . balqsos and radio - loud ( rl ) quasars were excluded as the former are known to be absorbed in the x - ray ( e.g. , * ? ? ? * ; * ? ? ? * ) and the latter are known to have jet - enhanced x - ray emission ( e.g. , * ? ? ? * ; * ? ? ? balqsos were taken from @xcite and rl quasars were defined according to @xmath10 ( e.g. , * ? ? ? * ) , or ( for the pg quasars ) @xmath11 ; see @xcite . the following criteria were applied to limit the samples to objects with robust measurements from sdss and that are not significantly reddened ( see @xcite for more details ) : @xmath12 , eqw@xmath13 5(rest frame ) , @xmath14 ( an error code ) , @xmath15 10 , fwhm@xmath13 1000 , fwhm@xmath16 , eqw@xmath17 , @xmath18 @xcite . we further excluded objects with error codes as noted in the papers below . finally , given the heterogeneous nature of the data set , we do not attempt to include objects with x - ray non - detections using partial correlation analysis ( e.g. , * ? ? ? * ) ; however , this choice imposes a bias that should not be completely neglected . table [ tab : tab1 ] lists all of the data sources , the number of objects matched to our database of sdss - dr7 spectral properties , the number removed because they are bals , rl , outside of the redshift range covered ( must be visible in the sdss spectra ) , or have optical properties that do not meet the above criteria . objects can be rejected for more than one reason . thus , we also tabulate the final number of objects kept ( after resolving duplicate entries [ in favor of the most recent data ] between the samples ) . finally , we tabulate the redshift and @xmath19 range of the objects kept from each sample . the objects from @xcite are not tabulated in table [ tab : tab1 ] as they are not matched to the sdss database ; see below . in the end , we have a sample of 409 unique quasars . figure [ fig : distribution ] shows the distribution of the objects in eqw - blueshift parameter space . of these objects , 164 have values for the x - ray photon index , @xmath3 ; their distribution is shown in figure [ fig : gamdistrb ] . in @xcite we used _ chandra _ to explore the x - ray properties of sdss quasars at the extrema of the blueshift distribution ; archival observations filled in the blueshift distribution and provided additional objects at the extrema . in the interim many more investigations of ( largely ) serendipitous quasar observations have been conducted by a number of groups ( culminating in the _ chandra _ source catalog project ) , greatly simplifying the use of archival data . as such , in addition to new _ observations described below , this investigation takes advantage of archival x - ray data from the following sources : @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . see @xcite for an investigation of objects even more extreme than considered herein . the values taken from the following papers were @xmath20 , @xmath6 , @xmath21 , and @xmath3 , where @xmath22 is the flux ratio between the x - ray portion of the sed ( at @xmath23 ) and the optical / uv portion ( at @xmath24 ) of the sed . we will follow standard convention and abbreviate @xmath25 as @xmath19 . @xmath21 is the luminosity - corrected value of @xmath6 , defined by @xmath26 using the @xmath27@xmath6 relationship from @xcite . our sign convention is such that for both @xmath6 and @xmath21 , more negative values indicate ( relatively ) weaker x - ray sources . @xmath3 is the x - ray photon index , which is related to the standard spectral index in the x - ray regime by @xmath28 . if @xmath20 was not given , we instead used @xmath29 from our database . for a power - law spectral index of @xmath30 @xcite , @xmath31 ( and @xmath32 ) . while our 2800 luminosities are continuum luminosities and @xmath20 typically includes emission - line flux , the difference is typically negligible when it comes to computing @xmath6 ( @xmath6 changes by only @xmath33 for a 10% line contribution ) . if @xmath21 was not given , we derived it from @xmath20 and @xmath6 . only some of the samples included measurements of @xmath3 . @xcite cataloged 155 sdss - dr2 quasars in _ rosat _ medium - deep fields . in all , 86% were detected in the x - ray . the _ rosat _ objects had an average exposure time of 16.7ks and were not targeted by sdss solely because of their x - ray detections ; only four _ rosat _ detections were from pointed observations of the quasars themselves . from their table 1 we extracted @xmath20 and @xmath6 . there is no information on @xmath3 for these data . all other parameters ( e.g. , and @xmath21 ) were determined as described above . @xcite extended the work of @xcite to lower luminosities using data from the combo-17 survey in the e - cdf - s @xcite and with 46 low-@xmath34 quasars from the pg sample . as @xcite provide the measurements for the pg sample , we were able to include the objects from the pg sample in our analysis ; however , these are not matched to the sdss spectral database and thus are not tabulated in table [ tab : tab1 ] . table 2 from @xcite provided values for @xmath35 and @xmath6 ; no information on @xmath3 is available . no sdss data exist for the combo-17 objects and they are not included here . in addition , while @xcite also utilize 19 @xmath36 quasars , many of these are outside of the sdss footprint as such , they lack other parameters needed for our analysis and they are also excluded . @xcite report x - ray properties for seven new high - redshift quasars and 167 archival quasar with @xmath37 ; of these 157 match to our sdss - dr7 spectral database . the values for @xmath6 , @xmath3 , and @xmath20 were extracted from their table 4 . the energy range for @xmath3 was 0.37.0kev ; most sources had insufficient counts to fit simultaneously for intrinsic @xmath38 . using these data , @xcite confirmed ( at the @xmath39 level ) the anti - correlation between @xmath3 and @xmath40 seen by @xcite . @xcite examine 59 highly luminous quasars with @xmath41 , including 21 objects with new _ observations ; 32 objects are sdss - dr3 quasars . from their table 4 , we extracted @xmath20 , @xmath6 , and @xmath21 . @xmath3 values are taken from their table 3 ; the energy range was 0.38.0kev and the signal - to - noise ratio was not high enough to justify additional constraints on @xmath42 for individual sources . for most sources no intrinsic absorption was allowed when fitting for @xmath3 ; however , sources with over 200 counts were simultaneously fit for @xmath3 and @xmath42 . @xcite catalog _ chandra _ and _ xmm - newton _ observations of 536 sdss - dr5 quasars covering @xmath43 . most of the observations are serendipitous ; less than 9% of the quasars were the targets of the x - ray observations . we obtained @xmath20 , @xmath6 and @xmath21 from their table 1 . no x - ray spectral indices ( @xmath3 ) are available . using these data @xcite argue that the spread of @xmath21 is mostly due to variability ( see also @xcite ) and that the fraction of quasars that are intrinsically x - ray weak by a factor of 10 or more is @xmath44% . @xcite present _ chandra _ data for 1135 spectroscopic and photometric sdss - dr6 quasars analyzed by the chandra multiwavelength project ( champ ; @xcite ) ; we include the fraction of these that have sdss spectroscopic redshifts . from their table 2 we extract values for @xmath3 , @xmath20 , and @xmath6 ( flipping the sign of @xmath6 to agree with our convention ) . for objects with more than 200 counts , @xmath42 was fit in addition to @xmath3 ; the energy range was 0.58.0kev . finally , @xcite tabulate serendipitous _ xmm - newton _ observations of 792 sdss - dr5 quasars . 473 of these quasars had sufficient s / n to determine @xmath3 ( and @xmath42 , if warranted ) . the spectral range for fitting was 0.510kev . @xmath3 and @xmath6 values are taken from their table 2 . while our goal is to understand the physics of quasars in the ensemble average , we often gain critical insight by first trying to understand the extrema . this is as true for blueshifts as it is for any other quasar parameter . as outliers in any distribution are rare , we have expanded the sample of high - blueshift quasars with x - ray data by observing seven new radio - quiet sources in _ chandra s _ cycle 9 . we also obtained additional time on three sources from @xcite to improve the s / n for those sources . the new targets were chosen to be the highest blueshift quasars ( @xmath45 ) from the sdss - dr5 sample that met the following criteria . redshift was limited to @xmath46 such that both civ and mgii are seen . balqsos and rl quasars were excluded as discussed above . finally , to maximize our x - ray counts , we further limited the sample to @xmath47 and galactic @xmath48 . the targets are summarized in table [ tab : chandra ] . exposure times were set so as to achieve 100 total counts for each object . all targets were detected with between 63 and 140 counts , and fluxes , luminosities , and photon indices were derived from the soft- and hard - band photometry following the methods described in @xcite . spectra with sufficient counts were analyzed with xspec @xcite using unbinned data and c - statistics to constrain @xmath42 and @xmath3 . the x - ray properties of the sample are summarized in table [ tab : xcalc ] and these objects are combined with the archival data above in our analysis . three sources in the @xcite sample , j0051@xmath490102 , j0147@xmath500001 , and j0208@xmath500022 , were reobserved in our new program with the goal of increasing the counts in the combined spectra for each object . between the observations , the sources showed significant flux variability of @xmath4952% , @xmath5026% , and @xmath5045% , respectively . while the @xmath51 values between epochs were consistent within the 1@xmath52 error bars , the low counts in the individual spectra could obscure significant spectral variability as well . combined fitting of the new and old data together was inconclusive ( in terms of the presence of absorption ) , likely because of the substantial variability between epochs . the key results from @xcite and @xcite were to provide a bridge between properties of the broad emission lines in quasars with their spectral energy distributions . with this in mind , we further consider the x - ray properties of the heterogeneous data set described above . we specifically consider the distributions of @xmath6 , @xmath21 , and @xmath53 in the eqw - blueshift parameter space as defined by @xcite . if the differences seen across the range of emission - line properties are indicative of a range of abilities to drive a strong accretion disk wind , and if radiation line driving plays a significant role in driving such a wind , then we might expect to see differences in x - ray properties of quasars across the eqw - blueshift parameter space . ultimately , our hope is that this work is a first step toward improving the effective resolution of quasar seds in the `` unseen '' part of the continuum that covers nearly two decades in frequency between the optical and the x - ray . in a disk - wind picture , quasars with radiation - pressure dominated winds should be _ intrinsically _ x - ray weak relative to the optical / uv and are likely to display x - ray warm absorbers @xcite that may _ additionally _ reduce the soft x - ray flux along our line of sight . our working hypothesis is that an object s location in the eqw - blueshift parameter space reflects its intrinsic ability to drive a strong accretion disk wind . assuming that radiation line driving dominates the wind component , we expect to see a reduction in @xmath6 across the eqw - blueshift parameter space ( from top left to bottom right in figure [ fig : distribution ] ) as the hypothetical wind strength increases . see ( * ? ? ? * figure 7 ) , for evidence of this behavior in nls1s . figure [ fig : aoxhist ] gives the range of @xmath6 values in the sample , while figure [ fig : aox ] shows how @xmath6 changes across the eqw - blueshift parameter space using two different diagnostics . first we show the location of the individual quasars in our heterogeneous sample color - coded by @xmath6 ( circles ) as indicated by the color - bar on the right - hand side . the dashed lines indicate the divisions used in @xcite to create composite spectra as a function of emission - line properties ( see also section [ sec : seds ] ) . there appears to be a general trend towards more negative @xmath6 values ( objects relatively weaker in the x - ray ) from the top - left to the lower - right . however , as there is considerable scatter in the diagram , we also bin the results to better discern the general trend . specifically the colored squares indicate the median values of eqw and blueshift after sorting the values of @xmath6 and grouping them into 10 bins . although the median values span a much smaller range in properties than the raw sample , they clearly reveal a general trend towards x - ray weaker objects in the lower right - hand corner . further evidence for a trend in @xmath6 with properties comes from performing a student s @xmath54-test for different means ( e.g. , * ? ? ? * ) using the objects in the upper - left and lower - right regions of the parameter space in figure [ fig : aox ] . we find that the mean values for @xmath6 in those bins are @xmath55 and @xmath56 , respectively , with a standard deviation of @xmath57 , yielding a @xmath54 value of @xmath58 with very high significance that the samples have different means . this trend of decreasing relative x - ray strength as we move from the upper - left ( large eqw , small blueshift ) to the lower - right ( small eqw , large blueshift ) in figure [ fig : aox ] is consistent with the expectations of a model of the belr where a radiation line driven wind is influencing the properties of . when the x - ray flux is high , a strong radiation line driven wind can not form ( though perhaps there is still an mhd - driven wind ; @xcite ) and the ionizing continuum photons can create a large population of triply ionized carbon atoms in the outer part of the accretion disk . if instead , the x - ray flux is weaker , a strong radiation line driven wind can form , shielding the outer accretion disk from an already weaker x - ray flux . in this case the line is formed mostly in the wind itself and may be blueshifted due to the outflowing nature of the source and lower in equivalent width due to the overall reduction in ionizing photons @xcite . it is less clear what is happening in the lower - left corner ( small eqw , and small blueshifts ) , but it would appear that those objects have more heterogeneous @xmath6 values than the objects in the upper - left or lower - right corners . third parameters , such as covering fraction , orientation , absorption , and/or variability effects could explain objects in the lower left - hand corner ( ( e.g. , * ? ? ? * ) ; bowler , allen , & hewett , in preparation ) . as noted in @xcite , objects with both large blueshift and large eqw do not seem to exist and this is unlikely to be a selection effect . our results are consistent with the results of @xcite who have done a multivariate regression of eqw against both @xmath19 and @xmath6 . they find that eqw correlates with @xmath6 such that harder spectra have stronger . what our work does is to further separate low eqw objects in another dimension . while there is no degeneracy for large eqw quasars , those with smaller eqw values can have a range of blueshifts . since @xmath6 is also correlated with blueshift ( in a way that is not trivially related to the eqw ) , it is important to explore trends in @xmath6 in this 2-d parameter space . student s t results : aox t = 9.85303744323 significance = 1.0071271342e-16 mean and s_d values : aox bin 1 mean : -1.49680769231 bin 8 mean : -1.70046551724 s_d : 0.0206695474474 as it is well known that @xmath6 is correlated with uv luminosity ( e.g. , * ? ? ? * ; * ? ? ? * ) , the above trends with @xmath6 in the previous section might instead be ascribed to luminosity . indeed , we find that uv luminosity is increasing from the upper - left to the lower - right in figure 1 . as such , it has become common to reference @xmath6 values to the mean @xmath6 values observed for objects with the same uv luminosity according to @xmath60 . here we adopt the expected value of @xmath6 from @xcite , specifically @xmath61 . @xmath59 indicates the relative amount of x - ray emission , whereby negative values of @xmath59 indicate a relative deficit of x - rays , while positive values indicate an x - ray surplus . figure [ fig : daoxhist ] shows the distribution of @xmath21 values in our sample . by using @xmath21 instead of @xmath6 , luminosity is marginalized and any remaining sed effects must instead be due to the _ shape _ of the sed rather than its absolute value . it is therefore significant that figure [ fig : daox ] shows the same general trend as does figure [ fig : aox ] . while taking out the luminosity dependence has apparently reduced the trend for individual objects , the median values of eqw and blueshift ( using 10 bins sorted in @xmath21 ) again show a significant trend from more x - ray luminous objects in the upper left - hand corner to relatively weaker x - ray sources in the lower right - hand corner . @xcite similarly find that low eqw objects and large blueshift objects have more negative values of @xmath21 ( their figures 8 and 10 ) , although it should be noted that half of our sample comes from their paper , so our results are not independent . as with @xmath6 , a student s @xmath54-test confirms that the mean values of @xmath21 in the upper - left and lower - right parts of the distribution in figure [ fig : daox ] are significantly different . we find respective mean values of @xmath62 and @xmath63 with a standard deviation of @xmath57 , yielding @xmath64 with high significance . unfortunately our understanding of the x - ray sed data is in roughly the same state as our understanding of the optical / uv sed was in the 1990s . @xcite were able to establish that the optical / uv spectral index was @xmath65 with an error of @xmath66 ; however , it was nt until @xcite that the photometric errors were small enough that the @xmath67 distribution could be shown to have an intrinsic spread rather than a single characteristic value smeared by measurement errors . in the case of the x - ray , variability likely plays more of a role than measurement errors especially since both the optical and x - ray measurements are subject to changes over time and are non - simultaneous . indeed , the majority of the width in the @xmath21 distribution can be attributed to variability @xcite . the width of the @xmath21 distribution notwithstanding , there does appear to be a trend towards relatively weaker x - ray sources in objects with large blueshifts . while the @xmath6 and @xmath21 trends with eqw and blueshift are expected in a radiation line driven wind model , we have less intuition regarding how the x - ray spectral indices ( @xmath68 ) should behave . at low redshift , where x - ray data sample the soft x - ray part of the spectrum , it is observed that @xmath69 correlates with the strength of ( relative to h@xmath70 ) and anti - correlates with the fwhm of h@xmath70 @xcite ; x - ray softer objects thus have stronger and narrower h@xmath70 @xcite . in the context of black hole binaries , such soft - spectrum objects ( including nls1s ; @xcite ) are thought to have high accretion rates @xcite . unfortunately there are relatively few objects with spectral coverage of both the h@xmath70 and line , so it is hard to know how these trends would propagate to higher redshift objects . however , using _ hubble space telescope _ data on for low-@xmath34 quasars , @xcite have shown that large blueshift objects tend towards softer @xmath69 . the @xmath3 values tabulated here ( fig . [ fig : gammahist ] ) generally refer to the harder part of the x - ray spectrum from 2kev to 10kev , so it does not necessarily follow that large blueshift objects should tend towards softer @xmath71 ; however , that does seem to be the case . if higher uv luminosity results in more compton cooling of the corona , then one indeed might expect @xmath71 to follow @xmath69 @xcite and that quasars with large blueshift might have high accretion rates . figure [ fig : gamma ] shows a weak trend of increasing ( softer ) @xmath3 with increasing blueshift and decreasing eqw . as with the @xmath6 and @xmath21 figures the median properties as a function of @xmath3 show a clearer trend than can be discerned from looking at the distribution of individual objects . we find a student s @xmath54 of @xmath72 with significance of @xmath73 , showing that there is a clear difference when comparing the means of objects in the top - left ( @xmath74 ) and the lower - right ( @xmath75 ) parts of figure [ fig : gamma ] ; the standard deviation is @xmath76 . clearly further work is needed to better understand whether the the harder @xmath3 values for strong quasars are indicative of the intrinsic x - ray spectrum . the values of @xmath6 , @xmath21 , and @xmath3 are perhaps more meaningful within the context of the overall sed . as such , in figure [ fig : sed ] we create typical seds spanning the eqw - blueshift bins used by @xcite ; the properties of each sed are shown in figure [ fig : sedpos ] . for this analysis , we imposed a @xmath77 limit such that both and are observed and a uv spectral index between the continua underlying those lines could be determined empirically . the seds were created using the _ median _ values of l@xmath78 , @xmath79 , @xmath6 and @xmath3 for each sub - sample , where the first two quantities were taken from our own work and the last two quantities were taken from the literature as described above ( with the exception of our new _ chandra _ data ) . to make the seds , the median value of @xmath80 ) was used with the median @xmath79 to calculate a value for log(l@xmath81 ) at 2500 . in turn , log(l@xmath82 ) was used with @xmath6 to determine a value for log(l@xmath83 ) . then @xmath53 was used to plot a point at @xmath84 . in addition to our optical and x - ray results , we have included the uv results of @xcite who found that the uv spectral index over @xmath856501100 is a function of luminosity . specifically , we have used the results of table 4 and figure 18 from @xcite and our value of @xmath86 ( as a proxy for @xmath87 used by @xcite ) to estimate the spectral slope over @xmath88 [ @xmath89})>15.39 $ ] ] . the luminosity dependences of @xmath40 and @xmath6 appear to be fairly comparable as the break at 500 is difficult to discern , which is consistent with the findings of @xcite . we emphasize that there is little information on the shape of the sed between 500 and @xmath90kev for luminous quasars ; @xmath6 only parametrizes the uv to x - ray flux ratio , not the actual shape of the sed , which has implications for bolometric corrections ; see section [ sec : bc ] . interestingly , those objects that are relatively stronger at 2kev relative to 2500 ( i.e. , have a harder ionizing spectrum ) are also apparently harder ( i.e. , flatter ) over 210kev . although the differences between the seds are subtle , the effects are clearly systematic . relative to their 2500 luminosities , the large eqw objects have `` harder '' seds than the large blueshift objects . we illustrate this more clearly in the inset of figure [ fig : sed ] by normalizing all of the seds to the same 2500 luminosity . it is important to realize that while the @xmath6 distribution for these sdss quasars nowhere near spans the extrema as characterized by the weak - lined quasar phl 1811 ( @xmath91 ; @xcite ) and narrow - line seyfert 1 galaxy re 1034 + 39 ( @xmath92 ; @xcite ) , nevertheless , for the same uv luminosity , a change in @xmath6 of just @xmath93 corresponds to a change in flux at 2kev of a factor of 3 . this difference is simply a reflection of the large lever arm between 2500 and 2kev ( 2.6 decades in frequency ) . as such it is not unreasonable to expect significant differences in the ability of quasars to drive a wind through radiation line pressure as we move across the eqw - blueshift parameter space . since there are few _ empirical _ constraints on the 500 to 0.2kev part of the sed , it is useful to compare our empirical seds to some example model seds . we specifically consider model seds similar to those described in @xcite in accordance with the agn spectrum in the cloudy package @xcite . the optical / uv portion of the sed is modeled as a power law with exponential cutoffs at high and low energies as parametrized by a cutoff temperature ( the temperature of the inner edge of the accretion disk for the high energy cutoff ) . the optical spectral index is taken to be @xmath94 and the x - ray spectral index is fixed to @xmath95 . the optical / uv part of the spectrum is related to the x - ray part of the spectrum by the well - known @xmath27@xmath6 relationship ( e.g. , * ? ? ? note that , in this model , there is a degeneracy between @xmath96 , @xmath27 , @xmath97 , and @xmath98 that means that the seds are uniquely described by a single parameter that can be characterized by @xmath6 . in figure [ fig : bb ] we show a model sed ( solid black line ) that best represents the average quasar in our sample . two examples that are meant to bracket the range of seds spanned by our sample ( in terms of @xmath27 and @xmath6 ) are also shown . the two extreme seds shown have cutoff temperatures of 20ev and 50ev and have , respectively , @xmath99 and @xmath100 , and @xmath101 of 31.8 and 30.0 . as noted by @xcite , such model seds with very `` hot '' accretion disks are apparently inconsistent with the uv part of empirical seds ( e.g. , * ? ? ? * ; * ? ? ? indeed , the extreme models shown in figure [ fig : bb ] have spectral indices of @xmath102 and @xmath103 at 800 , respectively , which are within the range observed by @xcite , but are too blue / hard for these luminosities as can be seen by comparison with our empirical seds ( dotted lines ) . however , there is evidence that the sed the emission - line gas sees may not be the same as the sed that we see @xcite and our work herein and in @xcite might be taken as further evidence of such a situation . if both the observer and the emitting gas _ do _ see the same sed then , if this model parametrization is accurate , our empirical data would suggest that a lower cutoff temperature might be more appropriate for the luminous quasars in our sample . we are currently pursuing a more detailed characterization of the near - uv part of the spectrum for sdss quasars ( krawczyk et al . 2011 , in preparation ) . aside from the near - uv slope mismatch , the 20ev peaked spectrum is representative of the most luminous quasars in our sample ( in terms of @xmath6 ) . the relatively soft spectrum ( x - ray weak relative to uv ) would be conducive to driving a strong wind through radiation line driving , yielding a large blueshift . on the other hand , the 50ev peaked spectrum is more representative of the less luminous quasars in our sample . for such quasars , the relatively hard spectrum is more likely to overionize the accretion disk atmosphere and has less uv flux , both of which would tend to inhibit a radiation line driven wind . this harder sed is one that we associate with quasars that have strong emission that is dominated by the disk component of the belr . although there is a lack of data to characterize the shape of the sed over much of the crucial ionizing part of the spectrum , our model seds represent extrema characterizing the ionizing sed for a given @xmath6 ( c.f . , the simple power - law model , dotted lines in figure [ fig : bb ] ) in that they have as much far - uv flux as one could imagine . thus , these extrema are particularly interesting to consider in terms of bolometric corrections ; see section [ sec : bc ] . an intermediate model might be one that has a cooler accretion disk with a comptonized extension to soft x - ray energies ( e.g. , * ? ? ? * ; * ? ? ? * ) , with another , hotter comptonized region producing the hard x - ray spectrum . for example , even if the accretion disk is not as hot as depicted by the models in figure [ fig : bb ] , @xcite suggest that quasars with strong ( which tend to have strong at the systemic redshift ; @xcite ) would require a factor of a few higher flux at @xmath104ev than most empirical seds ( including ours ) would suggest . x values for all seds : [ 15.079181246047625 , 15.286789556549371 , 15.778151250383644 , 17.68444107586626 , 18.383411080202279 ] y values : 1 : [ 45.824299789225691 , 45.895555533849368 , 45.71531315062834 , 44.521669874316373 , 44.591566874749972 ] 2 : [ 45.832601746814539 , 45.879568986049371 , 45.703175639659065 , 44.347603643817912 , 44.47691309462008 ] 3 : [ 46.05948146646201 , 46.143108022699373 , 45.90326311246357 , 44.274878483036247 , 44.072177181778798 ] 4 : [ 45.99986485309438 , 46.092615352549373 , 45.864927422639823 , 44.46276155350138 , 44.508194603783224 ] 5 : [ 45.975707955510615 , 46.064411513249368 , 45.843514143602853 , 44.308341664426678 , 44.25941376412316 ] 6 : [ 46.038919794266711 , 46.12732101114937 , 45.891277095954109 , 44.423658699779153 , 44.472586600082678 ] 7 : [ 46.106010890707509 , 46.210943690549371 , 45.954766174471573 , 44.3865394030272 , 44.407508503157288 ] 8 : [ 46.384655655710873 , 46.477307544749372 , 46.156998341329356 , 44.53492117653964 , 44.465024176106034 ] comparison of the shape of the model seds in figure [ fig : bb ] is relevant not only for bolometric corrections , but may also relate to the origin of soft x - ray excesses ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? specifically , since the uv through x - ray seds of radio - quiet quasars appear to change significantly from one corner of parameter space to the other , it is interesting to ask where soft x - ray excess objects lie in the eqw - blueshift parameter space . unfortunately the bandpasses of _ chandra _ and _ xmm - newton _ are such that it is difficult to identify soft x - ray excesses in high - redshift quasars . however , the 21 low - redshift pg quasars studied by @xcite have uv measurements of from @xcite , @xcite , and @xcite and can be placed in our parameter space . we find that the @xcite objects are generally in the upper - left of the parameter space ( i.e. , they tend to have large eqw and small blueshift , which , ironically , suggest that soft excess objects are actually x - ray hard [ relative to the uv ] ) . although the @xcite objects are much less luminous than the average quasars in our sample ( @xmath105 ) , they are actually well matched to the lowest luminosity ( and generally hard - spectrum ) objects of our sample ( being bright objects at low redshift as opposed to faint objects at high redshift ) . as such , it would be valuable to explore the possibility of correlations between the properties of quasars and the presence of a soft x - ray excess . we specifically predict that the average high - luminosity , high-@xmath34 quasar does not have an x - ray soft excess based on the relative locations in the 2d parameter space of low-@xmath34 soft excess objects . lastly , we note that a strict linear interpolation between 2500 and 2kev would predict a _ significant _ soft x - ray excess in all of our sources ( compare the empirical and model seds in figure [ fig : bb ] for energies higher than 0.2kev ) . a lack of such a strong soft x - ray excess in luminous quasars would mean that while @xmath6 may accurately indicate the 2kev flux density in quasars , it is a poor characterization of the shape of the true ionizing sed over @xmath1061000ev and we reiterate that the models shown in figure [ fig : bb ] essentially represent the opposite extrema . here we close by merely pointing out that the apparent hard - spectrum nature of disk - dominated belrs and soft - spectrum nature of wind - dominated belrs suggests that these two extrema may have different soft x - ray excess properties and that the presence or lack of a soft x - ray excess might provide a way to reverse engineer the detailed shape of the ionizing sed ( which is important to radiation line driving ) . agn luminosities are often derived from observations in one , or at best a few , narrow bandpasses , whereas one wants the luminosity integrated over the full sed ( or perhaps just the accretion disk s contribution ; @xcite ) , i.e. , the bolometric luminosity . for example , the bolometric luminosity is the relevant quantity needed to estimate the accretion rate through @xmath107 . considering our empirical seds and the range of @xmath6 values in our sample , we can see from figure [ fig : sed ] that while the difference in seds may not appear to be very large , it is clearly _ systematic_. if we assume the same bolometric correction for all quasars regardless of their seds , then there will be a systematic error in @xmath108 as a function of their location in eqw - blueshift parameter space . indeed , a change of 0.2 in @xmath6 corresponds to a 35% change in the integrated luminosity between 2500 and 2kev ( assuming that the sed is described only by @xmath6 ) . if there are additional features in the sed at far - uv wavelengths ( e.g. , * ? ? ? * ) or if the shape of the euv spectrum is more like the models in figure [ fig : bb ] than our power - law fits , then that is an additional source of systematic error in the bolometric correction that arises through the adoption of a single mean sed ( or even one parameterized by the @xmath27@xmath6 relation ) . indeed , @xcite found that the there is no single bolometric correction , but rather a roughly factor of four range around the mean value . while @xcite were unable to determine which classes of quasars should use lower / higher bolometric corrections , our work here would suggest that bolometric corrections may be systematically underestimated for large eqw - small blueshift quasars ( i.e. , those with disk - dominated belrs ) and systematically overestimated for small eqw - large blueshift quasars ( i.e , those with wind - dominated belrs ) . it is beyond the scope of this paper to specifically parameterize the bolometric correction as a function of the indirectly measured ionizing flux in the euv ; however , we hope to pursue that in a future publication ( krawczyk et al . 2011 , in preparation ) . given the trends with uv luminosity in the eqw - blueshift parameter space , we might expect that more accurate bolometric corrections would result in a narrower distribution of @xmath108 . as @xmath108 is generally used to estimate the eddington ratio ( i.e. , mass weighted accretion rate ) , @xmath109 , those values may also have systematic errors . given our findings on the systematic changes in sed over eqw - blueshift parameter space , we consider the issue of nomenclature for these differences . in @xcite we referred to the rq quasars with small blueshifts and large eqws as `` rl - like '' because rl quasars are predominantly found in that part of parameter space . based on our results above , we suggest a simpler , yet more descriptive nomenclature . as rq quasars in the rl part of parameter space have harder ( more ionizing ) spectra , such objects could be broadly categorized as hard - spectrum quasars ( hsqs ) . the high - blueshift , low - eqw quasars on the other hand have softer ( less ionizing ) spectra and could be considered to be soft - spectrum quasars ( ssqs ) . these distinctions are not dissimilar from the population a / b terminology used by @xcite , where the most extreme population a objects would correspond roughly to ssq and population b to the hsq objects . the hard / soft terminology simply has the advantage of being physically motivated . it is less clear what is happening in the low - eqw , low - blueshift quasars ( occupying the other part of the `` population a '' parameter space ) . weak lines could indicate a relatively weak ionizing continuum , but the lack of a strong blueshift might indicate that the ionizing continuum must be intrinsically weak rather than @xmath6 being soft . another possibility is that @xmath6 could be soft , but the uv flux is insufficient to drive a wind . further _ chandra _ and/or _ xmm - newton _ observations are needed to better understand the differences between sources with weak that is systemic and weak that is highly blueshifted . if the winds in high - blueshift , low - eqw quasars ( i.e. , ssqs ) are predominantly radiatively driven , their soft - spectrum nature can not be generally due to absorption along our line of sight , despite the fact that these objects appear to be the parent sample of balqsos . that is because the sed needs to be _ intrinsically _ soft in order to drive a strong radiation pressure driven wind . ( mhd driving could also be at work , and may be a source of scatter in our diagrams . ) rather , it must be that for balqsos , there is additional absorption of an sed that already intrinsically lies at the soft ( x - ray weaker ) extrema of the range of normal seds spanned by quasars ( e.g. , * ? ? ? * ; * ? ? ? much as the fwhm of h@xmath70 , strength of ( relative to h@xmath70 ) , and soft x - ray properties can be used to divide low-@xmath34 quasars into extrema in the `` eigenvector 1 '' context @xcite , so can the properties of the emission line at high redshift . in @xcite , we showed that the extrema of large ( small ) eqw and small ( large ) blueshift can be attributed to different components of the belr : a disk and a wind , respectively . here , we consider the uv through x - ray properties of radio - quiet quasars in the eqw - blueshift parameter space and argue that these extrema are likely due to a hard ionizing spectrum in the `` disk''-dominated systems and a soft ionizing spectrum in the `` wind''-dominated systems ( figures [ fig : aox ] , [ fig : daox ] , and [ fig : sed ] ) . indeed these extrema could instead be considered as hard - spectrum quasars ( hsq ) and soft - spectrum ( ssq ) quasars . more work is needed to understand the nature of quasars with weak that appears at the systemic redshift and the relationship between the optical to x - ray flux ratio ( @xmath6 ) and both the soft and hard x - ray spectral indices . we argue that differences in the properties of quasars could be used to better trace the structure of the sed in the two decades of frequency space between uv and x - ray measurements , allowing for more accurate bolometric corrections for individual quasars ( figures [ fig : sed ] and [ fig : bb ] ) . the emission line may further allow a quantification the differences in the `` unseen '' part of the ionizing sed ( e.g. , an excess in the helium continuum above the nominal assumed from @xmath6 ) and may help identify which objects are likely to show a soft x - ray excess feature . support for this work was provided by the national aeronautics and space administration through chandra award number g08 - 9103x issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics space administration under contract nas8 - 03060 . gtr acknowledges support from an alfred p. sloan research fellowship and nasa grant 07-adp07 - 0035 . scg thanks the national science and engineering research council of canada and an ontario early research award for support . kml acknowledges support by nsf ast-0707703 . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , cambridge university , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . ccccccccc chandra c9 & 10 & 0 & 1 & 0 & 0 & 6 & 1.651.90 & 31.4831.68 + gallagher 05 & 14 & 1 & 2 & 0 & 3 & 10 & 1.692.17 & 31.0031.73 + strateva 05 & 155 & 0 & 10 & 94 & 89 & 23 & 1.552.24 & 30.3531.54 + kelly 07 & 157 & 1 & 9 & 95 & 112 & 6 & 1.572.03 & 30.0730.93 + just 07 & 32 & 3 & 10 & 26 & 21 & 5 & 1.912.27 & 32.0532.23 + gibson 08 & 536 & 52 & 71 & 100 & 210 & 181 & 1.702.31 & 30.3432.11 + green 09 & 281 & 11 & 29 & 146 & 182 & 46 & 1.542.26 & 30.4931.79 + young 09 & 792 & 33 & 78 & 423 & 604 & 91 & 1.552.27 & 30.1031.75 + lcccccrl 005102.42@xmath49010244.4 & 1.889 & 17.366 & 31.571 & 1597 & 20.3 & 3.5 & 9224 + 014812.23 + 000153.2 & 1.708 & 17.386 & 31.411 & 1590 & 15.2 & 10.5 & 9225 + 020845.53 + 002236.0 & 1.898 & 16.723 & 31.854 & 1555 & 23.1 & 3.5 & 9223 + 090007.14 + 321921.9 & 1.851 & 17.095 & 31.650 & 1955 & 21.6 & 11.0 & 9323 + 100401.28 + 423123.0 & 1.666 & 16.733 & 31.725 & 2281 & 16.8 & 8.1 & 9322 + 102907.06 + 651024.6 & 2.171 & 16.730 & 31.968 & 1560 & 17.2 & 10.6 & 9228 + 115351.11 + 113649.2 & 1.681 & 17.262 & 31.463 & 2026 & 22.9 & 11.0 & 9230 + 141949.39 + 060654.0 & 1.649 & 17.119 & 31.385 & 1941 & 23.5 & 9.9 & 9226 + 150313.63 + 575151.6 & 1.721 & 17.075 & 31.593 & 1555 & 15.9 & 10.0 & 9227 + 162622.06 + 295237.4 & 1.902 & 17.017 & 31.700 & 1501 & 20.6 & 10.9 & 9229 005102.42@xmath49010244.4 & 1.889 & 17.366 & 31.571 & 1597.2 & 20.31 & 3.5 & 9224 + 014812.23 + 000153.2 & 1.708 & 17.386 & 31.411 & 1589.7 & 15.16 & 10.5 & 9225 + 020845.53 + 002236.0 & 1.898 & 16.723 & 31.854 & 1554.9 & 23.13 & 3.5 & 9223 + 090007.14 + 321921.9 & 1.851 & 17.095 & 31.650 & 1954.5 & 21.64 & 11.0 & 9323 + 100401.28 + 423123.0 & 1.666 & 16.733 & 31.725 & 2280.5 & 16.83 & 8.1 & 9322 + 102907.06 + 651024.6 & 2.171 & 16.730 & 31.968 & 1560.0 & 17.21 & 10.6 & 9228 + 115351.11 + 113649.2 & 1.681 & 17.262 & 31.463 & 2026.1 & 22.87 & 11.0 & 9230 + 141949.39 + 060654.0 & 1.649 & 17.119 & 31.385 & 1941.2 & 23.54 & 9.9 & 9226 + 150313.63 + 575151.6 & 1.721 & 17.075 & 31.593 & 1554.8 & 15.91 & 10.0 & 9227 + 162622.06 + 295237.4 & 1.902 & 17.017 & 31.700 & 1501.3 & 20.61 & 10.9 & 9229 lcccccccr 005102.42@xmath49010244.4 & 2.75@xmath110 & @xmath111 & @xmath112 & @xmath113 & 31.510 & @xmath114 & @xmath115 + 014812.23 + 000153.2 & 2.13@xmath116 & @xmath117 & @xmath118 & @xmath119 & 31.434 & @xmath120 & @xmath121 + 020845.53 + 002236.0 & 1.62@xmath122 & @xmath123 & @xmath124 & @xmath125 & 31.768 & @xmath126 & @xmath127 + 090007.14 + 321921.9 & 2.20@xmath128 & @xmath129 & @xmath130 & @xmath131 & 31.635 & @xmath132 & @xmath133 + 100401.28 + 423123.0 & 1.65@xmath134 & @xmath135 & @xmath136 & @xmath137 & 31.677 & @xmath138 & @xmath139 + 102907.06 + 651024.6 & 2.10@xmath140 & @xmath141 & @xmath142 & @xmath143 & 31.971 & @xmath144 & @xmath127 + 115351.11 + 113649.2 & 2.10@xmath140 & @xmath145 & @xmath146 & @xmath147 & 31.484 & @xmath148 & @xmath149 + 141949.39 + 060654.0 & 2.65@xmath150 & @xmath151 & @xmath152 & @xmath153 & 31.533 & @xmath154 & @xmath155 + 150313.63 + 575151.6 & 2.30@xmath156 & @xmath157 & @xmath158 & @xmath159 & 31.574 & @xmath160 & @xmath161 + 162622.06 + 295237.4 & 2.60@xmath162 & @xmath163 & @xmath164 & @xmath165 & 31.679 & @xmath120 & @xmath166 .... # obsid , ra , dec , soft , soft_err , hard , hard_err , tot , tot_err , rate_soft , rate_soft_err , rate_hard , rate_hard_err , rate_tot , rate_tot_err , hr_obs 9223 , 32.189583 , 0.376667 , 48.99 , 7.07 , 17.65 , 4.24 , 66.47 , 8.25 , 1.4061e-02 , 2.0296e-03 , 5.0658e-03 , 1.2178e-03 , 1.9079e-02 , 2.3670e-03 , -0.47 9224 , 12.760000 , -1.045667 , 26.38 , 5.20 , 0.00 , 0.00 , 29.25 , 5.48 , 7.3846e-03 , 1.4544e-03 , 0.0000e+00 , 0.0000e+00 , 8.1877e-03 , 1.5331e-03 , -- 9225 , 27.050833 , 0.031444 , 64.36 , 8.12 , 6.70 , 2.65 , 77.71 , 8.89 , 6.1131e-03 , 7.7161e-04 , 6.3632e-04 , 2.5132e-04 , 7.3813e-03 , 8.4420e-04 , -0.81 9226 , 214.955833 , 6.115000 , 93.50 , 9.80 , 10.86 , 3.32 , 106.24 , 10.34 , 9.4342e-03 , 9.8862e-04 , 1.0953e-03 , 3.3466e-04 , 1.0719e-02 , 1.0437e-03 , -0.79 9227 , 225.806667 , 57.864333 , 114.30 , 10.86 , 18.41 , 4.36 , 136.27 , 11.75 , 1.1404e-02 , 1.0838e-03 , 1.8371e-03 , 4.3489e-04 , 1.3596e-02 , 1.1721e-03 , -0.72 9228 , 157.279583 , 65.173500 , 111.99 , 10.77 , 25.43 , 5.10 , 140.03 , 11.92 , 1.0539e-02 , 1.0136e-03 , 2.3927e-03 , 4.7983e-04 , 1.3177e-02 , 1.1214e-03 , -0.63 9229 , 246.592083 , 29.877056 , 78.08 , 8.94 , 9.79 , 3.16 , 89.98 , 9.54 , 6.8972e-03 , 7.9014e-04 , 8.6453e-04 , 2.7935e-04 , 7.9489e-03 , 8.4270e-04 , -0.78 9230 , 178.462917 , 11.613667 , 112.09 , 10.68 , 24.62 , 5.00 , 137.65 , 11.79 , 1.0166e-02 , 9.6836e-04 , 2.2329e-03 , 4.5345e-04 , 1.2484e-02 , 1.0693e-03 , -0.64 9322 , 151.005417 , 42.523056 , 46.77 , 6.93 , 14.53 , 3.87 , 63.41 , 8.06 , 6.1342e-03 , 9.0868e-04 , 1.9058e-03 , 5.0804e-04 , 8.3172e-03 , 1.0574e-03 , -0.53 9323 , 135.029724 , 32.322716 , 90.38 , 9.59 , 17.54 , 4.24 , 108.60 , 10.49 , 8.3068e-03 , 8.8161e-04 , 1.6117e-03 , 3.8992e-04 , 9.9817e-03 , 9.6399e-04 , -0.67 ....

the algorithm for the dpd fluid , the dynamics of which is conceptually a combination of molecular dynamics , brownian dynamics and lattice gas automata , is designed for simulating rheological properties of complex fluids on hydrodynamic time scales . this paper calculates the equilibrium and transport properties ( viscosity , self - diffusion ) of the thermostated dpd fluid explicitly in terms of the system parameters . it is demonstrated that temperature gradients can not exist , and that there is therefore no heat conductivity . starting from the-particle fokker - planck , or kramers equation , we prove an-theorem for the free energy , obtain hydrodynamic equations , and derive a nonlinear kinetic equation ( the fokker - planck - boltzmann equation ) for the single particle distribution function . this kinetic equation is solved by the chapman - enskog method . the analytic results are compared with numerical simulations . keywords : dissipative dynamics , fokker - planck equation , transport coefficients , kinetic theory , computer simulation techniques .

cond-mat9702036

the interest of the last decade in dynamical and rheological properties of complex fluids has seen the introduction of several new numerical methods for carrying out computer simulations on hydrodynamic time scales , the simulation of which using molecular dynamics often results in intensive computational demands . these new techniques include : ( i ) lattice gas cellular automata ( lgca ) ; ( ii ) lattice boltzmann equation ( lbe ) and ( iii ) dissipative particle dynamics ( dpd ) . the last method was introduced by hoogerbrugge and koelman , and was modified by espaol and warren to ensure a proper thermal equilibrium state . the primary goal of this paper is a theoretical analysis and explicit calculation of transport and thermodynamic properties in terms of model parameters . this is highly relevant in view of the current interest in applications of dpd to systems such as flows past complex objects , concentrated colloidal suspensions , dilute polymer solutions and phase separation . the dpd algorithm models a fluid of @xmath0 interacting particles out of equilibrium and conserves mass and momentum . position and velocity variables are _ continuous _ , as in molecular dynamics ( md ) , but time is updated in _ steps @xmath2 , as in lgca and lbe . the algorithm is a mixture of molecular dynamics , brownian and stokesian dynamics and lgca s , with a _ collision _ and a _ propagation _ step . in the collision step each particle interacts with all the particles inside an action sphere of radius @xmath3 through _ conservative _ forces @xmath4 , _ dissipative _ forces @xmath5 , which are proportional to both the stepsize @xmath2 and a friction constant @xmath6 , and _ random _ forces @xmath7 , which supply the energy lost by the damping . here @xmath8 label the particles . in numerical simulations , this is implemented by simultaneously updating the velocities from their precollision value @xmath9 to their postcollision value @xmath10 according to the instantaneous forces exerted by all particles inside the action sphere . in the subsequent _ propagation _ step of fixed length @xmath2 all particles move freely to their new positions @xmath11 . usual forms of the conservative force mean that the particles may be considered as completely interpenetrable . these softer interactions have the computational advantage of allowing particle motion on the order of a mean free path @xmath12 during each time step of fixed length @xmath2 . this represents a substantial advantage over event driven md algorithms for hard sphere fluids , where the length @xmath2 of the free propagation interval is on average much shorter , especially at fluid densities . by ignoring some of the microscopic details of the interactions , which are presumably irrelevant for fluid dynamics , dpd has the advantages of lgcas , but avoids the disadvantages of lacking galilean invariance and of introducing spurious conservation laws . in fact , the `` point particles '' should not be considered as molecules in a fluid , but rather as clusters of particles that interact dissipatively . the introduction of noise and dissipation represents a coarse - grained mesoscopic level of description and hydrodynamic behaviour is expected at much smaller particle numbers than in conventional md . if @xmath13 denotes the characteristic kinetic timescale in dpd , with @xmath14 the number density , @xmath15 the number of dimensions and @xmath6 the friction constant , then @xmath16 is considered to be large compared to any molecular time scale . in this coarse - grained description the dominant interactions are the dissipative and random forces , whereas the conservative forces can be interpreted as weak forces of relatively long range and may be taken into account as a vlasov mean field term in the kinetic equations . in addition , they can have the spurious effect of tending to force the dpd particles into `` colloidal crystal '' configurations , unless friction and noise are sufficiently large to prevent cooling into a lattice configuration . in the second half of the paper , where we derive a kinetic equation for the single particle distribution function , the conservative force will be neglected . this corresponds to the strong damping limit ( @xmath6 large ) . the random forces act effectively as repulsive forces to prevent collapse of dpd particles . a substantial contribution towards the understanding of the dpd fluid was given by espaol and warren , who derived a fokker - planck equation for the @xmath0-particle distribution function in the limit of continuous time ( @xmath17 ) . these authors also modified the original algorithm by imposing the detailed balance conditions , which guarantee the existence of the proper thermal gibbs equilibrium , described by @xmath18 $ ] where @xmath1 is the hamiltonian of the corresponding conservative system and @xmath19 is the global equilibrium temperature . these results are briefly reviewed in section [ section2 ] to establish the notation . concerning the macroscopic evolution equations , espaol formally established the linearized navier - stokes equations and derived green - kubo formulae for the dpd transport coefficients using a mori - zwanzig projection operator technique . however , to date no quantitative evaluation of these formulae for dpd seems to exist . hence , little is known explicitly about the approach to equilibrium , the validity of standard hydrodynamics ( system size dependence , effects of generalised hydrodynamics ) or about transport coefficients . for the transport coefficients , hoogerbrugge and koelman have estimated the kinematic viscosity @xmath20 , where @xmath21 is the shear viscosity and @xmath22 is the mass density , as @xmath23 with @xmath24 . this result has recently been extended in to include the bulk viscosity by applying the `` continuum approximation '' to the discrete equations of motion for the dpd particles , following suggestions of hoogerbrugge and koelman . in section [ section3 ] we show how the free energy of the dpd fluid _ monotonically _ approaches its equilibrium value by proving an @xmath1-theorem for the fokker - planck equation of espaol and warren , and we make the connection with the detailed balance conditions derived in . as a first step towards establishing the full nonlinear hydrodynamic equations we derive in section [ section4 ] the full macroscopic conservation laws for mass and momentum density , as well as the energy balance equation ( details are given in appendix a ) . the conceptual basis for the existence of hydrodynamic equations is the _ local equilibrium _ state , which in dpd is very different from that in a molecular fluid , because of the unusual role of the temperature . in section [ section5 ] we study in a quantitative fashion the decay of the energy density @xmath25 and `` kinetic '' temperature @xmath26 towards thermal equilibrium with global temperature @xmath27 , and we assess in what sense and on what timescale the dpd fluid describes an _ isothermal _ fluid out of equilibrium . this is done on the basis of a nonlinear kinetic equation -referred to as fokker - planck - boltzmann ( fpb ) equation- for the single particle distribution function @xmath28 . it will be obtained from the first equation of the bbgky hierarchy for the dpd fluid in combination with the molecular chaos assumption . by solving in section [ section6 ] the fpb equation in the hydrodynamic stage , using the chapman - enskog method , we derive the constitutive relations and the navier - stokes equation . this enables us to calculate in section [ section7 ] the transport coefficients of shear and bulk viscosity , as well as the self - diffusion coefficient . so far , we have not discussed the _ discrete time version _ of dpd , as implemented in actual simulations . they show a sensitive dependence of thermodynamic and transport properties on time step @xmath2 . a promising step towards understanding the @xmath2-dependence was recently taken by marsh and yeomans , who calculated the equilibrium temperature as a function of the step size , determined stability criteria for the step size , and validated their result by extensive numerical simulations . we shall not attempt to present here a systematic study of the different @xmath29-corrections to equilibrium distributions and transport properties , but postpone this to a later publication . the paper ends in section [ section8 ] with comments on the most important results and future prospects for dpd . the dynamics of a dpd system defines the time evolution of an @xmath0-particle system , specified by a point @xmath30 in phase space , in terms of stochastic differential equations . for a theoretical description it is more convenient to consider the equivalent fokker - planck equation , derived by espaol and warren . to interpret the separate terms in the fokker - planck equation , it is instructive first to consider the analogous _ kramers equation _ for the probability @xmath31 of a single particle of mass @xmath32 , having a phase description @xmath33 at time @xmath34 : @xmath35 the three terms on the right can be interpreted as follows . the first term is an external conservative force @xmath36 . the term involving the damping constant @xmath6 corresponds to the langevin force @xmath37 and the diffusive term with diffusion coefficient @xmath38 results from the random force @xmath39 in the equivalent langevin description , which reads : @xmath40 where @xmath41 is gaussian white noise with amplitude @xmath42 and @xmath43 and @xmath44 , where @xmath45 is a @xmath15-dimensional unit tensor . if we impose that the stationary solution of the kramers equation be the gibbs distribution : @xmath46 , then the diffusion coefficient must satisfy the following _ detailed balance _ ( db ) condition : @xmath47 where @xmath19 is the temperature in thermal equilibrium , measured in energy units . the full fokker - planck equation derived by espaol and warren for the dpd system is a direct extension of the kramers equation to @xmath0 interacting particles . the time evolution of the @xmath0-particle distribution function @xmath48 is governed by : @xmath49 where the conservative , dissipative and random parts of the evolution operator are defined respectively as : @xmath50 the summations run over all particles and the only difference to the original is that the parameters @xmath6 and @xmath42 have been scaled by the mass @xmath32 such that @xmath6 has the dimensions of an inverse time . the three terms above are the @xmath0-particle extensions of the three terms on the rhs of ( [ eqn : kramers ] ) . * the conservative part @xmath51 results from the additive and central interparticle interactions due to a potential , @xmath52 where @xmath53 is the relative position and a hat denotes a unit vector . it is the liouville operator for the corresponding conservative system and in the limit of zero noise and friction , equation ( [ eqn : operators ] ) reduces to the liouville equation . * the second term is analogous to the dissipative term in the kramers equation . it accounts for the langevin damping force between the pair @xmath54 , which is proportional to the friction constant @xmath6 and to the component of the relvative velocity @xmath55 along the line of centres @xmath56 , and is of finite range . this last property is described by a positive weighting function @xmath57 that is only non - vanishing inside an action sphere of finite radius @xmath3 . * the last term in ( [ eqn : operators ] ) represents the random noise and should be compared with the diffusive term in ( [ eqn : kramers ] ) . the random force @xmath58 between the pair @xmath54 is directed along @xmath56 and is proportional to @xmath59 where the weighting function @xmath60 is again only non - vanishing within a finite action sphere . the ranges of the conservative , dissipative and random forces may all be different as the model stands . moreover , one of the essential properties of dpd is that its dynamics conserves total particle number , @xmath0 , and the total momentum , @xmath61 . consequently @xmath62 and @xmath63 are constants of the motion . the latter is always set equal to 0 as the total system is assumed to be at rest . here @xmath64 is the single particle distribution function . in addition , we want to emphasize that microscopic momentum conservation is an essential property of a fluid model if it is to have a momentum density @xmath65 that is slowly varying in space and time . in contrast , we note that the energy of the system is not strictly conserved under the dpd algorithm . the equations for the mass , momentum and energy density will be discussed more fully in section [ section4 ] . in typical applications the conservative force may be set equal to zero . the parameters @xmath6 and @xmath42 satisfy the equation ( [ eqn : db ] ) and the weighting functions are chosen such that @xmath66 , which constitute in combination with ( [ eqn : db ] ) the detailed balance conditions for the dpd system , as will be discussed in section [ section3 ] . the density is typically chosen such that there are 5 to 10 particles within an action sphere , which means that the instantaneous total force on any particle is small on average . figure 1 shows an enlargement of part of configuration space , showing a sequence of 20 consecutive particle positions evolving from a randomly chosen initial configuration . the trajectories are relatively smooth , in contrast to the discontinuous paths in hard core interactions , illustrating that the resultant force on each particle is relatively small at this parameter setting . consider the following functional of the @xmath0-particle distribution function @xmath48 : @xmath67 = \int d{\bf \gamma } p({\bf \gamma},t ) \left\ { h+\theta_0 \ln p({\bf \gamma},t ) \right\},\ ] ] where @xmath68 and @xmath1 is the hamiltonian of the corresponding conservative system : @xmath69 @xmath70 is the potential energy and @xmath71 is the pair interaction . the functional can be interpreted as a sort of _ free energy _ @xmath72 , where @xmath73 is the average total energy and @xmath74 yields the total entropy . the goal of this section is to show that @xmath75 is a lyapunov functional with @xmath76 and to investigate the implications of this result for the equilibrium solution of the fokker - planck equation . the time derivative of ( [ eqn : functional ] ) yields in combination with ( [ eqn : fp ] ) : @xmath77 we observe that the third term inside @xmath78 in ( [ eqn : f_evol ] ) vanishes due to total probability conservation . then consider the contribution @xmath79 to ( [ eqn : f_evol ] ) due to the liouville operator @xmath51 . partial integration with respect to @xmath80 and @xmath81 yields directly @xmath82 here @xmath83 because the curly brackets represent poisson brackets , as can easily be demonstrated . the second term in ( [ eqn : f_evol_c ] ) reduces to surface terms in @xmath80 and @xmath81 and therefore vanishes too . next , we combine the remaining terms in ( [ eqn : f_evol ] ) and perform a partial @xmath81 integration , also symmetrizing the result over @xmath84 and @xmath85 . the final result is : @xmath86 now we make the following observation . if we choose : @xmath87 where @xmath88 is an arbitrary positive function vanishing for @xmath89 , then @xmath76 , as the rhs of ( [ eqn : f_evol_full ] ) can be cast into the form : @xmath90 where @xmath91 is the same as in ( [ eqn : f_evol_full ] ) . note that the equality sign applies if and only if @xmath92 is the solution of ( [ eqn : eqm_constraint ] ) below . consequently , the free energy - type function @xmath93 $ ] is a monotonically decreasing function of time , until it reaches equilibrium where @xmath94 which is simply the solution of @xmath91=0 for every pair @xmath54 : @xmath95 changing variables to the relative velocities of the particles , it is easy to prove that the equilibrium distribution of the system is separable in the velocities , and has the general form : @xmath96 where @xmath97 is a constant independent of @xmath80 and @xmath34 . we will only consider macroscopic systems which are not in uniform motion at long times and consequently limit ourselves to @xmath98 . the function @xmath99 is also the stationary solution of the fokker - planck equation ( [ eqn : fp ] ) if @xmath100 satisfies @xmath101 . this yields the gibbs distribution as the equilibrium solution : @xmath102 where @xmath1 is the hamiltonian ( [ eqn : hamiltonian ] ) of the system and @xmath103 is a normalisation constant . we assume that @xmath104 is uniquely determined by the requirements that it satisfies ( [ eqn : eqm_constraint ] ) and ( [ eqn : fp ] ) . consequently the dpd system will always reach the same equilibrium state if left undriven , independent of the volume and number of particles . the temperature of this equilibrium state has a value @xmath105 , which _ only _ depends on the parameters of the model . so dpd describes a system , thermostated at @xmath27 and with a free energy @xmath106 $ ] at equilibrium . note that , in contrast to the @xmath1-theorem for the boltzmann equation ( see e.g. ) , no molecular chaos approximation is required to derive this result for the dpd system . in their original discussion , espaol and warren imposed that the gibbs distribution be the stationary solution of the fokker - planck equation . the consequences of this requirement can be seen by inserting ( [ eqn : gibbs ] ) into ( [ eqn : fp ] ) . it leads to the so - called _ detailed balance _ constraint @xmath107 consequently , the constraint imposed in ( [ eqn : weights ] ) is the detailed balance constraint for dpd . the important result from the @xmath1-theorem is that it demonstrates that the gibbs distribution ( [ eqn : gibbs ] ) is the _ inevitable _ equilibrium distribution . throughout the rest of the paper we shall restrict ourselves to dealing exclusively with dpd systems that obey the db condition ( [ eqn : db_constraint ] ) . if the db condition is violated , no @xmath1-theorem can be derived and the gibbs distribution is not a stationary solution of the fp equation for dpd . in this case , the stationary state of the system does not correspond to thermal equilibrium but to some driven state , which will in general exhibit long range spatial correlations ( see e.g. ) . the original version of dpd , introduced by hoogerbrugge and koelman violates the db requirement ( [ eqn : db_constraint ] ) and therefore its stationary distribution will not approach a gibbs state but may exhibit spatial correlations , e.g. algebraic correlations @xmath108 where @xmath15 is the number of dimensions , extending far beyond the ranges of the conservative , dissipative and stochastic forces . this absence of thermal equilibrium is likely to be the reason for difficulties and inconsistencies discussed in . in the previous section we have established the existence of and approach to a thermal equilibrium state for the dpd system that obeys the detailed balance condition ( [ eqn : db_constraint ] ) . in this section , we address the problem of how the quantities of macroscopic interest evolve in time towards the final equilibrium state , concentrating on the local mass density @xmath109 , the local momentum density @xmath65 and the local energy density @xmath25 . as discussed in section [ section2 ] the microscopic dynamics of dpd conserve mass and momentum , and the corresponding macroscopic densities obey local conservation laws . as the total energy is not conserved under dpd , the evolution equation for the macroscopic energy density does not have the form of a local conservation equation , but contains source and sink terms corresponding to the random and dissipative forces respectively . in the final equilibrium state , these will balance each other . consider a general macroscopic quantity @xmath110 , defined through @xmath111 its time evolution can be obtained from the fokker - planck equation ( [ eqn : fp ] ) combined with the detailed balance condition ( [ eqn : db_constraint ] ) , by multiplying the fokker - planck equation with @xmath112 , integrating over all @xmath113-space and performing one or two partial integrations with respect to @xmath81 and @xmath114 . the result is the general rate of change equation : @xmath115 for any dynamic variable @xmath116 . consider the conserved mass density @xmath109 and the momentum density @xmath65 defined through : @xmath117 it is convenient at this stage to introduce the single particle and pair distribution functions , defined as : @xmath118 application of ( [ eqn : grce ] ) to the conserved densities in ( [ eqn : field_defn ] ) yields the macroscopic conservation laws : @xmath119 where @xmath120 and @xmath121 is the local pressure tensor or momentum flux density in the local rest frame of the fluid . the continuity equation has been derived by setting @xmath122 into ( [ eqn : grce ] ) . the only non - vanishing term is the one containing @xmath123 , and the continuity equation follows at once . derivation of the conservation equation for the momentum density proceeds along similar lines by choosing @xmath124 . details of the latter derivation are given in the appendix [ appendix1 ] , where it is shown that @xmath125 with kinetic(k ) , collisional transfer(c ) and dissipative(d ) contributions : @xmath126 with @xmath127 . the kinetic flux contains the so - called _ peculiar velocity _ , @xmath128 , and @xmath129 is the spatially averaged pair distribution function : @xmath130 the kinetic and collisional transfer contributions to the momentum flux in ( [ eqn : pt_defn ] ) and ( [ eqn : pt_contribs ] ) are present in any particle model with conservative forces . they are dominant in systems with sufficiently high density - dense gases and liquids - where the potential energy contributions are non - negligible with respect to the kinetic fluxes . the explicit form for these collisional transfer contributions is given in the literature for several cases : smooth potentials , elastic hard spheres , or inelastic hard spheres . the dissipative contribution @xmath131 results from the langevin - type damping forces between the particles . the random forces do not contribute to the momentum flux . the @xmath1-theorem , derived in section [ section3 ] , guarantees the approach to thermal equilibrium , where the distribution functions take the form : @xmath132 where @xmath133 is the pair distribution function in thermal equilibrium , @xmath134 is the number density , and @xmath135 the maxwellian velocity distribution . the sum of the kinetic and collisional transfer contributions reduces to the equilibrium pressure and the dissipative contribution vanishes . thus , @xmath136 with @xmath137 given by the virial theorem : @xmath138 where @xmath139 away from global equilibrium , the pressure tensor @xmath140 will contain the local equilibrium pressure and terms involving the viscosities . however , before the navier - stokes equations , or more generally , the full set of hydrodynamic equations , can be derived the concept of _ local equilibrium _ - which forms the conceptual basis of slow hydrodynamic evolution - has to be re - examined , as the energy is no longer a conserved quantity . this can only be done after identifying the slow and fast relaxation modes in dpd , on the basis of a kinetic equation . this will be done in section [ section5 ] . next we consider the energy density , defined as : @xmath141 where @xmath142 is the microscopic energy per particle . use of the rate of change equation ( [ eqn : grce ] ) leads after some lengthy algebra to the energy balance equation , as discussed in appendix [ appendix1 ] . it reads : @xmath143 here the explicit form of the source term is : @xmath144 where the term proportional to @xmath27 is a source resulting from the random force , and the term with the minus sign is a sink resulting from the langevin - type damping forces . in global equilibrium , the source and sink terms balance one another and @xmath145 . the _ heat current _ @xmath146 , given explicitly in equations ( [ microenergy]-[energyflow ] ) of appendix [ appendix1 ] , contains the standard kinetic and collisional transfer contributions due to conservative forces , as well as dissipative contributions analagous to @xmath131 in ( [ eqn : pt_contribs ] ) , and @xmath146 vanishes in global equilibrium . if @xmath147 would be set equal to zero , equation ( [ eqn : e_flow ] ) would have the generic form of the energy balance equation in ordinary hydrodynamics , where the heat current would contain a term proportional to the temperature gradient . as will become apparent in later sections , this is not the case in the dpd system . although ( [ eqn : e_flow ] ) looks like a macroscopic equation for the energy balance in the presence of sources and sinks , it looses its physical significance after a relaxation time @xmath16 , in which @xmath148 , and ( [ eqn : e_flow ] ) reduces to the continuity equation . this will be discussed in section [ section6 ] , below equation ( [ eqn : a_defn ] ) . one may also derive a balance equation for the free energy density , which would be a local version of the @xmath1-theorem of section [ section3 ] or of the corresponding one of section [ section5 ] for the fokker - planck - boltzmann equation . it would enable one to identify the irreversible entropy production . a similar balance equation for the entropy density in a dilute gas can be derived from the boltzmann equation . in this section we derive an approximate kinetic equation , referred to as the fokker - planck - boltzmann ( fpb ) equation , for the single particle distribution function @xmath64 , which is based on the molecular chaos assumption and has a collision term which is quadratic in @xmath64 . moreover , from here on the _ conservative forces _ will be neglected , which corresponds to the strong damping limit ( @xmath6 large ) . this section is organised as follows . we start by deriving the first equation of the bbgky hierarchy , which relates @xmath149 to the pair function @xmath150 . then the _ molecular chaos _ assumption : @xmath151 yields a closed equation , the fpb equation , which again satisfies an @xmath1-theorem . next we analyse the _ local equilibrium _ solution of the kinetic equation , which provides the conceptual basis for the existence of hydrodynamic equations and transport coefficients , as well as the justification for solving this kinetic equation for finding the `` normal solution '' by means of the chapman - enskog method . the first equation of the bbgky hierarchy can be derived directly by applying equation ( [ eqn : grce ] ) to the @xmath152-space density : @xmath153 its average yields @xmath64 on account of ( [ eqn : f_defn ] ) . the resulting equation of motion is : @xmath154 where the @xmath155 contraction of tensors is defined by @xmath156 , with @xmath157 , @xmath158 denoting cartesian components of vectors or tensors . the equation can be further simplified to : @xmath159 where @xmath160 . performing the integrals over all variables except @xmath161 and @xmath162 leads to : @xmath163 this is the first equation of the bbgky hierarchy with the fokker - planck equation ( [ eqn : fp ] ) taking the place of the liouville equation as the evolution equation . under the molecular chaos approximation ( [ eqn : molecular_chaos ] ) we have the following closed equation for the one particle distribution function : @xmath164 the molecular chaos approximation is a mean - field approximation , which neglects dynamical correlations resulting from correlated multiple collisions taking place inside an action sphere . as we have set all conservative forces equal to zero , the molecular chaos assumption is exact in the global equilibrium state . indeed , simulation results show that this is in fact an excellent approximation in the small - time step limit ( as shown in figure 2 ) . it can be shown in a similar fashion to section [ section3 ] that the functional , @xmath165 satisfies an @xmath1-theorem ( @xmath76 ) where the equality only holds if @xmath64 is given by the equilibrium form @xmath166 of ( [ eqn : f_eqm ] ) , which establishes the existence of a unique global equilibrium state . the next problem is to solve the nonlinear fpb equation , and to analyse the approach to equilibrium using the _ chapman - enskog _ ( ce ) method . according to this method , one can distinguish two stages in the evolution of the single particle distribution function @xmath64 : a rapid kinetic stage and a slow hydrodynamic stage . in the _ kinetic _ stage , @xmath64 decays within a characteristic kinetic time @xmath16 to the so - called _ normal _ solution @xmath167 which depends on space and time only through the first few moments @xmath168 , the conserved densities , where @xmath169 are the collisional invariants . in fluid systems the time @xmath16 is the mean free time , whereas in a dpd system , @xmath16 is estimated from ( [ eqn : bbgky ] ) as @xmath170 . in the subsequent _ hydrodynamic _ stage , @xmath171 depends only on space and time through its dependence on the conserved densities . in this stage , the solution @xmath172 of the fpb equation can be determined perturbatively , @xmath173 , as an expansion in powers of a small parameter , @xmath174 , which measures the variation of the macroscopic parameters over a characteristic kinetic length scale , @xmath175 , where @xmath176 is a typical mean velocity . therefore the @xmath152-expansion is essentially an expansion in the small parameter @xmath177 ( c.f . solution to kramers equation in ) . in the remaining part of this section , we focus on determining the lowest order solution @xmath178 of ( [ eqn : bbgky ] ) , which is the _ local equilibrium _ distribution . we first observe that the lhs of ( [ eqn : bbgky ] ) is of @xmath179 , as @xmath149 is proportional to @xmath180 , and similarly for the gradient term . the rhs of ( [ eqn : bbgky ] ) is of @xmath181 . this requires that , to the dominant order in @xmath152 , @xmath171 should satisfy @xmath182 . to determine the solution @xmath178 , we delocalise the collision operator @xmath183 by replacing @xmath184 on the rhs of ( [ eqn : bbgky ] ) by @xmath185 . if we denote the _ delocalised _ collision operator by @xmath186 , then @xmath178 is the solution of : @xmath187 guided by the @xmath1-theorem , we assume the standard form for the local equilibrium distribution : @xmath188,\ ] ] where @xmath189 , @xmath190 and @xmath191 are arbitrary functions of @xmath192 and @xmath34 . substitution of ( [ eqn : local_eqm ] ) into ( [ eqn : f0 ] ) shows , however , that the above @xmath193 is _ only _ a solution if @xmath194 where @xmath27 is the constant model parameter introduced below equation ( [ eqn : functional ] ) , which equals the global equilibrium temperature . the parameters @xmath195 and @xmath196 in ( [ eqn : local_eqm ] ) are chosen to be the fluid density and flow velocity . hence , @xmath197 where @xmath198 is a collisional invariant . this observation ( [ eqn : temp_defn ] ) has a profound consequence on the physical processes occurring in the dpd system , and makes it very different from standard fluids with energy conservation . in fluids , there is a fast kinetic relaxation to a local equilibrium state specified by @xmath195 , @xmath199 and @xmath196 , and a subsequent slow hydrodynamic relaxation of these fields to global equilibrium . the dpd system distinguishes itself from standard fluids in the sense that there is a fast relaxation on a time scale @xmath16 to a local equilibrium state ( [ eqn : local_eqm])-([eqn : temp_defn ] ) , specified by @xmath195 , @xmath200 and a spatially uniform and constant temperature @xmath27 . the subsequent slow relaxation involves only the density @xmath195 and flow velocity @xmath196 . consequently , a dpd system is not able to sustain a temperature gradient on hydrodynamic time scales ; there is no heat current proportional to a temperature gradient ; and there is _ no heat conductivity_. thus , the dpd system describes a _ thermostated _ or _ isothermal _ process at a fixed temperature @xmath27 . it may only model physical systems where the temperature either relaxes very rapidly to an equilibrium value or where the temperature is irrelevant ( an athermal process ) . it is worthwhile to explore the differences between dpd and standard fluids somewhat further . recalling that conservative forces have been set equal to zero in the present situation , permits us to write the energy density , @xmath201 in terms of a _ kinetic temperature _ @xmath199 . clearly @xmath195 is a slowly changing variable , but what is the behaviour of @xmath199 ? to answer this question , it is sufficient to consider only _ small _ deviations from global equilibrium , @xmath202 , and to linearize the fpb equation around @xmath203 . from that equation we shall derive how @xmath204 decays to zero . linearisation of ( [ eqn : bbgky ] ) yields , after some algebra : @xmath205 where we have used the relation @xmath206 and introduced the coefficient , @xmath207 @xmath208 \mbox { .}\ ] ] the equation with its non - local integral operator on the rhs of ( [ eqn : linearized ] ) would be the starting point for studying generalised hydrodynamics with wave - number dependent transport coefficients . here , however , we shall only consider the decay of @xmath209 then the rate of change of @xmath210 can be calculated from ( [ eqn : linearized ] ) , and yields : @xmath211 the second term on the lhs of ( [ eqn : heat_mode ] ) is typically an @xmath179 correction to the dominant decay terms . so ( [ eqn : heat_mode ] ) is a simple relaxation equation which shows explicitly that the kinetic temperature @xmath199 decays within the kinetic stage to the global temperature @xmath27 with a relaxation time @xmath212 . the conclusion is that the energy density in the hydrodynamic stage , given by @xmath213 , is still a slow but _ not an independent _ variable . it is strictly proportional to the density . moreover , we can conclude that the free - energy - type functions ( [ eqn : functional ] ) and ( [ eqn : little_functional ] ) represent the actual _ free energy _ of the dpd system in the hydrodynamic stage . similarly , we can determine the _ local equilibrium _ part of the pressure tensor ( [ eqn : pt_contribs ] ) in absence of conservative forces , by replacing @xmath171 in @xmath214 by its local equilibrium form @xmath178 and @xmath215 in @xmath131 by @xmath216 according to the stosszahlansatz . to zeroth order in @xmath152 , the dissipative part @xmath131 vanishes , and the local equilibrium pressure is given by @xmath217 these results will be needed in the next section to solve the fpb equation to linear order in @xmath152 . in section [ section4 ] we have derived the fokker - planck - boltzmann equation for the dpd system and described the chapman - enskog method for obtaining its solution @xmath218 in the hydrodynamical stage . the method requires that the rhs and lhs of the fpb - equation ( [ eqn : bbgky ] ) are expanded in powers of @xmath219 , using the expansion @xmath220 every @xmath221 is replaced by @xmath222 and the derivative @xmath223 is eliminated using the macroscopic conservation laws . the lowest order solution , which is the local equilibrium distribution , @xmath224 \mbox { , } \label{localequi}\ ] ] has been determined in the previous section . to obtain @xmath225 we expand the fpb equation in powers of @xmath152 , yielding @xmath226 we start with the rhs of ( [ bm ] ) , which has been calculated exactly to @xmath179-terms included . one finds after some algebra that @xmath227 . in the previous section it has only been verified that @xmath228 . the latter result is _ not _ sufficient , whereas the former is sufficient for our present purpose . in the remaining terms on the rhs we replace the collision operator @xmath229 by its delocalised form @xmath186 , as defined below ( [ eqn : f0 ] ) . the rhs of ( [ bm ] ) then becomes : @xmath230 with @xmath231 defined in ( [ eqn : omega ] ) . to calculate the lhs of ( [ bm ] ) to @xmath179 we need the rate of change of @xmath189 and @xmath191 to lowest order in @xmath152 , which may be calculated from the conservation equations ( [ eqn : macro_cons ] ) with @xmath121 replaced by its local equilibrium part @xmath232 , calculated in ( [ eqn : local_eqm_p ] ) , i.e. @xmath233 they yield in combination with ( [ localequi ] ) @xmath234 \label{split}\ ] ] with @xmath235 and @xmath236 note that the density and temperature gradients are absent on the rhs of ( [ split ] ) , in contrast to the traditional chapman - enskog result @xcite . for convenience of notation we introduce the fokker - planck operator @xmath237 and its adjoint @xmath238 , defined as @xmath239\label{operators}\\ { \cal l}^{+ } & = & [ - { \bf v}+ \frac{\theta_0}{m } \frac{\partial}{\partial { \bf v}}]\cdot \frac{\partial}{\partial { \bf v } } , \nonumber\end{aligned}\ ] ] to write the final equation for @xmath225 as : @xmath240 \mbox{. } \label{finaleq}\ ] ] it is a second order pde with an inhomogeneity on the rhs . we first construct a special solution by recalling that the fokker - planck operator @xmath237 can be mapped onto the schrdinger equation for an isotropic @xmath15-dimensional harmonic oscillator . its eigenfunctions are the tensor hermite polynomials , usually called sonine polynomials in a kinetic theory context , and the microscopic fluxes @xmath241 and @xmath242 are among them , i.e. @xmath243 this can easily be verified . combination of ( [ finaleq ] ) and ( [ eigenfuncties ] ) yields the special solution @xmath244 \mbox{. } \label{specsol}\ ] ] the general solution is obtained by adding an arbitrary linear combination of collisional invariants @xmath245 , which are the solutions to the homogeneous equation @xmath246 . however , the constraint ( [ eqn : a_defn ] ) suppresses these terms and @xmath225 is the desired solution of the fpb equation to linear order in @xmath152 . the only slow macroscopic fields are the density @xmath195 and the flow velocity @xmath196 , leading to the continuity equation and navier - stokes equation . the energy density in the hydrodynamic stage , @xmath247 , is not an independent variable . the energy balance equation derived in the appendix [ appendix1 ] is only relevant in the kinetic stage , but has no physical significance in the hydrodynamical stage . the results for @xmath178 and @xmath225 are sufficient to obtain the hydrodynamic equations to navier - stokes order and to obtain explicit expressions for the transport coefficients . the @xmath179-correction @xmath225 contains only gradients of the flow field , @xmath248 , but _ no _ gradients of the temperature . therefore , there will be no heat current and vanishing heat conductivity . the @xmath179-terms in the pressure tensor @xmath249 will be proportional to @xmath248 and we define the viscosities as the coefficients of proportionality through the constitutive relation @xmath250 where @xmath21 and @xmath251 are respectively shear and bulk viscosity . combining ( [ defvis ] ) , ( [ eqn : local_eqm_p ] ) and ( [ eqn : macro_cons ] ) then yields the navier - stokes equation for the dpd system @xmath252 the explicit expressions will be obtained in the next section . there are two contributions to the pressure tensor : the _ kinetic _ part @xmath255 and the _ dissipative _ part @xmath131 , defined in ( [ eqn : pt_contribs ] ) and two corresponding viscosities . the kinetic part depends only on @xmath256 , which are given in ( [ localequi ] ) and ( [ specsol ] ) . then @xmath214 becomes @xmath257 where @xmath152 is a formal expansion parameter that will be set equal to unity at the end of the calculations and @xmath258 here the dyadic @xmath259 has been split up into a traceless tensor , @xmath260 , and a term @xmath261 , proportional to the unit tensor and we have used the relation @xmath262 ( see ( [ eqn : a_defn ] ) ) . inserting the explicit solution ( [ specsol ] ) into ( [ kinpart1 ] ) allows us to write ( [ kinpart ] ) in the form @xmath263 where we have used the relation @xmath264 ( see ( [ eqn : f_eqm ] ) ) and introduced the inner product @xmath265 moreover a crossproduct of a traceless tensor and a scalar vanishes , i.e. @xmath266 . the product @xmath267 involves a simple gaussian integral and yields @xmath268 the fourth rank tensor @xmath269 is isotropic , traceless and symmetric , which implies the general form @xmath270 \mbox{. } \nonumber\end{aligned}\ ] ] by taking double contractions and evaluating gaussian integrals the constant @xmath271 comes out to be equal to @xmath272 and the @xmath155 product in ( [ pressuretensor ] ) yields @xmath273 combination of ( [ pressuretensor ] ) , ( [ inprod ] ) and ( [ eigenvalue ] ) finally yields @xmath274 comparison with the constitutive relation ( [ defvis ] ) enables us to identify the coefficients as the _ kinetic parts _ of the viscosities @xmath275 } \qquad \zeta_{\rm k } = \frac{n \theta_0}{d \omega_0}= \frac{\theta_0}{\gamma [ w]}\mbox { , } \label{transport}\ ] ] where the definition ( [ eqn : omega ] ) of @xmath231 has been used . we note that the kinetic part is inversely proportional to @xmath6 and has been for the first time explicitly calculated . this section deals with the dissipative part @xmath121 of the pressure tensor in ( [ eqn : pt_contribs ] ) , which depends on the pair distribution function @xmath215 . this function has a local equilibrium part @xmath278 and a part @xmath279 , linear in the gradients . we start with the first part . in order to make a direct comparison with the work of espaol we retain the conservative forces , for the time being . then , the local equilibrium pair function has the form @xmath280 where @xmath281 is the spatial correlation function in local equilibrium . only at the end of the calculation we will set the conservative forces equal to zero , so that @xmath282 . substitution of ( [ loc_eq_pair ] ) into ( [ eqn : pt_contribs ] ) yields then @xmath283 \right ) n({\bf r } ) n({\bf r}-{\bf r } ) \label{dispart}\\ & \simeq & - { \textstyle{\frac{1}{2 } } } \gamma m n^2 \int d{\bf r } r^2 g_0(r ) w(r ) { \bf \hat{r } } { \bf \hat{r } } { \bf \hat{r } } { \bf \hat{r } } : \mbox{\boldmath $ \nabla$ } { \bf u } \mbox { , } \nonumber\end{aligned}\ ] ] where @xmath284 $ ] has been expanded to linear order in the gradients . calculation of the completely symmetric isotropic fourth rank tensor proceeds as in ( [ stream ] ) with the result @xmath285}{d(d+2)}[\delta_{\alpha \beta } \delta_{\gamma \delta}+ \delta_{\alpha \delta } \delta_{\beta \gamma } + \delta_{\alpha \gamma } \delta_{\beta \delta } ] \mbox { , } \label{contrac}\ ] ] where @xmath286 \equiv \int d{\bf r } r^{2 } g_0(r ) w(r ) \mbox{. } \label{moment}\ ] ] we note that the definition of @xmath131 in ( [ eqn : pt_contribs ] ) contains @xmath129 rather than @xmath287 . one easily verifies that the spatial averaging , denoted by the overline , makes no difference to linear order in the gradients . the final result for the dissipative part then becomes @xmath288}{d(d+2 ) } { \bf \sf d}- \frac{m \gamma n^{2 } [ r^2 w g_0]}{2d^{2 } } \mbox{\boldmath $ \nabla$}\cdot { \bf u } { { { \bbb i}}}. \label{resultdis}\ ] ] with the help of ( [ defvis ] ) the coefficients can be identified as the contributions to the viscosities due to the dissipative forces , i.e. @xmath289}{2d(d+2 ) } \qquad \zeta_{\rm d } = \frac{m \gamma n^2 [ r^2 w g_0]}{2d^2}\mbox{. } \label{transport2}\ ] ] the local equilibrium contribution ( [ resultdis ] ) to the dissipative pressure tensor turns out to be the dominant contribution to the viscosity of a dpd fluid , for large values of @xmath290 , as illustrated in figure 3 and confirmed by numerical simulation in . we also want to point out that an @xmath291-contribution to the pressure tensor , calculated in local equilibrium as in ( [ dispart ] ) , is not a novelty of this paper , but also occurs in all systems with impulsive ( hard core ) interactions that are _ not strictly local_. for instance , consider the collisional transfer contribution analogous to @xmath292 for elastic hard spheres , where @xmath293 in ( [ eqn : pt_contribs ] ) is ill - defined . this term is calculated in sections 16.4 and 16.5 of , where its local equilibrium contribution yields @xmath294 with @xmath295 , defined in eq . ( 16.5.7 ) of . these contributions in real fluids are the direct counterparts of @xmath296 in dpd . we return to the dpd system _ without _ conservative forces , where the gibbs distribution ( [ eqn : gibbs ] ) reduces to @xmath297 , and where the spatial correlations are absent , i.e. @xmath282 . this equality is also required here for consistency with the molecular chaos approximation ( [ eqn : molecular_chaos ] ) , used in section [ section5 ] and subsequent ones . so far , we have only considered local equilibrium contributions to @xmath298 in ( [ dispart ] ) . to obtain the complete contribution , consistent with the molecular chaos assumption , we substitute ( [ eqn : molecular_chaos ] ) into ( [ eqn : pt_contribs ] ) and use the definitions ( [ eqn : field_defn ] ) . surprisingly , the results ( [ dispart ] ) are recovered , showing that ( [ dispart])-([transport2 ] ) give the full contribution of @xmath299 to the navier - stokes equation , at least within the molecular chaos assumption . to facilitate the comparison with the original predictions of , we set @xmath300 in ( [ resultdis ] ) and introduce @xmath301/ [ w ] , \label{defweight}\ ] ] where @xmath302 $ ] denotes the spatial average introduced in ( [ eqn : omega ] ) , so that @xmath303 . the final result for the dissipative part of the viscosities is then @xmath304}{2d(d+2 ) } = \omega_0 t_w^2 n \theta_0/2(d+2 ) \label{dispartvis}\\ \zeta_{\rm d } & = & \frac{\gamma m n^2 < r^2>_{w } [ w]}{2d^2 } = \omega_0 t_w^2 n \theta_0/2d , \nonumber\end{aligned}\ ] ] where @xmath305/d$ ] is the characteristic relaxation rate introduced in ( [ eqn : omega ] ) , and @xmath306 , defined through @xmath307 , is the average trasversal time of an action sphere with @xmath308 the thermal velocity . these results are in fact the theoretical predictions for the _ total _ shear and bulk viscosity of the dpd fluid , as obtained in on the basis of the `` continuum approximation '' to the equations of motion of the dpd particles . in the present context of non - equilibrium statistical mechanics and kinetic theory , these contributions have been identified as the local equilibrium contributions to the transport coefficients in order to make the connection with hoogerbrugge and koelman s expression for the kinematic viscosity : @xmath309 with their friction constant @xmath310 , proportional to @xmath2 , as a proper friction should . we recall that the range function @xmath88 in is normalised as , @xmath311= n \int d{\bf r } w(r ) = 1 . \label{norm}\ ] ] so , the results ( [ dispartvis ] ) and ( [ kmhb ] ) are identical . hoogerbrugge and koelman have also shown that the viscosity found in their numerical simulations approaches ( [ dispartvis ] ) and ( [ kmhb ] ) for large @xmath290 . simulations carried out with the modified dpd algorithm show the same properties . we conclude this subsection by listing the full results ( [ transport ] ) and ( [ dispartvis ] ) for the shear and bulk viscosity in a dpd fluid with continuous time ( @xmath17 ) : @xmath312 they involve the two intrinsic time scales of the dpd fluid : the characteristic kinetic time @xmath313 ( see ( [ eqn : omega ] ) ) and the traversal time @xmath306 of an action sphere , as defined below ( [ dispartvis ] ) , which is of order @xmath314 . in the parameter range @xmath315 the estimates @xmath316 and @xmath317 of dominate , and in the range @xmath318 the kinematic viscosities do , as illustrated in figure 3 . as a simple test , the shear viscosity @xmath21 of the dpd system was measured in two dimensions using a physical method . a linear velocity gradient was established between two moving plates and the force required to maintain this system was measured once equilibrium had been attained . by means of these simulations , we have measured the viscosity of the dpd fluid as a function of @xmath290 at different temperatures @xmath319 . results are shown in figure 3 for a higher temperature to emphasize the importance of the kinematic contribution . at large @xmath290 the measured viscosity approaches the theoretical prediction when the time step @xmath2 is reduced . in this range of parameters , the viscosity is dominated by its dissipative part ( [ dispartvis ] ) , corresponding to the original estimates of hoogerbrugge and koelman . at small @xmath290 and high temperature @xmath27 the viscosity is dominated by the kinetic viscosity . at small @xmath290 there are sizeable differences between predicted and simulated results , which do not decrease with decreasing time step size . the breakdown of the theory in this range of parameters has a fundamental reason . inspection of the collision term on the rhs of ( [ eqn : bbgky ] ) or ( [ enskog1 m ] ) shows that with @xmath320 and @xmath321 small the typical size of the collision term @xmath322 may not be large compared to the propagation terms on the rhs of ( [ eqn : bbgky ] ) . consequently , the chapman - enskog expansion will be poorly convergent or even divergent , because the kinetic and hydrodynamc time regimes are no longer well separated , or , equivalently , because the change of the macroscopic flow velocity over the characteristic kinetic length scale becomes large . to be consistent with the physical requirement of well separated time scales in this range of parameters , the imposed velocity gradients have to be reduced . remaining differences between theory and simulations may be caused by a breakdown of the molecular chaos assumption , which neglects the dynamical correlations that may have to be taken into account . the coefficient of self - diffusion can be obtained by considering a dpd fluid that is in equilibrium , except for the probability distribution @xmath324 of a tagged particle , labeled as @xmath325 . following the arguments of section [ section5 ] and choosing the @xmath152-space density @xmath326 , instead of ( [ eqn : mu_density ] ) , one arrives at an equation similar to ( [ bbgky1 ] ) , with @xmath171 and @xmath215 replaced by @xmath324 and @xmath327 respectively , defined as @xmath328 the molecular chaos assumption ( [ eqn : molecular_chaos ] ) now takes the form @xmath329 as the fluid particles are in thermal equilibrium with the maxwellian @xmath330 defined in ( [ eqn : f_eqm ] ) , and the resulting fpb equation is linear , i.e. @xmath331 f_s . \label{fpe2}\ ] ] it is identical to the kramers equation ( [ eqn : kramers ] ) with @xmath332 . the continuity equation takes the form @xmath333 with tagged particle density and current defined as @xmath334 application of the chapman - enskog method to ( [ fpe2 ] ) yields the `` local equilibrium '' distribution function @xmath335 and following equation for @xmath336 , @xmath337 as @xmath338 at the lhs is again an eigenfunction of @xmath237 with eigenvalue @xmath339 , we find @xmath340 the coefficient of _ self - diffusion _ @xmath323 , defined through the constitutive equation : @xmath341 becomes for the dpd fluid : @xmath342 } , \label{selfdif}\ ] ] where @xmath343 is the mass density of the fluid . the above result is new . there is only a kinetic contribution and no dissipative one . the main results of this paper are the derivation and solution of the fokker - planck - boltzmann ( fpb ) equation for the dpd fluid , providing explicit results for the thermodynamic and transport properties in terms of the system parameters : density @xmath189 , friction constant @xmath6 , temperature @xmath344 and range function @xmath88 with range @xmath3 . there are two intrinsic time scales : the kinetic relaxation time @xmath345 determined by the collision term , and the traversal time @xmath346 of an action sphere , where @xmath347 is the average velocity . we highlight the most important new results and future prospects in a number of comments . 1 . the dpd fluid for continuous time ( step size @xmath17 ) , described by the @xmath0-particle fokker - planck equation of espaol and warren , obeys an @xmath1-theorem for the free energy @xmath348 . the indispensable role of the detailed balance condition in establishing such a theorem is demonstrated . it guarantees a monotonic approach of @xmath348 towards a unique thermal equilibrium , described by the gibbs distribution with a temperature @xmath349 . the local conservation laws for mass density @xmath350 and momentum density are the essential prerequisites for the validity of the navier - stokes equations . the temperature , however , plays a very peculiar role . on the one hand the detailed balance condition guarantees the existence of a well - defined thermal equilibrium with a global equilibrium @xmath27 , in which energy is conserved on average . on the other hand , the local equilibrium state depends only on @xmath195 and @xmath196 , but not on a local equilibrium temperature @xmath351 , which relaxes in a time @xmath16 ( kinetic stage ) towards its uniform equilibrium value @xmath27 . in the subsequent hydrodynamic stage the dpd fluid is not able to sustain a temperature gradient , there is no heat conduction and all processes occur isothermally . 3 . in the coarse - grained mesoscopic interpretation of dpd particles as `` lumps of fluids '' , the microscopic conservative forces between the dpd particles are small compared to the mesoscopic friction and random noise ( large @xmath6 limit ) , and have been neglected in deriving the fpb equation . at sufficiently low temperature conservative forces can have the effect of forcing the dpd particles into crystalline configurations . the fpb equation is derived from the first equation of the bbgky - hierarchy for the distribution functions , obtained from the @xmath0-particle fokker - planck equation , playing the role of the liouville equation . in addition , the molecular chaos assumption , @xmath352 has been used . the chapman - enskog solution to the continuous time fpb equation yields two types of contributions to the viscosities : ( i ) _ dissipative _ parts @xmath276 and @xmath277 , accounting for the collisional transfer through the nonlocal dissipative interactions . they are determined by the local equilibrium distribution . ( ii ) kinetic parts @xmath253 and @xmath254 , coming from the collision operator and determined by the chapman - enskog solution of the fpb equation . if @xmath315 the dissipative viscosities are dominant ; if @xmath353 the kinetic viscosities are dominant . in the total viscosity is estimated by @xmath276 and @xmath277 , which is correct for @xmath354 . the simulated results for the kinematic viscosity are in reasonably good agreement with the predictions within the assessed theoretical regions of validity of the theory . we also calculated the coefficient of self - diffusion , which only has a kinetic part . it is also of interest to consider the green - kubo formulae for the viscosities in a dpd fluid as derived in , where the linear @xmath6-dependence of the viscosity in the limit of _ large _ dissipation @xmath6 is questioned . to make the connection we observe that @xmath276 and @xmath355 of should be identified respectively with @xmath276 and @xmath253 of the present paper . the time correlation functions of for @xmath276 and @xmath355 are formally proportional to @xmath356 and @xmath272 respectively . both time integrals appearing in the green - kubo formulae extend over the characteristic kinetic time to @xmath357 . consequently , @xmath358 and @xmath359 in the limit of large @xmath6 , in complete agreement with the detailed calculation of the present paper . the validity of the kinetic transport coefficients @xmath253 and @xmath360 and the convergence of the chapman - enskog expansion require that spatial variations ( @xmath174 ) are small over a characteristic kinetic lengthscale @xmath361 . the convergence of the gradient expansion in ( [ dispart ] ) and the validity of the dissipative viscosities @xmath276 and @xmath277 require in addition that spatial variations are small over the diameter of an action sphere , @xmath362 . both criteria pose bounds on the shear rates , imposed in the simulations , as well as on the validity of the chapman - enskog expansion . . an interesting extension of the present theory would be towards generalised hydrodynamics . such a region exists if @xmath354 or @xmath363 . then , the hydrodynamic modes with wave numbers @xmath364 in the range @xmath365 have @xmath364-dependent dissipative viscosities @xmath276 and @xmath277 . they may be calculated by studying the eigenmodes of the linearised fpb equation ( [ eqn : linearized ] ) . a similar wavevector - range to generalised hydrodynamics occurs in dense hard sphere fluids , where @xmath12 is small compared to the hard sphere diameter @xmath3 . such theories have been used successfully to describe neutron scattering experiments on liquid argon and liquid sodium . generalised hydrodynamics in dpd might therefore be of interest in explaining light and neutron experiments on concentrated colloidal suspensions . the equilibrium properties ( see figure 2 and ) and transport coefficients of dpd ( see ) depend sensitively on the step size @xmath2 . the fokker - planck equation ( [ eqn : fp ] ) , the detailed balance condition ( [ eqn : db_constraint ] ) , the fpb equation ( [ eqn : bbgky ] ) , the hydrodynamic equations ( [ ns ] ) and corresponding transport coeffcients in subsections 1,2,and 3 of section [ section7 ] only hold for the continuous time model @xmath366 . the only analytic study , available on dpd at finite @xmath2 , calculates the equilibrium temperature @xmath367 , and derives criteria , imposed on @xmath2 , for the stability of the equilibrium distribution @xmath368 . the most important open problem on dpd is a systematic analysis of all @xmath2-corrections to equilibrium and transport properties , such as an explanation of figure 2 and 3 , suggesting that the current form of the modified dpd algorithm for _ finite _ step size @xmath2 does not obey the detailed balance conditions , which implies that its stationary state is not the thermal equilibrium state described by the gibbs distribution . c.m . acknowledges financial support from epsrc(uk ) , erasmus(eu ) and unilever research(uk ) and the hospitality of universiteit utrecht where most of this work was carried out . g.b . acknowledges the financial support of the foundation `` fundamental onderzoek der materie '' ( fom ) , which is financially supported by the dutch national science foundation ( nwo ) . we would also like to thank p. espaol , d. ten bosch , j. van den berg , n. van kampen , t. van noije , p. warren and j. yeomans for helpful discussions and correspondence is the kinetic part of the pressure tensor , as listed in ( [ eqn : pt_contribs ] ) . the second term on the rhs of ( [ eqn : mom1 ] ) , which will be called rhs2 , involves the conservative interparticle forces @xmath374 . symmetrising over @xmath84 and @xmath85 yields then : @xmath375 \right > \nonumber \\ & = & -\mbox{\boldmath $ \nabla$ } \cdot \left < \frac{1}{2 } \sum_{i , j \neq i } { \bf f}({\bf r}_{ij } ) { \bf r}_{ij } \int^{1}_{0 } \ ! d\lambda \ : \delta({\bf r}-{\bf r}_i + \lambda { \bf r}_{ij } ) \right > \nonumber \\ & \equiv & -\mbox{\boldmath $ \nabla$ } \cdot { \bf \pi}_{\rm c}.\end{aligned}\ ] ] the third term on the rhs of ( [ eqn : mom1 ] ) , referred to as rhs3 , is due to dissipative particle interactions and can be treated in a similar fashion . symmetrising over @xmath84 and @xmath85 , and replacing @xmath377 by @xmath378 $ ] we obtain : where ( [ rhs2.1 ] ) , ( [ rhs3.1 ] ) and ( [ usef2 ] ) yield respectively @xmath381 and @xmath131 as listed in ( [ eqn : pt_contribs ] ) with @xmath129 defined in ( [ eqn : f2bar ] ) . combination of ( [ eqn : mom1 ] ) , ( [ rhs1.1 ] ) , ( [ rhs2.1 ] ) and ( [ rhs3.1 ] ) gives the macroscopic equation for the momentum density , the expression for [ ( a2)+(b2 ) ] has been symmetrized over @xmath84 and @xmath85 and ( [ eqn : ik_result ] ) has been used . the term ( a4 ) represents the energy source @xmath392 caused by the random forces . in ( a3 ) we split @xmath81 into @xmath393 . the first term containing @xmath55 gives the energy sink @xmath394 resulting from the damping forces . the second term containing @xmath55 is again symmetrized and combined with ( [ eqn : ik_result ] ) to give the dissipative part of the energy current @xmath395 . combination of these terms then gives :

we present a detailed discussion of how to obtain precise stellar photometry in crowded fields using images obtained with multi - conjugate adaptive optics ( mcao ) , with the intent of informing the scientific development of this key technology for the extremely large telescopes . we use deep j and k exposures of ngc 1851 obtained using the gemini multi - conjugate adaptive optics system ( gems ) on gemini south to quantify the performance of the system and to develop an optimal strategy for extracting precise stellar photometry from the images using well - known psf - fitting techniques . we judge the success of the various techniques we employ by using science - based metrics , particularly the width of the main sequence turn - off region . we also compare the gems photometry with the exquisite hst data of the same target in the visible . we show that the psf produced by gems possesses significant spatial and temporal variability that must be accounted for during the photometric analysis by allowing the psf model a reasonably high degree of spatial variability and by treating all exposures independently . as with seeing - limited observations , steps need to be taken to minimize contamination of the psf model and sky estimates from unidentified faint neighboring stars . we show that standard star observations are insufficient to provide accurate photometric calibration of mcao data . instead , we cross - match sources in our mcao observations to a calibrated seeing - limited catalog with a shallower depth but sufficient to include a large number of common sources . however , done in this way , it is critical to consider the different spatial resolutions of the two datasets by accounting for stars that are multiple sources in the mcao dataset but which appear as single sources in the seeing - limited catalog . we identify photometric calibration as a critical issue for next generation mcao systems such as those on tmt and the e - elt . our final cmds reach k , below the main sequence knee of this cluster , and theoretical isochrones provide remarkable agreement with the stellar locus in our data from the main sequence knee to the red giant branch .

1611.00422

the extremely large telescopes ( elt)the european extremely large telescope ( e - elt ) , the thirty meter telescope ( tmt ) , and the giant magellan telescope ( gmt)will reach unprecedented depths at optical and infrared wavelengths and have unprecedented spatial resolutions on account of their enormous apertures . to fully exploit the potential `` @xmath2 '' benefit of these facilities , taking advantage of both depth and resolution , requires adaptive optics ( ao ; see @xcite for a comprehensive review ; see @xcite for discussion relating to tmt and @xcite for discussion relating to e - elt ) . this technology improves the strehl ratio of the point - spread function ( psf ) , a measurement of the peak of the psf relative to the theoretical diffraction - limited value . it does this by correcting the wavefront distortions in the science light caused by the atmosphere . adaptive optics has a better performance at longer wavelengths and so they generally operate at near - infrared ( nir ) wavelengths . in classic adaptive optics , optical aberrations are first measured in the direction of a point - like source above the turbulence by using a wavefront sensor . natural guide stars ( ngs ) are the best sources to use since they are distant and point - like . however , there can be not enough suitable ( i.e. , bright ) ngss in a given field of view , leading to the use of artificial laser guide star ( lgs ) . these are created by a laser beam launched from the proximity of the telescope in the direction of the science target . a return of photons toward the telescope aperture is produced by either rayleigh scattering from molecules in the atmosphere or by excitement of sodium atoms at high altitudes . since the lgs wavefront sensors are blind to the tip - tilt aberration @xcite , ngss are still necessary but they can be fainter because they are required only to determine tip and tilt . a control system acquires the data from the wavefront sensors and calculates the correction that the ao system has to apply . a deformable mirror then injects an equal and opposite distortion to correct the image . the main shortcoming of classical ao is the limited size of the corrected field , typically of the order of a few arcseconds . a single guide star can probe only the cylinder of atmosphere in its direction ; the more distant a target is from it , the less the guide star probes the relevant aberrations and so the less effective the correction . this effect , known as anisoplanatism , is improved by multi - conjugate adaptive optics ( mcao ) . here , multiple guide stars are employed to probe the atmosphere in different directions and reconstruct the atmospheric distortion in a larger volume . multiple deformable mirrors , optically conjugated to different altitudes , apply the correction to the layers of turbulence associated to those heights . while an mcao system can work using only ngss , such as the mcao demonstrator mad on vlt @xcite , the use of lgss is advantageous because it alleviates the requirements for an asterism of bright stars close to the scientific target , thus increasing significantly the sky coverage . the gemini multi - conjugate adaptive optics system ( gems ) on gemini south is the first facility class mcao instrument , and had first light in 2012 . high resolution imaging of extended crowded stellar fields has generally been a prerogative of space telescopes , but gems now routinely enables this capability from the ground . we have observed the galactic globular clusters ngc 1851 , ngc 2808 , ngc 5904 , ngc 6652 , ngc 6681 , and ngc 6723 in the j and k@xmath0 bands . these clusters were selected as southern hemisphere targets which had preexisting , high quality , space - based imaging from the hst acs survey of galactic globular clusters of @xcite . first results for ngc 1851 are presented in @xcite and results for ngc 2808 and ngc 6681 are published in @xcite . in addition to the study of their stellar populations and dynamics , a major goal of our observing program is to better understand how best to exploit mcao data for precision photometry and astrometry , in order to prepare for science observations with the elts . the large number of relatively bright point sources distributed over the field of view allows our globular clusters to be used as a training ground for measurement techniques on mcao data . the success of mcao as a technology relies on its ability to produce science - ready astronomical measurements particularly photometry and astrometry and so it is important to use science - based metrics to judge the success or otherwise of any techniques . in this paper , we examine the critical question of how to extract the most precise photometry from mcao data , using our observations of ngc 1851 . like most globular clusters , ngc 1851 has a thin main sequence and giant branch but has a split subgiant branch @xcite . examination of the quality of the color magnitude diagrams will allow us to determine the utility of the strategies that we employ to obtain the photometry . while our conclusions are specifically valid for gems , they are also expected to be of relevance for any mcao dataset , and it is our expectation that such studies will prove useful for the development of the mcao systems and observing strategies for the tmt and e - elt . a subsequent contribution will present similar details for astrometric measurements , noting that the capacity of gems to produce precise astrometry has already been shown in @xcite . the format of this paper is as follows : section [ sec : observations ] discusses the details of our observations of ngc 1851 ; section [ sec : images ] presents a summary of the image processing ; section [ sec : performance ] overviews the photometric performance of gems / gsaoi , especially regarding the uniformity of the shape of the psf ; section [ sec : photometry ] presents an overview of our photometric analysis procedure and a detailed discussion of the steps taken to optimize the process to obtain the most precise photometry ; section [ sec : cmds ] presents our final color - magnitude diagrams and a brief discussion of their most prominent features ; section [ sec : conclusion ] summarises our main conclusions . gems , the gemini multi - conjugate adaptive optics system @xcite , is the first facility - class mcao system . it started science operations at the end of 2012 on the 8-meter gemini south telescope on cerro pachn ( chile ) . it measures the aberrations caused by the volume of atmosphere in the direction of the observed field using five laser beams , tuned to the sodium d2 line at 589 nm , that are launched from the back of the secondary mirror . by exciting the layer of atmospheric neutral sodium at 90100 km @xcite , the system generates five lgss arranged in a 60 square asterism with the fifth star in the center . a wavefront sensor ( wfs ) for each lgs measures the wavefront distortions along these five directions . since the lgs wfss are blind to tip - tilt aberrations @xcite , 13 bright natural guide stars ( ngss ) are required to determine this low - order mode , each with their own wfs . all wavefront measurements are combined to calculate the commands given to the deformable mirrors that provide the correction . the two deformable mirrors are optically conjugated to 0 and 8.5 km to correct for the turbulence at the ground and at high altitudes , respectively . the corrected wavefront is imaged by gsaoi , the gemini south adaptive optics imager @xcite . this camera has four teledyne hawaii-2rg nir detectors @xcite arranged in a 2 x 2 array , each chip with 2040 x 2040 active pixels . the pixel scale is 0.0196px@xmath3 and the total field of view is 82.5 x 82.5 ( with a gap of 2.5 between the chips ) . the detectors are read out using fowler sampling @xcite , which consists of a series of nondestructive reads of the pixels at the beginning and at the end of each exposure . the two sets are averaged and their difference constitutes the signal accumulated . the larger the number of samples taken , the lower the read noise . during the gems / gsaoi science verification period , we observed the galactic globular cluster ngc 1851 ( 05@xmath414@xmath506@xmath6.76 , -400247.6 ) with the nir broad band filters j ( @xmath7 ) and k@xmath0 ( @xmath8 ) on the nights of 2012 december 31 , 2013 january 27 , and 2013 january 30 ( gemini program gs-2012b - sv-406 , pi : mcconnachie ) . sub - exposures of 160 seconds were used and dithered in a pattern of 3 steps to cover the gap between the four chips ( figure [ fig : dithering ] ) . seventeen exposures were taken in the j band and twelve in the k@xmath0 . two shorter exposures were also taken in each band to obtain non - saturated photometry of the brightest stars . table [ tab : exposures ] provides a summary of our exposures . we used the `` faint object '' read mode with four fowler samples per set ( read noise of 13.6 @xmath9 ) . three natural guide stars in the field of view have been used for the tip - tilt correction , with magnitudes of 13.3 , 13.7 , and 14.4 in the r band . unfortunately , reliable measurements of the seeing conditions at the telescope site during our individual observations were not recorded . however , measurements made on cerro tololo by the ctio robodimm indicate relatively poor seeing conditions , with median values of approximately 1.1 and 0.8 when taking the j and k@xmath0 exposures , respectively . one of the j band images ( exposure # 10 ) has not been used in the data analysis because of its particularly poor mcao correction . a single k@xmath0 band exposure is shown in figure [ fig : mosaic ] with the positions of the natural and laser guide stars highlighted . the same pattern was used for the 17 j band long exposures , with 5 of the pointings repeated.,scaledwidth=60.0% ] .observing log of ngc 1851 with gems [ cols="^,^,^,^,^",options="header " , ] exposure of 160 seconds of ngc 1851 with gems / gsaoi . the orange star symbols show the position of the five lgss . the three ngss used are circled in red.[fig : mosaic],scaledwidth=40.0% ] prior to performing the photometric analysis , we process the raw images with the gemini / gsaoi iraf package correcting for bad pixels , linearity and flat fielding . * _ bad pixels _ the gsaoi detector has a large number of both dead and bright pixels . the defective elements have been flagged using the bad pixel map provided by the gemini iraf package . most of the hot pixels also cause an uncorrected leak of signal towards adjacent pixels that have lower counts , producing a `` cross - like '' pattern as shown in figure [ fig : capac ] . these are evident especially in the j band exposures due to the lower sky emission . this effect is caused by inter - pixel capacitance @xcite , an additional voltage on a pixel during its reading that is generated by the large difference in collected charges with respect to an adjacent one . * _ linearity _ hawaii-2rg chips deviate appreciably at high signal levels from a linear response because of the change in the reverse bias voltage caused by the accumulated charges @xcite . to compensate for this behavior , the values of every pixel are corrected with a quadratic polynomial whose coefficients were measured during the gsaoi commissioning @xcite . * _ flat field _ to create our flat fields in each band , we first produced a master dome flat and a master twilight flat by median - combing individual exposures using average sigma clipping . the dome flats have the benefit of a high signal to noise , although their spectral signature and uniformity is inferior to the sky flats . the latter are therefore preferred but also have a lower signal to noise than the dome flats . thus , we combine the dome and twilight master flats together to create a `` super '' flat in the following way . we first divide the master twilight by the master dome and we apply on the result a top - hat median circular filter to keep only the large - scale spatial variation . this creates a low - pass filtered image that is then multiplied by the dome master flat that has information on the pixel - to - pixel difference in response . the resulting `` super '' flat has the high signal - to - noise of the dome exposures and the uniform illumination and spectral signature of the twilight exposures , making it our choice for the flat fielding of the science exposures . , scaledwidth=40.0% ] we note that the detectors have a very low dark current ( 0.01 @xmath9 s@xmath3 px@xmath3 ) . the scale of this effect is of the order of one electron per pixel in our longest exposures ; hence , no dark frame has been subtracted . we also highlight that we do not apply sky frames . the focus of our analysis is point source photometry and , on the spatial scale of a psf , the flux from the sky is expected to be constant . applying a sky frame introduces additional systematic uncertainties and contributes to the photon noise . instead , we evaluate the sky at the position of each star using an annulus around them during the profile fitting , as discussed in section [ sec : photometry ] . an example region ( 4 x 3.3 ) of one of the k@xmath0 exposures before and after processing is shown in figure [ fig : iraf ] . the purpose of mcao is to provide diffraction - limited imaging over the largest field of view possible . in the case of a perfect correction , the psfs across the field would have the same size , shape , and flux concentration . globular clusters provide an ideal testing ground of mcao systems in this regard , with a large number of stars to examine . we now characterize the quality and stability of the mcao correction in our exposures to inform the more detailed photometric measurements that are described in section [ sec : photometry ] . we use the large number of bright and isolated psf stars ( see discussion in section [ sec : psfestimate ] ) , distributed across the field of view . we subtract all other stars from the images ( using the technique described in section [ sec : photometry ] ) and we apply a median filter on the bad pixels to remove them from the images . an example of a stellar image is shown in the left panel of figure [ fig : interp ] . the stellar images are then interpolated onto a finer grid by zero - padding their fourier transform by a factor of 3 , an example of which is shown in the right panel of figure [ fig : interp ] . the strehl ratio ( sr ) is the ratio of the peak flux of the actual psf to the peak flux of the diffraction - limited psf . to measure it , we adopt the peak value of the interpolated image as the peak of the psf . the diffraction - limited psf is generated by the fourier transform of the gemini south telescope aperture , which we model as a circle with a diameter of 8.1 m and a central circular obstruction of 1 m. we omit the spider of the secondary mirror since the change in the psf peak by including this more complex structure is negligible . the diffraction peak is normalized to the flux of the star using the instrumental magnitudes measured by psf fitting by the techniques in section [ sec : photometry ] . however , the psf model we use here has a radius of 45 pixels that is larger than the radius used in section [ sec : psfrad ] this is so that we include in each measurement the majority of flux from the star ( see figure [ fig : psfradstat ] ) . this improves the accuracy of the flux , at a cost to the precision , by avoiding larger systematic errors . we provide a simple quantification of the two dimensional shape of the psf by fitting an ellipse to the contour that defines the half - maximum of the psf . an illustration of this is shown in the right hand panel of figure [ fig : interp ] . the fwhm is then defined from the geometric mean of the semi - major ( @xmath10 ) and semi - minor ( @xmath11 ) axes , such that fwhm @xmath12 , and the ellipticity is defined as @xmath13 . we also measure the position angle of each ellipse . this allows us to create spatial performance maps of the mcao correction , an example of which is shown in figure [ fig : spline ] for one of the long k@xmath0 exposures ( exposure 10 ) . the left panel shows the distribution of raw sr measurements for each of the psf stars . note that there is a lack of psf stars in the central region of the image due to extreme crowding . the right panel shows the same performance map but interpolated across the full field using a smoothing spline and shown as a contour map . the average values of the fwhm , strehl ratio , and ellipticity of each long exposure image for both the j and k@xmath0 bands are shown in figure [ fig : perfstats ] . these exposures are numbered sequentially in the order that they were taken . the average fwhm in the j band is @xmath14 and in the k@xmath0 band is @xmath15 . for the strehl ratio , the average value for the j band is @xmath16 , whereas for the k@xmath0 band it is @xmath17 . it is expected that adaptive optics performs in k@xmath0 band better than in j since the aberrated wavefront has a larger spatial and temporal coherence at longer wavelengths @xcite . bands ( blue and red , respectively ) . the diffraction - limit fwhm values are shown in the top panel for comparison . we have not analyzed exposure # 10 in the j band due to a particularly poor mcao correction . these exposures are numbered sequentially in the order they were observed ; all k@xmath0 band exposures were observed on the same night , whereas the j band exposures were observed on two different nights , with the first ten observed on the first night.[fig : perfstats],title="fig:",scaledwidth=50.0% ] bands ( blue and red , respectively ) . the diffraction - limit fwhm values are shown in the top panel for comparison . we have not analyzed exposure # 10 in the j band due to a particularly poor mcao correction . these exposures are numbered sequentially in the order they were observed ; all k@xmath0 band exposures were observed on the same night , whereas the j band exposures were observed on two different nights , with the first ten observed on the first night.[fig : perfstats],title="fig:",scaledwidth=50.0% ] bands ( blue and red , respectively ) . the diffraction - limit fwhm values are shown in the top panel for comparison . we have not analyzed exposure # 10 in the j band due to a particularly poor mcao correction . these exposures are numbered sequentially in the order they were observed ; all k@xmath0 band exposures were observed on the same night , whereas the j band exposures were observed on two different nights , with the first ten observed on the first night.[fig : perfstats],title="fig:",scaledwidth=50.0% ] in figure [ fig : perfmaps ] , we show the performance maps for the best of our exposure ( # 8 in the k@xmath0 band ) , which has the smallest fwhm , the largest strehl ratios , and the most circular psfs . the first three panels show the interpolated maps of the fwhm , sr , and ellipticity . the fourth panel indicates the shape of each fwhm ellipse , where the angle of the dash indicates the position angle and the length indicates the ellipticity ( short dashes are more circular ellipses and thus the position angle is not as well measured as for longer dashes ) . in all four panels , it is clear that the performance is relatively uniform but certainly not constant across the field . in line with expectations , the best values of fwhm and sr are near the center of the lgs asterism . but we also notice that the shape of the psf , described by its ellipticity , has a more complex spatial pattern and does not perform better in the center ( thus our best exposure has smaller but more elliptical psfs in the center of the asterism compared to elsewhere ) . indeed , we find that the behavior of the ellipticity and position angles varies abruptly in consecutive exposures , as demonstrated in figure [ fig : elliptmaps ] for three consecutive k@xmath0 band exposures . note that this does not correlate with the movement of the ngs asterism caused by the dithering of the telescope . one of the main problems in delivering a better performance was the incorrect conjugation in altitude of the optical elements in the lgs wavefront sensors , as documented in a gemini observatory internal report ( g. sivo , private communication ) . this was identified and corrected in the year after our observations . despite these few issues , our mcao observations show a substantial improvement from seeing - limited images . our estimation of the performance of gems are consistent with previous findings @xcite . we now demonstrate that they can produce unprecedented levels of photometric precision in a crowded field from the ground . it is well know that precision photometry in a crowded stellar field can not be achieved by aperture photometry due to crowding , where starlight from one star compromises the measurement of the flux of any nearby star . a more robust method is profile fitting photometry , especially if the psf can be estimated empirically on the images . this is the case for our observations of ngc 1851 , where we have a large number of bright point - like sources in every part of the field of view . for the ensuing photometric analysis we have used the daophot ii suite of programs @xcite . we begin by providing a broad overview of the process that we use to obtain our photometry prior to discussing how certain critical stages of our analysis are optimized for best performance . * the stars are first detected ( using the find routine ) by convolving each individual image with a truncated gaussian profile , lowered to have a zero integral and that has a fwhm similar to that of the real psf . local maxima in the convolved images are identified as stars if they are above a chosen noise threshold ( @xmath18 for j and @xmath19 for k@xmath0 ) and within a range of `` sharpness '' and `` roundness '' parameters , selected to discriminate a star from extended objects , cosmic rays or artifacts of the detector . * for each catalog corresponding to each of the original images , we transform the coordinate system to a master frame of reference ( using the daomaster program ) . the master frame is defined by the seeing - limited catalog used for the photometric calibration ( section [ sec : calibration ] ) . a polynomial transformation of order 3 * we additionally stack the long exposure images ( using montage ) by assigning to each pixel of the stacked image the median value of its nearest pixel from each of the transformed frames , and identifying all sources ( using find ) . while the resulting image has a larger psf than any of the single exposures ( due to the different psfs of each image , see section [ sec : performance ] , and the nearest - pixel approach instead of interpolation ) , these stacked images allow us to detect and measure a large number of faint stars that are not detected in the individual exposures ( figure [ fig : stack ] ) . * the master catalog is created in each band by combining all detections found in each of the short exposures , the stacked images , and in at least three of the dithered long exposures . the short exposures are primarily used to find the brightest unsaturated stars , the stacked images are primarily for the faintest stars , and the long exposures are for all the other sources , with the requirement of three independent detections to remove most spurious objects . * we select 500 `` psf stars '' in the entire field of view , approximately 90 in each quadrant of the detector ( left panel of figure [ fig : psf_stars ] ) , that are used to model the psf independently for every chip of every exposure . these are stars that are bright but not saturated and that do not have obvious close companions . they are chosen to be distributed as uniformly as possible across the observed field of view , except in the cluster core where many saturated stars cause large flux gradients that render reliable background estimates difficult . the significant number of psf stars ensures that the photon noise in the psf model is reduced and that systematic effects on individual psf stars ( such as detector cosmetics or contamination from unrecognized companion stars ) are minimized . a constant background is estimated for each psf star as the modal value of the pixels in a surrounding annulus and is then subtracted from each psf star . we choose the inner edge of this sky region to start from the pixel immediately outside the psf model and the outer edge to contain approximately 2000 pixels in the annulus . * the psf is modeled in two parts . the first component is a simple analytic function , for which we have chosen a lorentzian . the second component is a two dimensional grid of empirical corrections calculated from the residuals of the psf stars after subtracting the analytic component . this look - up table can be set to have either the same values everywhere in the field of view or to vary spatially . in the spatially varying case , each element of the table is associated with a set of coefficients that describe a bivariate polynomial in x and y , the position on the field of view . in section [ sec : performance ] , we demonstrated that there is significant spatial variation in the psf in our images , and so we have chosen to use a cubic polynomial , the highest degree of variability available in daophot . the possibility to calculate such a high variability is given by the large number of psf stars available in the images . photometric measurements are obtained for objects in each image , including the stacked images , by fitting the appropriate psf model to each object in the master source catalog ( using allstar ) . at each iteration , the flux of a star is calculated by first subtracting the profiles of its neighbors using their positions and magnitudes derived from the previous iteration ( starting with estimates from aperture photometry ) . the star s background ( as estimated from the previous iteration ) is then subtracted and the instrumental magnitude is measured by fitting the psf model to the star s central pixels within three fwhma , where the s / n is higher . the sky beneath the star is then updated by removing the star itself and measuring the values in an annulus starting at two pixels from the peak ( to avoid the large residuals of the core ) out to the edge of the psf model . the photometric calibration of the instrumental magnitudes is performed with respect to the 2mass photometric system . however , we have had to use an auxiliary catalog to connect our observations to the 2mass point source catalog , since the latter has only about 150 stars in the field and most of them are too bright for our exposures . this auxiliary catalog has been obtained by one of the authors ( pbs ) from 256 seeing - limited archival images of ngc 1851 ( median fwhm of 1.32 ) taken with the newfirm instrument on the blanco telescope on cerro tololo ( chile ) , and has a depth intermediate between our gemini images and the 2mass catalog . by matching the stars in the auxiliary catalog to the corresponding stars in our gemini dataset , we are able to determine the zeropoints of each chip in each exposure in both bands by using the technique described in section [ sec : calibration ] . the final step is to clean the calibrated j and k@xmath0 catalogs of spurious objects , extended objects , or poor photometry . to do this , we use the various photometric parameters calculated by daophot . specifically , we use : @xmath20 , the goodness of the photometric fit that is measured by the ratio between the fitting residuals and the expected scatter caused by noise only ; sharpness , which is derived from the difference in width between the object and the psf ( objects that are too small are usually bad pixels or cosmic rays , objects that are too large are usually galaxies ) ; the magnitude error , which is a combination of the estimated random noise and the fitting residuals ( weighted toward the bigger of the two ) . the left panels of figure [ fig : cmdclean ] show the cuts applied to these three parameters in both bands . the right panel of figure [ fig : cmdclean ] shows the two dimensional distribution of the difference in position of matched objects between the two bands . we remove from the final catalog those stars with more than 0.01 of discrepancy , corresponding to half a pixel . thus , our final catalog contains only high quality detections in both bands . the preceding subsection provided a high level overview of the methodology for obtaining the final photometry from our processed images . we have followed a standard procedure for psf photometry that has been implemented on optical and near ir observations of stellar fields for many years . however , science quality mcao observations of crowded fields are a relatively new capability and a major focus of this work has been to understand what modifications and subtleties are required at various stages in the process to have confidence that the final photometry is optimal . in particular , our analysis in section [ sec : performance ] demonstrates that the images still have significant distortions present that change the shape of the psf across the field and between exposures in a complex fashion . what follows is a discussion of what we have identified as critical steps and decisions in the method described in the previous subsection , to obtain the highest quality photometry . specifically : 1 . size of the psf model 2 . variability of the psf 3 . selection of the sky annuli 4 . minimizing contamination light in the psf stars 5 . independent profile fitting most of these items are judged by science - based metrics based on the quality of the final color magnitude diagram , instead of using parameters like fwhm and strehl ratio that are difficult to measure and do not translate easily into an impact for the science . calibration is discussed independently in section [ sec : calibration ] . in this discussion we show results stemming from the analysis in the k@xmath0 band , but we note similar results with the j filter . the psf model in daophot is defined within a radius ( termed the `` psf radius '' ) measured from the centroid . in the case of no variation in the psf across the field , this aperture will contain the same fraction of flux for every star . in this way , the flux measured within the psf radius for each star can be translated into a total flux via an additional term in the photometric zeropoint measured during the calibration ( section [ sec : calibration ] ) . in the case of a spatially variable psf , the psf radius should ideally contain most of the spatially variable component , so that the flux within the aperture is still a constant fraction of the total . ideally , the psf radius should also be large enough to include most of the real psf . in this way , when the star is subtracted during the profile fitting , all its flux is removed from the image and it does nt contaminate the profile of nearby stars . most of the spatial variability in the psf of mcao images is caused by an inhomegeneous correction applied over the field of view . the maximum spatial frequency of the aberrated wavefront that an mcao system can correct is @xmath21 , where @xmath22 is the pitch of the highest density of actuators in the deformable mirrors projected on to the aperture of the telescope . for gems it is @xmath23 @xcite . the mcao system behaves as a high - pass filter for the wavefront , such that only light within the `` control radius '' , @xmath24 , of any psf is affected by the correction , leaving the wings of the profile mostly unchanged with respect to the seeing - limited psf . since the actuators are actually arranged on a square grid , the psf region influenced by the adaptive optics for gems is a square with a side of @xmath25 . for gems , @xmath26 in k@xmath0 band and the psf radius used for our images should therefore , in principle , be at least this value . examination of a typical psf in our images suggests that this argument is valid . figure [ fig : rad ] presents an example of a stellar radial profile from our data . the left panel shows the two dimensional image , with the centroid marked by the green cross and a blue square superimposed with sides of @xmath27 . shown for reference is also a circle with a radius of 50 pixels . the right panel shows the radial profile of the psf , where each dot corresponds to a pixel in the image . firstly , we see that the profile does not level off to a constant value until around 50 pixels . secondly , the spread in the profile at small radius is significant and increases with decreasing radius . this is to be expected if the psf is significantly distorted ( e.g. , has significant ellipticity ) . as the inset panel shows , however , the spread in the points becomes broadly constant at a radius greater than 2122 pixels . this is in line with expectations from our argument above and confirms that the asymmetric part of the psf is mostly within the control radius . we have tested the effect of modifying the psf radius on the resulting photometry from our images . we have produced several different photometric catalogs , where each version uses a different value for the psf radius . all other steps in the photometric analysis are identical . we have combined the resulting instrumental magnitudes with the high precision photometry from the hst / acs catalogue of ngc 1851 in the v band ( f606w ) presented in @xcite . since this catalog is extremely precise , examination of the relative quality of the resulting optical - nir cmds allows us to develop a science based metric to judge the quality of our nir photometry . figure [ fig : psfradcmd ] shows the cmds corresponding to each version of our k@xmath0 photometry ( a similar analysis was made for the j band ) . we use the width of the main sequence turn - off ( msto ) , in the region where the sequence is nearly vertical , as a metric on the quality of the photometry . we quantify the width of this sequence via the fwhm of a gaussian curve fitted to the histogram of the color distribution of the stars . photometric errors and systematics will act to increase the measured width of the main sequence , even in the presence of multiple populations ( as is the case for ngc 1851 , see section [ sec : cmds ] ) , thus higher quality photometry will produce a smaller fwhm . we note that figure [ fig : psfradcmd ] uses uncalibrated k@xmath0 magnitudes . therefore , each cmd is shifted horizontally with respect to the others since the instrumental magnitudes are `` fainter '' when a smaller psf radius is used . the histograms in figure [ fig : psfradcmd ] demonstrate that the width of the msto in the cmd is minimized for psf radii of @xmath28 ( a psf radius of 24 pixels also appears to produce a narrower red giant branch ) . we find that this result holds for all exposures , independent of the quality of the ao correction applied . it is clear , therefore , that the optimal psf model radius for our analysis is close to the value of the control radius @xmath29 , which itself is considerably smaller than the `` overall '' size of the psf . figure [ fig : psfradstat ] plots the uncalibrated color of the msto as a function of the psf radius , and it is clear that the color does not change significantly ( i.e. , the instrumental k@xmath0 magnitude is constant ) for a psf radius between 40 and 50 pixels ; this value is also consistent with our discussion relating to figure [ fig : rad ] . image . the width of the sequence is represented by the error bars.[fig : psfradstat],scaledwidth=60.0% ] the flux outside the control radius is the same fraction for every star since all the variability in the shape of the psf is interior to this value . moreover , at large radii , where the intrinsic flux from the star is low , contamination due to noise , cosmic rays , and nearby stars can be significant ( for example , note the presence of several structures at large radii from the center of the bright star shown in figure [ fig : rad ] ) . adoption of a large psf radius means that these spurious structures are considered part of the psf model , leading to an overall reduction in the precision of the photometry . this is illustrated in figure [ fig : psfs ] , where the first panel shows the psf of a star observed in the k@xmath0 filter . the second and third panels are the two psf models for this star , where the former adopts a psf radius of 24 pixels and the latter a radius of 40 pixels . the latter model has a large amount of small scale structure at large radii that are not intrinsic to the halo , expected to be smooth for these long exposure times . thus , a psf radius close to the control radius is large enough to include all the variability of the psf but small enough to exclude the majority of spurious structure , leading to more precise photometry . an unwanted side effect of using a smaller value of the psf radius is that when a star is subtracted from the image during the profile fitting , it leaves an `` orphan '' halo with a hole in the center where the core of the psf was . the inner edge of the halo of a bright star looks like a doughnut , as shown in figure [ fig : halo ] . this discontinuity in brightness can lead to spurious detections , as is visible in figure [ fig : halo ] . however , these spurious detections are easily identified and removed from the final catalog during the quality control phase ( section [ sec : quality ] ) . , scaledwidth=30.0% ] in seeing - limited and space - based images , the psf is generally uniform across the field . any variability that does exist is usually a result of static and quasi - static distortions introduced by the optics , with no significant variation between observations . in contrast , in mcao data the psf can present a complex spatial and temporal variability in terms of size and overall shape , as demonstrated in section [ sec : performance ] , especially figure [ fig : elliptmaps ] . such distortions are much less over large fields than using classic ao but they still necessitate careful modeling to obtain accurate photometry . daophot uses for the psf a constant analytical function for all stars in combination with a two dimensional lookup table for the empirical corrections to create the model psf . each pixel in this grid can be set to have either the same values everywhere in the field of view or to vary spatially . in the spatially varying case , each element of the grid is associated with a set of coefficients that together describe a bivariate polynomial in x and y , the coordinates of the star in the field . the polynomial can be up to the third order , making the psf vary linearly , quadratically or cubically . we also note that each grid element varies independently of every other one . the top panel of figure [ fig : psfvarsub ] shows examples of six psf stars in a k@xmath0 image from different parts of a frame . the other panels show the same regions after profiles with different variability have been fitted and subtracted . the constant and linear models present large residuals , while the quadratic and cubic seem to fit similarly well the profiles . to determine which produces the most accurate photometry , we have created uncalibrated cmds ( figure [ fig : psfvarcmd ] ) as in section [ sec : psfrad ] to analyze the width of their mstos . on all the tested images we find that the cubic variability has similar or slightly better performance than the quadratic , which is unsurprising given the large distortions of the psfs that we observed in section [ sec : performance ] . when testing the psf radius we have also modified two related sets of radii . the first is the annulus of pixels that we haveused to measure and remove the sky value of the psf stars , measured beyond the psf radius ( see section [ sec : photometry ] ) . the second set is the annulus to calculate the sky during the profile fitting that is computed at every iteration after the profile subtraction and which is measured interior to the psf radius . the inner radius of this annulus is set at two pixels from the center to avoid the large residuals that could be near the peak . we have first tested the influence of the psf model s sky annulus on the photometry by varying its distance from the edge of the psf radius and its width . we have tried three annuli with inner and outer radii of [ 25 , 36 ] , [ 41 , 49 ] , and [ 25 , 49 ] , each of which include at least 2000 pixels to have a good signal . the first and third of these options measure the sky relatively close to where the psf is measured ( but with a larger area used in the latter option ) , whereas the second option measures the sky away from the psf to reduce any contribution to the flux from the psf star . using the same cmd dispersion criterion of section [ sec : psfrad ] , we found no measurable change in the photometric precision for any of the three photometric catalogs . we therefore adopted the first of the three choices . for the sky annulus used during profile fitting , we have tested an outer radius of 16 , 20 , and 24 pixels . for most of the images in k@xmath0 filter , the photometric precision does not measurably change . however , for the k@xmath0 exposures with poorer mcao correction and for all the j band exposures we tested , we find that a small outer radius is preferable , as is shown in figure [ fig : psfskycmd ] . we have therefore adopted a sky annulus with an inner radius of 2 pixels and an outer radius of 16 pixels . we postulate that a small outer radius is preferred because , when the strehl ratio is lower , more faint stars are left undetected . a larger outer sky radius then increases the chances of these faint stars contributing to the `` sky '' measurement , reducing the accuracy of this measurement . in a crowded image , the psf is inevitably going to be measured using stars that have their profiles contaminated by nearby sources , causing inaccuracies in its estimation ( for example , see the left panel of figure [ fig : rad ] ) . to improve the resulting model , we have performed an iterative estimation of the psf . namely , we measure a first estimate of the psf on the crowded image ( first iteration ) and with it we subtract all the stars from the image except for the psf stars . we then use this `` cleaned '' image to redetermine a better psf ( second iteration ) . figure [ fig : psf_stars ] shows an example of an original reduced frame and a `` cleaned '' one , where the psf stars are marked . by measuring the width of the msto in the usual way , we determine that there is a gain in photometric precision when using the psf estimated in the second iteration compared to the first ( figure [ fig : psfitercmd ] ) . no benefit has been observed from increasing the number of iterations that we use beyond two . the allframe task in the daophot suite can perform the profile fitting photometry on multiple images at the same time . at each iteration , this program uses the geometric transformations of each frame on the master catalog to determine the position of the stars on the images . then the psf models are centered on these positions and are fitted to the image by subtracting the nearby stars determined in the previous iteration . the estimated residuals in the positions of the stars are minimized in the following iterations by adjusting both the stars positions in the master catalog and the coefficients of the geometric transformation . by fixing the positions of the stars in the reference frame for all the images , the combined photometry is potentially more precise by reducing the uncertainty caused by the noise in positioning the fitted psf . in figure [ fig : alsalf ] , the first panel shows a region of a k@xmath0 band image near the edge of the field . the middle panel shows the subtracted image once all the psfs have been fitted without placing constraints on their positions from the master catalog . the right panel shows the subtracted image where the positions of the stars are fixed using the allframe transformation . clearly , the latter image displays strong residuals caused by a similar misplacement of the psf centers . we note that the region selected for this example is one where the effect is particularly dramatic and that in most of the images it is more subtle . while not a focus of this paper , the resulting astrometry is also affected : figure [ fig : alsalfpos ] shows the difference between the positions of psf stars in a k@xmath0 band image determined with the two methods . arrows are magnified by a factor of 2000 for clarity and the histogram of the offset of the positions is also shown . image . the arrows lengths are magnified by 2000 . _ bottom _ : histogram of the differences in position between the two methods.[fig : alsalfpos],scaledwidth=40.0% ] clearly , figure [ fig : alsalfpos ] indicates a problem using daophot in the multi - frame mode for gems images . we note that allframe uses a polynomial fit of maximum third order . it has been shown by @xcite that gems delivers a field of view with geometric distortions well described by a fifth order polynomial with an rms of about 0.01 px respect to the third order . for the exposure in figure [ fig : alsalfpos ] , we find a median difference in positions of 0.03 px between no constrains and the third order . however , the difference between polynomials could be more severe near the edges . by relying on the daophot transformations , we are potentially producing differences between the real position on the image of the stars and where the psf is fitted by allframe . by analyzing the test cmds for the two cases in the usual way ( figure [ fig : alsalfcmd ] ) , we do not find substantial differences around the msto between the two methods . however , to reduce astrometric systematic errors and eliminate residuals such as those in figure [ fig : alsalf ] , we use the single - frame method in the profile fitting . the instrumental magnitudes of stars measured in our data are calibrated to our reference catalog described in section [ sec : calibrationover ] by applying a zeropoint ( @xmath30 ) and a color correction ( @xmath31 ) . to determine these coefficients , we have matched our psf stars to the reference catalog obtaining more than 50 matches in every exposure / chip . we begin with an initial estimation of the zeropoint @xmath30 defined as the median of the difference between the instrumental and reference magnitude for all stars in our exposures that are matched to the reference catalog . zeropoints are worked out for each chip in each exposure , independently . the resulting values are shown in the left panel of figure [ fig : zerop ] . it is worthwhile to examine the distribution of values of the difference between the instrumental and reference magnitudes for matched stars in one of the chips / exposures , on which this initial estimate of the zeropoint is based . these values are shown in figure [ fig : seeingmagerr ] and are typical of all the chips / exposures . the range of values is very large , of the order of several magnitudes . while the median operation reduces the effect of outliers , the number of them is so large that it implies that there is an issue with the matched magnitudes . the origin of this issue is that we are comparing two catalogs with very different angular resolutions : our reference catalog is seeing - limited while our observations with gems are diffraction - limited . therefore , our reference stars include multiple stars that are resolved in our images but not in the reference catalog . this can be clearly seen in figure [ fig : seeingcalib ] , where the left image is from our reference dataset and the right image is the corresponding part of the image from our data . image . the large spread does not correlate with the magnitude errors estimated by daophot.[fig : seeingmagerr],scaledwidth=50.0% ] it is therefore necessary to reformulate our estimate of @xmath30 for each chip / exposure . for a star in our reference catalog , its true magnitude @xmath32 is related to its reported magnitude @xmath33 via @xmath34 , where @xmath35 is a correction that accounts for unresolved companions located within a distance equal to the spatial resolution of our reference catalog . normally , this would equal the seeing of the observation . however , given our reference catalog is based on a composite dataset , we determine its spatial resolution by making a histogram of the distances of every star to their closest neighbor in the catalog , shown in figure [ fig : seeingrad ] . the low cut - off to this distribution is at @xmath36 ( close to the median seeing value of 1.32 of the images that make this composite dataset ) . , scaledwidth=60.0% ] for a star @xmath37 with true magnitude @xmath38 that is present both in the reference and the gems catalog , we can identify all neighbors , that are found within a radius of 1.2 in our gemini data . further , we know the instrumental magnitudes @xmath39 of each of these neighboring stars . therefore , @xmath40 assuming small color corrections , @xmath41 , where the lowercase @xmath42 denotes instrumental magnitudes . figure [ fig : seeingmag ] plots @xmath35 against @xmath43 for all matched stars in a single chip / exposure . there is a clearly defined linear relationship between the reference magnitude and the instrumental magnitude corrected for the nearby neighbors that are unresolved in the reference catalog . accounting for this contribution naturally explains the large spread in the difference between instrumental and reference magnitudes that we observed in figure [ fig : seeingmagerr ] . the appropriate value for the zeropoint is then the y - intercept of the best fit linear model with a slope equal to 1 , shown by the dashed line in figure [ fig : seeingmag ] . we note that to determine the zeropoint to high accuracy as is done here requires us to eliminate all spurious detections in the vicinity of bright stars , especially those relating to the `` orphan '' halos discussed previously in section [ sec : psfrad ] . and the diffrence between the instrumental magnitudes @xmath42 in our data and the corresponding reference magnitudes @xmath33 in the seeing limited catalog ( see text for details ) . the best fit linear relationship with a slope of 1 is shown by the dashed line.[fig : seeingmag],scaledwidth=50.0% ] the k@xmath0 zeropoints calculated in this way are shown in the right panel of figure [ fig : zerop ] and are highly correlated between chips . the improvement of this calibration can be quantified by comparing the left and middle panels in figure [ fig : seeingcmd ] , where we have used the same metric of the width of the msto as used earlier . indeed , the improvement is also visible by eye in comparing the widths of the red giant branch . we do not find a significant spatial variation of the zeropoint in any of the images . we note that it is a common photometric calibration strategy to observe a standard star or stars near in time and space to the science observation . however , the variation in the zeropoint of each chip between subsequent exposures , that is visible in figure [ fig : zerop ] , warns against this strategy for mcao observations . adopting a zeropoint for a given image based on observations in a different image , even one taken immediately before in the same direction , could be a source of error for precision mcao photometric observations . there are many ways to simulate this ; we demonstrate this here by averaging the zeropoints in one band of each quadrant of gsaoi , and adopting these as the appropriate zeropoint corrections for all exposures . the result is a widening of the sequences in the cmd , as seen in the right panel of figure [ fig : seeingcmd ] . having estimated the zeropoints , we can now determine the color terms @xmath31 . we assume that these do not change as a function of exposure / time . we attempt to calculate only a single term @xmath31 for each chip and can therefore use all the exposures . to simplify our analysis , we consider only those matched stars that have little contamination from nearby stars ( @xmath44 ) , thus we can assume that the color reported in the reference catalog is close to the real one . for every of those stars , we calculate @xmath45 as a function of the color @xmath46 . where the star is measured in multiple exposures , we average these measurement and associate an error bar that is a combination of the reported magnitude error and the standard error in the mean . figure [ fig : colterm ] shows the results for each chip . for each chip , it is clear that the scatter is significant and it is unclear if there is a trend with color . to determine if there is a significant correlation for any of the chips , we perform a pearson correlation test . the pearson correlation coefficients of the k@xmath0 band are 0.34 , 0.27 , 0.23 , and 0.11 for the four chips , indicating that there is not a significant correlation with color . therefore , we assume the color terms are zero . all the stars found in each image are used to calculate the geometric transformation , weighted for their magnitude error . they are then matched to create the final catalog and the magnitude of each is determined from all its available calibrated measurements in each of the different exposures . these magnitudes are combined using the method of `` artificial skepticism '' @xcite . figure [ fig : cmds ] shows the final , fully calibrated and cleaned by photometric parameters and proper motions j vs ( v - j ) , k@xmath0 vs ( v - k@xmath0 ) , and k@xmath0 vs ( j - k@xmath0 ) cmds for ngc 1851 . the ( v - j ) and ( v - k@xmath0 ) cmds on the left have inset panels that show a zoom - in around the split subgiant branch where this feature is visible . the right panels show the same cmds with theoretical isochrones overlaid . the isochrones are generated with the dartmouth models of stellar evolution @xcite using @xmath47=-1.16 $ ] , @xmath48=0.4 $ ] @xcite . @xcite suggest the hypothesis that the cluster is composed of two discrete stellar populations , one of 11 gyr @xcite , the other 1 gyr older . as such , and for comparison purposes only , we show three different isochrones spanning this age range . these have been corrected for distance and reddening by @xmath49 @xmath50 @xmath51 where the extinction ratios @xmath52 , @xmath53 , @xmath54 are from @xcite and the visual distance modulus @xmath55 and the color excess @xmath56 are from @xcite . figure [ fig : pos ] shows the spatial positions of the stars in the final j - k@xmath0 cmd . holes correspond to regions of extreme crowding , as well as the positions of several very bright and saturated stars that prevent accurate photometry from being obtained for neighboring objects . cmd of figure [ fig : cmds].[fig : pos],scaledwidth=50.0% ] the overall depth of these data is impressive , reaching 23 and 22 magnitudes in j and k@xmath0 band , respectively , with formal uncertainties of 0.16 and 0.26 magnitudes . the knee of the main sequence @xcite is clearly visible in the nir - only cmd , some three magnitudes fainter than the msto . the extended horizontal branch is divided in two by the rr lyrae gap and , at around the same magnitude , we can observe the red giant bump on the red giant branch . we can also notice a weak population of blue stragglers . in all three cmds , the agreement between the theoretical isochrones and the actual stellar locus is quite remarkable considering the isochrones were derived solely on the preexisting relevant values in the literature and no effort has been made to match them to these new nir data . in each case , the isochrone appears to very nearly follow the central spine of the stellar locus , from the main sequence knee through the msto , through the ( split ) sub - giant branch , and up to the red giant branch . this gives us further confidence that the strategies that we have employed in our analysis , especially regarding the photometric calibration , are valid across the full magnitude and color range of the target stars . @xcite presented an earlier version of our k@xmath0 band photometry , in combination with the hst acs data , that was of sufficient quality to exhibit the double subgiant branch of this globular cluster . here again the split is present and we also see weak evidence for this feature in the j band data , when combined with the hst acs data . we do not see any evidence of the split subgiant branch in the nir - only cmd . the combined uncertainties on the j and k@xmath0 data would be too large for such structure . in this paper , we have presented a detailed analysis of the photometric performance of gems / gsaoi using data obtained during its science verification phase . in particular , we have focused on developing the optimal strategies for obtaining precise stellar photometry from mcao images using science based metrics to quantify the success or otherwise of our approaches for crowded fields . our strategy is based on the well known technique of psf fitting , specifically with the daophot ii suite of programs , but modified to account for the challenges posed by mcao data , especially the significant temporal and spatial variation in the psf . efficient exploitation of mcao imaging in the nir is absolutely essential to the future of ground - based astronomy given the development of the next generation of oir telescopes , especially tmt and the e - elt . the following is a list of our primary conclusions : * mcao delivers a more uniform correction than classical ao over the moderately wide field of view . however , the shape of the psf can still vary significantly in a complex manner , both spatially and temporally ( i.e. , between exposures ; section [ sec : performance ] ) , as also found by @xcite . it is possible that some of this variation has been removed since gems has undergone several realignments of the optics , but residual variability is still expected . as a consequence , this should be reflected in the creation of the psf model by allowing it to have a high degree of spatial variability , and by treating all exposures independantly . * the mcao correction is applied on a scale equal to the `` control radius '' of the system . the majority of the spatial variability of the psf is found within this radius . we demonstrate that we obtain the most precise photometry when the radius of the psf model is set approximately equal to the control radius of the ao system ( section [ sec : psfrad ] ) . if the psf radius is considerably smaller than the control radius , part of the variable psf controlled by the mcao system is not measured , causing a systematic error dependent on the position of the stars . if the psf radius is much larger than the control radius , then the resulting model attempts to include the halo of the psf . this region is generally not varying and contains a fixed fraction of the flux in the psf , and can therefore be accounted for during the calibration procedure . if the psf model includes these outer regions , then the low flux levels mean that other features , including noise artifacts and faint stars , are included and identified as intrinsic features of the psf . this results in an incorrect psf model for these regions and a reduction in the precision of the photometry . * there is an appreciable improvement in photometric precision when the psf model is based on a second iteration of the derivation procedure , i.e. , derived from an image which has had non - psf stars subtracted ( section [ sec : psfclean ] ) . this practice is likely less important with sparse fields . * during the profile fitting , the sky flux under every star is measured interior to the psf radius . in exposures with a large fwhm , we have noticed that using an area significantly smaller than the psf radius increases the precision of the photometry ( section [ sec : psfsky ] ) . we postulate that this is because a larger sky region is more likely to have contamination from unidentified stars , leading to inaccuracies in the sky measurements . * simultaneous profile fitting of all the observations , such as is done with allframe , should be avoided because of the difficulty in finding the proper geometric transformation for the edges of some of the images . treating the images independently leads to more robust photometric / astrometric solutions ( section [ sec : psfindep ] ) . * calibration of mcao photometry is complex . standard star observations are insufficient to estimate the zeropoint since the zeropoint changes for each exposure as a result of the mcao correction . it is therefore advisable to match every exposure and chip independently to a reference catalog of the same field , that is itself calibrated to the nir 2mass system . the measurement of the zeropoints must account for the different spatial resolutions of the catalogs ( section [ sec : calibration ] ) . for gems / gsaoi , we do not measure a significant color term . the procedures described in this paper have allowed us to derive extremely deep and precise nir photometry for the globular cluster ngc 1851 . comparison to isochrones suggest remarkable agreement between our observational data and the theoretical expectations for the color and magnitude distribution of the stars ( from the main sequence knee , through the main sequence turn - off to the subgiant and giant branch ) given our understanding of the age , metallicity , and distance of this cluster as derived from optical data . when combined with optical hst acs data , the ( v - j ) and ( v - k@xmath0 ) cmds allow for the identification of the split subgiant branch ( already demonstrated for the k@xmath0 band in @xcite . with precision photometry and high quality psf modeling in hand , a future contribution will discuss the astrometric performance and precision of our data using a similar approach of the identification of science - based metrics . mcao is a key technology for the future of ground - based nir astronomy , especially as we approach the era of tmt and e - elt , where the large apertures will provide diffraction limits that will significantly surpass even the james webb space telescope . gems / gsaoi allows us to explore the utility of this technology for science observations and develop solutions to the new challenges that are encountered . for example , nfiraos , the mcao system on tmt @xcite , is still expected to present noticeable psf variability even though it will be considerably more uniform and stable than gems . the instrument will be thermally controlled , on a nasmyth platform , and equipped with internal calibration tools . the geometry of the system will also improve the quality of the correction thanks to the use of more laser guide stars , a larger overlapping of the volumes probed by the guide stars , and a smaller scientific field of view . a comparable fraction of the flux will still reside outside the control radius compared to gems , since the density of actuators on the deformable mirrors will be similar . our current analysis suggests that the photometric calibration of tmt observations of crowded fields will be a significant challenge . our present technique of using a calibrated reference catalog of the field would require relatively deep , relatively high spatial resolution photometry if we are to identify large numbers of stars in common with ultra - deep , ultra - high resolution imaging from tmt . by exploring and solving these issues now , before the new generation of mcao instruments are on - sky , we can expect to improve significantly their science return at first light . based on observations for the program gs-2012b - sv-406 ( p.i . : mcconnachie ) at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the national research council ( canada ) , conicyt ( chile ) , ministerio de ciencia , tecnologa e innovacin productiva ( argentina ) , and ministrio da cincia , tecnologia e inovao ( brazil ) . ascenso , j. , neichel , b. , silva , m. , fusco , t. & garcia , p. 2015 , ao4elt 4 conference proceedings , 31540 bec , m. , rigaut , f. j. , galvez , r. , et al . 2008 , , 7015 , 701568 blank , r. , anglin , s. , beletic , j. w. , et al . 2012 , , 8453 , 845310 bono , g. , stetson , p. b. , vandenberg , d. a. , et al . 2010 , , 708 , l74 boyer , c. & ellerbroek , b. 2016 , , 9909 , 990908 carrasco , e. r. , edwards , m. l. , mcgregor , p. j. , et al . 2012 , , 8447 , 84470n carretta , e. , lucatello , s. , gratton , r. g. , bragaglia , a. & dorazi , v. 2011 , , 533 , a69 davies , r. & kasper , m. 2012 , , 50 , 305 diolaiti , e. , ciliegi , p. , abicca , r. , et al . 2016 , , 9909 , 99092d dotter , a. , chaboyer , b. , jevremovi , d. , et al . 2008 , , 178 , 89 ferraro , f. r. , dalessandro , e. , mucciarelli , a. , et al . 2009 , , 462 , 483 figer , d. f. , rauscher , b. j. , regan , m. w. , et al . 2004 , , 5167 , 270 finger , g. , dorn , r. , meyer , m. , et al . 2006 , , 6276 , 62760f fiorentino , g. , tolstoy , e. , diolaiti , e. , et al . 2011 , , 535 , a63 fowler , a. m. & gatley , i. 1990 , , 353 , l33 harris , w. e. 1996 , , 112 , 1487 herriot , g. , andersen , d. r. , atwood , j. , et al . 2014 , , 9148 , 914810 marchetti , e. , hubin , n. n. , fedrigo , e. , et al . 2003 , , 4839 , 317 massari , d. , fiorentino , g. , mcconnachie , a. , et al . 2016a , , 586 , a51 massari , d. , fiorentino , g. , mcconnachie , a. , et al . 2016b , , in press massari , d. , fiorentino , g. , tolstoy , e. , et al . 2016c , , 9909 , 99091 g moretti , a. , piotto , g. , arcidiacono , c. , et al . 2009 , , 493 , 539 mcgregor , p. j. , hart , j. , stevanovic , d. , et al . 2004 , , 5492 , 1033 milone , a. p. , bedin , l. r. , piotto , g. , et al . 2008 , , 673 , 241 neichel , b. , dorgeville , c. , callingham , j. , et al . 2013 , , 429 , 3522 neichel , b. , rigaut , f. j. , vidal , f. , et al . 2014 , , 440 , 1002 rigaut , f. j. & gendron , e. 1992 , , 261 , 677 rigaut , f. j. , neichel , b. , boccas , m. , et al . 2014 , , 437 , 2361 saracino , s. , dalessandro , e. , ferraro , f. r. , et al . 2015 , , 806 , 152 saracino , s. , dalessandro , e. , ferraro , f. r. , et al . 2016 , , in press sarajedini , a. , bedin , l. r. , chaboyer , b. , et al . 2007 , , 133 , 1658 stetson , p. b. 1987 , , 99 , 191 stetson , p. b. 1994 , , 106 , 250 stetson , p. b. & harris , w. e. 1988 , , 96 , 909 stetson , p. b. 1989 , the techniques of least squares and stellar photometry with ccds lectures at v escola avanada de astrofsica , aguas de so pedro , brazil turri , p. , mcconnachie , a. w. , stetson , p. b. , et al . 2015 , , 811 , l15 vandenberg , d. a. , brogaard , k. , leaman , r. & casagrande , l. 2013 , , 775 , 134 yuan , h. b. , liu , x. w. & xiang , m. s. 2013 , , 430 , 2188

we prove novel convergence results for a stochastic proximal gradient algorithm suitable for solving a large class of convex optimization problems , where a convex objective function is given by the sum of a smooth and a possibly non - smooth component . we consider the iterates convergence and derive non asymptotic bounds in expectation in the strongly convex case , as well as almost sure convergence results under weaker assumptions . our approach allows to avoid averaging and weaken boundedness assumptions which are often considered in theoretical studies and might not be satisfied in practice . proximal methods , forward - backward splitting algorithm , stochastic optimization , online learning algorithms .

1403.5074

first order methods have recently been widely applied to solve convex optimization problems in a variety of areas including machine learning and signal processing . in particular , proximal gradient algorithms ( a.k.a . forward - backward splitting algorithms ) and their accelerated variants have received considerable attention ( see @xcite and references therein ) . these algorithms are easy to implement and suitable for solving high dimensional problems thanks to the low memory requirement of each iteration . moreover , they are particularly suitable for composite optimization , that is when a convex objective function is the sum of a smooth and a non - smooth component . this class of optimization problems arises naturally in regularization schemes where one component is a data fitting term and the other a regularizer , see for example @xcite . interestingly , proximal splitting algorithms separate the contribution of each component at every iteration : the proximal operator defined by the non smooth term is applied to a gradient descent step for the smooth term . in practice it is often relevant to consider situations where the latter operation can not be perfomed _ exactly_. for example the case where the proximal operator is known only up - to an error have been considered in @xcite . in this paper we are interested in the complementary situation where it is the gradient of the smooth term to be know up - to an error . more precisely , we consider the case where only stochastic estimates of the gradient are available and develop stochastic versions of proximal splitting methods . this latter situation is particularly relevant in statistical learning , where we have to minimize an expected objective function from random samples . in this context , iterative algorithms , where only one gradient estimate is used in each step , are often referred to as online learning algorithms . more generally , the situation where only stochastic gradient estimates are available is important in stochastic optimization , where iterative algorithms can be seen as a form of stochastic approximation . finally , stochastic gradient approaches are considered in the incremental optimization of an objective function which is the sum of many terms , e.g. the empirical risk in machine learning @xcite , see section [ ss : amm ] for a detailed discussion . in the next section we describe our contribution in the context of the state of the art . the study of stochastic approximation methods originates in the classical work of @xcite , and assumes the objective function to be smooth and strongly convex ; the related literature is vast ( see e.g. @xcite and references therein ) . an improvement of the original stochastic approximation method , based on averaging of the trajectories and larger step - sizes , is proposed by @xcite and @xcite . more recently , one can recognize two main approaches to solve general nonsmooth convex stochastic optimization . the first one uses different versions of mirror descent stochastic approximation , based on projected subgradient averaging techniques @xcite . similar methods have been extensively studied also in the machine learning community in the context of online learning , where the proof of convergence of the average of the iterates is often based on regret analysis and , the so called , online - to - batch conversion @xcite . the second line of research is based on stochastic variants of accelerated proximal gradient descent @xcite . the algorithm we consider is also a stochastic extension of proximal gradient descent , but corresponds to its basic version with no acceleration . indeed , as discussed below , a main question we consider is if accelerated methods yield any advantage in the stochastic case . the fobos algorithm in @xcite is the closest approach to the one we consider , the main two differences being 1 ) we consider an additional relaxation step which may lead to accelerations , and especially 2 ) we do not consider averaging of the iterates . this latter point is important , since averaging can have a detrimental effect . indeed , non - smooth problems often arise in applications where sparsity of the solution is of interest , and it is easy to see that averaging prevent the solution to be sparse @xcite . moreover , as noted in @xcite and @xcite , averaging can have a negative impact on the convergence rate in the strongly convex case . indeed , in this paper we improve the error bound in @xcite in this latter case . our study is developed in an infinite dimensional setting , where we focus on almost sure convergence of the iterates and non asymptotic bounds on their expectation . considering iterates convergence is standard in optimization theory and often considered in machine learning when sparsity based learning is studied @xcite . the theoretical analysis in the paper is divided in two parts . in the first , we study convergence in expectation in the strongly convex case , generalizing the results in ( * ? ? ? * section 3 ) to the nonsmooth case . we provide a non - asymptotic analysis of stochastic proximal gradient descent where the bounds depend explicitly on the parameters of the problem . interestingly , we obtain , in the strongly convex case , the same @xmath0 error bound that can be obtained from the optimal rate of convergence for function values as achieved by accelerated methods , see e.g. @xcite . this result ( confirmed by numerical simulations ) suggests that , unlike in the deterministic setting , in stochastic optimization acceleration does not have an impact on the rate of convergence . in the second part , we establish almost sure convergence . our results generalize to the composite case the analysis of the stochastic projected subgradient algorithm in a hilbert space @xcite ( see also @xcite ) . our analysis is based on a novel extension of the analysis of proximal methods with exact gradient , based on considering random quasi - fejr sequences @xcite . this approach allows to consider assumptions on the stochastic estimates of the gradients which are more general than those considered in previous work , and does not require boundedness of the iterates . we note that a recent technical report @xcite also analyzes a stochastic proximal gradient method ( without the relaxation step ) and its accelerated variant . almost sure convergence of the iterates ( without averaging ) is proved under uniqueness of the minimizer , but under assumptions different from ours : continuity of the objective function thus excluding constrained smooth optimization and boundedness the iterates . convergence rates for the iterates without averaging are derived , but only for the accelerated method . finally , we note that convergence of the iterates of stochastic proximal gradient has been recently obtained from the analysis of convergence of stochastic fixed point algorithms presented in the recent preprint @xcite . however , this latter results is derived from summability assumptions on the errors of the stochastic estimates which are usually not satisfied in the machine learning setting . the paper is organized as follows . in section [ sec : psae ] we introduce composite optimization and the stochastic proximal gradient algorithm , along with some relevant special cases . in section [ s : alconv ] , we study convergence in expectation and almost surely that we prove in section [ sec : proofs ] . section [ s : exp ] describes some numerical tests comparing the stochastic projected gradient algorithm with state of the art stochastic first order methods . the proofs of auxiliary results are found in appendix [ s : bg ] . [ [ notation - and - basic - definitions ] ] * notation and basic definitions * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + throughout , @xmath1 is a probability space , @xmath2 , and @xmath3 is a real separable hilbert space . we use the notation @xmath4 and @xmath5 for the scalar product and the associated norm in @xmath3 . the symbols @xmath6 and @xmath7 denote , respectively , weak and strong convergence . the class of lower semicontinuous convex functions @xmath8-\infty,+\infty\right]$ ] such that @xmath9 , is denoted by @xmath10 . the proximity operator of @xmath11 is @xmath12 throughout this paper , we assume implicitly that the closed - form expressions of the proximity operators to be available . we refer to @xcite for the closed - form expression of a wide class of functions , see @xcite for examples in machine learning . given a random variable @xmath13 , we denote by @xmath14 $ ] its expected value , and by @xmath15 the @xmath16-field generated by @xmath13 . the conditional expectation of @xmath13 given a @xmath16-algebra @xmath17 is denoted by @xmath18 $ ] . the conditional expectation of @xmath13 given @xmath19 is denoted by @xmath20 $ ] . a filtration of @xmath21 is an increasing sequence @xmath22 of sub-@xmath16-algebras of @xmath21 . a @xmath3-valued random process is a sequence of random variables @xmath23 taking values in @xmath3 . the shorthand notation ` a.s . ' stands for ` almost sure ' . in this section , we introduce the composite convex optimization problem , the stochastic proximal method we study , and discuss some special cases of the framework we consider . composite optimization problems are defined as the problem of minimizing the sum of a smooth convex function and a possibly nonsmooth convex function . here we assume that the latter is proximable , that is the proximity operator is available in closed form or can be easily computed . [ prob2 ] let @xmath24 , let @xmath250,\infty[$ ] , and let @xmath26 be convex and differentiable , with a @xmath27-lipschitz continuous gradient . the problem is to @xmath28 under the assumption that the set of solutions to is non - empty . as mentioned in the introduction , problems with this composite structure has been recently extensively studied in convex optimization . in particular , the class of splitting methods , which decouple the contribution of the smooth term and the nonsmooth one , received a lot of attention @xcite . within the class of splitting methods , in this paper we study the following stochastic proximal gradient ( spg ) algorithm . [ a : main ] let @xmath29 be a strictly positive sequence , let @xmath30 be a sequence in @xmath31 $ ] , and let @xmath32 be a @xmath3-valued random process such that @xmath33 @xmath34<+\infty$ ] . fix @xmath35 a @xmath3-valued integrable vector with @xmath36 < + \infty$ ] and set @xmath37 \end{array}\ ] ] algorithm [ a : main ] is a stochastic version of the proximal forward - backward splitting @xcite , where we replace the exact gradient by a stochastic element . more specifically , if , for every @xmath38 , @xmath39 , our algorithm reduces to the one in @xcite . a stochastic proximal forward - backward splitting ( fobos ) was firstly proposed in @xcite for minimizing the sum of two functions where one of them is proximable , and the other is convex and subdifferentiable . algorithm [ a : main ] generalizes the fobos algorithm , by including a relaxation step , while assuming the first component in to be smooth . as it is the standard , to ensure convergence of the proposed algorithm , we need additional conditions on the random process @xmath40 as well as on the sequence of step - sizes @xmath29 . the following conditions will be considered for the filtration @xmath22 with @xmath41 . 1 . [ e : csss1 ] for every @xmath38 , @xmath42 = \nabla { l}(w_n)$ ] . 2 . [ e : csss2 ] for every @xmath38 , there exist @xmath430,+\infty\right[$ ] and @xmath440,+\infty\right[$ ] such that @xmath45 \leq \sigma^2(1 + \alpha_n\|\nabla { l}(w_n)\|^2 ) \ ] ] 3 . there exists @xmath460,+\infty\right[$ ] such that @xmath47 4 . for any solution @xmath48 of the problem , set @xmath49 . assume that @xmath50 condition ( a1 ) means that , at each iteration @xmath51 , @xmath52 is an unbiased estimate of the gradient of the smooth term . condition ( a2 ) has been considered in @xcite . it is weaker than typical conditions used in the analysis of stochastic ( sub)gradient algorithms , namely boundedness of the sequence @xmath53)_{n\in{\ensuremath{\mathbb n}}^*}$ ] ( see @xcite ) or even boundedness of @xmath54 ( see @xcite ) . we note that this last requirement on the entire space is not compatible with the assumption of strong convexity , because the gradient is necessarily not uniformly bounded , therefore the use of the more general condition ( a2 ) is needed in this case . conditions such as ( a3 ) and ( a4 ) are , respectively , widely used in the deterministic setting and in stochastic optimization . assumption ( a3 ) is more restrictive that the one usually assumed in the deterministic setting , that is @xmath55 . we also note that when @xmath30 is bounded away from zero , and @xmath56 is bounded , ( a4 ) implies ( a3 ) for @xmath51 large enough . the condition @xmath57 in assumption @xmath58 is satisfied if @xmath59 is summable . moreover , if @xmath60 , it reduces to @xmath61 , since in this case @xmath62 for every solution @xmath63 . finally , in our case , the step - size is required to converge to zero , while it is typically bounded away from zero in the study of deterministic proximal forward - backward splitting algorithm @xcite . problem [ prob2 ] covers a wide class of deterministic as well as stochastic convex optimization problems , especially from machine learning and signal processing , see e.g. @xcite and references therein . the simplest case is when @xmath64 is identically equal to 0 , so that problem [ prob2 ] reduces to the classic problem of finding a minimizer of a convex differentiable function from unbiased estimates of its gradients . in the case when @xmath64 is the indicator function of a nonempty , convex , closed set @xmath65 , i.e. @xmath66 then problem reduces to a constrained minimization problem of the form @xmath67 which is well studied in the literature , as mentioned in the introduction . below , we discuss in more detail some special cases of interest . ( * minimization of an expectation*).[ex : expmin ] let @xmath68 be a random vector with probability distribution @xmath69 supported on @xmath70 and @xmath71 . stochastic gradient descent methods are usually studied in the case where @xmath3 is an euclidean space and @xmath72 = \int_{{\ensuremath{\boldsymbol{e } } } } f(w,\xi)dp(\xi),\ ] ] under the assumption that @xmath73 @xmath74 is a convex differentiable function with lipschitz continuous gradient @xcite . let @xmath75 be independent copies of the random vector @xmath68 . assume that there is an oracle that , for each @xmath76 , returns a vector @xmath77 such that @xmath78 $ ] . by setting @xmath79 and @xmath80 , then ( a2 ) holds . this latter assertion follows from standard properties of conditional expectation , see e.g. ( * ? ? ? * example 5.1.5 ) . ( * minimization of a sum of functions*)[ex : sumoffunctions ] let @xmath24 , let @xmath81 be a strictly positive integer . for every @xmath82 , let @xmath83 be convex and differentiable , such that @xmath84 has a @xmath27-lipschitz continuous gradient , for some @xmath850,+\infty\right[$ ] . the problem is to @xmath86 this problem is a special case of problem [ e : prob2 ] with @xmath87 , and is especially of interest when @xmath81 is very large and we know the exact gradient of each component @xmath88 . the stochastic estimate of the gradient of @xmath89 is then defined as @xmath90 where @xmath91 is a random process of independent random variables uniformly distributed on @xmath92 , see @xcite . clearly @xmath93 holds . assumption @xmath94 specializes in this case to @xmath95 if the latter is satisfied , then spg algorithm can be applied with a suitable choice of the stepsize . finally , in the next section , we discuss how the above setting specializes to the context of machine learning . consider two measurable spaces @xmath96 and @xmath97 and assume there is a probability measure @xmath98 on @xmath99 . the measure @xmath98 is fixed but known only through a training set @xmath100 of samples i.i.d with respect to @xmath98 . consider a loss function @xmath101 and a hypothesis space @xmath3 of functions from @xmath96 to @xmath97 , e.g. a reproducing kernel hilbert space . a key problem in this context is ( regularized ) empirical risk minimization , @xmath102 the above problem can be seen as an approximation of the problem , @xmath103 the analysis , in this paper , can be adapted to the machine learning setting in two different ways . the first , following example [ ex : expmin ] , is to apply the spg algorithm to directly solve the regularized _ expected _ loss minimization problem . the second , following example [ ex : sumoffunctions ] , is to apply the spg algorithm to solve the regularized _ empirical _ risk minimization problem . in either one of the above two problems , the first term is differentiable if the loss functions is differentiable with respect to its second argument , examples being the squared or the logistic loss . for these latter loss functions , and more generally for loss functions which are twice differentiable in their second argument , it easy to see that the lipschitz continuity of the gradient is satisfied if the maximum eigenvalue of the hessian is bounded . the second term @xmath104 can be seen as a regularizer / penalty encoding some prior information about the learning problem . examples of convex , non - differentiable penalties include sparsity inducing penalties such as the @xmath105 norm , as well as more complex structured sparsity penalties @xcite . stronger convexity properties can be obtained considering an _ elastic net penalty _ @xcite , that is adding a small strongly convex term to the sparsity inducing penalty . clearly , the latter term would not be necessary if the risk in problem [ e : learn1 ] ( or the empirical risk in ) is strongly convex . however , this latter requirement depends on the probability measure @xmath98 and is typically not satisfied when considering high ( possibly infinite ) dimensional settings . in this section , we state and discuss the main results of the paper . we derive convergence rates of the proximal gradient algorithm ( with relaxation ) for stochastic minimization . the section is divided in two parts . in the first one , section [ sec : exp ] , we focus on convergence in expectation . in the second one , section [ sec : as ] , we study almost sure convergence of the sequence of iterates . in both cases , additional convexity conditions on the objective function are required to derive convergence results . the proofs are deferred to section [ sec : proofs ] . in this section , we denote by @xmath48 a solution of problem [ e : prob2 ] and provide an explicit non - asymptotic bound on @xmath106 $ ] . this result generalizes to the nonsmooth case the bound obtained in ( * ? ? ? * theorem 1 ) for stochastic gradient descent . the following assumption is considered throughout this section . [ a : strcon ] the function @xmath89 is @xmath107-strongly convex and @xmath64 is @xmath108-strongly convex , for some @xmath109 and @xmath110 , with @xmath111 . note that , we do not assume both @xmath89 and @xmath64 to be strongly convex , indeed the constants @xmath107 and @xmath108 can be zero , but require that only one of the two is . this implies that @xmath48 is the unique solution of problem [ e : prob2 ] . in the statement of the following theorem , we will use the family of functions @xmath112 defined by setting , for every @xmath113 , @xmath1140,{\ensuremath{+\infty}}\right[\to { \ensuremath{\mathbb r}}\colon t\mapsto \begin{cases } ( t^{c}-1)/c & \text{if $ c \not=0$};\\ \log t & \text{if $ c = 0$}. \end{cases}\ ] ] this family of functions arises in lemma [ l : ocs ] in the appendix and are useful to bound the sum of the stepsizes . [ t:1 ] assume that conditions @xmath115 , @xmath116 and assumption [ a : strcon ] are satisfied . suppose that there exist @xmath1170,+\infty\right[$ ] and @xmath1180,+\infty\right[$ ] such that @xmath119 let @xmath1200,+\infty[$ ] and let @xmath1210,1\right]$ ] . suppose that , for every @xmath122 , @xmath123 . set @xmath124 let @xmath125 be the smallest integer such that @xmath126 , and @xmath127 then , by setting @xmath128,\ ] ] we have , for every @xmath129 , @xmath1300,1\right[$,}\\ s_{n_0}\big(\dfrac{n_0}{n+1}\big)^{c}+ \dfrac{2^c\tau c^2 } { ( n+1)^{c}}\varphi_{c-1}(n ) & \text{if $ \theta = 1$. } \end{cases}\ ] ] in theorem [ t:1 ] , the dependence on the strong convexity constants is hidden in the constant @xmath131 . taking into account , we can write more explicitly the asymptotic behavior of the sequence @xmath132 . [ cor:1 ] under the same assumptions and with the same notation of theorem [ t:1 ] , the following holds @xmath133=\begin{cases}o(n^{-\theta } ) & \text{if $ \theta\in\,]0,1[$},\\ o(n^{-c})+o(n^{-1 } ) & \text{if $ \theta=1$}. \end{cases}\ ] ] thus , if @xmath134 and @xmath135 is chosen such that @xmath136 , then @xmath137 = o(n^{-1})$ ] . in particular , if @xmath138 , @xmath139 for every @xmath38 , and @xmath140 , then @xmath141 , @xmath142 , and @xmath133\leq \dfrac{n_{0}^2{\ensuremath{\mathbb{e}}}[\|w_{n_0}-\overline{w}\|^2 ] } { ( n+1)^2}+ \dfrac{8\sigma^2(1+\overline{\alpha } \|\nabla { l}(\overline{w})\|)(1+\nu)^4}{\underline{\lambda}^2\left(\mu \epsilon+\nu\right)^2}\ ] ] theorem [ t:1 ] is the extension to the nonsmooth case of ( * ? ? ? * theorem 1 ) , in particular , when @xmath143 , we obtain the same bounds . note however that the assumptions on the stochastic approximations of the gradient of the smooth part are different . in particular , we replace the boundedness condition at the solution and the lipschitz continuity assumption on @xmath40 with assumption ( a2 ) . as can be seen from corollary [ cor:1 ] , the fastest asymptotic rate corresponds to @xmath138 and it is the same obtained in the smooth case in ( * ? ? ? * theorem 2 ) . note that this rate depends on the asymptotic behavior of the step - size , but also on the constant @xmath131 , which in turns depends on @xmath135 . as pointed out in @xcite , see also in @xcite , this choice is critical , because too small choices of @xmath135 affect the convergence rates , and too big choices influence significantly the value of the constants in the first term of . in particular , as can be readily seen in corollary [ cor:1 ] , the choice is determined by the strong convexity constants . moreover , the dependence on the strong convexity constant shown in corollary [ cor:1 ] is of the same type of the one obtained in the regret minimization framework by @xcite . there are other stochastic first order methods achieving the same rate of convergence for the iterates in the strongly convex case , see e.g. @xcite . indeed , the rate we obtain is the rate that can be obtained by the optimal ( in the sense of @xcite ) convergence rate on the function values . among the mentioned methods those in @xcite belong to the class of accelerated proximal gradient methods . our result shows that , in the strongly convex case , the rate of convergence of the iterates is the same in the accelerated and non accelerated case . in addition , if sparsity is the main interest , we highlight that many of the algorithms discussed above ( e.g. @xcite ) involve some form of averaging or linear combination which prevent sparsity of the iterates , as it is discussed in @xcite . our result shows that in this case averaging is not needed , since the iterates themselves are convergent . we next compare in some detail our results with those obtained for the fobos algorithm in @xcite and to the stochastic proximal gradient in @xcite . there are a few difference in the settings considered . in particular , convergence of the average of the iterates with respect to the function values is considered in @xcite assuming uniform boundedness of the iterations and the subdifferentials . the space @xmath3 is assumed to be finite dimensional , though the analysis might be extended to infinite dimensional spaces . finally , the optimal stepsize in @xcite depends explicitly on the radius of the ball containing the iterates , which in general might not be available . our convergence results consider convergence of the iterates ( with no averaging ) and hold in an infinite dimensional setting , without boundedness assumptions . the non asymptotic rate @xmath144 which we obtain for the iterates improves the @xmath145 rate derived from ( * ? ? ? * corollary 10 ) for the average of the iterates . however , it should be noted that convergence of the objective values is studied in @xcite also for the non strongly convex case . spg ( without relaxation ) has been recently studied in @xcite . also in this case the authors assume a priori boundedness of the iterates and prove convergence of the averaged sequence . theorem [ t:1 ] is also comparable with deterministic stochastic proximal forward - backward algorithm with errors @xcite . on the one hand , we allow the errors to satisfy assumption ( a2 ) , while in the deterministic case the errors in the computation of the gradient should decrease to zero sufficiently fast . on the other hand , we require asymptotically vanishing ( and smaller , according to ( a3 ) ) step - sizes , while , in the deterministic case , the step - size is bounded from below . finally , if @xmath146 is continuous , in the setting of theorem [ t:1 ] , it holds @xmath147 . moreover , if @xmath146 is lipschitz continuous , then @xmath148 and if @xmath146 is differentiable with lipschitz continuous gradient , @xmath149 . in this section , we focus on almost sure convergence of spg algorithm . this kind of convergence of the iterates is the one traditionally studied in the stochastic optimization literature . depending on the convexity properties of the function @xmath89 , we get two different convergence properties . the first theorem requires uniform convexity of @xmath89 at the solution . [ t:2 ] suppose that the conditions @xmath115 , @xmath116 , and @xmath58 are satisfied . let @xmath150 be a sequence generated by algorithm [ a : main ] and assume that @xmath89 is uniformly convex at @xmath48 . then @xmath151 a.s . if we relax the strong convexity assumption , we can still prove weak convergence of a subsequence in the strictly convex case , provided an additional regularity assumption holds . [ t:3 ] suppose that the conditions @xmath115 , @xmath116 , and @xmath58 are satisfied . let @xmath150 be a sequence generated by algorithm [ a : main ] . assume that @xmath89 is strictly convex , and let @xmath48 be the unique solution of problem [ e : prob2 ] . if @xmath152 is weakly continuous , then there exists a subsequence @xmath153 such that @xmath154 a.s . with respect to the previous section , here we make the additional assumption @xmath58 on the summability of the sequence of step - sizes multiplied by the relaxation parameters . for stochastic gradient algorithm without relaxation , i.e , @xmath143 and , for every @xmath38 @xmath155 , assumption ( a4 ) coincides with the classical step - size condition @xmath156 and @xmath157 which guarantees a sufficient but not too fast decrease of the step - size ( see e.g. @xcite ) . assumption @xmath94 has been considered in the context of stochastic gradient descent in @xcite . note that under such a condition , the variance of the stochastic approximation is allowed to grow with @xmath158 . as mentioned in the introduction , the study of almost sure convergence is classical . an analysis of a stochastic projected subgradient algorithm in an infinite dimensional hilbert space can be found in @xcite . theorem [ t:2 ] can be seen as an extension of ( * ? ? ? * theorem 3.1 ) , where the case where @xmath64 is an indicator function is considered . our approach is based on random quasi - fejr sequences , and on probabilistic quasi martingale techniques @xcite . [ r:3 ] if @xmath89 is assumed to be only strictly convex and its gradient is not weakly continuous , theorem [ t:3 ] does not ensure weak convergence of any subsequence of @xmath159 . however , if the sequence of function values @xmath160 converges to the minimum of @xmath161 , then @xmath162 a.s . this happens ( see @xcite ) when @xmath163 for some closed subspace @xmath164 of @xmath3 , or when @xmath165 for some non - empty closed convex @xmath65 of @xmath3 , and there exists a bounded function @xmath166 such that @xmath167 \leq h(\|\nabla { l}(w_n)\|).$ ] the proof of remark [ r:3 ] can be found in the next section . we start by recalling the firmly non - expansiveness of the proximity operator and the baillon - haddad theorem ( see ( * ? ? ? * theorem 18.15 ) ) . * lemma 2.4 ) [ l : firm ] let @xmath168 . then the proximity of @xmath64 is firmly non - expansive , i.e. , @xmath169 ( * ? ? ? * definition 4.4 ) [ d : coco ] let @xmath170 , and let @xmath1710,+\infty\right[}}$ ] . then @xmath172 is @xmath173-cocoercive if @xmath174 let @xmath175 be a convex differentiable function with @xmath27 lipschitz gradient . then , @xmath152 is @xmath176-cocoercive . we next state the following lemma ; see also ( * ? ? ? * lemma 5 , chapter 2.2 ) and @xcite . we will use the family of functions @xmath177 defined in . for completeness , the proof is given in the appendix . [ l : ocs ] let @xmath1780,1\right]$ ] , and let @xmath131 and @xmath179 be in @xmath1800,+\infty[$ ] , let @xmath181 be a strictly positive sequence defined by @xmath182 . let @xmath183 be such that @xmath184 let @xmath125 be the smallest integer such that @xmath185 and set @xmath186 . then , for every @xmath129 , @xmath1870,1\right[$,}\\ s_{n_0}\big(\frac{n_0}{n+1}\big)^{c}+ \frac{\tau c^2}{(n+1)^{c}}(1 + \frac{1}{n_0})^{c}\varphi_{c-1}(n ) & \text{if $ \alpha = 1$. } \end{cases}\ ] ] we start with a technical result , giving some bounds that will be repeatedly used . [ p : bb ] consider the setting of the spg algorithm and let @xmath48 be a solution of problem [ e : prob2 ] . suppose that conditions ( a1 ) , ( a2 ) , and ( a3 ) are satisfied . then the following hold : 1 . [ e : est1 t ] @xmath188 2 . [ p : bbii ] set @xmath189 then , for every @xmath190 @xmath191 3 . [ p : bbiii ] for every @xmath190 @xmath192 \leq & \left({\ensuremath{\mathbb{e}}}\left[\|w_n-\overline{w}\|^2\right]-2\gamma_n\left(1-\gamma_n\beta(1 + 2\sigma^2\alpha_n)\right)\cdot \right.\\ & \cdot{\ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{\nabla{l}(w_n)-\nabla { l}(\overline{w } ) } \right\rangle } ] + 2\gamma_n^2\sigma^2 ( 1+\alpha_n\|\nabla{l}(\overline{w})\|^2)\bigg).\end{aligned}\ ] ] [ e : est1 t ] : follows from convexity of @xmath193 . [ p : bbii ] : we have @xmath194 moreover , since @xmath195 is firmly non - expansive by lemma [ l : firm ] @xmath196 and the statement follows . [ p : bbiii ] : note that , for every @xmath38 , we have that @xmath197 and @xmath52 are measurable with respect to @xmath198 since they are @xmath199 measurable and by definition @xmath200 . the same holds for @xmath201 , for it is the difference of two measurable functions . we next show by induction that @xmath33 @xmath202 is integrable . first , @xmath203 is integrable by assumption . then , assume by inductive hypothesis that @xmath202 is integrable . then so is @xmath204 , for @xmath52 is square integrable by assumption . moreover , @xmath205 , because @xmath206 is nonexpansive . therefore @xmath207 is integrable and hence so is @xmath208 . this implies that @xmath209<+\infty$ ] and @xmath210<+\infty$ ] . therefore , using assumption ( a1 ) , we obtain @xmath211 & = { \ensuremath{\mathbb{e}}}[{\ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{\mathrm{g}_n - \nabla{l}(\overline{w } ) } \right\rangle}|\mathcal{a}_n ] \notag\\ & = { \ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{{\ensuremath{\mathbb{e } } } [ \mathrm{g}_n - \nabla{l}(\overline{w})|\mathcal{a}_n ] } \right\rangle}]\notag\\ & = { \ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) } \right\rangle}].\end{aligned}\ ] ] moreover , using the assumption ( a2 ) , we have @xmath212 \leq 2 { \ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2 ] + 2 { \ensuremath{\mathbb{e}}}[\|\mathrm{g}_n - \nabla{l}(w_n)\|^2]\notag\\ & \leq 2 { \ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2 ] + 2\sigma^2(1 + \alpha_n{\ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n)\|^2])\notag\\ & \leq ( 2 + 4\sigma^2\alpha_n){\ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2 ] + 2\sigma^2(1 + 2\alpha_n\|\nabla{l}(\overline{w})\|^2)\notag\\ & \leq { ( 2 + 4\sigma^2\alpha_n)}{\beta}{\ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) } \right\rangle } ] + 2\sigma^2(1 + 2\alpha_n)\|\nabla{l}(\overline{w})\|^2 , \end{aligned}\ ] ] where the last inequality follows from the fact that @xmath213 is cocoercive since it is lipschitz - continuous ( by the baillon - haddad theorem ) . the statement then follows from , , and . we are now ready to prove theorem [ t:1 ] . since @xmath111 , then @xmath214 is strongly convex . hence , problem has a unique minimizer @xmath48 . since @xmath215 is @xmath216-strongly convex , by ( * ? ? ? * proposition 23.11 ) @xmath206 is @xmath217-cocoercive , and then @xmath218 next , proceeding as in the proof of proposition [ p : bb ] , we get an inequality analogue to , that is @xmath219 \leq & \frac{1}{(1+\gamma_n\nu)^2 } \left({\ensuremath{\mathbb{e}}}\left[\|w_n-\overline{w}\|^2\right]-2\gamma_n\left(1-\gamma_n\beta(1 + 2\sigma^2\alpha_n)\right)\cdot \right.\\ \label{eq : ymenw } & \cdot{\ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{\nabla{l}(w_n)-\nabla { l}(\overline{w } ) } \right\rangle } ] + 2\gamma_n^2\sigma^2 ( 1+\alpha_n\|\nabla{l}(\overline{w})\|^2)\bigg).\end{aligned}\ ] ] since @xmath89 is strongly convex of parameter @xmath107 , it holds @xmath220 . therefore , from , using the @xmath107-strong convexity of @xmath89 and ( a3 ) , we get @xmath221 & \leq \frac{1}{(1+\gamma_n\nu)^2}\bigg((1 - 2\gamma_n\mu\epsilon){\ensuremath{\mathbb{e}}}\left[\|w_n-\overline{w}\|^2\right ] + 2\sigma^2\chi_n^2\bigg).\end{aligned}\ ] ] hence , by definition of @xmath222 , @xmath223 & \leq \bigg(1-\frac{\lambda_n\gamma_n(2\nu + \gamma_{n}\nu^2 + 2\mu\epsilon)}{(1+\gamma_n\nu)^2}\bigg){\ensuremath{\mathbb{e}}}[\|w_n-\overline{w}\|^2]+ \frac { 2\sigma^2\chi^{2}_n}{(1+\gamma_n\nu)^2}.\end{aligned}\ ] ] let @xmath224 and fix @xmath225 . since @xmath226 , we have @xmath227 where we set @xmath228 . on the other hand , @xmath229 then , putting together , , and , we get @xmath230 \leq ( 1-\eta_n){\ensuremath{\mathbb{e}}}[\|w_n-\overline{w}\|^2]+ \tau\eta_n^2 $ ] , with @xmath231 and @xmath232 . finally , follows from lemma [ l : ocs ] . in order to prove theorem [ t:2 ] , we start by giving the definition of deterministic and random quasi - fejr sequences . we denote by @xmath233 the set of summable sequences in @xmath234 . @xcite let @xmath235 be a non - empty subset of @xmath3 and let @xmath236 be a sequence in @xmath237 . then , * a sequence @xmath159 in @xmath3 is deterministic quasi - fejr monotone with respect to the target set @xmath235 if @xmath238 * a sequence of random vectors @xmath159 in @xmath3 is stochastic quasi - fejr monotone with respect to the target set @xmath235 if @xmath36 < + \infty$ ] and @xmath239 \leq \|w_n - w\|^2 + \varepsilon_n.\ ] ] the following result has been stated in @xcite without a proof . for the sake of completeness , a proof is given in the appendix . * lemma 2.3 ) [ p : fejer ] let @xmath235 be a non - empty closed subset of @xmath3 , let @xmath240 . let @xmath150 be a sequence of random vectors in @xmath3 such that @xmath36 < + \infty$ ] , and let @xmath241 . assume that @xmath242 \leq \|w_n - w\|^2 + \varepsilon_n \quad \text{a.s}.\ ] ] then the following hold . 1 . [ p : fejeri ] let @xmath243 . then , @xmath244)_{n\in{\ensuremath{\mathbb n}}^{*}}$ ] converges to some @xmath245 and @xmath246 converges a.s . to an integrable random vector @xmath247 . [ p : fejerii ] @xmath150 is bounded a.s . [ p : fejeriii ] the set of weak cluster points of @xmath150 is non - empty a.s . we next collect some convergence results that will be useful in the proof of the main theorem [ t:2 ] . [ p:1 ] suppose that ( a1 ) , ( a2 ) , ( a3 ) , and ( a4 ) are satisfied . let @xmath150 be a sequence generated by algorithm [ a : main ] . then , for any solution @xmath48 of the problem , the following hold : 1 . [ p:1i ] the sequence @xmath248)_{n\in{\ensuremath{\mathbb n}}^{*}}$ ] converges to a finite value . [ t:2i ] the sequence @xmath249 converges a.s to some integrable random variable @xmath250 . [ p:1iii ] @xmath251 < { \ensuremath{+\infty}}$ ] . consequently , @xmath252 = 0 \quad \text{and } \quad \varliminf_{n\to\infty}{\ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n)-\nabla{l}(\overline{w})\|^2 ] = 0.\ ] ] 4 . [ p:1ii ] @xmath253 < { \ensuremath{+\infty}}$ ] and @xmath254 < + \infty$ ] . by proposition [ p : bb][e : est1t]-[p : bbiii ] , and by condition ( a3 ) , we get @xmath255 \leq ( 1-\lambda_n ) { \ensuremath{\mathbb{e}}}[\|w_n-\overline{w}\|^2 ] + \lambda_n { \ensuremath{\mathbb{e}}}[\|y_n-\overline{w}\|^2 ] \notag\\ & \leq { \ensuremath{\mathbb{e}}}[\|w_n-\overline{w}\|^2 ] - 2\varepsilon \gamma_n\lambda_n { \ensuremath{\mathbb{e}}}[{\left\langle{w_n-\overline{w}},{\nabla{l}(w_n)-\nabla{l}(\overline{w } ) } \right\rangle } ] + 2\sigma^2\chi^{2}_n -\lambda_n { \ensuremath{\mathbb{e}}}[\|u_n\|^2]\notag\\ & \leq { \ensuremath{\mathbb{e}}}[\|w_n-\overline{w}\|^2 ] + 2\sigma^2\chi^{2}_n,\end{aligned}\ ] ] where the last inequality follows by the monotonicity of @xmath152 . [ p:1i ] : since the sequence @xmath256 is summable by assumption ( a4 ) , we derive from that @xmath257)_{n\in{\ensuremath{\mathbb n}}^{*}}$ ] converges to a finite value . [ t:2i ] : we estimate the conditional expectation with respect to @xmath199 of each term in the right hand side of . since @xmath197 is @xmath199-measurable , we have @xmath258 = \|w_n-\overline{w}\|^2.\ ] ] using assumption ( a1 ) , @xmath259 & = { \left\langle{w_n-\overline{w}},{{\ensuremath{\mathbb{e } } } [ \mathrm{g}_n - \nabla{l}(\overline{w})| \mathcal{a}_n } \right\rangle}\notag\\ & = { \left\langle{w_n-\overline{w}},{\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) } \right\rangle}. \end{aligned}\ ] ] next , note that @xmath260 is @xmath199-measurable by ( a1 ) , and therefore by @xmath94 , we get @xmath261 \leq 2 { \ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2| \mathcal{a}_n ] + 2 { \ensuremath{\mathbb{e}}}[\|\mathrm{g}_n - \nabla{l}(w_n)\|^2 |\mathcal{a}_n ] \notag\\ & \leq 2\|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2 + 2\sigma^2(1+\alpha_n\|\nabla{l}(w_n)\|^2)\notag\\ & \leq 2\|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2 + 2\sigma^2(1 + 2\alpha_n \|\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) \|^2 + 2\alpha_n\|\nabla{l}(\overline{w})\|^2 ) \notag\\ & \leq { ( 2 + 4\sigma^2\alpha_n ) } { \beta } { \left\langle{w_n-\overline{w}},{\nabla{l}(w_n ) - \nabla{l}(\overline{w } ) } \right\rangle } + 2\sigma^2(1 + 2\alpha_n\|\nabla{l}(\overline{w})\|^2 ) , \end{aligned}\ ] ] where the last inequality follows from the cocoercivity of @xmath213 . taking the conditional expectation with respect to @xmath199 , and invoking , , , and , we obtain , @xmath262 \leq ( 1-\lambda_n ) \|w_n-\overline{w}\|^2 + \lambda_n { \ensuremath{\mathbb{e}}}[\|y_n-\overline{w}\|^2|\mathcal{a}_n ] \notag\\ & \leq \|w_n-\overline{w}\|^2 - 2\gamma_n\lambda_n(1-{\beta\gamma_n(1 + 2\sigma^2\alpha_n)}){\left\langle{\nabla{l}(w_n)-\nabla{l}(\overline{w } ) } , { w_n-\overline{w } } \right\rangle } \notag\\ & \quad + 2\sigma^2\chi^{2}_n-\lambda_n{\ensuremath{\mathbb{e } } } [ \|u_n\|^2|\mathcal{a}_n]\notag\\ & \leq \|w_n-\overline{w}\|^2 - 2\varepsilon \gamma_n\lambda_n { \left\langle{\nabla{l}(w_n)-\nabla{l}(\overline{w } ) } , { w_n-\overline{w } } \right\rangle } + 2\sigma^2\chi^{2}_n-\lambda_n{\ensuremath{\mathbb{e } } } [ \|u_n\|^2|\mathcal{a}_n]\notag\\ & \leq \|w_n-\overline{w}\|^2 + 2\sigma^2\chi^{2}_n\,.\end{aligned}\ ] ] hence , @xmath150 is a random quasi - fejr sequence with respect to the nonempty closed and convex set @xmath263 . taking into account that @xmath36<+\infty$ ] by assumption , it follows from proposition [ p : fejer][p : fejeri ] that @xmath249 converges a.s to some integrable random variable @xmath250 . [ p:1iii ] : we derive from that @xmath264 < + \infty.\ ] ] since @xmath265 , we obtain @xmath266 = 0 \quad \rightarrow \varliminf_{n\to\infty}{\ensuremath{\mathbb{e}}}[\|\nabla{l}(w_n)-\nabla{l}(\overline{w } ) \|^2 ] = 0,\ ] ] using again the cocoercivity of @xmath152 . [ p:1ii ] we directly get from that @xmath267 . since @xmath213 is lipschitz - continuous , and @xmath268)_{n\in{\ensuremath{\mathbb n}}^*}$ ] is convergent by [ p:1i ] , there exists @xmath2690,+\infty[$ ] such that @xmath270 \leq \beta { \ensuremath{\mathbb{e}}}[\|w_n-\overline{w}\|^2 ] \leq m < + \infty.\end{aligned}\ ] ] hence , we derive from and that @xmath271 < + \infty.\ ] ] now , recalling the definition of @xmath272 in , using and , we obtain @xmath273 \leq 2\sum_{n\in{\ensuremath{\mathbb n } } } \lambda_n { \ensuremath{\mathbb{e}}}[\|u_n\|^2 ] + 2\sum_{n\in{\ensuremath{\mathbb n}}^ { * } } \lambda_n\gamma^{2}_n { \ensuremath{\mathbb{e}}}[\|\mathrm{g}_n-\nabla{l}(\overline{w})\|^2 ] < + \infty.\ ] ] since @xmath89 is uniformly convex at @xmath48 , there exists @xmath274 increasing and vanishing only at @xmath275 such that @xmath276 therefore , we derive from proposition [ p:1 ] [ p:1iii ] that @xmath277 < \infty,$ ] and hence @xmath278 since @xmath279 is not summable , we have @xmath280 a.s . consequently , there exists a subsequence @xmath281 such that @xmath282 a.s , which implies that @xmath283 a.s . in view of proposition [ p:1][t:2i ] , we get @xmath284 a.s . by proposition [ p:1][p:1i ] , @xmath285 converges to an integrable random variable , hence it is uniformly bounded . moreover , @xmath286= 0 $ ] , and hence there exists a subsequence @xmath281 such that @xmath287= 0.$ ] thus , there exists a subsequence @xmath288 of @xmath281 such that @xmath289 let @xmath290 be a weak cluster point of @xmath291 , then there exists a subsequence @xmath292 such that for almost all @xmath293 , @xmath294 . since @xmath213 is weakly continuous , for almost all @xmath293 , @xmath295 . therefore , for almost every @xmath293 , by , @xmath296 , and hence @xmath297 since @xmath89 is strictly convex , @xmath152 is strictly monotone , we obtain @xmath298 . this shows that @xmath299 a.s . let @xmath300 be a weak cluster point of @xmath150 , i.e. , there exists a subsequence @xmath301 such that @xmath302 a.s . since @xmath303 is convex and lower semicontinous , it is weakly lower semicontinous , hence @xmath304 which shows that @xmath305 a.s . we therefore conclude that @xmath150 converges weakly to an optimal solution a.s . in this section we first present numerical experiments aimed at studying the computational performance of the spg algorithm ( see algorithm [ a : main ] ) , with respect to the step - size , the strong convexity constant , and the noise level . then we compare the proposed method with other state - of - the - art stochastic first order methods : an accelerated stochastic proximal gradient method , called sage ( * ? ? ? * theorem 2 ) and the fobos algorithm @xcite . in order to study the behavior of the spg algorithm with respect to the relevant parameters of the optimization problem , we focus on a toy example , where the exact solution is known . more specifically , we consider the following minimization problem on the real line : @xmath306 it is clear that @xmath307 is @xmath107-strongly convex function with @xmath308 and the optimal value @xmath309 . we consider a stochastic perturbation of the exact gradient of the function @xmath310 of the form @xmath311 where @xmath312 is a realization of a gaussian random variable with @xmath275 mean and @xmath313 variance . we apply spg one hundred times for 100 independent realizations of the random process @xmath314 to problem with @xmath33 @xmath155 and @xmath315 for some constant @xmath316 . we evaluate the average performance of spg over the first 100 iterations for different values of the strong convexity parameter @xmath107 , and several values of @xmath16 and @xmath65 , and by measuring @xmath317 . the results are displayed in figure [ fig : orbit1pcabc ] . as can be seen by visual inspection , the convergence is faster when @xmath107 is bigger and when the noise variance is smaller . moreover , the constant @xmath65 in the step - size heavily influence the convergence behavior . the latter is a well - known phenomenon in the context of stochastic optimization @xcite , [ cols="^,^,^ " , ] as a last experiment , we focus on the problem of recovering an ideal signal @xmath48 from a noisy observation of the form @xmath318 where @xmath319 and @xmath320 is a gaussian kernel . , width=302 ] to find an approximation of the ideal signal , we solve the following variational problem @xmath321 an approximation @xmath48 of the exact solution is found by running the forward - backward splitting method in @xcite for @xmath322 iterations . then , we run spg , sage , and fobos with the same initialization , for @xmath323 iterations using at the @xmath51-th iteration a stochastic gradient of the form @xmath324 where @xmath325 and @xmath326 . fobos is run with @xmath327 . this is not the theoretically optimal choice , but gave better results in practice . in spg we set @xmath155 and @xmath328 . convergence of @xmath329 for the three algorithms is presented in figure [ fig : orbit1pc1 ] . in this case sage is the fastest , and spg shows slightly worse convergence . fobos is again slower . , width=302 ] finally , we address the problem of the iterations sparsity . we generate the data according to the model in , starting from an original signal with @xmath330 zero components . in figure [ fig : orbit1pc1a ] we display the number of zero components of the iterates . as it can be readily seen by visual inspection , after few iteratons both sage ans spg generate sparse iterations . on this example this does not hold for the fobos algorithm , for which the sparsity of the iterates is a decreasing function of the number of iterations . the number of zero components of the last iterate of spg , sage , and fobos is 937 , 937 , and 438 , respectively . with the same initial point @xmath275 for spg and sage . [ fig : orbit1pc1a ] , width=302 ] in this paper we proposed and studied a stochastic approach to the problem of minimizing a strongly convex non - smooth function . in particular , we have considered the case of composite minimization ( the objective function is the sum of a smooth and a convex term ) and proposed a stochastic extension of proximal splitting methods which have become widely popular in a deterministic setting . these latter approaches are based on recursively computing the gradient of the smooth term and then applying the proximity operator defined by the convex term . the starting point of the paper is considering the case where only stochastic estimates of the gradient are available . this latter situation is relevant in online approaches to learning and more generally in stochastic and incremental optimization . the main contributions of this paper are to provide convergence rates in expectation and to establish almost sure convergence of the proposed stochastic proximal gradient method . a further contribution regards our proving techniques , which differ from many previous approaches based on regret analysis and online - to batch conversion . indeed , our approach is based on extending convex optimization techniques from the deterministic to the stochastic setting . such extensions are interesting in their own right and could lead to further applications . an outcome of our analysis is that , unlike in the deterministic case , in the stochastic case acceleration does not yield any improvement in the rates . the analysis in the paper suggests a few venues for future work . a main one is the relaxation of the strong convexity assumption , considering in particular objective functions which are only convex ; steps in this direction have been taken in @xcite . deriving high probability , rather than expectation bounds , would also be interesting . this material is based upon work supported by the center for brains , minds and machines ( cbmm ) , funded by nsf stc award ccf-1231216 . l. r. acknowledges the financial support of the italian ministry of education , university and research firb project rbfr12m3ac . s. v. is member of the gruppo nazionale per lanalisi matematica , la probabilit e le loro applicazioni ( gnampa ) of the istituto nazionale di alta matematica ( indam ) . here we prove the statements about random quasi - fejr sequences and an upper bound on a numerical sequence satisfying the assumptions of lemma [ l : ics ] . [ p : fejeri ] : since the sequence @xmath236 is summable and @xmath332 $ ] is finite , we derive from that @xmath333)_{n\in{\ensuremath{\mathbb n}}^*}$ ] is a real positive quasi - fejr sequence , and therefore it converges to some @xmath334 by ( * ? ? ? * lemma 3.1 ) . set @xmath335 then , it follows from that @xmath336 & = { \ensuremath{\mathbb{e}}}[\|w_{n+1}-w\|^2 |\mathcal{a}_n ] + \sum_{k = n+1}^{\infty}\varepsilon_n\notag\\ & \leq \|w_n - w\|^2 + \sum_{k = n}^{\infty}\varepsilon_n\notag\\ & = r_n.\end{aligned}\ ] ] therefore @xmath337 is a ( real ) supermartingale . since @xmath338 < + \infty$ ] by , @xmath339 converges a.s to an integrable random variable ( * ? ? ? * theorem 9.4 ) , that we denote by @xmath340 . note that , for every @xmath341 , @xmath38 , @xmath342 : @xmath343 where @xmath344 is defined by . since all terms in are positive for @xmath225 , by applying the recursion @xmath345 times we have @xmath346 let us estimate the first term in the right hand side of . since @xmath347 for every @xmath348 , from , we derive @xmath349 s_{n_0 } \exp\big(\frac{c}{1-\alpha}(n_{0}^{1-\alpha } - ( n+1)^{1-\alpha } ) \big ) & \text{if $ 0 < \alpha < 1$. } \end{cases}\end{aligned}\ ] ] to estimate the second term in the right hand side of , let us first consider the case @xmath350 , let @xmath351 such that @xmath352 . we have @xmath353 hence , combining and , for @xmath1780,1\right[$ ] we get @xmath354 we next estimate the second term in the right hand side of in the case @xmath355 . we have @xmath356 therefore , for @xmath355 , we obtain , @xmath357 which completes the proof . p. l. combettes . quasi - fejrian analysis of some optimization algorithms . in _ inherently parallel algorithms in feasibility and optimization and their applications ( haifa , 2000 ) _ , volume 8 of _ stud . comput . math . _ , pages 115152 . north - holland , amsterdam , 2001 . l. combettes and jean - christophe pesquet . proximal splitting methods in signal processing . in _ fixed - point algorithms for inverse problems in science and engineering _ , volume 49 of _ springer optim . _ , pages 185212 . springer , new york , 2011 . s. mosci , l. rosasco , m. santoro , a. verri , and s. villa . solving structured sparsity regularization with proximal methods . in _ machine learning and knowledge discovery in databases european conference , ecml pkdd 2010 _ , pages 418433 , barcelona , spain , 2010 . springer . o. shamir and t. zhang . stochastic gradient descent for non - smooth optimization : convergence results and optimal averaging schemes . in _ proceedings of the 30th international conference on machine learning _

experiments with superconducting microwave cavities have been performed in our laboratory for more than two decades . the purpose of the present article is to recapitulate some of the highlights achieved . we briefly review ( i ) results obtained with flat , cylindrical microwave resonators , so - called microwave billiards , concerning the universal fluctuation properties of the eigenvalues of classically chaotic systems with no , a threefold and a broken symmetry ; ( ii ) summarize our findings concerning the wave - dynamical chaos in three - dimensional microwave cavities ; ( iii ) present a new approach for the understanding of the phenomenon of dynamical tunneling which was developed on the basis of experiments that were performed recently with unprecedented precision , and finally , ( iv ) give an insight into an ongoing project , where we investigate universal properties of ( artificial ) graphene with superconducting microwave photonic crystals that are enclosed in a microwave resonator , i.e. , so - called dirac billiards . * quantum chaos , a field that focuses on the quantum manifestations of classical chaos , attracted a lot of attention during the last years . the signatures of classical chaos in universal properties of the corresponding quantum system are indeed of relevance in nuclear physics , atomic physics , condensed matter physics , optics and in acoustics . they were observed in the statistical properties of the eigenvalues and the wave functions , in the fluctuation properties of the scattering matrix elements of chaotic scattering processes , in the transport properties of quantum dots and also in systems , where time - reversal invariance is broken , e.g. , by a magnetic field . billiards provide a most appropriate system for the study of features from the field of quantum chaos , because the degree of chaoticity of their classical dynamics only depends on their shape . the eigenvalues and wave functions of two - dimensional quantum billiards are measured since two decades in our laboratory using flat cylindrical , superconducting and normal conducting microwave resonators , respectively . thereby large sequences of approximately 1000 eigenvalues were determined in the first experiments with a hitherto never achieved high precision yielding new insight into quantum chaos phenomena , because at that time their computation was either not possible or only feasible with a large numerical effort . we will present the results of these and subsequent experiments and also concerning our latest studies of dynamical tunneling . the latter demonstrate that even nowadays the measurements with superconducting microwave resonators may provide valuable new information that is indispensable for the development of a theoretical description of quantum phenomena like dynamical tunneling , even though eigenvalues may now be computed with high accuracy . most of the works on quantum chaos concentrate on quantum systems described by the non - relativistic schrdinger equation . only a few studies focus on the properties of the eigenvalues and the wave functions of relativistic billiards that are governed by the dirac equation of a massless spin- particle . we started lately with their experimental investigation and will present the system and recent results on the statistical properties of the associated eigenvalues in this article . *

1504.04160

versus the excitation frequency , of a microwave billiard with the shape of a quarter stadium measured at room temperature ( upper panel ) and at 2 k ( lower panel ) , where it is normal conducting and superconducting , respectively . the shape of the resonator and the positions of the antennas are illustrated in the inset . reprinted from phys . rev . lett . * 69 * , 1296 ( 1992 ) . ] this article provides a brief survey on groundbreaking experiments that were performed in the last two decades in our group with superconducting microwave billiards . they focused on the understanding of the generic properties of quantum systems like nuclei and of condensed matter structures like graphene or fullerene . in the first case , we exploited the equivalence of the scalar helmholtz equation governing flat , metallic microwave resonators , so - called microwave billiards , and the non - relativistic schrdinger equation of the corresponding quantum billiard , to address problems from the fields of quantum chaos and compound nucleus reaction theory @xcite . in the second case , the analogy between the scalar helmholtz equation describing microwave photonic crystals squeezed between two metal plates or enclosed in a microwave billiard , so - called dirac billiards , and the dirac equation was used to experimentally investigate relativistic phenomena occurring in graphene . some details on the experimental setups for the measurements of resonance spectra at superconducting conditions will be provided in . a classical billiard @xcite is a bounded domain , in which a pointlike particle moves freely and is reflected elastically on impact with the boundary . since its dynamics only depends on its shape , such systems are widely used to investigate the effects of the classical dynamics on properties of the associated quantum system @xcite . the correspondence between the latter , i.e. , a quantum billiard and a microwave billiard of the same shape , holds below a certain excitation frequency @xmath2 of the microwaves that are sent into the resonator , @xmath3 , where the electric field vector is parallel to the cylinder axis . here , @xmath4 is the velocity of light and @xmath5 is the height of the resonator . originally , this analogy was used to study universal fluctuation properties of the eigenvalues of a quantum billiard with microwave billiards at room temperature @xcite . the resolution of the resonances in the spectra , however , was very limited due to the relatively low quality factor @xmath6 for normal conducting microwave billiards . we obtained the eigenvalues of ( non ) relativistic quantum billiards by measuring resonance spectra to a very high accuracy with superconducting microwave billiards with quality factors of up to @xmath7 . this is an indispensable requirement for the determination of complete sequences of several hundreds and even thousands of eigenvalues , as illustrated in fig . [ spektren2dstadium ] . shown are transmission spectra of a microwave billiard made out of niobium metal with the shape of a quarter stadium ( see ) measured at room temperature ( upper panel ) , where the resonator is normal conducting and at @xmath8 k , where it is superconducting . its eigenfrequencies , and thus the eigenvalues of the corresponding quantum billiard , are obtained from the positions of the resonances . in the upper spectrum in fig . [ spektren2dstadium ] many resonances overlap , whereas all resonances are well resolved in the lower one , so that the determination of the complete sequence of eigenvalues is achievable only in the superconducting case . because of this capability we were able to tackle a wide range of problems that are far from feasible with normal conducting microwave billiards . since the resonance spectra are measured by emitting microwave power into the resonator via one antenna and receiving it at the same or another one , microwave billiards can also be viewed as scattering systems . we utilized this capability and studied the universal properties of quantum scattering systems with an unprecedented accuracy . due to limitation in space we focus here only on a few of our experiments . in secs . [ 2dstadium ] and [ triangular ] we present results on the spectral properties of classically chaotic systems . in we review our studies concerning the effect of symmetry breaking on the properties of the eigenvalues and the wave functions of chaotic systems . we also investigated the spectral properties of three - dimensional resonators that do not have a quantum analogue . these experiments provided additional important information to previously performed ones @xcite of , due to the superconductivity , hitherto never achieved precision . one example will be discussed in . the analysis of the spectral properties addresses the manifestation of the gross structure of the poincar surface of section ( psos ) , namely the relative sizes of the regions of regular and chaotic dynamics , in the associated quantum system . the phenomenon of quantum dynamical tunneling @xcite , that is , the classically forbidden transition between states localized in separated parts of the classical phase space has been studied theoretically intensively , but only in a few experiments @xcite , because the measurements require an exceptionally high precision . in sec . [ tunneling ] we will review the main results of our investigations of chaos - assisted tunneling @xcite and of dynamical tunneling @xcite . only lately we started experimental investigations of relativistic phenomena associated with graphene and will present some results of our most recent experiments in . as emphasized in and illustrated in fig . [ spektren2dstadium ] for the determination of complete sequences of eigenvalues of a quantum billiard from the resonance spectra of the corresponding microwave billiard , high - precision measurements are indispensable . therefore , we used microwave billiards that were superconducting at the temperature of liquid helium @xmath9 k. the first billiards were constructed in the cern workshop in geneva . they were made from niobium and quality factors @xmath10 were achieved . in the very first experiments a microwave billiard with the shape of a quarter stadium , reported on in , was inserted into one of the cryostats of the superconducting darmstadt electron linear accelerator s - dalinac @xcite and the measurements were performed at @xmath8 k. a part of a spectrum is shown in the lower panel of fig . [ spektren2dstadium ] . in the subsequent experiments the microwave billiards were cooled down to @xmath11 k with liquid helium in a separate lhe bath cryostat . the niobium cavities were made from one piece , so it was not possible , e.g. , to vary their shapes or to introduce a scatterer into them . therefore , since about the year @xmath12 the resonators were constructed from three copper parts , a bottom plate , a middle one with a hole of the shape of the billiard and a lid . recently , we have constructed them from two brass plates , where a bassin in the shape of the billiard was milled out of the bottom one . to attain superconductivity all parts of the resonator were lead plated and then squeezed on top of each other tigthly with screws along the boundary of the billiard . for the measurement of the resonance spectra microwave power was emitted into the resonator via one wire antenna and coupled out either via the same antenna or via a second one , thus yielding a reflection and a transmission spectrum , respectively . a vectorial network analyzer determined the relative phase and amplitude of the output and input signal . this provided the scattering matrix elements @xmath13 for the scattering from antenna @xmath14 to antena @xmath15 , with @xmath16 depending on the number of antennas introduced into the resonator via holes in the top plates . as already mentioned above , the height @xmath5 of the resonator determines the maximum frequency @xmath17 below which the analogy to the quantum billiard of corresponding shape holds . the billiard s size , and thus the number of attainable eigenmodes @xcite , is bounded by that of the cryostat . accordingly , in the first measurements with microwave billiards that had a height @xmath18 mm approximately @xmath19 eigenvalues could be determined below @xmath20 ghz , the most recent measurements yielded several thousand eigenvalues using resonators with a height of @xmath21 mm which corresponds to a maximum frequency @xmath22 ghz . in sections iii , vi , viiamd viii we will present results from all types of resonators . a multitude of the experiments addressed the statistical properties of the eigenvalues and the wave functions of billiards of different shapes . they focused on problems related to the field of quantum chaos , as , e.g. the validity of the `` bohigas - giannoni - schmit '' conjecture @xcite for chaotic systems . it states that the spectral fluctuation properties of a chaotic system coincide with those of random matrices from the gaussian orthogonal ensemble ( goe ) , if time - reversal ( @xmath23 ) is conserved , and with those of random matrices from the gaussian unitary ensemble ( gue ) , if it is violated . if the classical dynamics is regular they follow the poisson statistics . there also exist random - matrix theory ( rmt ) predictions for systems with mixed regular - chaotic dynamics , and for chaotic systems with a partially broken symmetry , as , e.g. , the isospin mixing observed in nuclear physics ( see ) , and with partially violated @xmath23invariance . the first experiments were performed with a niobium microwave billiard with the shape of a quarter bunimovich stadium @xcite shown in fig . [ photo2dstadium ] . the radius of the quarter circle equaled @xmath24 mm and the length of the rectangular part @xmath25 mm . k. its height is @xmath26 mm so the system is governed by the scalar helmholtz equation below the frequency @xmath27 ghz . the microwave billiard was cooled down to @xmath8 k in the superconducting darmstadt electron linear accelerator s - dalinac and quality factors of @xmath10 were achieved . ] a complete sequence of @xmath28 resonance frequencies was identified . for the study of the spectral properties the spectra were unfolded to mean resonance spacing unity with weyl s formula @xcite which provides an analytical expression for the smooth part of the integrated resonance density @xmath29 in terms of a polynomial of second order , @xmath30 , with @xmath31 the area and @xmath32 the perimeter of the billiard . figure [ nfluc2dstadium ] shows the fluctuating part of the integrated resonance density , @xmath33 which exhibits fast fluctuations and smooth oscillations . the latter are due to non - generic periodic orbits ( pos ) , so - called bouncing - ball orbits ( bbos ) , that bounce back and forth between the two parallel side walls of the rectangular part of the stadium billiard . the red ( gray ) full line shows the analytical result for their contribution , @xmath34 it was derived on the basis of a semiclassical method @xcite . while the bouncing - ball orbits are of measure zero in the classical dynamics they have a strong effect on the spectral propertes of the quantum billiard , as demonstrated in fig . [ delta32dstadium ] . shown is the dyson - mehta @xmath35 statistics @xcite , which gives the least - square deviation of the integrated density of the unfolded resonance frequencies from the best straight - line fit in the interval of length @xmath36 . since the classical dynamics of a bunimovich stadium is completely chaotic @xcite it is expected to coincide with the rmt results for the goe . the circles and the diamonds show the experimental results obtained when unfolding the resonance frequencies @xmath2 by replacing them as usually by @xmath37 , and by @xmath38 taking into account the effect of the bbos , respectively . it is clearly visible that the agreement between the experimental @xmath39 statistics and the goe is very good only in the latter case . ( black line ) and the contributions of the bouncing ball orbits ( red line ) deduced from eq . ( [ semicl ] ) . reprinted from phys . 69 * , 1296 ( 1992 ) . ] statistics of the resonance frequencies of the microwave billiard in fig . [ photo2dstadium ] before ( diamonds ) and after ( circles ) extraction of the contribution from the bouncing - ball orbits compared to the rmt result for the goe ( full line ) . reprinted from phys . lett . * 69 * , 1296 ( 1992 ) . ] as was noted above , for the measurement of the resonance spectra microwave power is emitted into the resonator via one antenna thereby exciting an electric field mode in its interior and an output signal is received at the same or another antenna . hence , the resonator is an open scattering system with the antennas acting as single scattering channels . the associated @xmath40 matrix comprises the scattering between the input and the output antennas and dissipation into the walls , which is accounted for by introducing a large number of weakly coupled fictitious scattering channels . the @xmath40-matrix formalism for compound nucleus reactions @xcite is equivalent to that for microwave resonators @xcite . we used this analogy to investigate universal properties of the associated @xmath40 matrix and more generally for quantum scattering processes with intrinsic chaotic dynamics . in regions of isolated resonances , as is the case in the measurements with the superconducting quarter - stadium billiard , the @xmath40-matrix element for the scattering from antenna @xmath14 to antenna @xmath15 is given as @xmath41 in the vicinity of the resonance @xmath42 . here , @xmath43 and @xmath44 are the partial widths related to the emitting and the receiving antennas @xmath14 and @xmath15 , respectively , and @xmath45 is the total width of the resonance , @xmath46 . it includes the partial widths of the @xmath47 antennas attached to the microwave billiard ( see fig . [ photo2dstadium ] ) and the dissipation in the walls . note , that only for the spectra that were measured at @xmath8 k , i.e. , for those of the quarter - stadium billiard @xmath48 resonance parameters could be determined from a fit of the lorentzian in eq . ( [ sab ] ) to the resonances @xcite . the reason is , that the contributions @xmath49 from the losses due to the dissipation in the walls were negligible as compared to those from the partial widths , whereas this is no longer the case in the measurements at liquid helium temperatures of @xmath11 k , see below . figure [ breiten ] shows the total widths @xmath45 and the partial widths @xmath50 associated with antenna @xmath8 versus the resonance frequencies @xmath51 . the data fluctuate strongly around a slow secular variation which is well approximated by a polynomial of @xmath52th order and was removed for the analysis of their statistical properties by scaling the data with the latter . the distribution of the partial widths is expected to coincide with that of a @xmath53 distribution with the number of degrees of freedom @xmath54 , which is a porter - thomas distribution . the sum of @xmath8 and @xmath47 partial widths should coincide with that of @xmath8 and @xmath47 degrees of freedom , respectively , if the partial widths indeed are uncorrelated . as demonstrated in fig . [ zerfall ] the agreement between the experimental and the expected results is very good @xcite . versus their frequencies . the full lines are polynomial fits to the data , that describe the secular dependence of the widths on the frequency . reprinted from phys . rev . lett . * 74 * , 62 ( 1995 ) . ] associated with antennas @xmath55 normalized by their respective mean . those of @xmath56 ( upper panel ) and of the sum of @xmath57 ( middle panel ) and of all ( lower panel ) partial widths ( histograms ) of the resonances measured with the superconducting microwave billiard in fig . [ photo2dstadium ] , respectively , are displayed . the full curves show the @xmath53 distributions of @xmath58 and @xmath47 degrees of freedom from top to bottom . adopted from ref . @xcite . with kind permission from springer science and business media . ] furthermore , the @xmath40-matrix autocorrelation function was computed using the experimental resonance spectra and the rmt predictions for those of isolated resonances @xcite were confirmed . the fourier tranforms of the @xmath40-matrix autocorrelation function revealed an algebraic decay . this actually is expected for isolated and weakly overlapping resonances @xcite . we would like to close this section with a few remarks concerning our experimental tests of analytical results within the theory of chaotic scattering which has been largely developed in the framework of the statistical theory of compound - nucleus reactions @xcite . predictions for the fluctuation properties of the associated @xmath40 matrix were obtained based on rmt and on the supersymmetry method @xcite . to be more precise , analytic expressions for @xmath40-matrix correlation functions and the distribution of the @xmath40-matrix elements have been derived which are valid from the regime of isolated resonances to that of strongly overlapping ones . for their experimental verification , scattering data where taken at room temperature using a microwave resonator with the shape of a chaotic tilted stadium billiard allowing tests also with unprecedented accuracy in the regions of isolated and weakly overlapping resonances @xcite . experiments were furthermore performed with microwave billiards where a partial @xmath23violation was induced by a magnetized ferrite @xcite . for the analysis of the latter we extended predictions derived for @xmath23-invariant systems by verbaarschot , weidenmller and zirnbauer @xcite for the @xmath40-matrix autocorrelation functions to systems with partial @xmath23violation . due to the large sets of scattering data measured as function of the excitation frequency we could also study higher order correlation functions , as e.g. the cross - section autocorrelation function to test rmt predictions on cross - section fluctuations @xcite . conservation laws , that is , the invariance of physical quantities with respect to certain transformations , play an important role in nearly all fields of physics . in 1918 , emmy noether formulated a theorem @xcite , which states that any differentiable symmetry of the action of a physical system has a corresponding conservation law . in a quantum system each conserved quantity corresponds to a quantum number . we investigated the symmetry breaking in a system that is characterized by two quantum numbers , like , e.g. , that of the isospin symmetry in @xmath59al or @xmath60p , which is broken due to the coulomb interaction , see @xcite and references cited therein . in @xmath59al a _ complete _ sequence of @xmath61 states of positive parity from the ground state up to about @xmath26 mev excitation energy was identified . among these @xmath62 had isospin quantum number @xmath63 and @xmath64 @xmath65 , so one has the situation of two symmetry classes with each characterized by a goe , since the corresponding classical dynamics is chaotic . the upper panel of fig . [ abstgekop ] shows their nearest - neighbor spacing distribution ( histogram ) . it should coincide with that of two uncoupled goes , plotted as dashed line in fig . [ abstgekop ] , if the isospin were a good quantum number . if , on the other hand , the symmetry were completely broken , that is , if there were no isospin symmetry , the distribution should coincide with that of one goe , shown as dotted line . al ( upper panel ) and of the resonance frequencies of the uncoupled stadium billiards ( lower panel ) . the dashed lines correpond to the curve for two uncoupled goes , the dotted ones to that of one goe . ] the fact , that the nearest - neighbor spacing distribution coincides with neither of the two clearly indicates that the isospin is partially broken in @xmath59al , or equivalently , that the two sets of states with isospin quantum numbers @xmath63 and @xmath65 are mixed . this mixing of the states of two chaotic systems is modelled by an ensemble of random matrices of the form @xmath66 which is a special case of the rosenzweig - porter model @xcite . here , the first part preserves the symmetry , where , in the case of @xmath59al , @xmath67 and @xmath68 are goe matrices and describe the spectral properties of the states with @xmath63 and @xmath65 , respectively . the symmetry breaking effect of the coulomb interaction , that is , the mixing of the isospin states is accounted for by the second term , with @xmath69 denoting the mean level spacing . for a non - vanishing symmetry - breaking parameter @xmath70 it yields a coupling of @xmath67 and @xmath68 . the symmetry breaking actually could be tested experimentally by mapping this rmt model onto a system consisting of two coupled bunimovich stadium billiards @xcite shown schematically in fig . [ gekop ] . a smaller microwave stadium billiard with the ratio @xmath71 between the length of the rectangle and the radius of the quarter circle was coupled to the one presented in the previous section iii and shown in fig . [ photo2dstadium ] which had a ratio @xmath72 . thus , the number of resonances identified up to @xmath73 ghz in the former , @xmath74 , is smaller than that in the latter , @xmath75 . the dynamics in the two uncoupled cavities is simulated in the rmt model [ eq . ( [ eq : mixedh ] ) ] by goe matrices @xmath67 and @xmath68 with the dimensions equal to @xmath76 and @xmath77 , respectively . for the coupling , the microwave billiards were put on top of each other and a niobium pin , @xmath8 mm in diameter , penetrated through holes in their walls into both of them . its size was controlled by the associated penetration length . this led to the formation of a tem mode between the walls and the pin . the mixing of the @xmath76 and @xmath77 modes of both microwave billiards via the tem mode was modelled by an @xmath76 dimensional vector @xmath78 and an @xmath77 dimensional vector @xmath79 with gaussian - distributed random numbers as entries , respectively , thus yielding a dyadic structure for @xmath80 in eq . ( [ eq : mixedh ] ) . the distribution of its matrix elements is given by a modified @xmath81 bessel function . for a more detailed description see refs . @xcite . altogether @xmath82 transmission spectra were measured with three antennas for @xmath52 different couplings . the effect of the coupling on the spectra is illustrated in fig . [ spektrengekop ] . the change of the coupling led to a shift of resonances , whose associated electric field strength distribution , i.e. , wave function , was non - vanishing at the position of the pin . we investigated the influence of the symmetry breaking on the eigenvalues of a chaotic system in terms of the spectral fluctuation properties of the resonance frequencies , and that on the wave functions in terms of the resonance strengths @xmath83 in eq . ( [ sab ] ) . the latter provide information on the electric field strengths , i.e. , the wave function components at the positions of the antennas . the lower panel of fig . [ abstgekop ] shows the nearest - neighbor spacing distribution of the unfolded resonance frequencies for the case of no coupling . it is well described by the rmt model [ eq . ( [ eq : mixedh ] ) ] with the coupling parameter set to zero , @xmath84 ( dashed line ) . in order to determine the symmetry - breaking parameter for the different experimental realizations of the coupling strength we , firstly , fitted as outlined in ref . @xcite the analytical result @xcite for the number variance @xmath85 @xcite of the eigenvalues of the random matrices given in eq . ( [ eq : mixedh ] ) to that of the unfolded resonance frequencies . in fig . [ sigma2gekop ] the experimental data points with error bars and the analytical results ( full lines ) are shown for three different couplings together with those for the limiting cases of no symmetry , i.e. , for @xmath86 goe ( dashed line ) , and of symmetry conservation , i.e. , @xmath8 goes ( dotted line ) . statistics of the measured resonance frequencies of the coupled stadium billiards for no coupling , a weak coupling and the strongest achieved coupling from top to bottom ( filled circles ) . the dashed and the dotted lines show the rmt results for two uncoupled goes and one goe , respectively . the full line results from a fit of the analytical result for the @xmath85 statistics of the model [ eq.([eq : mixedh ] ) ] to the experimental one . this yielded @xmath87 , @xmath88 and @xmath89 from top to bottom . reprinted from phys . lett . * 81 * , 4847 ( 1998 ) . ] secondly , in ref . @xcite , we compared the distributions of the resonance strengths @xmath83 obtained from the measured resonances with a qualitative model developed based on the eigenvectors of the random matrices in eq . ( [ eq : mixedh ] ) . figure [ strengthgekop ] shows the experimental ( histograms ) and analytical ( dashed lines ) results for one antenna combination and for four different couplings . here , we took into account the experimental threshold of detection for narrow resonances and resonances with small strengths @xcite . ) including the threshold of detection , best fitting the experimental results are shown as dashed lines . the symmetry breaking parameters deduced from the fits equal @xmath90 , @xmath91 , @xmath92 and @xmath93 from the upper left to the lower right panel . ] the symmetry - breaking parameters @xmath70 obtained from both procedures agree within the experimental errors . the symmetry breaking actually turned out to be relatively weak even in the case of the strongest coupling , but nevertheless to be of the same size as the isospin symmetry breaking in @xmath59al , which yielded for the spreading width @xmath94 with respect to the mean level spacing , @xmath95 , i.e. , approximately every fourth resonance is influenced by the coupling . another important issue besides the spectral fluctuation properties of a quantum system within the field of quantum chaos concerns the semiclassical approach . in this context , one of the most important achievements was the periodic orbit theory ( pot ) pioneered by gutzwiller @xcite , which paved the way for an understanding of the impact of the classical dynamics on the properties of the corresponding quantum system in terms of purely classical quantities . in contrast , rmt addresses exclusively system independent , i.e. , universal properties . yet , only recently , the `` bohigas - giannoni - schmit '' conjecture @xcite was proven on the basis of gutzwiller s trace formula , which provides a semiclassical approximation for the fluctuating part of the resonance density of a quantum system in terms of a sum over the pos of the corresponding classical system @xcite . we have performed experimental tests of gutzwiller s trace formula and its analogue for systems of chaotic , mixed or regular dynamics with metallic and dielectric resonators @xcite . we also tested a gutzwiller - like trace formula derived by balian and duplantier for three - dimensional electromagnetic resonators @xcite . here the main difficulty to overcome was the design of a microwave resonator that is completely chaotic and permits only a few non - generic modes . they were even present in our latest experiments with the resonator shown in fig . [ photos3d ] which had the shape of a three - dimensional stadium billiard . note , that there is no analogy between three - dimensional microwave resonators and quantum billiards . still , the former are most suitable for the study of wave - dynamical phenomena in chaotic systems . the three - dimensional desymmetrized stadium billiard consisted of two quarter cylinders with radii @xmath96 mm and @xmath97 mm that were rotated by @xmath98 with respect to each other so that the height of each cylinder coincided with the radius of the other one . the resonator was constructed from niobium and the measurements were performed at liquid helium temperature , yielding exceptionally high quality factors of up to @xmath7 . consequently , a complete sequence of @xmath99 resonance frequencies could be identified for excitation frequencies of up to @xmath100 ghz . parts of one of the altogether @xmath82 measured transmission spectra are shown in fig . [ spektrum3dstadion ] . for two different frequency ranges . due to the high quality factors the resonances were well isolated up to @xmath100 ghz yielding a complete sequence of altogether 18764 resonance frequencies . ] for the investigation of the fluctuation properties of the resonance frequencies these had to be unfolded with weyl s formula for the smooth part @xmath101 of the integrated resonance density @xmath29 of electromagnetic cavities @xcite . it is a polynomial of third order in the frequency , where , in distinction to that for the corresponding quantum billiard , the quadratic term is absent . the fluctuating part , @xmath33 , is shown in fig . [ nfluc3dstadion ] . like in the case of the two - dimensional stadium billiard in fig . [ nfluc2dstadium ] , it exhibits slow oscillations , that again are due to non - generic pos . these consist of marginally stable bouncing - ball orbits with lengths @xmath102 and @xmath103 that bounce back and forth between the top and bottom walls of the quarter cylinders . furthermore there are orbits in their common rectangular plane that are linearly stable within the plane and unstable with respect to perpendicular disturbances . an analytical expression for the contribution of the former to the resonance density was derived in @xcite , that for the entire non - generic one in @xcite on the basis of a quantum adiabatic method . the latter is plotted as red ( gray ) line in fig . [ nfluc3dstadion ] . compared to the contribution of the non - generic orbits ( red [ gray ] line ) . reprinted from phys . lett . * 89 * , 064101 ( 2002 ) . ] figure [ fft3dstad ] exhibits the experimental length spectrum , deduced from the fourier transform of the fluctuating part of the density of the resonance frequencies as black full line . the red dashed one was obtained from the sum of the fourier transforms of the analytical results for the resonance density of the non - generic pos and the unstable ones . the latter is provided by the semiclassical trace formula of balian and duplantier @xcite , @xmath104 the factor @xmath105 accounts for the polarization , that is , the vectorial character of the helmholtz equation for three - dimensional resonators . it vanishes for orbits with an odd number of reflections . note , that for separable systems , where the transversal electric and transversal magnetic modes decouple completely , @xmath106 . the quantities @xmath107 and @xmath108 denote the length of the periodic orbit @xmath109 , and its stability matrix , respectively . they are identical with those entering gutzwiller s trace formula @xcite . the diamonds on the abscissa in fig . [ fft3dstad ] mark the peaks that are exclucively assigned to pos in the common plane of the quarter cylinders . the other peaks below @xmath110 m are at the lengths of the bouncing - ball orbits . these indeed are comparatively large . in conclusion , the contributions from non - generic orbits could not be neglected and had to be extracted for the study of the spectral fluctuation properties of the unfolded resonance frequencies . ( black full line ) and for the non - generic modes @xcite plus the trace formula eq . ( [ bd ] ) of balian and duplantier ( red [ gray ] dashed line ) . both curves exhibit peaks at the lengths of the pos . the diamonds mark the lengths of the pos in the common rectangular plane of the quarter cylinders . ] these were expected to coincide with the rmt result for the goe , since the dynamics of the corresponding classical billiard was shown to be chaotic @xcite . in fig . [ delta33dstadion ] the experimental result for the @xmath39 statistics is plotted ( diamonds ) for three frequency ranges . even though the non - generic contributions were extracted by proceeding as in the case of the two - dimensional stadium billiard ( see ) , we observe deviations from the goe result ( dotted lines ) . instead , the @xmath39 statistics is very well decribed by that of random matrices of the form in eq . ( [ eq : mixedh ] ) , applicable to sytems consisting of two coupled chaotic ones ( full lines ) . actually , in the frequency range of @xmath111 ghz it is close to that of two uncoupled goes ( dashed line ) . the reason for the deviations from the goe results is not yet fully understood . numerical studies @xcite of the spectral properties of three - dimensional quantum stadium billiards demonstrated that they can not be attributed to the partial breaking of the reflection symmetry present , when the radii @xmath112 and @xmath113 of the two quarter cylinders are equal . furthermore , our interpretation of the good agreement of the spectral properties with those of the model in eq . ( [ eq : mixedh ] ) as a partial decoupling of the electric and the magnetic field modes could not yet be confirmed , neither experimentally nor numerically . however , the fact that a considerable fraction of the pos have @xmath106 provides a supportive indicator for its correctness . in the following section we will present another system whose spectral properties deviate from the rmt predictions . statistics of the resonance frequencies of the resonator shown in fig . [ photos3d ] evaluated in the three frequency intervals indicated in the insets ( diamonds ) . the dotted curve corresponds to that for one goe , the dashed one to that of two uncoupled goes , i.e. , to the case @xmath84 in eq . ( [ eq : mixedh ] ) , and the full one to the curve deduced from eq . ( [ eq : mixedh ] ) best fitting the experimental one yielding @xmath114 from top to bottom . ] as has been already stated above , according to the `` bohigas - giannoni - schmit '' conjecture @xcite the spectral properties of time - reversal ( @xmath23 ) invariant chaotic systems coincide with those of random matrices drawn from the goe . this , however , is not true for systems with an unidirectional classical dynamics ( see ) or for billiards that have a shape with a threefold symmetry and no further one like , e.g. , a mirror symmetry @xcite . in this context , we investigated the spectral properties of the microwave billiard presented in fig . [ phototriangular ] . its shape is invariant under rotations of @xmath115 . the eigenfunctions @xmath116 of the hamiltonian of such a quantum billiard can be classified according to their transformation properties with respect to a rotation @xmath117 by @xmath115 , @xmath118 . due to @xmath23invariance the eigenvalues associated with the classes @xmath119 are degenerate . accordingly , the eigenvalue spectrum is composed of singlets corresponding to the case @xmath120 and doublets of identical ones @xcite . the spectral properties of the singlets and the doublets have been predicted to coincide with those of random matrices from the goe and the gue , respectively @xcite . we determined the eigenvalues of the billiard with the shape shown in fig . [ phototriangular ] by performing measurements at superconducting conditions . still , since the degeneracy of the doublet states was only partly split due to mechanical imperfection @xcite a more sophisticated method had to be developed for their identification @xcite . it utilized the dependence of the behavior of the eigenmodes under the variation of the relative phase between two input antennas on their symmetry properties . in fig . [ absttriangular ] the nearest - neighbor spacing distributions of the singlets and doublets ( histograms ) are compared with the rmt result for the goe ( black line ) and for the gue ( red [ gray ] line ) , respectively . similarly , the @xmath39 statistics of the singlets and doublets ( diamonds ) is presented in fig . [ delta3triangular ] together with the goe ( black line ) and the gue ( red [ gray ] line ) , respectively . these figures demonstrate that the experimental results were in accordance with the theory . ( histogram ) . the rmt results for the goe and the gue are shown as black and red ( gray ) lines , respectively . ] statistics of the resonance frequencies of the singlets and the doublets obtained for the microwave billiard shown in fig . [ phototriangular ] ( diamonds ) . the rmt results for the goe and the gue are plotted as black and as red ( gray ) lines , respectively . reprinted from phys . lett . * 90 * , 014102 ( 2003 ) . ] recently , we investigated the phenomenon of dynamical tunneling in two microwave billiards with the shapes of constant - width billiards @xcite . a photograph of one is shown in fig . [ photocw ] . such billiards possess the particular property of unidirectionality , that is , when a particle is launched into the billiard with a certain rotational direction then it will continue to rotate in that direction forever . accordingly , when constructing a poincar surface of section ( psos ) by launching many particles into the billiard in clockwise direction , then only its upper half will be filled , as illustrated in fig . [ psscw ] . here , @xmath121 where @xmath122 is the angle between the particle trajectory and the normal to the boundary at the point of impact , and @xmath123 is the location of the latter in terms of the arclength along the boundary . the psos consists of a large chaotic sea and two regions of regular kam tori . firstly , there is a barrier region around the diameter orbit at @xmath124 , shown in blue ( light gray ) . secondly , there is a region of whispering gallery modes around @xmath125 . examples of both types of orbits are plotted in the inset in the corresponding colors . furthermore , we found very tiny regular islands inside the chaotic sea for one of the two investigated billiards @xcite . the fact , that only one half of the psos is filled is a clear indication that classically the transition from clockwise to anticlockwise motion , i.e. , the transition from positive to negative @xmath109 values is strightly forbidden . in contrast , in the corresponding quantum billiard this is possible via tunneling through the barrier region around @xmath124 . this manifests itself in a partial lifting of the degeneracies of pairs of eigenenergies associated with the clockwise and anticlockwise quasimodes @xcite . accordingly , the size of the tunneling effect is reflected in that of the splittings of the eigenenergies . plotted in red ( gray ) and one of orbits around the diameter orbit plotted in blue ( light gray ) . the inset shows examples for the associated orbits . reprinted from phys . e * 90 * , 022903 ( 2014 ) . ] the objective of the experiments with the two microwave billiards with the shapes of constant - width billiards , referred to as billiards b1 amd b2 in the sequel , was to investigate this phenomenon with unprecedented accuracy using superconducting microwave billiards . both billiards had the same diameter @xmath126 cm and their shapes were barely distinguishable , whereas the widths of the barrier and the whispering gallery orbit regions were considerably larger in the psos of b1 than in that of b2 . more details concerning the measurements are provided in ref . @xcite . to demonstrate the high precision of the measurements and also its indispensability for the identification of the doublets we show in fig . [ spectrumcw ] two parts of the transmission spectrum of billiard b1 . the resonances visible in the upper panel seem to be sharp and well isolated . however , a zoom into the region of the triplett observed at the left side of the spectrum , reveals that they all correspond to doublets of nearly degenerate resonance frequencies . in total , @xmath127 doublets and @xmath128 singlets were detected in billiards b1 and b2 , respectively . the latter correspond to diameter orbits , that are analogues of those of the circle billiard of the same diameter @xmath129 , the eigenvalues @xmath130 are the zeroes of the bessel functions , @xmath131 . we actually identified _ all _ singlets in the list of resonance frequencies of b1 and b2 by comparison with these zero - momentum modes . in order to check the completeness of the sequence of doublets and to study the spectral properties of the constant - width billiards , we separated the resonance frequencies into singlets , and the smaller and larger ones of the doublets , i.e. , the left , @xmath132 , and the right , @xmath133 , doublet partners in the resonance spectra ( see lower panel of fig . [ spectrumcw ] ) , respectively . this was possible , because the correlations between the doublet partners and those between the singlets and the doublets were much weaker than those amongst the members of each of the sequences . consequently , we could simply disregard the singlets and investigate the fluctuation properties of the left and the right doublet partners separately . for this , we first unfolded them with weyl s formula @xcite ( see ) . the comparison of the smooth part of the integrated density of the resonance frequencies with the latter and also the shape of its fluctuating part confirmed that we identified the _ complete _ sequence of doublets . due to the unidirectionality of the classical dynamics , the spectral properties were predicted to coincide with the rmt result for the gue , even though the system is @xmath23invariant @xcite . we did nt find any agreement with the gue and also not with the goe results , but instead found very good agreement with those of a random - matrix model for systems with violated @xmath23invariance and mixed dynamics . within this model the form of the random matrices is given by eq . ( [ eq : mixedh ] ) , with @xmath67 replaced by a random matrix from the gue , and @xmath68 by a diagonal matrix with random poissonian numbers as entries . the elements of the matrix @xmath134 , which describes the mixing of the chaotic and the regular states modeled by @xmath67 and @xmath68 , respectively , were chosen as gaussian - distributed random numbers . thus , even though the classical dynamics of the investigated billiards was not fully chaotic , the predictions of ref . @xcite could be confirmed . , for b1 ( upper panel ) and b2 ( lower panel ) versus their average resonance frequencies . they are by a factor of @xmath135 smaller than their average resonance frequency . this underlines again the necessity of high - precision experiments to identify a complete sequence of doublets . ] the splittings @xmath136 of the resonance frequencies of the doublet partners are plotted versus their average resonance frequency for b1 in the upper panel and for b2 in the lower one of fig . [ splitcw ] , respectively . the typical splitting size is smaller for b1 than for b2 . this is in agreement with the expectation , that the tunneling rate through the barrier region in the psos , which is broader for b1 than for b2 , decreases with increasing width of that region . furthermore , the average splitting size diminishes algebraically with increasing frequency , not exponentially as might be expected semiclassically . this can be attributed to the fact , that the psos consists of a chaotic region and several regular regions , where the splitting sizes diminish exponentially with different rates . we considered all splittings , i.e. , the sum over the exponentials which yields an algebraical decay @xcite . and b2 [ lower panel ] with that obtained from the random - matrix model of eq . ( [ eq:1 ] ) shown as red ( gray ) histograms and with the analytical result of eq . ( [ spl_distr ] ) plotted as green ( light gray ) line . here , the tunneling parameters deduced from the fit of the rmt model to the experimental distributions are @xmath137 for b1 and @xmath138 for b2 . reprinted from phys . e * 90 * , 022903 ( 2014).,title="fig : " ] and b2 [ lower panel ] with that obtained from the random - matrix model of eq . ( [ eq:1 ] ) shown as red ( gray ) histograms and with the analytical result of eq . ( [ spl_distr ] ) plotted as green ( light gray ) line . here , the tunneling parameters deduced from the fit of the rmt model to the experimental distributions are @xmath137 for b1 and @xmath138 for b2 . reprinted from phys . e * 90 * , 022903 ( 2014).,title="fig : " ] for a better understanding of the tunneling we investigated also the distribution of the splittings . the black histograms in the upper and lower panel in fig . [ splitdistrcw ] correspond to the experimental results for billiards b1 and b2 , respectively . in order to extract system specific quantities , like the area and the perimeter of the billiards , we used the splittings @xmath139 of the unfolded left and right doublet partners . we demonstrated in ref . @xcite that the shapes of the splitting distributions are not affected by states that are associated with the regular parts of the psos , that is , the chaotic component dominates . they , indeed , are well described by a random - matrix model of the form @xmath140 which is applicable to systems with broken @xmath23invariance exhibiting tunneling between two regions with chaotic dynamics . here , @xmath141 and @xmath142 are @xmath143 random matrices from the gue , that are associated with the parts of the hilbert space corresponding to clockwise and anticlockwise motion , respectively , and @xmath144 is a random poissonian number which represents a singlet state . the tunneling from the upper to the lower part of the psos via the barrier region , that is , the coupling of @xmath141 and @xmath142 via the singlet state , is caused by the off - diagonal matrix @xmath134 , which has gaussian - distributed random numbers as entries . the strength of the coupling is determined by the tunneling parameter @xmath145 . it was obtained by fitting the splitting distribution of the rmt model of eq . ( [ eq:1 ] ) to the experimental ones . on the basis of the latter we furthermore derived an analytical expression for the splitting distribution , @xmath146 , \label{spl_distr}\ ] ] which we believe is generally applicable to systems with violated @xmath23invariance , where dynamical tunneling takes place between two regions with a chaotic dynamics . the green ( light gray ) lines in fig . [ splitdistrcw ] correspond to the analyticial results computed for the parameter values of @xmath145 given in the captions . its value is smaller for billiard b1 than for b2 , as expected , because the psos of the former comprises a broader barrier region . the agreement between the experimental , the rmt and the analytical results for the splitting distributions is very good . remarkably , the rmt model and the analytical formula for the splitting distribution depend only on one parameter which provides information on the tunneling rate . we would like to close this section by mentioning another experiment on dynamical tunneling , which we performed @xmath147 years ago with superconducting microwave billiards with the shapes of annular billiards @xcite . there , the dynamical tunneling between quasimodes corresponding to clockwise and anticlockwise moving whispering gallery orbits , respectively , was investigated . as recognizable in the upper panels of figs . [ splitannular ] and [ splitbeachannular ] , such a billiard corresponds to a circular one of radius @xmath117 that contains a circular disk of radius @xmath148 , where the circle centers are displaced by a value @xmath149 with respect to each other . the classical dynamics is characterized by the value @xmath150 which was chosen equal to @xmath151 . then the regions of neutrally stable clockwise and anticlockwise moving whispering gallery orbits are located in the psos at angular momentum values @xmath152 and @xmath153 in units of @xmath117 , respectively , where @xmath154 corresponds to @xmath109 in fig . [ psscw ] . for @xmath155 the psos consists of a large chaotic sea with a regular island at its center . the objective of the experiments was the investigation of chaos - assisted tunneling @xcite between quasimodes localized in the regions @xmath156 via the chaotic sea , that is , the enhancement of the tunneling through a coupling of these modes to chaotic ones . the coupling is largest for modes located close to the borders @xmath157 that separate the three regions . as in the case of the constant - width billiard the tunneling manifests itself in the splitting of the degenerate quasimodes corresponding to clockwise and anticlockwise moving whispering gallery orbits into doublets . thus , if chaos - assisted tunneling takes place , its size should depend on their location in the psos , i.e. , on their angular momentum values . according to the einstein - brillouin - keller quantization the quantum analogue of the angular momentum @xmath154 is given as @xmath158 for a pair of whispering gallery modes . here , @xmath159 denotes their mean wave number and their angular and radial quantum numbers are @xmath160 and @xmath161 , respectively . they actually coincide to a good approximation with those of the integrable annular billiard with @xmath162 . to determine them the electric field distributions , i.e. the wave functions of the annular billiard , were measured in a normal conducting microwave billiard made from copper using the perturbation method @xcite . the mean wave numbers @xmath159 of the whispering gallery modes and the associated splittings were obtained from resonance spectra that were taken with superconducting billiards constructed from niobium . figure [ splitannular ] shows a small part of a transmission spectrum which contains two singlets and one doublet , which is assigned to a pair of whispering gallery modes with the angular quantum number @xmath163 given by half the number of maxima of the field intensity shown in the middle panel above the spectra . furthermore , it has radial quantum numer @xmath164 , since the latter exhibits only one maximum in the radial direction . the singlets correspond to modes that are localized in the region with @xmath165 . to @xmath166 ghz obtained at room temperature ( upper panel ) and at 4.2 k ( lower panel ) . the combined measurement of field distributions with the normal conducting microwave billiard and of resonance spectra with the superconducting one allows the identification of the quasidoublets and the assigment of quantum numbers @xmath167 and of a parity with respect to the symmetry axis like for the three modes associated with the resonances , shown in the top panels . reprinted from phys . * 84 * , 867 ( 2000 ) . ] for the study of the dependence of the tunneling rates , that is , of the sizes of the splittings on the location of the associated doublet modes in the psos , i.e. , on their angular momentum values , we considered only modes with radial quantum number @xmath164 . figure [ splitbeachannular ] shows the resulting splittings versus the angular momentum @xmath168 , with @xmath169 denoting the wave number corresponding to the resonance frequency @xmath51 of the mode . the curve exhibits a clear peak in the splittings , i.e. , an enhancement of the tunneling in the beach region around the border @xmath170 between the regular coast and the chaotic sea . to the best of our knowledge , this corroborated for the first time the occurrence of chaos - assisted tunneling . ) towards very small ones in the regular coastal region ( @xmath152 ) is visible . the appearance of a pronounced maximum in the direct vicinity of the beach ( @xmath171 ) is a clear indicator for the occurrence of chaos - assisted tunneling . reprinted from phys . 84 * , 867 ( 2000 ) . ] only recently we have started a series of experiments with microwave photonic crystals @xcite that model properties of graphene @xcite , a monoatomic layer of carbon atoms arranged on a honeycomb lattice . due to its particular electronic properties , graphene attracted a lot of attention over the last years . its conductance and its valence band form conically shaped valleys that touch each other at the six corners of the hexagonal brillouin zone . in their vicinity the electron energy depends linearly on the quasimomentum , implying an energy independent velocity . thus , the electrons and holes behave like massless relativistic particles governed by the dirac equation for spin-@xmath1 particles . accordingly , the touching points are referred to as dirac points @xcite . this extraordinary feature of the band structure of graphene results in a number of electronic properties which have an analogue in relativistic quantum mechanics . it actually stems from the symmetry properties of its honeycomb structure which is formed by two interpenetrating triangular lattices with threefold symmetry . this led to the manufacturing of artificial graphene using two - dimensional electron gases , molecular assemblies , ultracold atoms @xcite or photonic crystals @xcite . we exploited this fact to simulate the spectral properties of graphene with microwave resonators that were constructed by squeezing a photonic crystal , which is composed of several hundreds of metallic cylinders arranged on a triangular lattice , between two metal plates . the shape of its band structure , i.e. , of its frequencies of wave propagation as function of the two components of the quasimomentum vector , is equivalent to that of the conductance and the valence band in graphene , e.g. , for its 1st and its 2nd band . accordingly , around the dirac points the propagation of the microwaves in the photonic crystal can be effectively described by the dirac equation for massless spin-@xmath1 particles . such a region is identified in the resonance spectrum as one of an exceptionally low resonance density , which increases linearly with the distance from the frequency of the dirac point , the so - called dirac frequency @xcite . in order to investigate this behavior in detail , we performed high - precision experiments with superconducting dirac billiards with the shapes of a rectangle and the african continent @xcite , respectively . photographs are shown in the left parts of figs . [ photosdiracrect ] and [ photosdiracafr ] . we use them to simulate properties of graphene billiards or graphene flakes . both billiards consist of a bassin with the shape of the billiard and a lid , that were made from brass plates and then lead plated to attain superconductivity at liquid helium temperature . for the construction of the photonic crystals located inside the bassins @xmath172 cylinders were milled out of a brass plate . the height of the resonators was @xmath21 mm so for @xmath173 ghz the associated helmholtz equation is mathematically identical to the schrdinger equation of the quantum billiard containing circular scatterers at the positions of the cylinders . the great advantage of dirac billiards as compared to graphene flakes is that we can control the boundary conditions . here , those for the wave functions at the voids that are located at the edges of the photonic crystal are of relevance . they can be manipulated by varying the relative position between these voids and the walls of the microwave billiard , where the electric field strength vanishes , i.e. , obeys the dirichlet boundary condition . metal cylinders arranged on a triangular grid . the top plate is shifted with respect to the bottom one . reprinted from phys . b * 88 * , 104101 ( 2013 ) . right part : schematic view of the triangular lattice of the rectangular dirac billiard . the red ( gray ) and blue ( dark gray ) dots mark the voids between the cylinders . they form a honeycomb lattice.,title="fig : " ] metal cylinders arranged on a triangular grid . the top plate is shifted with respect to the bottom one . reprinted from phys . b * 88 * , 104101 ( 2013 ) . right part : schematic view of the triangular lattice of the rectangular dirac billiard . the red ( gray ) and blue ( dark gray ) dots mark the voids between the cylinders . they form a honeycomb lattice.,title="fig : " ] but for an africa billiard . the top plate is removed . ] the cylinders were arranged on a triangular grid , as indicated in the schematic views in figs . [ photosdiracrect ] and [ photosdiracafr ] , so the voids at the centers of the cells formed by , respectively , three of them yield a hexagonal configuration similar to the honeycomb structure in graphene . each of the cells may host a quasibound state and thus can be considered as one of the atoms forming the two independent triangular sublattices in graphene . accordingly , both graphene and microwave photonic crystals can be treated in a similar way within the tight - binding model ( tbm ) . we determined altogether @xmath174 and @xmath175 resonance frequencies in the first two bands of the rectangular and the africa dirac billiard , respectively . figure [ rhosdirac ] shows their resonance densities ( black curves ) . they clearly differ from those expected for the corresponding _ empty _ microwave billiards , that , according to weyl s law , would increase linearly with the resonance frequency . both exhibit a minimum at the frequency of the dirac point , which separates the two bands . it is bordered by two sharp peaks corresponding to van hove singularities , that generally occur in two - dimensional crystals with a periodic structure @xcite . there , the resonance density diverges logarithmically for infintely extended crystals , whereas in the case of the dirac billiards the peak heights increase logarithmically with the size of the crystal , i.e. with the number of voids formed by the cylinders @xcite . the red ( gray ) dashed lines in both panels of fig . [ rhosdirac ] correspond to fits of the tbm @xcite for finite - size photonic crystals to the experimental ones . it incorporates nearest - neighbor , as well as second- and third - nearest - neighbor couplings and in addition the corresponding overlaps of the wave functions centered at the different voids . furthermore , the resonance density of the africa dirac billiard exhibits a slight bump to the right of the dirac frequency , which is attributed to the zigzag edge parts of the void structure formed by the associated photonic crystal @xcite . the voids of the rectangular billiard also have a zigzag structure along its longer sides , see fig . [ photosdiracrect ] , however , the photonic crystal can be reproduced periodically via reflections at all its sides so no edge states are generated @xcite . and of the african continent [ lower panel ] . a slight bump is visible in the latter to the right of the dirac frequency . it is attributed to edge states occurring along the zigzag edge parts of the void structure , see also the right hand side of fig . [ photosdiracafr].,title="fig : " ] and of the african continent [ lower panel ] . a slight bump is visible in the latter to the right of the dirac frequency . it is attributed to edge states occurring along the zigzag edge parts of the void structure , see also the right hand side of fig . [ photosdiracafr].,title="fig : " ] we demonstrated recently in ref . @xcite that the van hove singularities divide the two bands framing the dirac point into two frequency ranges below and above them , where the system is described by the nonrelativistic schrdinger equation of the _ empty _ quantum billiard , and one between them , where the underlying wave equation coincides with the dirac equation of the graphene billiard or flake of corresponding shape . actually , the tbm does not only provide the eigenvalues , but also the wave functions of a bounded photonic crystal . in the regions close to the lower and the upper band edges , they are _ identical _ with those of the corresponding quantum billiard . moreover , we showed that this also holds for the fluctuating part of the integrated resonance density and for the spectral fluctuation properties . furthermore , the computation of the momentum distributions using the wavefunctions and also the direct comparison of the band structure of the photonic crystals with the resonance frequencies plotted versus the eigenvalues of the quantum billiard yielded , that in the vicinity of the lower and the upper band edge the latter maybe identified with the quasimomenta . accordingly , we call the regions below and above the van hove singularities the nonrelativistic schrdinger region . in the region between the van hove singularities , i.e. , around the dirac point , the electromagnetic waves inside the billiards are governed by the dirac equation . accordingly , there we compared the resonance frequencies and their fluctuation properties with those of the eigenvalues of the corresponding graphene billiard @xcite . the latter were computed by solving the dirac equation with appropriate boundary conditions depending on the edge structure . we arrived at the conclusion , that they and the resonance frequencies interdepend linearly and furthermore , that they coincide with the quasimomenta in the vicinity of the dirac point . thus , at the van hove singularities a transition takes place between a nonrelativistic schrdinger region and the relativistic dirac region , where the dirac billiard exhibits the same features as the corresponding _ empty _ quantum billiard and the graphene billiard , respectively . in ref . @xcite we provided evidence that it can be interpreted as a lifshitz and an excited state quantum phase transition . presently , we perform more investigations concerning the spectral properties of the dirac billiards at the van hove singularities in order to corroborate this assumption . the spectral properties are similar in the different regions and agree with those of the corresponding _ empty _ quantum billiard . the classical dynamics of the _ empty _ rectangular and africa billiards is regular , respectively , chaotic @xcite . the nearest - neighbor spacing distribution and the @xmath39 statistics of the rectangular dirac billiard shown in the upper panels of fig . [ abstdirac ] ( histogram ) and fig . [ delta3dirac ] ( diamonds ) , respectively , are well described by those of poissonian random numbers ( red [ gray ] lines ) whereas those of the africa one plotted in the lower panels coincide with the goe results ( black line ) . line ) , the latter with the rmt result for the goe ( black line ) . the upper figure has been reprinted from phys . b * 91 * , 035411 ( 2015 ) . ] statistics of the experimentally determined resonance frequencies of the dirac billiards ( diamonds ) with the shapes of a rectangle ( upper panel ) and of the african continent ( lower panel ) . the former agrees well with that of poissononian random numbers , the latter with the rmt result for the goe . the upper figure has been reprinted phys . b * 91 * , 035411 ( 2015 ) . ] the objective of this article was to present some of the highlights of our experimental investigations concerning the quantum manifestations of classical chaos . the focus has been set on spectral fluctuation properties , on symmetry breaking showing up in particular properties of the eigenvalues and the wave functions , on the phenomenon of dynamical tunneling , and , moreover , the universal properties of graphene . due to the limited space , we refered to only a few examples of the numerous experiments performed with superconducting microwave cavities in the last two decades . accordingly , we selected some that provided groundbreaking new insight into physical phenomena . future projects concern the realization of scattering systems by using microwave billiards with one or several openings in the side walls @xcite . the objective is the measurement of the temporal decay of microwave power through these openings and a comparison with rmt and other predictions . such experiments yield information on the universal decay behavior of open systems like excited nuclei . in order to ensure that the system decays exclusively via the openings , the dissipation into the walls of the resonator needs to be eliminated . this is only possible by performing the measurements of the @xmath40-matrix elements at superconducting conditions . there , however , a calibrated measurement , an essential requirement for the analysis of properties of the @xmath40 matrix , is not yet feasible . such a method was developed only recently @xcite . currently we implement and adjust it to our superconducting measurements of the @xmath40 matrix with open microwave resonators with the shapes of regular and chaotic billiards . the aim is to obtain information on the differences in their decay behavior . furthermore , we construct superconducting quantum graphs @xcite , a network - shaped structure of vertices bonded by rectangular waveguides . the objective of these experiments is the experimental test of rmt predictions concerning the correlation functions of the @xmath40 matrix of open quantum graphs @xcite in high - precision experiments . finally , a further current project concerns the experimental verification of the so - called atiyah - singer index theorem @xcite , which relates the number of zero modes of the dirac operator to the topology of the associated graphene sheet . the resonator consists of a network of rectangular waveguides that connect circular cavities . it was milled out of a brass sphere and has the form of a fullerene c@xmath176 molecule @xcite . summarizing , we have with our macroscopic , superconducting microwave resonators systems available that might be tailored to model the universal properties of a large range of microscopic quantum systems in high precision experiments . this work was supported by the deutsche forschungsgemeinschaft ( dfg ) within the collaborative research center 634 . we thank all the co - authors of our publications quoted in the list of references for a fruitful collaboration and for their many contributions to the scientific output of the experiments presented in this article . from the beginning the late oriol bohigas and martin gutzwiller have accompanied our work on superconducting billiards with great interest and support . we thus gratefully dedicate this review to their memory .

we present the results obtained from an all - sky search for gravitational - wave ( gw ) bursts in the 642000hz frequency range in data collected by the ligo detectors during the first year ( november 2005 november 2006 ) of their fifth science run.the total analyzed livetime was 268.6 days . multiple hierarchical data analysis methods were invoked in this search . the overall sensitivity expressed in terms of the root - sum - square ( rss ) strain amplitude for gravitational - wave bursts with various morphologies was in the range of to a few . no gw signals were observed and a frequentist upper limit of 3.75 events per year on the rate of strong gw bursts was placed at the 90% confidence level . as in our previous searches , we also combined this rate limit with the detection efficiency for selected waveform morphologies to obtain event rate versus strength exclusion curves . in sensitivity , these exclusion curves are the most stringent to date .

0905.0020

after many years of preparation , interferometric gravitational wave ( gw ) detectors have now begun an era of long - duration observing . the three detectors of the laser interferometer gravitational - wave observatory ( ligo ) @xcite reached their design sensitivity levels in 2005 and began a `` science run '' that collected data through late 2007 . this run is called `` s5 '' since it followed a sequence of four shorter science runs that began in 2002 . the german / british geo600 detector @xcite joined the s5 run in january 2006 , and the italian / french virgo detector @xcite began its first science run ( denoted vsr1 ) in may 2007 , overlapping the last @xmath3 months of the s5 run . the data collected by these detectors provide the best opportunity yet to identify a gw signal though detection is still far from certain and is a baseline for future coordinated data collection with upgraded detectors . gravitational waves in the frequency band of ligo and the other ground - based detectors may be produced by a variety of astrophysical processes @xcite . see for example @xcite for inspiralling compact binaries , @xcite for spinning neutron stars , @xcite for binary mergers , and @xcite for core - collapse supernovae . the gw waveform emitted by a compact binary system during the inspiral phase can be calculated accurately in many cases , allowing searches with optimal matched filtering ; see , for example , @xcite . the waveform from the subsequent merger of two black holes is being modeled with ever - increasing success using numerical relativity calculations , but is highly dependent on physical parameters and the properties of strong - field gravity . the uncertainties for the waveforms of other transient sources are even larger . it is thus desirable to explore more generic search algorithms capable of detecting a wide range of short - duration gw signals from poorly - modeled sources such as stellar core collapse to a neutron star or black hole or unanticipated sources . as gw detectors extend the sensitivity frontier , it is important to not rely too heavily on assumptions about source astrophysics or about the true nature of strong - field gravity , and to search as broadly as possible . in this paper , we report on a search for gw `` burst '' signals in the ligo data that were collected during the first 12 months of the s5 science run . a search for gw bursts in the remainder of the s5 data set , along with the virgo vsr1 data , will be published jointly by the lsc and virgo collaborations at a later date . the gw burst signals targeted are assumed to have signal power within ligo s frequency band and durations shorter than @xmath41s , but are otherwise arbitrary . this analysis , like most of our previously published searches for gw bursts , focuses on low frequencies in this case 64hz to 2000hz where the detectors are the most sensitive . a dedicated search for bursts above 2000hz is presented in a companion paper @xcite . interferometric gw detectors collect stable , high - sensitivity ( `` science mode '' ) data typically for several hours at a time , with interruptions due to adverse environmental conditions , maintenance , diagnostics , and the need to occasionally regain the `` locked '' state of the servo controls . in this analysis we searched the data at all times when two or more ligo detectors were operating , a departure from the all - sky gw burst searches from earlier science runs @xcite , which required coincidence among three ( or more ) detectors . in this paper , the term `` network '' is used to describe a set of detectors operating in science mode at a given time . a network may include any combination of the hanford 4 km ( h1 ) and 2 km ( h2 ) detectors , the livingston 4 km ( l1 ) detector and geo600 . because the geo600 detector was significantly less sensitive than ligo during the s5 run ( a factor of 3 at 1000hz , and almost two orders of magnitude at 100hz ) , we do not use its data in the initial search but reserve it for evaluating any event candidates found in the ligo data . this paper presents results from three different `` analysis pipelines '' , each representing a complete search . while the pipelines analyzed the data independently , they began with a common selection of good - quality data and applied a common set of vetoes to reject identifiable artifacts . each pipeline was tuned to maximize the sensitivity to simulated gw signals while maintaining a fixed , low false alarm rate . the tuning of the pipelines , the choice of good data and the decision on the veto procedure were made before looking at potential candidates . no gw signal candidates were identified by any of the analysis pipelines with the chosen thresholds . in order to interpret this non - detection , we evaluate the sensitivity of each pipeline for simulated signals of various morphologies , randomly distributed over the sky and over time . as expected , there are some sensitivity differences among the pipelines , although the sensitivities rarely differ by more than a factor of 2 ( see section [ sec : simulations ] ) and no single pipeline performs best for all of the simulated signals considered . we combine the results of the pipelines to calculate upper limits on the rate of gw bursts as a function of signal morphology and strength . the rest of the paper is organized as follows : after specifying the periods of data , forming the first year of the s5 science run in sec . [ sec : ligos5run ] , sec . [ sec : s5detectors ] describes the state of the detectors during that period . section iv summarizes the elements of this gw burst search which are common to all of the analysis pipelines . the analysis pipelines themselves are detailed in sec . [ sec : algorithms ] and appendices c , d and e. section [ sec : tuning ] describes how each pipeline is tuned , while sec . [ sec : simulations ] presents the sensitivity curves for simulated signals and sec . [ sec : systematics ] describes the systematic errors in these sensitivity curves . the results of the search are given in sec . [ sec : results ] , and some discussion including estimates of the astrophysical reach for burst candidates in sec . [ sec : summaries ] . the search described in this paper uses data from approximately the first calendar year of s5 , specifically from november 4 , 2005 at 16:00 utc through november 14 , 2006 at 18:00 utc . + figure [ figure : venn ] shows the amount of science - mode data collected ( `` livetime '' ) for each mutually - exclusive network of detectors along with percentages of the experiment calendar duration ( duty cycle ) . the top venn diagram represents the data with basic data quality and veto conditions ( see sec . iv and appendices [ sec : dataquality ] and [ sec : vetoes ] ) , including @xmath5 days of data during which two or more ligo detectors were in science mode ; this is the sample which is searched for gw burst signals . an explicit list of the analyzed intervals after category 2 dqfs is available at @xcite . the bottom venn diagram shows the livetimes after the application of additional data quality cuts and vetoes that provide somewhat cleaner data for establishing upper limits on gw burst event rates . in practice , only the h1h2l1 and h1h2 ( not l1 ) networks encompassing most of the livetime , 224 days are used to set upper limits . the high sensitivity ( see fig . [ fg : ligo - noise ] ) and duty cycles ( 78.0% for h1 , 78.5% for h2 , and 66.9% for l1 ) achieved during the s5 run were the result of a number of improvements made prior to the run @xcite . the major changes were the successful operation at livingston of a hydraulic external pre - isolator ( hepi ) to suppress seismic disturbances , and the implementation at both sites of a thermal compensation system ( tcs ) to reduce thermal lensing effects in the interferometer arm cavities due to optical absorption in mirror coatings and substrates . the hepi system provides a reduction of the seismic noise by an order of magnitude in the band 0.22.0hz , and thus significantly improves the duty cycle of the l1 detector . another significant improvement was the extension of the wave - front sensing ( wfs ) subsystem to control all alignment degrees of freedom of the core interferometer optics , leading to significantly reduced alignment fluctations . several improvements were made to the length sensing and control subsystem , enabling the photodetectors to take more power without saturation and thus allowing the laser power to be increased . a new method to calibrate the detectors was introduced , based on direct actuation of the test masses via radiation pressure from an auxiliary laser beam . unlike the traditional coil - drive calibration method @xcite , which requires rather large test mass displacements , the new technique allows calibration of the detectors at a level closer to the anticipated signal strength . other improvements included modifications to acoustic and seismic isolation of optical tables with detection photodiodes , changes to the safety shutters to protect photodiodes from damage when interferometers fall out of lock , and improved detection of impending saturation of photodiodes to prevent lock losses . finally , a number of physical effects which led to spurious transients and spectral lines in the data during previous science runs have been diagnosed and mitigated . the geo600 detector , located near hannover , germany , was also operational during the s5 run , though with a lower sensitivity than the ligo detectors . the geo600 data were not used in the current study as the modest gains in the sensitivity to gw signals would not have offset the increased complexity of the analysis . the geo600 data were held in reserve , and could have been used to follow up on detection candidates from the ligo - only analysis . geo600 began its participation in s5 on january 21 , 2006 , operating in a night - and - weekend mode . in this mode , science data were acquired during nights and weekends while commissioning work was performed during the day time . the commissioning work focused mainly on gaining a better understanding of the detector and improving data quality . it was performed in a manner that avoided disrupting science periods and allowed for well - calibrated data to be acquired . between may 1 and october 6 , 2006 , geo600 operated in so - called 24/7-mode , during which the detector s duty cycle in science - mode operation was maximized and only very short maintenance periods took place . overall in 24/7-mode an instrumental duty cycle of about 95% and a science - mode duty cycle of more than 90% were achieved . geo600 returned to night - and - weekend mode on october 16 , 2006 , and work began on further improving the reliability of the instrumentation and reducing the glitch rate . the detector was operated in night - and - weekend mode until the end of s5 in october 2007 . overall , geo600 collected about 415 days of well - calibrated and characterized science data in the period between january 2006 and october 2007 . in this search for gw bursts , three independent end - to - end analysis pipelines have been used to analyze the data . these pipelines were developed and implemented separately , building upon many of the techniques that were used in previous searches for bursts in the s1 , s2 , s3 and s4 runs of ligo and geo600 @xcite , and prove to have comparable sensitivities ( within a factor of @xmath4@xmath6 ; see sec . [ sec : simulations ] ) . one of these pipelines is fully coherent in the sense of combining data ( amplitude and phase ) from all detectors and accounting appropriately for time delays and antenna responses for a hypothetical gravitational - wave burst impinging upon the network . this provides a powerful test to distinguish gw signals from noise fluctuations . here we give an overview of the basic building blocks common to all of the pipelines . the detailed operation of each pipeline will be described later . gravitational - wave burst searches are occasionally affected by instrumental or data acquisition problems as well as periods of degraded sensitivity or nonstationary noise due to bad weather or other environmental conditions . these may produce transient signals in the data and/or may complicate the evaluation of the significance of other candidate events . conditions which may adversely affect the quality of the data are catalogued during and after the run by defining `` data quality flags '' ( dqfs ) for lists of time intervals . dqfs are categorized according to their seriousness ; some are used immediately to select the data to be processed by the analysis pipelines ( a subset of the nominal science - mode data ) , while others are applied during post - processing . these categories are described in more detail in appendix a. in all cases the dqfs were defined and categorized before analyzing unshifted data to identify event candidates . data that satisfies the initial selection criteria are passed to algorithms that perform the signal - processing part of the search , described in the following section and in three appendices . these algorithms decompose the data stream into a time - frequency representation and look for statistically significant transients , or `` triggers '' . triggers are accepted over a frequency band that spans from 64hz to 2000hz . the lower frequency cut - off is imposed by seismic noise which sharply reduces sensitivity at low frequencies , while the upper cut - off corresponds roughly to the frequency at which the sensitivity degrades to the level found at the low frequency cut - off . ( a dedicated search for bursts with frequency content above 2000hz is presented in a companion paper @xcite . ) after gravitational - wave triggers have been identified by an analysis pipeline , they are checked against additional dqfs and `` veto '' conditions to see if they occurred within a time interval which should be excluded from the search . the dqfs applied at this stage consist of many short intervals which would have fragmented the data set if applied in the initial data selection stage . event - by - event veto conditions are based on a statistical correlation between the rate of transients in the gw channel and noise transients , or `` glitches '' , in environmental and interferometric auxiliary channels . the performance of vetoes ( as well as dqfs ) are evaluated by the extent to which they remove the gw channel transients of each interferometer , as identified by the kleinewelle ( kw ) @xcite algorithm . kw looks for excess signal energy by decomposing a timeseries into the haar wavelet domain . for each transient , kw calculates a significance defined as the negative of the natural logarithm of the probability , in gaussian noise , of observing an event as energetic or more than the one in consideration . the veto conditions , like the dqfs , were completely defined before unshifted data was analyzed to identify gravitational - wave event candidates . a detailed description of the implementation of the vetoes is given in appendix b. in order to estimate the false trigger rate from detector noise fluctuations and artifacts , data from the various detectors are artificially shifted in time so as to remove any coincident signals . these time - shifts have strides much longer than the intersite time - of - flight for a true gravitational - wave signal and thus are unlikely to preserve any reconstructable astrophysical signal when analyzed . we refer to these as time - shifted data . both unshifted and time - shifted data are analyzed by identical procedures , yielding the candidate sample and the estimated background of the search , respectively . in order to avoid any biases , no unshifted data are used in the tuning of the methods . instead , combined with simulations ( see below ) , background data are used as the test set over which all analysis cuts are defined * prior * to examining the unshifted data - set . in this way , our analyses are `` blind '' . during the s5 run , simulated gw signals were occasionally injected into the data by applying an actuation to the mirrors at the ends of the interferometer arms . the waveforms and times of the injections were cataloged for later study . these were analyzed as an end - to - end validation of the interferometer readout , calibration , and detection algorithms . in addition to analyzing the recorded data stream in its original form , many simulated signals are injected in software by adding the signal to the digital data stream in order to to simulate the passage of gravitational - wave bursts through the network of detectors . the same simulated signals are analyzed by all three analysis pipelines . this provides a means for establishing the sensitivity of the search by measuring the probability of detection as a function of the signal morphology and strength . these will also be referred to as efficiency curves . unmodeled gw bursts can be distinguished from instrumental noise if they show consistency in time , frequency , shape , and amplitude among the ligo detectors . the time constraints , for example , follow from the maximum possible propagation delay between the hanford and livingston sites which is 10ms . this s5 analysis employs three algorithms to search for gw bursts : blocknormal @xcite , qpipeline @xcite , and coherent waveburst @xcite . a detailed description of each algorithm can be found in the appendices . here we limit ourselves to a brief summary of the three techniques . all three algorithms essentially look for excess power @xcite in a time - frequency decomposition of the data stream . events are ranked and checked for temporal coincidence and coherence ( defined differently for the different algorithms ) across the network of detectors . the three techniques differ in the details of how the time - frequency decompositions are performed , how the excess power is computed , and how coherence is assessed . each analysis pipeline was independently developed , coded and tuned . because the three pipelines have different sensitivities to different types of gw signals and instrumental artifacts , the results of the three searches can be combined to produce stronger statements about event candidates and upper limits . blocknormal ( bn ) performs a time - frequency decomposition by taking short segments of data and applying a heterodyne basebanding procedure to divide each segment into frequency bands . a change - point analysis is used to identify events with excess power in each frequency band for each detector , and events are clustered to form single - interferometer triggers . triggers from the various interferometers that fall within a certain coincidence window are then combined to compute the `` combined power '' , @xmath7 , across the network . these coincident triggers are then checked for coherence using corrpower , which calculates a cross - correlation statistic @xmath8 that was also used in the s4 search @xcite . a detailed description of the bn algorithm can be found in appendix [ sec : blocknormal ] . qpipeline ( qp ) performs a time - frequency decomposition by filtering the data against bisquare - enveloped sine waves , in what amounts to an over - sampled wavelet transform . the filtering procedure yields a standard matched filter signal to noise ratio ( snr ) , @xmath9 , which is used to identify excess power events in each interferometer ( quoted in terms of the quantity @xmath10 ) . triggers from the various interferometers are combined to give candidate events if they have consistent central times and frequencies . qpipeline also looks for coherence in the response of the h1 and h2 interferometers by comparing the excess power of sums ( the coherent combination @xmath11 ) and differences ( the null combination @xmath12 ) of the data . rather than using the single - interferometer h1 , h2 , l1 , signal to noise ratios , the qpipeline analysis uses the snrs in the transformed channels @xmath11 , @xmath12 , and l1 . a detailed description of the qpipeline algorithm can be found in appendix [ sec : qpipeline ] . coherent waveburst ( cwb ) performs a time - frequency decomposition using critically sampled meyer wavelets . the cwb version used in s5 replaces the separate coincidence and correlation test ( corrpower ) used in the s4 analysis @xcite by a single coherent search statistic based on a gaussian likelihood function . constrained waveform reconstruction is used to compute the _ network _ likelihood and a coherent network amplitude . this coherent analysis has the advantage that it is not limited by the performance of the least sensitive detector in the network . in the cwb analysis , various signal combinations are used to measure the signal consistency among different sites : a network correlation statistic @xmath13 , network energy disbalance @xmath14 , h1-h2 disbalance @xmath15 and a penalty factor @xmath16 . these quantities are used in concert with the coherent network amplitude @xmath17 to develop efficient selection cuts that can eliminate spurious events with a very limited impact on the sensitivity . it is worth noting that the version of cwb used in the s5 search is more advanced than the one used on ligo and geo data in s4 @xcite . a detailed description of the cwb algorithm can be found in appendix [ sec : cwb ] . both qpipeline and coherent waveburst use the freedom to form linear combinations of the data to construct `` null streams '' that are insensitive to gws . these null streams provide a powerful tool for distinguishing between genuine gw signals and instrument artifacts @xcite . as mentioned in sec . iv , the statistical properties of the noise triggers ( background ) are studied for all network combinations by analyzing time - shifted data , while the detection capabilities of the search pipelines for various types of gw signals are studied by analyzing simulated signals ( described in the following section ) injected into actual detector noise . plots of the parameters for noise triggers and signal injections are then examined to tune the searches . thresholds on the parameters are chosen to maximize the efficiency in detecting gws for a predetermined , conservative false alarm rate of roughly 5 events for every 100 time shifts of the full data set , _ @xmath4@xmath18 events expected for the duration of the data set . for a given energy threshold , all three pipelines observed a much larger rate of triggers with frequencies below 200hz than at higher frequencies . therefore , each pipeline set separate thresholds for triggers above and below 200hz , maintaining good sensitivity for higher - frequency signals at the expense of some sensitivity for low - frequency signals . the thresholds were tuned separately for each detector network , and the cwb pipeline also distinguished among a few distinct epochs with different noise properties during the run . a more detailed description of the tuning process can be found in appendices c , d , and e. in this section we present the efficiencies of the different algorithms in detecting simulated gws . as in previous science runs , we do not attempt to survey the complete spectrum of astrophysically motivated signals . instead , we use a limited number of ad hoc waveforms that probe the range of frequencies of interest , different signal durations , and different gw polarizations . we choose three families of waveforms : sine - gaussians , gaussians , and `` white noise bursts '' . an isotropic sky distribution was generated in all cases . the gaussian and sine - gaussian signals have a uniformly distributed random linear polarization , while the white noise bursts contain approximately equal power in both polarizations . we define the amplitude of an injection in terms of the total signal energy at the earth observable by an ideal optimally oriented detector able to independently measure both signal polarizations : @xmath19 in reality , the signal observed at an individual detector depends on the direction @xmath20 to the source and the polarization angle @xmath21 through `` antenna factors '' @xmath22 and @xmath23 : @xmath24 in order to estimate the detection efficiency as a function of signal strength , the simulated signals were injected at 22 logarithmically spaced values of @xmath25 ranging from @xmath26 to @xmath27 , stepping by factors of @xmath4@xmath28 . injections were performed at quasi - random times regardless of data quality or detector state , with an average rate of one injection every 100 seconds . the efficiency of a method is then defined as the fraction of waveforms that are detected out of all that were injected into the data analyzed by the method . the first family of injected signals are sine - gaussians . these are sinusoids with a central frequency @xmath29 , dimensionless width @xmath30 and arrival time @xmath31 , defined by : @xmath32.\ ] ] more specifically @xmath29 was chosen to be one of ( 70 , 100 , 153 , 235 , 361 , 554 , 849 , 945 , 1053 , 1172 , 1304 , 1451 , 1615 , 1797 , 2000 ) hz ; and @xmath30 to be one of 3 , 9 , or 100 . the second family consists of gaussian pulses described by the following expression : @xmath33 where @xmath34 is chosen to be one of ( 0.05 , 0.1 , 0.25 , 0.5 , 1.0 , 2.5 , 4.0 , 6.0 , 8.0 ) ms . the third family are the `` white noise bursts '' ( wnbs ) . these were generated by bandpassing white noise in frequency bands starting at 100hz , 250hz , or 1000hz , with bandwidth 10hz , 100hz , or 1000hz , and by time windowing with gaussian profiles of duration ( half of the interval between the inflection points ) equal to 100ms , 10ms , or 1ms . for each waveform type ( a choice of central frequency , bandwidth , and duration ) , 30 waveform files with random data content were created . the injections for each waveform type use random pairs selected from the 30 created waveforms for the @xmath35 and @xmath36 polarizations ( the selection avoids pairs with identical waveforms ) . this results in unpolarized injections with equal amounts of power on average in each polarization state . combined efficiencies of the three pipelines and two networks ( h1h2l1 and h1h2 ) used in the upper limit analysis for selected sine - gaussian waveforms with ( a ) @xmath37 , ( b ) @xmath38 , ( c ) @xmath39 . these efficiencies have been calculated using the logical or of the pipelines and networks for the subset of simulated signals that were injected in time intervals that were actually analyzed , and thus approach unity for large amplitudes . , title="fig:",width=288 ] + combined efficiencies of the three pipelines and two networks ( h1h2l1 and h1h2 ) used in the upper limit analysis for selected sine - gaussian waveforms with ( a ) @xmath37 , ( b ) @xmath38 , ( c ) @xmath39 . these efficiencies have been calculated using the logical or of the pipelines and networks for the subset of simulated signals that were injected in time intervals that were actually analyzed , and thus approach unity for large amplitudes . , title="fig:",width=288 ] + combined efficiencies of the three pipelines and two networks ( h1h2l1 and h1h2 ) used in the upper limit analysis for selected sine - gaussian waveforms with ( a ) @xmath37 , ( b ) @xmath38 , ( c ) @xmath39 . these efficiencies have been calculated using the logical or of the pipelines and networks for the subset of simulated signals that were injected in time intervals that were actually analyzed , and thus approach unity for large amplitudes . , title="fig:",width=288 ] combined efficiency of the three pipelines and two networks ( h1h2l1 and h1h2 ) used in the upper limit analysis for ( a ) selected linearly - polarized gaussian waveforms ; ( b ) selected band - limited white - noise bursts with two independent polarization components . these efficiencies have been calculated using the logical or of the pipelines and networks for the subset of simulated signals that were injected in time intervals that were actually analyzed , and thus approach unity for large amplitudes . , title="fig:",width=288 ] + combined efficiency of the three pipelines and two networks ( h1h2l1 and h1h2 ) used in the upper limit analysis for ( a ) selected linearly - polarized gaussian waveforms ; ( b ) selected band - limited white - noise bursts with two independent polarization components . these efficiencies have been calculated using the logical or of the pipelines and networks for the subset of simulated signals that were injected in time intervals that were actually analyzed , and thus approach unity for large amplitudes . , title="fig:",width=288 ] each efficiency curve , consisting of the efficiencies determined for a given signal morphology at each of the 22 @xmath25 values , was fitted with an empirical four - parameter function . the efficiency curves for the logical or combination of the three pipelines and for the combined h1h2 and h1h2l1 networks are shown for selected waveforms in figs . [ figure : efficiencies_combined_1 ] and [ figure : efficiencies_combined_2 ] . the @xmath25 values yielding 50% detection efficiency , @xmath40 , are shown in tables i and ii for sine - gaussians with @xmath38 and for white noise bursts injected and analyzed in h1h2l1 data . the study of the efficiency for all the waveforms shows that the combination of the methods is slightly more sensitive than the best performing one , which is qpipeline for some of the sine - gaussians , and cwb for all other waveforms considered . .@xmath25 values yielding 50% detection efficiency , in units of @xmath41 , for different sine - gaussian waveforms and pipelines in the h1h2l1 network . the first column is the central frequency , the second the quality factor , the third the @xmath40 of the logical or of the pipelines , and the remaining three columns the @xmath40 of the individual pipelines . these @xmath42 values include an adjustment of 11.1% to take into account calibration and statistical uncertainties as explained in sec . [ sec : systematics ] . [ cols=">,>,>,>,>,>",options="header " , ] the division into the twelve frequency bands was done using a basebanding procedure . any calibration lines within the band were removed by low - order regression filtering against the calibration line injection channel data . a final whitening filter of modest order was applied in each band to satisfy the blocknormal statistic s assumption of gaussianity in the background noise . the data conditioning procedures also had to minimize mixing noise characteristics between different time periods for the change - point analysis , and thus could not rely on predictive filtering . the blocknormal algorithm uses a bayesian statistic termed @xmath43 to perform a change - point analysis using the noise characteristics of time - series data . for an interval of @xmath44 time - series samples @xmath45 $ ] , this statistic measures the statistical likelihood ( at each sample @xmath46 within that interval ) that the data prior to that point are more consistent with a different gaussian - distributed ( or normal ) noise source than are the data following that point . it is defined as @xmath47\\ \nonumber & & \times \left[\frac{y_{1,k}^{-k/2 } y_{k+1,n}^{-(n - k)/2}}{y_{1,n}^{-n/2}}\right ] \left[\frac{\gamma(k/2 ) \gamma((n - k)/2)}{\gamma(n/2)}\right]\end{aligned}\ ] ] where @xmath48 \\ \overline{x_{i , j}^2 } : = \frac{1}{j - i+1 } \sum_{l = i}^{j } x[l]^2 \,.\end{aligned}\ ] ] the quantity @xmath49 is a constant proportional to @xmath50 , where @xmath51 is the prior probability , r the desired rate of blocks , and @xmath52 the sample rate . in fact each interval is searched for all change - points where @xmath53 exceeds a threshold value @xmath54 , where @xmath54 is implemented as a number times @xmath49 . the sub - intervals between change - points are termed `` blocks '' . the statistical significance of each such block is based on its `` excess power '' @xmath55 defined as @xmath56 where the block has mean @xmath57 and variance @xmath58 against a background of mean @xmath59 and variance @xmath60 . events were selected by requiring the negative - log - likelihood of @xmath55 ( termed @xmath61 ) to exceed a threshold . here @xmath62)\ ] ] where @xmath63 = \gamma(n/2,\xi^{*}/2)/\gamma(n/2).\ ] ] the variance - weighted time centroid , @xmath64 , of each event of @xmath65 samples of amplitude @xmath66 and time @xmath67 was calculated : @xmath68 the calibrated band - limited strain energy of each event was estimated using the frequency - averaged response @xmath69 over that band : @xmath70 the blocknormal algorithm was applied separately to the data in each frequency band ( table [ table : freqbands ] ) to select candidate gw burst events . the burst event generation was done on relatively long - duration epochs ( up to 1200s ) of continuous data to provide the best measure of the background noise characteristics . prior to the network coincidence step , events within each frequency band that are nearly adjacent were clustered into composite events . then , events between adjacent frequency bands whose time centroids were close were clustered into composite multiband events . all events were then characterized by their frequency coverage . for composite events , the effective time centroid was the energy - weighted average of the time centroid of the constituent events . the band - limited energy for composite events was simply the sum of the per - event energies . the central frequency for events in a single band was estimated by the average frequency of that band . for multi - band events , the energy - weighted average of these central frequencies was used . the signals from actual gw bursts in the ligo interferometers should be separated in time by no more than the maximum transit time ( @xmath71ms ) for gw between the hanford and livingston sites . for the co - located interferometers at hanford , there should be no time separation . the separation observed in the reconstructed events is larger due to limited time resolution , phase - delays in filtering , etc . for a candidate trigger , the time difference between candidate events in each pair of interferometers , @xmath72 , was required to fall within a fixed coincidence window , @xmath73 , for that pair of interferometers . this coincidence window had to be much broader than the transit time to account for limited time resolution and skewing of the time distributions from differential antenna response to @xmath74 and @xmath75 waveforms . the signals from actual gw bursts should also have similar strain amplitude ( and hence statistical significance ) in each interferometer . we derived a measure of coincident significance from the excess power significance @xmath61 in each candidate event in the trigger . this measure must correct for the lower significance for gw signals in the shorter h2 interferometer ( as compared to the h1 interferometer ) as well as the fluctuation of the relative gw signal strengths at the two ligo sites due to modulation from the antenna factors . the chosen metric for coincident significance , termed `` combined power '' or @xmath7 , was defined as @xmath76 this formulation was found to have the best performance in optimizing sensitivity to gw burst signals as a function of the background trigger rate . the coincidence procedure first identified events from each of the three detectors that had overlapping frequency coverage . these events then had to have time centroids whose difference @xmath77 was less than @xmath78ms . such time - coincidence events were retained as gw burst triggers if their combined power @xmath7 was above a threshold of 22 . the signals from gw bursts in each interferometer result from the same parent waveforms , and thus should have a large correlation sample - by - sample ( after correction for propagation delay ) . the cross - correlation statistic @xmath8 reported by the corrpower @xcite package is the maximum of the average correlation confidence of pair - wise correlation tests . it is positive - definite . larger values denote greater statistical certainty of coherence . the corrpower package was run on the list of candidate trigger times produced in the coincidence step . it retrieved the full time - series data from each interferometer around that time , calibrated the data , and calculated the @xmath8 cross - correlation statistic . for the three ligo interferometers , cuts were also made on the three pair - wise correlation tests . additional selection criteria took advantage of the special relationship for gw signals from the co - located interferometers h1 and h2 . one was the signed correlation factor between the h1 and h2 interferometers from the corrpower processing , termed @xmath79 . for triggers from gw bursts , this correlation factor should be positive . for triggers from a background of random coincidences , there should be an equal number of positive and negative correlation factors . also , since the h1 and h2 interferometers receive the same gw signal , the ratio of @xmath80 to @xmath81 should be close to one for a true gw burst . in contrast , for triggers from a random background this ratio will be centered around one - half . this arises because the h2 interferometer is approximately half as sensitive as h1 , so signals of the same statistical significance ( near the threshold ) will have only one - half the amplitude in h2 as they do in h1 . to simplify thresholding , the absolute value of the logarithm of the ratio was calculated @xmath82 for later use . the choices of tuning parameters are described in table [ tab : tuningcpbn ] . figure [ fg : tuningbn ] illustrates an example of plots used to tune the figures of merit for the h1h2l1 network . l + @xmath83 + @xmath84 for @xmath85hz + @xmath86 for @xmath87hz + @xmath88 , @xmath89 , @xmath90 + @xmath91 + @xmath92 + . the narrow black histogram represents the background ( noise ) triggers while the broader histogram represents the distribution of the injections . these triggers were generated in the h1h2l1 network and contain frequencies below 200hz . the vertical line indicates the cut made on this quantity.,height=226 ] qpipeline is an analysis pipeline for the detection of gw bursts in data from interferometric gravitational wave detectors @xcite . it is based on the q transform @xcite , a multi - resolution time - frequency transform that projects the data under test onto the space of bisquare - windowed complex exponentials characterized by central time @xmath34 , central frequency @xmath29 , and quality factor @xmath30 : @xmath93 where the bisquare window @xmath94 is @xmath95 ^ 2 & \text{for}~\displaystyle f < \frac{f_0\sqrt{5.5}}{q } \\ 0 & \text{otherwise } \end{array } \right.\ ] ] with @xmath96 the bisquare window is a close approximation to a gaussian window in frequency space ; the qpipeline is effectively a templated matched filter search @xcite for signals that are gaussian enveloped sinusoids in the whitened signal space . before applying the q transform , the data are first whitened by zero - phase linear predictive filtering @xcite . in linear predictive whitening , the @xmath65th sample of a discrete data sequence is assumed to be well modeled by a linear combination of the previous @xmath97 samples : @xmath98 = \sum_{m=1}^{m } c[m ] x[n - m ] \ , .\ ] ] the resulting whitened data stream is the prediction error sequence @xmath99 = \hat{x}[n ] - x[n]$ ] that remains after selecting the coefficients @xmath100 $ ] to minimize the error in the least - squares sense . the prediction error length @xmath97 is taken to be equal to the length of the longest basis function under test , which is approximately 1 second . this ensures that the data are uncorrelated on the time scales of the analysis . in order to avoid introducing phase errors between detectors , a modified zero - phase whitening filter is constructed by zero - padding the initial filter , converting to the frequency domain , and discarding all phase information . the space of gaussian enveloped complex exponentials is an over - complete basis of waveforms , whose duration @xmath101 and bandwidth @xmath102 have the minimum possible time - frequency uncertainty , @xmath103 , where @xmath104 . as a result , they provide the tightest possible constraints on the time - frequency area of a signal , maximizing the measured signal to noise ratio ( snr ) and minimizing the probability that false triggers are coincident in time and frequency between multiple detectors . in practice , the q transform is evaluated only for a finite number of basis functions , which are more commonly referred to as templates or tiles . these templates are selected to cover a targeted region of signal space , and are spaced such that the fractional signal energy loss @xmath105 due to the mismatch @xmath106 , @xmath107 , and @xmath108 between an arbitrary basis function and the nearest measurement template , @xmath109 is no larger than @xmath110 . this naturally leads to a tiling of the signal space that is logarithmic in @xmath30 , logarithmic in frequency , and linear in time . for this search , the qpipeline was applied to search the space of sinusoidal gaussians with central frequency from 48hz to 2048hz , and with @xmath30 from @xmath111 to @xmath112 . the statistical significance of q transform projections are given by their normalized energy @xmath113 , defined as the ratio of squared projection magnitude to the mean squared projection magnitude of other templates with the same central frequency and @xmath30 . for the case of ideal white noise , @xmath113 is exponentially distributed and is related to the matched filter snr quantity @xmath9 @xcite by the relation @xmath114 = \rho^2 / 2 \ , . \label{zdef}\ ] ] the q transform is applied to the whitened data and normalized energies are computed for each measurement template as a function of time . templates with statistically significant signal content are then identified by applying a threshold on the normalized energy . finally , since a single event may potentially produce multiple overlapping triggers due to the overlap between measurement templates , only the most significant of overlapping templates are reported as triggers . clustering of nearby triggers is not used in evaluating the significance of events . as a result , the detectability of gw burst signals depends on their maximum projection onto the space of gaussian enveloped sinusoids . for this search , the qpipeline took advantage of the co - located nature of the two ligo hanford detectors to form two linear combinations of the data streams from the two detectors . this coherent analysis makes use of correlations in the data to distinguish true gw signals from instrumental glitches . the first combination is the coherent signal stream , @xmath11 , a frequency dependent weighted sum of the data from the hanford detectors which maximizes the effective snr . the weighting is inversely proportional to the noise power spectral density , @xmath115 : @xmath116 the resulting combination is treated as the output of a new hybrid , `` coherent '' detector . under the assumption that the power spectral density is approximately flat across the window bandwidth , applying the q transform to this data stream leads to a coherent energy value , @xmath117 , which takes the following form : @xmath118 where @xmath119 , @xmath120 , and @xmath121 are functions of @xmath34 , @xmath29 , and @xmath30 , and the asterisk denotes complex conjugation . the last term represents the contribution of the cross - term , and is conceptually similar to a frequency domain representation of a cross - correlation of the h1 and h2 data streams . the energy expected in the coherent data stream if there were no correlations in the data can be characterized by the `` incoherent '' terms in eq . ( [ cohdef ] ) : @xmath122 the coherent and incoherent energies can then be normalized in the manner of eq . ( [ zdef ] ) : @xmath123 the correlation between the detectors can then be measured by the correlated energy , @xmath124 , given by @xmath125 the second combination is the difference between the calibrated data from the two detectors , known as the null stream , and is defined as @xmath126 by subtracting the co - located streams , any true gravitational wave signal should be canceled . the resulting combination is treated as the output of a new hybrid `` @xmath12 '' detector , which shows significant energy content in the presence of instrumental glitches but does not respond to gravitational waves . glitches are identified by thresholding on the corresponding normalized `` null energy '' , @xmath127 , calculated in an analogous manner to @xmath128 . signal tiles found to be in coincidence with significant null stream tiles are vetoed as instrumental glitches , and are not considered as candidate events . the threshold on @xmath127 can be expressed as @xmath129 where @xmath130 is chosen to limit the veto rate in gaussian noise to @xmath131 per 2048 tiles and @xmath51 is a parameter corresponding to the allowed tolerance in calibration uncertainty . this is an energy factor , and corresponds to an amplitude calibration uncertainty of approximately 22 percent . we expect that highly energetic instrumental glitches could leak energy into adjacent time - frequency bins , so the veto coincidence requirement between signal and null streams is scaled to give more - significant null stream tiles more area of veto influence in time - frequency space : @xmath132 @xmath133 where @xmath34 and @xmath29 are the central time and frequency of a tile , @xmath106 and @xmath134 are the duration and bandwidth of a tile , and the inflated null stream tile duration and bandwidth are defined as : @xmath135 @xmath136 coherent triggers from the two ligo hanford detectors were also tested for time - frequency coincidence with triggers from the ligo livingston detector using the following criteria , where @xmath137 is the speed of light travel time of 10ms between the two ligo sites : @xmath138 @xmath139 coincidence between the ligo hanford and livingston sites is not a requirement for detection , even if detectors at both sites are operational . the final trigger set is the union of triggers from the coherent h1h2 trigger set and the coincident h1h2l1 trigger set . the additional requirement of coincidence permits a lower threshold , and therefore greater detection efficiency , for the h1h2l1 data set . the choices of tuning parameters are described in table [ tab : tuningq ] . figure [ fg : tuningq ] an example scatter plot used to tune the figures of merit for the h1h2l1 network . @cll@c + & @xmath140 & & + & @xmath141{\frac{12.5}{z_{\text{l1}}}}\right)$ ] & for @xmath85hz & + & @xmath142 & for @xmath87hz & + & @xmath143 & & + + & @xmath140 & & + & @xmath144 & for @xmath85hz & + & @xmath145 & for @xmath87hz & , [ defined in eq . ( [ cor_energy ] ) ] , which measures the correlation of the strain at the two hanford interferometers , versus the l1 normalized energy [ defined in eq . ( [ zdef ] ) ] . the distribution of the background triggers is displayed in black while the distribution of simulated gw signals in gray . this example tuning plot is for triggers generated for the h1h2l1 network and containing frequencies below 200hz . the cuts on these quantities are displayed on the plot as thick lines . , height=226 ] coherent waveburst ( cwb ) is an analysis pipeline for the detection and reconstruction of gw burst signals from a network of detectors . the reconstructed gravitational waveform @xmath146 that best describes the response of the network is used to compute the maximum likelihood ratio of the putative gw signal , which forms the main detection statistic for the search . in effect , cwb is equivalent to a matched filter search with a very large template bank representing all possible time - domain signals with short duration . the cwb pipeline is divided into three main stages : the generation of coherent triggers , the reconstruction of the gw signal and the computation of the maximum likelihood ratio , and a post - production stage where additional detection cuts are applied . by using weighted coherent combinations of the data streams , cwb is not limited by the least sensitive detector in the network . the waveform reconstruction allows various physical properties of the signal to be estimated , including the sky location of the source . the coherent approach also allows for other statistics to be constructed , such as the null stream and coherent energy , to distinguish genuine gw signals from environmental and instrumental artifacts . the cwb analysis is performed in the wavelet domain . a discrete meyer wavelet transformation is applied to the sampled detector output to produce a discrete wavelet series @xmath147 $ ] , where @xmath148 is the time index , @xmath149 is the scale index and @xmath46 is the detector index . an important property of meyer wavelets is that they form an orthonormal basis that allow for the construction of wavelet filters with small spectral leakage @xcite . wavelet series give a time - scale representation of data where each wavelet scale can be associated with a certain frequency band of the initial time series . therefore a wavelet time - scale spectrum can be displayed as a time - frequency ( tf ) scalogram , where the scale is replaced with the central frequency @xmath150 of the band . the time series sampling rate @xmath151 and the scale number @xmath149 determine the time resolution @xmath152 at this scale . the frequency resolution @xmath153 is defined as @xmath154 and determines the data bandwidth at the scale @xmath149 . the time - frequency resolution defines the tiling of the tf plane . the individual tiles ( pixels ) represent data samples in the wavelet domain . in the cwb pipeline a uniform tiling is used @xmath155 , where @xmath65 is the wavelet decomposition depth ) , which is obtained with the meyer packet transformation @xcite . in this case the tf resolution is the same for all wavelet scales . for optimal localization of the gw energy in the tf plane , the cwb analysis is performed at six different frequency resolutions : 8 , 16 , 32 , 64 , 128 and 256 hz . before the coherent analysis is performed , two data conditioning algorithms are applied to the data in the wavelet domain : a linear prediction error ( lpe ) filter and a wavelet estimator of the power spectral density @xmath156 $ ] . lpe filters are used to remove `` predictable '' components from an input data series . in the cwb pipeline they are constructed individually for each wavelet layer and remove such components in the data as power line harmonics and violin - mode lines . a more detailed description of the lpe filters can be found elsewhere @xcite . the wavelet estimator of the one - sided power spectral density associated with each wavelet layer @xmath149 is @xmath157 = 2\frac{\sigma^2_k[j]}{r}\ ] ] where @xmath158 $ ] is the variance of the detector noise . in the analysis we assume that the detector noise is gaussian and quasi - stationary . the variance estimator may vary with time and therefore it is calculated for each sample in the wavelet layer : @xmath159 $ ] . the estimation of the noise variance is performed on data segments of length 60 seconds , with 40 seconds overlap . linear interpolation is used between two measurements to obtain @xmath159 $ ] . the first step in the analysis is to identify segments of data that may contain a signal . the triggers are evaluated using the whitened data vector @xmath160 $ ] @xmath161(\theta,\phi ) = \left ( \frac{a_1[i , j,\tau_1(\theta,\phi)]}{\sigma_1[i , j ] } , .. , \frac{a_k[i , j,\tau_k(\theta,\phi)]}{\sigma_k[i , j ] } \right ) \;.\ ] ] the sampled detector amplitudes in the wavelet domain @xmath162 $ ] take into account the time delays @xmath163 due to the time - of - flight between the detectors , which in turn depend on the source coordinates @xmath164 and @xmath165 . coherent triggers are generated for the entire network by maximizing the norm @xmath166|$ ] over the entire sky for each time - frequency location @xmath167 $ ] . to do this , the sky is divided into square degree patches and the quantity @xmath168 is calculated for each patch from the delayed detector amplitudes @xmath169 $ ] . by selecting clusters of pixels with the @xmath170 above some threshold , one can identify coherent triggers in the time - frequency plane . the data pixels @xmath171 $ ] selected by this procedure are then used to reconstruct the gw signal and compute the maximum likelihood statistic . for the case of gaussian quasi - stationary noise , the likelihood that data @xmath172 is purely instrumental noise is proportional to @xmath173 , while the likelihood that a gw signal @xmath146 is present is proportional to @xmath174 . the ratio of these likelihoods can be used as a detection statistic . here @xmath175 defines a noise weighted inner product , which for @xmath176 detectors with uncorrelated noise can be written in the wavelet domain as @xmath177 y_k[i , j]}{\sigma^2_k[i , j ] } \ , .\ ] ] where time @xmath148 and frequency @xmath149 indices run over some time - frequency area @xmath178 selected for the analysis . the coherent waveburst pipeline defines @xmath179 as twice the ( log ) likelihood ratio , and treats it as a functional in @xmath180 @xcite : @xmath181 = 2(a \vert h_{\text{det}})- ( h_{\text{det } } \vert h_{\text{det } } ) \ , , \ ] ] where @xmath182 $ ] are the detector responses ( eq . [ eq : hdet ] ) . the network sensitivity is characterized by the noise - scaled antenna pattern vectors @xmath183 and @xmath184 : @xmath185 = \left ( \frac{f_{1,+(\times)}(\vec\omega , \psi ) } { \sigma_1[i , j]}, .. ,\frac{f_{k,+(\times)}(\vec\omega , \psi)}{\sigma_k[i , j ] } \right ) \;.\ ] ] since the detector responses @xmath186 are independent of rotation by an arbitrary polarization angle in the wave frame , it is convenient to perform calculations in the dominant polarization frame ( dpf ) @xcite . in this frame the antenna pattern vectors @xmath183 and @xmath184 are orthogonal to each other : @xmath187 and we refer to them as @xmath188 and @xmath189 respectively . the corresponding solutions for the gw waveforms , @xmath190 and @xmath191 , are found by variation of the likelihood functional ( eq . ( [ eq : like ] ) ) that can be written as the sum of two terms , @xmath192 , where @xmath193 \ ; , \\ \label{eq : like2 } { \cal{l}}_2 = \sum_{\omega_{tf } } \left [ 2({\bf{w}}\cdot{\bf{f_2 } } ) h_2 - |{\bf{f_2}}|^2 h^2_2 \right ] \;. \end{aligned}\ ] ] the estimators of the gw waveforms for a particular sky location are then the solutions of the equations @xmath194 and @xmath195 : @xmath196 note , the norms @xmath197 and @xmath198 characterize the network sensitivity to the @xmath190 and @xmath199 polarizations respectively . the maximum likelihood ratio statistic for sky location @xmath200 is calculated by substituting the solution for @xmath146 into @xmath201 $ ] . the result can be written as @xmath202 where the matrix @xmath203 is the projection constructed from the components of the unit vectors @xmath204 and @xmath205 along the directions of the @xmath188 and @xmath189 respectively : @xmath206 the kernel of the projection @xmath203 is the _ signal _ plane defined by these two vectors . the null space of the projection @xmath203 defines the reconstructed detector noise which is referred to as the null stream . the projection matrix is invariant with respect to the rotation in the signal plane where any two orthogonal unit vectors can be used for construction of the @xmath207 . therefore one can select vectors @xmath208 and @xmath209 such that @xmath210 and @xmath211 the unit vector @xmath208 defines the vector @xmath212 whose components are estimators of the noise - scaled detector responses @xmath182/\sigma_k[i , j]$ ] . in principle the likelihood approach outlined above can be used for the reconstruction of the gw waveforms and calculation of the maximum likelihood statistic . in practice the formal solutions ( [ eq : syst1 ] ) , ( [ eq : syst2 ] ) need to be regularized by constraints that account for the way the network responds to a generic gw signal @xcite . for example , the network may be insensitive to gw signals with a particular sky location or polarization , resulting in an ill - posed inversion problem . these problems are addressed by using regulators and sky - dependent penalty factors . a classical example of a singular inversion problem is a network of aligned detectors where the detector responses @xmath186 are identical . in this case the algorithm can be constrained to search for one unknown function rather than for the two gw polarizations @xmath190 and @xmath191 , which span a larger parameter space . note that in this case @xmath213 , eq . ( [ eq : syst2 ] ) is ill - conditioned and the solution for the @xmath191 waveform can not be found . regulators are important not only for aligned detectors , but also for networks of misaligned detectors , for example , the ligo and virgo network . depending on the source location , the network can be much less sensitive to the second gw component ( @xmath214 ) and the @xmath191 waveform may not be reconstructable from the noisy data . in the coherent waveburst analysis we introduce a regulator by modifying the norm of the @xmath189 vector : @xmath215 is a tunable parameter . for example , if @xmath216 , the second gw component is entirely suppressed and the regulator corresponds to the `` hard constraint '' described in ref . @xcite . in this case the unit vector @xmath208 ( see eq . ( [ eq : u ] ) ) is pointing along the @xmath183 direction . in the cwb analysis the parameter @xmath217 is chosen to be @xmath218 } } \,.\ ] ] this regulator is more stringent for weak events which are generated by the pipeline at much higher rate than the loud events . the introduction of the regulator creates an obvious problem for the construction of the projection matrix . namely , the vector @xmath219 and the corresponding vector @xmath220 obtained by rotation of @xmath204 and @xmath221 in the signal plane are not unit vectors if @xmath222 . to fix this problem we re - normalize the vector @xmath220 to unity and use it for calculation of the maximum likelihood ratio and other coherent statistics . when the detector noise is gaussian and stationary , the maximum likelihood @xmath223 is the only statistic required for detection and selection of the gw events . in this case the false alarm and the false dismissal probabilities are controlled by the threshold on @xmath223 which is an estimator of the total snr detected by the network . however , the real data are contaminated with instrumental and environmental artifacts and additional selection cuts should be applied to separate them from genuine gw signals @xcite . in the coherent waveburst method these selection cuts are based on coherent statistics constructed from the elements of the likelihood and the null matrices . the diagonal terms of the matrix @xmath224 describe the reconstructed normalized incoherent energy . the sum of the off - diagonal terms is the coherent energy @xmath225 detected by the network . the next step is to optimize the solution over the sky . often , depending on the network configuration , the reconstruction of source coordinates is ambiguous . for example , for two separated detectors the relative time delay that yields maximum correlation between the data streams corresponds to an annulus on the sky . in this case , an `` optimal '' source location is selected , where the reconstructed detector responses are the most consistent with the output detector data streams . to properly account for the directional sensitivity of the network the optimization over sky locations has to be more than a simple maximization of @xmath226 . in the cwb analysis the statistic that is maximized has the form @xmath227 where @xmath16 is the penalty factor and @xmath228 is the network correlation coefficient . @xmath16 and @xmath228 are defined below in terms of the matrix @xmath229 and the diagonal matrices @xmath230 and @xmath231 which describe the normalized energy in the detectors , and the normalized reconstructed signal energy ( see eq . ( [ eq : xi ] ) ) , with @xmath232 ideally , the reconstructed signal energy in each detector @xmath233 should not significantly exceed the energy @xmath234 . this requirement can be enforced by the constraint @xmath235 for each detector in the network . these constraints can be applied during the signal reconstruction by way of lagrange multipliers in the variational analysis , however this greatly increases the computational complexity of the algorithm . a simpler alternative is to introduce a penalty factor @xmath16 that penalizes sky locations violating the constraint equation ( [ eq : ed ] ) : @xmath236 in addition to serving as a penalty factor in the position reconstruction , the ratio of reconstructed and detector energy were also used as a post - production cut . events with @xmath237 were discarded , as were events with large values of the network energy disbalance @xmath238 and the h1-h2 energy disbalance @xmath239 the latter cut was found to be particularly effective at rejecting correlated glitches in the two hanford interferometers . the network correlation coefficient is also used to weight the overall likelihood for each sky location . it is defined as @xmath240 where @xmath241 is the sum of all elements in the null matrix @xmath242 which represents the normalized energy of the reconstructed noise . usually for glitches little coherent energy is detected and the reconstructed detector responses are inconsistent with the detector output , which results in a large value for the null energy . in addition to helping select the optimal sky location , the correlation coefficients @xmath228 are used for a signal consistency test based on the comparison of the null energy and the coherent energy . the coherent terms of the likelihood matrix can be also used to calculate the correlation coefficients @xmath243 which represent pearson s correlation coefficients in the case of aligned detectors . we use the coefficients @xmath244 to construct the reduced coherent energy @xmath245 combined with the network correlation coefficient @xmath228 and the number of detectors in the network , @xmath176 , it yields a quantity which we call the _ coherent network amplitude _ , @xmath246 figure [ fg : tuningcwb ] shows the @xmath17@xmath13 distribution of the background events ( see sec . [ sec : tuning ] ) and simulated gw events ( see sec . [ sec : simulations ] ) for the l1h1h2 network . loud background events due to detector glitches with low values of the network correlation coefficient are rejected by a threshold on @xmath228 . relatively weak background events are rejected by a threshold on @xmath17 . table [ tab : tuningcwb ] describes the full set of tuning parameters for cwb . l + @xmath247 , @xmath248 + @xmath249 for @xmath150@xmath250@xmath251 hz , up to dec 12 2005 03:19:29 + + @xmath252 for @xmath150@xmath250@xmath251 hz , between dec 12 2005 03:19:29 + + @xmath253 for @xmath150@xmath254@xmath251 hz + @xmath255 , @xmath256 + + @xmath257 , @xmath248 + @xmath249 for @xmath150@xmath250@xmath251 hz + @xmath258 for @xmath150@xmath254@xmath251 hz , up to jul 17 2006 11:50:37 + @xmath253 for @xmath150@xmath254@xmath251 hz , after jul 17 2006 11:50:37 + @xmath255 , @xmath256 + + @xmath247 , @xmath248 + @xmath259 for @xmath150@xmath250@xmath251 hz , up to oct 07 2006 08:58:06 + @xmath260 for @xmath150@xmath250@xmath251 hz , after oct 07 2006 08:58:06 + @xmath261 for @xmath150@xmath254@xmath251 hz + @xmath256 + + @xmath247 , @xmath248 + @xmath259 for @xmath150@xmath250@xmath251 hz , up to mar 28 2006 04:23:06 + + @xmath261 for @xmath150@xmath250@xmath251 hz , between mar 28 2006 04:23:06 + + @xmath261 for @xmath150@xmath254@xmath251 hz + @xmath256 +

we have used the hubble space telescope ( _ hst _ ) and the faint object spectrograph ( fos ) to study faint uv stars in the core of the nearby globular cluster ngc 6397 . we confirm the presence of a 4th cataclysmic variable ( cv ) in ngc 6397 ( hereafter cv 4 ) , and we use the photometry of cool et al . ( 1998 ) to present evidence that cvs 14 all have faint disks and probably low accretion rates . by combining these results with new uv spectra of cv 1 and the published spectra of grindlay et al . ( 1995 ) we present new evidence that cvs 13 may be dq her systems , as originally suggested by grindlay et al . ( 1995 ) , and we show that cv 4 may either be a dwarf nova or another magnetic system . another possibility is that the cvs could be old novae in hibernation between nova eruptions ( shara et al . 1986 ) . we also present the first spectrum of a member of a new class of uv bright stars in ngc 6397 ( cool et al . 1998 ) . these faint , hot stars do not vary , unlike the cvs , and are thus denoted as `` non - flickerers '' ( nfs ) . like the cvs , their spatial concentration is strongly concentrated toward the cluster center . using detailed comparisons with stellar atmosphere models we have determined log g = 6.25 1.0 , and = 17,500 5,000 k for this nf . using these line parameters and the luminosity of the nf we show that the nf spectrum is consistent with a helium wd having a mass of.25 and an age between 0.1 and 0.5 gyr ( depending on the models used ) . the nf spectrum appears to be significantly doppler shifted from the expected wavelength , suggesting the presence of a dark , massive companion , probably a carbon - oxygen ( co ) wd . 2

astro-ph9811138

binaries have long been thought to have a crucial impact on globular cluster dynamics and evolution ( hut et al . 1992 ) , but only recently ( especially with the use of _ hst _ ) have large numbers of them been found in globular cluster cores where they are expected to act as the central energy source that drives cluster expansion . discoveries of binary millisecond pulsars ( e.g. manchester et al . 1991 ) and multiple low - luminosity x - ray sources ( hertz & grindlay 1983 ) have recently been supplemented by discoveries of large numbers of eclipsing binaries in globulars ( e.g. 47 tuc ; edmonds et al . 1996 and kaluzny et al . 1998 ) and a significant population of main sequence main sequence binaries in ngc 6752 ( rubenstein and bailyn 1997 ) . another recent breakthrough has been the discovery of cataclysmic variables ( cvs ) in the cores of globular clusters , using either dwarf nova ( dn ) outbursts in m5 ( oosterhoff 1941 ) , 47 tuc ( paresce & de marchi 1994 ) and ngc 6624 ( shara , zurek & rich 1996 ) , uv excess to recover an old nova in m80 ( shara and drissen 1995 ) or narrow - band h@xmath2 emission . using the latter technique 3 cvs have been reported in ngc 6397 by cool et al . ( 1995 ) and grindlay et al . ( 1995 ; hereafter gc95 ) , and a fourth cv candidate by cool et al . ( 1998 ; hereafter cg98 ) . also , 2 probable cvs have been reported in ngc 6752 by bailyn et al . these cvs appear to be the long - sought optical counterparts of the low - luminosity x - ray sources found in globular cluster cores . in particular , the 3 brightest optical emission line objects in ngc 6397 ( gc95 ) are the probable counterparts of the 3 brightest x - ray sources found by cool et al . ( 1993 ; see also cool et al . 1995 ) . observations of cvs in clusters can be used for a variety of studies including : ( 1 ) cv formation and evolution in low - metallicity environments , ( 2 ) stellar interactions in high - density environments , and ( 3 ) cluster dynamical evolution . probable formation mechanisms for globular cluster cvs include tidal capture and exchange collisions between main sequence ( ms ) stars and white dwarfs ( wds ) , complementing studies of the ms star - ms star interactions that produce blue stragglers . since these formation mechanisms differ from those for field cvs , and the stellar environment is different , it would not be surprising to find systematic differences between globular cluster and field cvs . in particular , gc95 and grindlay ( 1996 ) have suggested that the cvs in ngc 6397 might have a much higher percentage of magnetic wds than field cvs . the dense , collapsed core of ngc 6397 is a prime region to study the effects of stellar interactions because its high central density makes interaction rates large and its relative proximity at 2.2 kpc makes it possible to probe the core with high spatial resolution ( cg98 ) . this paper presents new _ hst_/fos spectra of cv 1 ( from gc95 ) and cv 4 ( from cg98 ) in ngc 6397 . by combining all available spectra for cvs 14 with the photometry of cg98 and comparing with field cvs we show that the cv disks are consistent with those of faint quiescent dne , given their expected periods , but that their 2 4686 lines are unusually strong for dne ( such systems would probably have long recurrence times between outbursts ) . instead , we argue that cvs 13 may be magnetic cvs ( or perhaps old novae ) , and that cv 4 is either a low accretion rate dn or a magnetic cv . it is possible that these objects may even be quiescent lmxbs , although this is unlikely based on detailed comparisons with the x - ray and optical properties ( grindlay 1996 , 1998 ) . in any case we present good evidence that there _ are _ systematic differences between populations of globular cluster and field cvs . along with the 3 previously known classes of uv bright stars in ngc 6397 ( blue stragglers , cvs and wds ) , cg98 have discovered another class of uv bright stars . three faint , hot stars have been found within only @xmath116@xmath3 of the cluster center , all of them non - variable ( unlike the flickering cvs ) . cg98 have argued that these non - flickering ( nf ) stars are unlikely to be cvs , `` normal '' co wds ( recently evolved from single red giants ) , extended horizontal branch stars or field stars , but instead that they are good candidates for low - mass helium wds . helium wds have masses @xmath4 0.49 , and in the field are usually found in binary systems containing either another wd or a neutron star ( marsh , dhillon & duck 1995 , and rappaport et al . these double degenerates are thought to form by roche lobe overflow ( and usually common envelope events ) in primordial binaries containing red giants , if he ignition in the red giant core , a proto helium wd , is avoided ( iben , tutukov and yungelson 1997 discuss detailed formation scenarios ) . several low - mass wds have been found or inferred in open and globular clusters including the helium wd red giant binary s1040 in m67 ( landsman et al . 1997 ) , and the ultra short period x - ray binary systems 4u 1820 - 30 in ngc 6624 ( anderson et al . 1997 ) and star s in ngc 6712 ( anderson et al . 1993 ) . s1040 in m67 probably formed after a subgiant underwent roche lobe overflow in a primordial binary ( landsman et al . 1997 ) , but in denser globular clusters , primordial binary evolution may be less important than interactions involving subgiants or red giants . for example , red giant / wd or red giant / ms star direct collisions should cause a helium wd to be left behind in a binary system ( davies , benz & hills 1991 ) . systems such as 4u 1820 - 30 and star s probably result from either neutron star / red giant collisions ( verbunt 1987 ) or neutron star / ms star capture and delayed mass transfer ( bailyn & grindlay 1987 ) . here , we report the first spectrum of one of the ngc 6397 nfs . the lack of emission lines in the spectrum provides extra evidence against the cv possibility and the log g value presented here argues against a co wd or extreme horizontal branch identification . by comparing with published model atmospheres we determine log g and for the nf and we then compare these parameters ( along with the luminosity ) with wd evolutionary models to show that a low - mass helium wd is , indeed , a plausible explanation for the nf . we also present evidence that the nf spectrum is significantly doppler shifted from the expected wavelength , suggesting that the nf is in a binary system with a massive dark companion . spectroscopic observations with _ hst _ were made of cv 1 , the brightest cv of the 3 studied by gc95 , on october 2nd , 1996 . ultraviolet and optical observations were obtained with the fos / prism , giving a spectral coverage from 18508950 with variable dispersion and resolution decreasing from the blue to the red . ultraviolet observations were obtained with the fos / g160l grating , with a spectral coverage from 11502510 ( at a spectral resolution of 9.2 ) , including a small contribution from second - order geocoronal ly@xmath2 near the red end . the data were originally reduced using the normal pipeline processing but were then recalibrated using updated flat - fields and inverse sensitivity files ( removing some flat - field features at around the 5% level ) . figure [ fig.allcv1 ] summarizes all of the fos data available for cv 1 . the full g160l spectrum is shown and then the prism spectrum from 2510red - ward , along with the red g570h spectrum ( with a resolution of 4.5 ) from cycle 4 . no scaling was used for either the g160l or prism spectra , showing good self - consistency in the spectral calibration and the photometric states of ( variable ) cv 1 between the two separate observations . despite the low spectral resolution of the prism in the blue , the h@xmath5 and h@xmath6 emission lines are visible , along with some absorption lines from the secondary ( for example mgii 2800 and the ca h / k doublet ) . since ngc 6397 lies close to the galactic plane its reddening is significant and therefore figure [ fig.allcv1 ] also shows a dereddened version of the smoothed prism spectrum ( using @xmath7 from alcaino et al . 1997 ) . using the wfpc2 photometry from cg98 we estimated the uv and optical contribution from the secondary component in cv 1 . we began by noting that cv 1 is almost on the ms in the @xmath8 vs @xmath9 cmd , which implies a bright secondary and relatively faint disk and @xmath8 passbands . ] for this system . then , assuming that all of the flux in @xmath10 comes from the secondary we used the position of the ms in cg98 to estimate @xmath11 and @xmath8 for the secondary ( showing the advantage of observing binaries in a cluster ) . we then chose the reddened kurucz atmosphere ( log z = 2.0 ) which best matched the @xmath12 photometry for the secondary . figure [ fig.allcv1 ] shows the kurucz stellar atmosphere with a dotted line . a close up of the g160l spectrum is shown in figure [ fig.g160 ] . the dashed line shows a reddened blackbody fit to the uv spectrum ( temperature = 12850 k , without removing the small contribution of the secondary ) and the solid line shows dereddening of this fit using @xmath7 . the detection of ly@xmath2 in second order and a marginal ( 3.5@xmath13 ) detection of 2 1640 are labeled . no other uv lines are detected , and we set 3@xmath13 upper limits on the equivalent width ( ew ) of nv 1246 ( 449 ) , siv / oiv 1402 ( 109 ) and civ 1553 ( 58 ) . the most striking feature of the g160l and prism spectra is the relatively low uv flux . we have compared the measured flux distribution of cv 1 with that of various subclasses of field cvs studied by verbunt ( 1987 ) with iue . this comparison has limitations , as pointed out by the referee , since verbunt s sample may not include the full range of cv properties ( including highly magnetic wd primaries and a possible lack of disks in some dq hers ) . however , the data set and analysis is homogeneous and covers most cv classes , with the notable exception of am her systems ( see below ) . verbunt ( 1987 ) determined the fluxes of cvs in 80 width bins centered on 1460 , 1800 , 2140 and 2880 . we first normalized each ( quiescent ) system in verbunt s study to have the same flux at 1460 and then averaged the fluxes at other wavelengths over each class . we then predicted , for each cv class , the average uv fluxes expected for @xmath8=18.27 , the magnitude of cv 1 ( @xmath8 magnitudes are also given in verbunt 1987 ) . we found reddened fluxes of @xmath14ergs @xmath15s@xmath16@xmath16 ( dq hers ) , @xmath17 ( u gems ) and @xmath18 ( z cams ) , and even brighter for other cv classes ( for example @xmath19 for old novae ) . these fluxes are all over ten times brighter than the observed ( reddened ) @xmath20 for cv 1 ( see figures [ fig.allcv1 ] and [ fig.g160 ] ) . however , scaling by the @xmath8 magnitude of the system has limited usefulness because the secondary is more massive and brighter than most field cvs . a better comparison is to calculate relative fluxes in the uv for cv 1 , where the contribution from the secondary is less important . the dereddened uv fluxes for cv 1 are shown in figure [ fig.uv ] along with fluxes for various cv classes ( the error bars combine uncertainties in the continuum level estimation and the reddening ) . not surprisingly , the reddest flux distributions are found for quiescent dwarf novae ( dne ) , characterized by low accretion rates , and dq her systems , with inner disks truncated by the magnetic field of the wd . clearly , cv 1 has a much redder uv flux distribution than the field cvs in verbunt s sample . figure [ fig.uv ] also plots the flux distribution for the cv 1 disk after subtracting off the secondary and shows that the disk is marginally redder , given the large errors , than all classes of field cv ( without the secondaries subtracted ) . we have obtained an optical spectrum of a fourth cv candidate discovered near the center of ngc 6397 ( cg98 ) . figure [ fig.cv4 ] shows the g570h spectrum of this star , after correcting for diffuse light ( from a combination of extended psf halos from bright stars such as giants and diffuse light from faint cluster stars ) . for comparison , we also show an average of the spectra of cvs 13 from gc95 , where the continua of cvs 2 and 3 were normalized to that of cv 1 . strong balmer emission lines are present in the spectrum of the new cv candidate along with several hei lines and 2 4686 . the spectrum confirms that this star is a cv , as suggested by its uv excess and flickering ( cg98 ) . the relative weakness of the 2 line is the most obvious difference between cv 4 and the average spectrum of cvs 13 ( see a quantitative analysis below ) . the spectrum of cv 4 has the highest signal to noise ( s / n ) ratio of the 4 cv spectra , enabling easy detection of relatively weak lines such as hei 4713 . the dotted line shows the estimated contribution of the secondary , using the procedure outlined above for cv 1 . to improve the spectra , we applied boxcar smoothing over 7 channels ( this improved the s / n ratio per independent channel by a factor of 2.6 at , but in simulations had a negligible effect on measured line parameters ) . we found that a voigt profile gave optimal fits to the emission lines , after also experimenting with gaussians and lorentzians . before deriving ews ( see section [ sec.he2 ] ) and integrated magnitudes from the spectrum of this faint star ( @xmath8=20.81 ) , we considered 2 possible additional sources of light in the 0.43@xmath3 aperture used : ( 1 ) possible light from individual neighboring stars and ( 2 ) the diffuse light contribution . by examining the high s / n wfpc2 images of cg98 we found that light from neighbors is negligible , since cv 4 has no measurable companions @xmath21 away , but that diffuse light is an important factor for this star . to measure the spectral contribution of the diffuse light we used kurucz ( 1993 ) stellar atmospheres to fit a single temperature model to the diffuse light components in @xmath11 , @xmath8 and @xmath10 derived by cg98 . a 6,000 k kurucz model gave an excellent fit and showed that relatively hot stars dominated the diffuse light at the position of cv 4 . around 45% of the flux at 5500 comes from the diffuse light showing the importance of the above correction for accurate ews . by comparing the cv 4 spectrum with the kurucz spectrum of the secondary ( see figure [ fig.cv4 ] ) we see some evidence for a small residual red component . this component is too red to be caused by the cv disk ( or wd ) , and is perhaps caused by earthshine ( a red component also appears , for data near the earth limb , to be present in the spectrum of the nf ; see below ) . an important diagnostic for cvs are the line fluxes , especially the relative strength of 2 4686 . table 1 gives values of the wavelength , line flux , ew and fwhm ( gaussian and lorentzian ) of the , h@xmath2 , hei 5876 and 2 4686 emission lines for cvs 14 ( 1@xmath13 errors are quoted ) . we have not undertaken a complete analysis for cvs 13 , which requires taking account of light contaminants in the fos aperture ( cv 2 , for example , has considerable contamination from a bright neighboring star ) . however , the line ratio between 2 4686 and will be much less sensitive to contamination errors than the ews and so we derive these ratios from the uncorrected spectra , after taking into account blends from hei 4713 . since multiline fitting in iraf / splot was unable to separate the heavily blended 24686 and hei 4713 lines in cvs 13 , we used the relative strength of hei 4713 compared to hei 5876 for cv 4 ( where the blend from the weak 2 4686 line was small ; see figure [ fig.cv4 ] ) to remove the hei 4713 line from these systems . the 2 4686/ line ratio for cv 4 is only 0.07 @xmath0 0.01 , compared to 0.32 @xmath0 0.04 , 0.34 @xmath0 0.05 and 0.25 @xmath0 0.03 for cvs 13 respectively ( these more accurate determination replace those given in gc95 ) . two other notable differences are that the lines of cv 4 are narrower ( smaller fwhms ) than those of cvs 13 and the ews of most of the cv 4 lines are greater than those of cvs 13 . this latter result is part of a very clear trend that the ews increase going from cv 1 to cv 4 , mainly because the secondaries ( which dominate the optical flux ) get fainter , without significant changes in the line fluxes . for example , comparing cv 4 with cv 1 , the ratio of ews of is 8.8 , with a factor of 0.08 difference in the continuum levels and only a factor of 0.7 difference in line fluxes . to compare the measured 2 4686/ line ratios with those of field cvs we used the emission line data of williams ( 1983 ) and echevarria ( 1988 ) for various cv classes ( after confirming the cv classification using ritter & kolb 1998 ) . because 2 4686 is often quite a weak line and it is highly variable only a subset of the cvs in echevarria s sample include 2 4686 , but even with this relatively small sample , some clear trends emerge . the average 2 4686/ line ratios for the non - magnetic cv classes are 0.16 @xmath0 0.05 ( 17 dne ) , 0.94 @xmath0 0.50 ( 8 old novae ) and 0.76@xmath0 0.51 ( 13 nova likes , excluding magnetic systems ) . these values are probably overestimates ( although we lack information about upper limits ) , especially for dne where only 17 out of 39 systems in the above sample have measurable 2 4686 ( the completeness for the other systems is 8/10 for old novae and 13/17 for nova likes ) . for magnetic systems we used the classifications by patterson ( 1994 ) and ritter & kolb ( 1998 ) and the spectra of williams ( 1983 ) to calculate an average 2 4686/ line ratio of 0.57 @xmath0 0.46 ( measurable 2 4686 for 7/8 dq her systems ) and 0.59 @xmath0 0.21 ( 4/4 am her systems ) . the high accretion rate novae and nova likes can have relatively strong 24686 ( probably because of very hot inner disks ) , while the magnetic systems also usually have strong 2 4686 , thought to be because the hot accretion `` curtains '' along the magnetic field lines of the wd are directly visible . only the lower accretion rate dne have relatively weak 2 4686 compared to . for comparison with field cvs we have also estimated the brightness of the disk in cvs 24 , using the techniques described in section [ sec.cv1 ] to subtract off the secondary . table 2 shows the total absolute magnitudes of cvs 14 , plus estimated absolute magnitudes and masses ( @xmath22 ) of the secondaries ( from bergbusch and vandenberg 1992 ) , and absolute magnitudes of the disks ( @xmath23 ( disk ) ) . since smith and dhillon ( 1998 ) have shown that secondaries in cvs with orbital periods below @xmath17 h , as likely here ( see discussion below ) are typically not evolved , the mass estimates for main sequence stars should be reasonable . note that , using this technique , we found that the @xmath8 magnitude of the secondary in cv 1 was 0.02 mag _ brighter _ than the @xmath8 magnitude of the total system , clearly an impossibility . variability is the likely explanation , plus errors will also play a part ; to investigate the sensitivity of these results to overestimating the brightness of the secondary ( and other errors ) we made the estimated @xmath10 magnitude of the secondary fainter by 0.1 mag and 0.2 mag and rederived the expected @xmath23 ( disk ) ( see table 2 ) . we also show in table 2 , estimates of upper limits for periods of cvs 14 , using the relationship given in warner ( 1995 ) between the mass of a roche - lobe filling secondary and the cv period , namely @xmath24(h ) . since this equation applies to pop i stars and pop ii stars have smaller radii for a given mass than pop i stars ( implying a smaller period for roche - lobe overflow ) , we have used the study by stehle , kolb & ritter ( 1997 ) to estimate the period correction required for pop ii stars ( @xmath25 1 h for all masses @xmath25 0.9 ) . as a guide to the validity of this procedure for popii stars we note that the two cv candidates in ngc 6752 ( bailyn et al . 1996 ) , have absolute @xmath8 and @xmath10 magnitudes only a few tenths of a magnitude different from cvs 2 and 3 in ngc 6397 . the estimated periods of 4.4 h and 3.8 h for cvs 2 and 3 compare well with the observed periods of 5.1 h and 3.7 h for the ngc 6752 cv candidates . the g570h spectrum of the nf is shown in figure [ fig.allnf ] . the blue continuum and broad line suggest that this star is a high gravity object ( see below ) . the spectrum suffers from both diffuse light contamination and from having a much brighter ( @xmath26v = 3.5 mag ) neighbor only 0.3@xmath3 distant ( a star near the ms turnoff ) . with perfect pointing during the fos observations we would simply be able to use the high s / n wfpc2 images from cg98 to determine the exact amount of turnoff star contamination . however , there are 0.08@xmath3 uncertainties in the pointing of _ hst _ which can potentially make a significant difference to the amount of light contamination . there is also a source of variable light , since the apparent flux of the star increases by about 15% towards the end of an orbit ( probably due to increased scattered light or earthshine as noted above ) . since it is difficult to estimate the `` clean '' , uncontaminated spectrum of this star we forced its continuum to equal that of a reddened kurucz atmosphere constrained by the @xmath11 , @xmath8 and @xmath10 measurements of cg98 . a log z = 2.0 kurucz spectrum was normalized so that the equivalent @xmath8 magnitude agreed with cool s @xmath8 magnitude , and the temperature was varied to give the best agreement at @xmath11 and @xmath10 . the best fit spectrum had a temperature of 22,000 @xmath0 7,000 k , where the error is 1@xmath13 . use of this technique means that errors in the continuum determination are dominated by errors in the photometry only ( normalization from a separate observation is valid because cg98 have shown that the star does not vary ) without incurring larger ( unknown ) errors by attempting a poorly constrained independent measurement . having determined the continuum level , the shape of the line for the nf must be determined . since both the neighboring turnoff star and diffuse light contributions have their own spectral components , estimates of these aperture contaminants are required before the decontaminated line profile can be determined ( the contribution of the neighboring turnoff star is most important because its relatively hot temperature implies a reasonably strong line ) . therefore , we still made an approximate determination of the telescope pointing during the fos observations . we compared the flux in the fos spectrum with the flux derived from a 0.43@xmath3 test aperture centered over the nf in the @xmath8 image from cg98 ( thus including all the aperture light , not just that from the nf ) . we first subtracted away from the original spectrum the earthshine component derived earlier ( @xmath8=21.7 ) , leaving only constant components . the equivalent @xmath8 magnitude of this `` constant '' fos component is 19.44 mag , in excellent agreement with the @xmath8 magnitude derived from the wfpc2 image ( 19.45 ) . since the nf does not vary and the pointing of _ hst _ was effectively constant during the fos observations ( no drifts above the @xmath10.001@xmath3 level ) , the contamination from the turnoff star in the fos aperture must be approximately the same as in the test wfpc2 aperture . using cool s photometry we decomposed the fos spectrum into its separate components of nf , turnoff star and diffuse light ( all fitted by kurucz atmospheres ) . the resulting residual is shown in figure [ fig.allnf ] and the close agreement with zero shows we made a reasonable , self consistent determination of the various aperture components . since the residual is so red , its contribution to the nf line profile should be negligible ( the source of this residual may be incomplete removal of the earthshine component mentioned earlier ) . as noted above , the high temperature and the broad line appear consistent with the hypothesis that this star is a high gravity object such as a wd . to test this hypothesis we used the pure hydrogen atmosphere models of wesemael et al . modern refinements of these models do not offer any significant advantages in the analysis of this low s / n ratio spectrum . the advantages of using these models are that they include detailed line profiles and cover a large range in log g ( 4 @xmath25 log g @xmath25 9 ) and ( 20,000 k @xmath25 @xmath25 150,000 k ) . since the only obvious line present ( in absorption ) is broader than the emission lines for cvs 14 , and because the s / n ratio in this line is low , we increased the length scale of the smoothing to 11 channels and applied the same smoothing factor to the models . we resampled the wesemael line profiles to the resolution of the fos data , applied smoothing and then experimented with different model profile fits ( lorentzians and voigt profiles gave almost identical results ) . to fit the line profile the icfit algorithm within iraf / splot was used to fit the continuum and a lorentzian was used to fit the line profile . the line depth and ew were found to be 0.55 @xmath0 0.05 and 23.9 @xmath0 2.4 respectively , where the errors are a combination of random errors in the parameter measurements ( determined by monte carlo experiments ) and systematic errors in the continuum and line profile measurements . for simplicity we defined each model line with two parameters , the line depth and the ew . we generated a complete grid of these two parameters for all of the available wesemael models , interpolated the log g = 1 spacing to log g = 0.25 and interpolated the temperature scale where necessary ( for some log g values fewer models were available ) . since the wesemael models are for @xmath27 20,000 k , we used the kurucz models ( with 4 @xmath25 log g @xmath25 5 and @xmath25 20,000 k ) to extend our line depth / ew grid below 20,000 k down to 10,000 k. ( encouragingly , excellent agreement was found between the wesemael and kurucz models in the overlap region of = 20,000 k , 4 @xmath25 log g @xmath25 5 ; differences of at most 2% were found in the two line parameters . ) we extrapolated the kurucz models to higher log g values ( from log g = 5 to log g = 9 ) , constrained by the functional form of the wesemael models for the line depth and ew at 20,000 k , using the same smoothing and interpolation as before ( see comments below about the validity of this extrapolation ) . finally , the measured line depth and ew were used to search for a solution in log g and . the optimal solution was found to be = 17,500 @xmath0 5,000 k and log g = 6.25 @xmath0 1.0 ( 1@xmath13 errors ) . the large errors in the gravity and temperature are caused by the limited information present in one noisy line , in particular the inability of the spectrum to trace possible narrow line cores . figure [ fig.contnf ] shows the total @xmath28 for the above solution as a function of and log g. the optimal solution with @xmath29 is marked ( `` l '' ) . the first , second and fourth contour levels correspond roughly to 1@xmath13 , 2@xmath13 and 3@xmath13 respectively . figure [ fig.contnf ] also shows the regions where the wesemael and kurucz models have been used , and where they were extrapolated . although extrapolation may incur extra uncertainties , figure [ fig.contnf ] shows that a wesemael model having = 20,000 and log g = 6.5 ( with @xmath30 ) is close to an optimal solution , without requiring any extrapolation . the kurucz models are useful because they show that a low gravity solution is probably not consistent with the data ( besides independently checking the low temperature , low gravity wesemael models ) . since the errors for log g and from the line measurement are considerable , we briefly discuss extra constraints on these two parameters . first , we note from figure [ fig.contnf ] that relative to the optimal solution ( `` l '' ) only high temperature / high gravity or low temperature / low gravity solutions are allowed . the high gravity solution with log g = 7.25 seems ruled out by the photometry of cool , sosin & king ( 1997 ) , since the nfs clearly have lower gravities than log g = 7 , the low gravity limit of the models used . the low gravity solution with log g = 5.25 and = 12,500 k has a temperature that is probably inconsistent with the determination from the photometry of cg98 ( 22,000 @xmath0 7,000 k ) . to determine the wavelength of the line we adopted a two - step procedure : ( 1 ) we used the lorentzian fit to the line as a first - order solution and then ( 2 ) used a spectral model ( template ) with fixed continuum , line depth and ew [ determined in ( 1 ) ] but with variable wavelength to refine the initial estimate . using template shifts of up to 10 in 0.1 steps we selected the shift which minimized the difference between the template and measured spectrum . this procedure is formally similar to a cross correlation , but it gives sub - pixel resolution without polynomial fitting and allows us to easily weight the first - order solution to optimize sensitivity to doppler shifts . using this procedure the line was determined to have a wavelength of 4865.8 , noticeably redder than the laboratory value of 4861.3 [ step ( 1 ) alone gave 4864.5 ] . the random error given by iraf / splot is only 0.4 , so we investigated the possibility of systematic effects . according to the _ hst _ data handbook , the overall 1@xmath13 random uncertainty is 0.7 ( or 43 ) for the accuracy with which the wavelength scale is known in an individual fos spectrum . however , the possibility of filter grating wheel ( fgw ) displacements means that a worst case disagreement in wavelength of over 4 is possible unless we have independent confirmation of the wavelength scale ( when the wavelength scale is fixed the overall 1@xmath13 random uncertainty falls to 0.5 or 31 ) . our only independent constraints on the wavelength scale are the g570h spectra of cv 4 ( obtained over 3 orbits ) , since no movement of the fgw was made between the observations of the nf and cv 4 . the wavelength of the line for cv 4 was determined to be 4861.8 . the 0.5 shift from the laboratory value is likely to consist of the velocity of the emission region in the binary , the cluster radial velocity of 21 ( 0.34 ) , orbital motion of the earth around the sun , and the motion of the telescope itself ( the last 2 effects being negligible given the resolution ) , plus systematic effects because of the fgw position . while cv emission lines are hardly ideal radial velocity standards in general , we believe that cv 4 provides a useful wavelength reference for several reasons : ( 1 ) the measured wavelength for is only 0.15 red - ward of the laboratory value , taking the cluster radial velocity into account , ( 2 ) the cv 4 wavelength measurement appears very stable since the 3 sub - observations for the 3 separate orbits ( separated in time by almost 2.5 hours , well over half the expected period of cv 4 ) give wavelengths of 4861.7 , 4861.8 , and 4862.0 , ( 3 ) the line is very symmetrical ( unlike cvs 1 and 2 ) so that the wavelength measurement is unlikely to have been skewed , and ( 4 ) the wavelength of h@xmath2 for cv 4 is 0.4 red - ward of the laboratory value , in excellent agreement with the 0.5 red shift for . also , two of the 3 strongest hei lines give consistent doppler shifts red - ward of the laboratory value : hei 4921 , shift = 0.35 @xmath0 0.45 and hei 5876 , shift = 0.32 @xmath0 0.21 . only hei 6678 gives a different shift , 0.72 @xmath0 0.25 relative to the laboratory value , however this line is over 1800 red - ward of and therefore has less value as a velocity reference . using the wavelength of for cv 4 as a reference , the doppler shift of the nf was measured to be 247 @xmath0 50 . to derive the error we added together in quadrature the random error ( 25 ) , the systematic instrumental error quoted above ( 31 ) and an estimate of other systematic errors , including the use of cv 4 as a velocity reference ( 0.5 or 31 ) . a closeup of the line profile is shown in figure [ fig.closenf ] . note , in particular , the vertical lines showing the difference in central wavelengths between the average nf and cv 4 lines . the significance of the apparent shift of for the nf is also clearly shown by comparing the two lorentzians ( the astrophysical significance of this 247 shift will be discussed in section [ sec.hewd ] ) . the variable flux but constant wavelength of the line for cv 4 is clearly shown by the 3 emission profiles ( the apparently constant velocity will be discussed in a future publication ) . one of the crucial advantages of studying cluster cvs , besides their known distance , is the opportunity to study cvs with different metallicities from those of field cvs ( see la dous , 1991 for examples of uv spectra of field cvs ) . although the bluest portion of the g160l spectrum is noisy , the detection of 2 1640 coupled with the non - detection of civ 1550 is probably a direct reflection of the low metallicity for ngc 6397 . the lack of obvious emission at mgii 2800 may be another measure of the low cluster metallicity , but we defer detailed spectral modeling incorporating the cluster s low metallicity to a future publication . we now discuss the significance of the uv distribution of cv 1 , the disk brightnesses of cvs 14 and their 2 line ratios . regarding the uv flux distribution , one obvious explanation for the redness of the disk explained in section [ sec.cv1 ] , given our sample size of one ( for g160l data ) , is an inclination effect . a second possible explanation is the relatively large contribution from the bright secondary . this explanation seems plausible in that : ( 1 ) our predicted period for cv 1 is 5.1 h , while the average period of the 8 dq hers in verbunt s sample is only 3.8 h ( also , popii secondaries tend to be brighter than popi secondaries at a given mass - stehle , kolb & ritter 1997 ) , and ( 2 ) if we replaced cv 1 by any one of the other cvs the flux distribution would be much bluer . for example , while the estimated secondary for cv 2 is @xmath11.3 mag fainter than the secondary for cv 1 , the disk in cv 2 is _ brighter _ ( in @xmath31 ) than the disk in cv 1 . if we then ignore the contribution of the secondary ( see figure [ fig.uv ] ) a simple interpretation of the flux distribution for cv 1 compared to the field cvs is that it has either : ( 1 ) a relatively faint and/or cool disk because of a low accretion rate , or ( 2 ) a wd with a moderately strong magnetic field . as noted earlier , am her systems ( with strong magnetic fields ) are not included in verbunt s sample . these systems , lacking disks , can be quite red , however the likely periods of the ngc 6397 cvs are longer than most am her systems ( see ritter & kolb 1998 ) . also , the line emission of the ngc 6397 cvs differs from field am her systems ( see below ) . to investigate ( 1 ) we compared @xmath23 ( disk ) and the cv periods from table 2 with figures 3.5 and 9.8 of warner ( 1995 ) , plots of @xmath23 ( disk ) versus orbital period for _ non_-magnetic field cvs . ( the two principal error sources for field cvs are estimates of the secondary component and distance estimates , errors that are considerably reduced by studying cluster cvs ) . clearly , the ngc 6397 cvs fall near the faint limit for cv disks when their expected period is considered . among field cvs only faint dne in quiescence have disks this faint , and these systems have much longer recurrence times for outbursts than brighter dne ( warner 1987 ) , possibly explaining the observed paucity of dn outbursts in globular clusters ( shara et al . generally cvs with periods above the period gap have disks that are brighter than @xmath32 . indeed , a large number of observational and theoretical studies have concluded that cvs with periods above the period gap generally have high accretion rates , with correspondingly bright disks while cvs with periods below the period gap have lower accretion rates with fainter disks ( patterson 1984 , di stefano & rappaport 1994 and stehle , kolb & ritter 1997 , the latter two studies specifically for popii cvs ) . another way to emphasize the unusually faint disks of the ngc 6397 cvs is to compare globular cluster and open cluster cvs ( minimizing uncertainties in @xmath23 ( disk ) estimates ) . there are now 6 probable cvs in globular clusters with well - measured @xmath33 magnitudes _ all _ lying at most @xmath10.2 magnitudes away from the ms in the @xmath8 vs @xmath9 cmd ( 4 in ngc 6397 and 2 in ngc 6752 ) . we contrast this result with the 3 cvs detected in open clusters that have quiescent @xmath9 colors blue - ward of the ms by @xmath34 mag ( m67 ; gilliland et al . 1991 ) , @xmath10.7 mag and @xmath11.0 mag ( both ngc 6791 ; kaluzny et al . since m67 and ngc 6791 are much less dense and dynamically evolved than ngc 6397 , their cvs are expected to have evolved from primordial binary systems , just as with field cvs . hence , the detection of a nova like and z cam ( relatively high accretion rate dn ) system in ngc 6791 is not surprising , since in any sample of field cvs these are among the brightest , intrinsically , because of their high accretion rate ( resulting in blue @xmath9 colors ) . the much lower metallicity of ngc 6397 compared to metal - rich field or open cluster cvs should , in general , mean that the ngc 6397 cvs have _ higher _ accretion rates ( for given binary parameters ) , according to stehle , kolb & ritter ( 1997 ) , implying that even brighter disks should be present . a possible explanation we have already suggested ( gc95 ) , is that these cvs are mostly magnetic systems with truncated and thus relatively faint disks . our original suggestion was based on the relative strength of 2 4686 compared to for cvs 13 . to summarise the results in section [ sec.he2 ] , for the am her systems ( with their strong magnetic fields ) , a strong 2 4686 line is a well known feature ( warner 1995 ) . however , known dq hers show a large range in 2 4686/ line ratios with values ranging from over one ( v533 her ) to zero ( ae aqr ) . to discriminate between magnetic and non - magnetic systems ( and eliminate high accretion systems like nova likes ) , silber ( 1992 ) has shown that line ratios of 2 4686/ @xmath27 0.4 and ew(h@xmath35 20 are reasonable signatures of magnetic systems . by combining our knowledge of @xmath23 ( disk ) and the 2 line ratios of cvs 14 , we can attempt identification of these systems . with its weak 2line and faint disk , cv 4 is a reasonable candidate for a quiescent dn system . since there is remarkable similarity between the 24686/ line ratios for cvs 13 and also similarities in the balmer line fluxes themselves ( see table 1 ) , cvs 13 may have very similar properties , as originally suggested by gc95 . they do not appear to be recent old novae or nova likes because of their faint disks ( with extra evidence from their 2 4686 line ratios ) , nor do they appear to be dne because they have moderately strong 2 lines . the final option is magnetic systems . cvs 13 do not have 2 ratios as large as am her systems , but while their ratios are also weaker than an `` average '' dq her system , there is considerable scatter for the dq her systems , as pointed out above . for example , cvs 13 have 2 ratios stronger than 4 of the 8 dq her systems in the sample of williams ( 1983 ) , and cvs 1 and 2 only just fail the magnetic criteria of silber ( 1992 ) . to conclude , cvs 13 do not appear to be dne , but they could be dq her type systems . is it true that dq her systems tend to have disks as faint as those found in cvs 14 ? extra information about dq her disks compared to other cv classes comes from the continuum slopes in williams ( 1983 ) . an estimate of the disk contribution ( and temperature ) relative to the secondary comes from analyzing the continuum ratio between two different wavelengths ( equivalent to a color ) . the two most convenient continuum points for both cvs 14 and the spectra of williams ( 1983 ) are at and h@xmath2 . figure [ fig.he2disk ] shows plots of this continuum ratio versus the 2/ line ratio for several different classes of field cv ( from williams 1983 , updated by ritter & kolb 1998 ) along with these ratios for cvs 14 . we examined the linear correlation between the continuum and line ratios for the individual cv classes shown , finding a linear correlation with absolute value @xmath27 0.5 in 3 cases : dq her systems ( linear correlation = 0.77 ) , nova likes ( 0.69 ) and old novae ( 0.57 ) . the correlation for dq hers implies that small 2 line ratios imply continuum ratios of @xmath11.0 ( meaning that the secondary dominates unless the disk is very cool ) . a notable element of figure [ fig.he2disk ] is that all of cvs 14 lie close to the best fit line for dq hers , and all of them have continuum ratios of @xmath11.0 . the dominance of the secondary for cvs 14 is therefore exactly what we expect for field dq her systems with weak 2 lines . using estimates of @xmath8 ( disk ) and distances for dq her systems ( from patterson 1994 ) we have estimated @xmath23 ( disk ) for dq hers , neglecting reddening ( the @xmath36 errors in the distances dominate the errors ) and assuming that the systems are in their faint , low accretion state . we found a strong linear correlation ( 0.83 ) between the 2/ line ratio and @xmath23 ( disk ) , as shown in the upper panel of figure [ fig.he2disk ] , in the sense that higher 2/ implies a brighter disk ( consistent with the bluer colors given above ) . we then used this correlation to _ predict _ @xmath23 ( disk ) using the measured 2/ line ratios for cvs 14 given in section [ sec.cv4 ] . the results are shown in the final column of table 2 , and agree nicely with the @xmath23 ( disk ) values given in table 2 for cvs 14 , especially given the large uncertainties in @xmath23 for the field cvs . ( we excluded gk per from the sample , with its @xmath12 day period , evolved companion and thus high accretion rate . including gk per gave a linear correlation of 0.59 between @xmath23 ( disk ) and 2/ and @xmath23 ( disk ) values @xmath10.6 mag brighter . ) these results are the best evidence that these cluster cvs are indeed largely dq her type systems . a possible alternative to the dq her hypothesis is that some of the 6397 cvs are old novae ( possibly in deep hibernation between outbursts ; see shara et al . this explanation is consistent with the correlation noted above between the continuum ratios and 2/ line ratios for old novae , plus the proximity of the 6397 cvs to the best - fit line for old novae . although the hibernation scenario is difficult to test ( and yet to be confirmed ) , upcoming _ hst _ spectra of the probable old nova in m80 will provide a critical comparison with the spectra of the ngc 6397 cvs . it is also possible that some of the cluster cvs fall in both categories , since , for example , 3 of the dq hers in the sample from williams ( 1983 ) are also old novae . the photometry of the nfs ( cg98 ) and the spectroscopy presented above place important constraints on the possible properties and evolution of these systems , which appear to be helium wds . from the spectroscopy we have already determined and log g and from the photometry we calculated the luminosity of the nf ( log(@xmath37 ) using the bolometric corrections presented in bergeron , wesemael & beauchamp ( 1995 ) . armed with these three quantities we have examined evolutionary models of wds to estimate masses and lifetimes for the nfs . we began by using the wd evolutionary models presented by bergeron et al . models with log g = 7 ( the limit of bergeron s models ) , = 17,500 k and with thick hydrogen layers have @xmath38 . extrapolating to log g = 6.25 , models should have @xmath39 . using the study by webbink ( 1975 ) of the evolution of helium wds in close binaries , we used the measured log(@xmath40 ) and to derive a mass between 0.2 and 0.25 for the nf , with cooling ages for these models between @xmath41 yr ( 0.25 ) and @xmath42 yr ( 0.20 ) . using log g , and log(@xmath40 ) , a similar mass is determined from the he wd study of althaus & benvenuto ( 1997 ) , although they do not include a hydrogen envelope . since the only significant line present in the nf is , a hydrogen envelope must be present , and therefore we have made comparisons with the models of benvenuto & althaus ( 1998 ) on the effects of hydrogen envelopes on the structure and evolution of low and intermediate they computed the evolution of wds with masses from 0.15 0.5 , while treating the mass of the hydrogen envelope as a free parameter within the range @xmath43 . because only a representative sample of the models are presented in benvenuto & althaus ( 1998 ) , l. althaus has kindly performed a dedicated search for models consistent with our log g and . two good solutions were found for a 0.24 model with @xmath44 and a 0.235 model with @xmath45 , with both of these solutions having log(@xmath46 , in excellent agreement with the measured luminosity . these solutions are close to the valid bright limit of the models of benvenuto & althaus ( 1998 ) , so the cooling age for these solutions of @xmath47 yr is highly uncertain ( and likely to be an underestimate ) since benvenuto & althaus ( 1998 ) did not attempt a detailed treatment of the formation of helium wds in binary systems . a detailed treatment of the evolution of a 0.3 wd in a binary system was given by iben & tutukov ( 1986 ; hereafter it86 ) . their model experiences two hydrogen shell flashes before cooling to @xmath1 16,000 k at an age of @xmath48 yr ( a comparable age to that found by benvenuto & althaus ( 1998 ) , but smaller than webbink s values ) . using their cooling curve , it86 construct a number luminosity distribution function for helium wds . this function has a similar slope to that for co wds , but at most magnitudes is about a factor of 4 smaller . one exception to this behavior is the region with @xmath49log@xmath50 . at log@xmath51 the predicted helium wd number - luminosity function shows a `` bump '' where it increases to be roughly equal to the co wd function . it then drops quickly with decreasing luminosity by about an order of magnitude before returning to the average value at log@xmath52 . this behavior is caused by the very slow rate of decline in luminosity following the first hydrogen shell flash , when t = @xmath53 yr . although it86 warn that a complete theoretical distribution function requires superposition of contributions from wds of many different masses ( each experiencing a hydrogen shell flash at a different luminosity ) , this oscillatory behavior may cause an enhancement in the number of helium wds at relatively high luminosity and a dearth for @xmath11.5 mag below this . this could be consistent with the observed cmd distribution in cg98 , perhaps explaining the lack of obvious low - mass wds further down a cooling sequence , particularly since log@xmath51 is not far from the measured luminosities for the two faintest helium wd candidates ( eventually close double degenerate systems should merge after a few gyr , forming a more massive wd ) . however , clearly more data from other clusters are needed to test this hypothesis . an obvious question remains : are the possible formation mechanisms listed in section [ sec.int ] consistent with the likely significant red shift of the line ? we reject the possibility of the red shift being an ejection velocity , since at 247 , the star would have moved 2.2 pc in only @xmath54 yr , and thus would be well outside the core . gravitational red shift is likely to contribute only a small red shift for this low - mass object , since @xmath55 ( reid 1996 ) , and using the mass and radius estimates from benvenuto and althaus models , @xmath56 3 . instead , we argue that the red shift may be a doppler shift from the helium wd orbiting a massive but faint companion , probably a wd . this is consistent with the primary formation mechanism listed above . also , a binary nature for this star would hardly be surprising for dynamical reasons alone , since the detection of 3 nfs near the center of ngc 6397 is good evidence of a binary origin for these stars ( through mass segregation ) , and most known low - mass wds are in binary systems . finally , we note the possibility that these low - mass wds may have neutron star companions , likely to be binary millisecond pulsars ( although none have been found yet in ngc 6397 ) as found in helium wd / ns binary systems in the field . if we have measured an orbital doppler shift its size is an important consideration . of the 8 double degenerate systems listed in iben et al . ( 1997 ) , all with a helium wd as the primary ( the brighter component ) , only 2 have @xmath57 . little is known about the masses of the secondaries in these systems , but models by iben et al . ( 1997 ) show that @xmath58 should be quite common for high - inclination systems with a helium wd primary and a co wd secondary ( from cool , piotto & king 1996 and cg98 , ngc 6397 clearly has a large reservoir of @xmath59 co wds , and it should also have many higher mass wds ) . for more massive secondaries such as neutron stars , larger values of @xmath60 can be found , for example @xmath61 for the helium wd msp binary j1012 + 5307 ( van kerkwijk , bergeron & kulkarni 1996 ) . adopting a range of possible mass functions for a binary system containing the nf and a unknown companion , and assuming @xmath60 = 247 and helium wd mass = 0.25 , we have derived expected periods for the system as a function of orbital inclination ( @xmath62 ) . although the nf observations were spread over 3 ( 96 minute ) orbits , @xmath142 min of spectra were taken on the middle orbit , but only @xmath110 min at the end of the first orbit and @xmath114 min at the start of the third . therefore , we can not rule out a @xmath14 - 5 h period , and for @xmath63 we require a wd companion with mass @xmath64 0.7 , slightly higher than the mass of wds currently being produced in the cluster ( low - mass ms stars are ruled out , consistent with the photometry of cg98 ) . since we have measured only a lower limit for @xmath60 , we have also experimented with values that are 25% higher than observed , maintaining the other assumptions . in this case a high mass wd ( mass @xmath64 1.1 ) or neutron star companion is required ( clearly longer observations may provide powerful constraints on the mass of the companion ) . to determine whether 3 systems with ages of @xmath65 yr ( it86 and webbink 1975 ) are likely to be present in ngc 6397 , we used expected encounter rates between various combinations of ms stars , wds , neutron stars and red giants ( davies & benz 1995 ) to make rough estimates of expected numbers of binaries or merger products . the calculations by davies & benz ( 1995 ) are specifically for @xmath66 cen and 47 tuc - we used their numbers for the denser cluster 47 tuc and divided the interaction rates by 5 to account for the smaller mass of ngc 6397 ( using cluster masses from pryor & meylan , 1993 ) . assuming a lifetime of 4 gyr ( sandquist , bolte , & hernquist 1997 ) for blue stragglers less than @xmath11.5 mag brighter than the ms turnoff ( where most of the observed blue stragglers are found ) and assuming all ms star collisions result in mergers , we used the calculations of davies & benz ( 1995 ) to estimate that 75 blue stragglers should currently be found in ngc 6397 ( from both 2- and 3-body interactions ) , comparing favorably with the 54 blue straggler candidates found by kaluzny ( 1997 ) . an overestimate of total blue stragglers is not surprising because merged stars with total masses at or below the turnoff are , of course , not included in kaluzny s sample . for cvs , we assumed that half the collisions between wds and ms stars result in the formation of binaries ( davies & benz 1995 ) , and used a conservative limit on the mass ratio of 1 for stable mass transfer ( davies & benz 1995 ) . then , assuming the average cv lifetime is 1.5 gyr for cvs above the period gap ( kolb & stehle 1996 ) , we expect 6 cvs from 2- and 3-body interactions , close to the observed number . finally , assuming ( 1 ) ages of 0.1 0.5 gyr for the observed helium wds , ( 2 ) that all of the red giant collisions with either neutron stars or wds result in the formation of he wds ( see davies , benz & hills 1991 ) , and ( 3 ) that 3-body interactions result in as many helium wds as 2-body interactions , we predict 0.7 3.5 bright helium wds to currently be found in ngc 6397 . the reasonable agreement between observed and estimated numbers of blue stragglers and cvs is comforting given the large uncertainties in the interaction rates used , the influence of possible _ differences _ in interaction rates between ngc 6397 and 47 tuc ( e.g. ngc 6397 is core - collapsed and 47 tuc may or may not be ) , and the effect of the unknown binary content ( davies & benz 1995 assume 10% binaries ) . since the lifetimes for blue stragglers are known reasonably well and the number of observed systems is probably fairly complete , the possible differences listed above must either be small or cancel each other out . we also have reasonable agreement between the numbers of expected and observed helium wds when using webbink s cooling ages . although our assumptions may appear somewhat generous , there may be other mechanisms for the formation of he wds : for example , red giant ms star collisions , binary binary collisions and primordial binary evolution . it is the expected lifetimes for the helium wds that are one of the largest sources of uncertainty in the above number estimates . the ( unknown ) thickness of the hydrogen envelope can make a significant difference to the lifetimes especially if hydrogen burning occurs . for example , a model with @xmath67 and @xmath68 ( the thickest envelope considered by benvenuto & althaus ( 1998 ) , with log g and consistent with our low - mass nf within the errors ) has an age @xmath150% greater than the 0.3 model with @xmath44 . it has also been suggested ( alberts et al . 1996 ) that the cooling time scales of very low - mass wds ( mass @xmath25 0.25 ) can be considerably underestimated by the traditional wd cooling curves of it86 and others for higher mass wds ( mass @xmath27 0.3 ) . alberts et al . ( 1996 ) predict that hydrogen shell flashes do not occur on wds with mass @xmath25 0.2 , but that these wds experience long lasting phases of hydrogen burning which significantly prolong their lifetimes ( however this difference in behavior may be caused by differences in time resolution between the models of alberts et al . and it86 ) . finally , sarna , antipova & muslimov ( 1998 ) find much greater cooling ages for their 0.16 wd compared to it86 because of differences in the formation mechanisms . while in it86 the donor star fills its roche lobe when it is on the red giant branch , forming a @xmath10.3 helium wd with a relatively thin hydrogen envelope , the donor star in sarna et al.s calculations fills its roche lobe while it is evolving through the hertzsprung gap , resulting in a 0.16 wd with a much thicker hydrogen envelope . since bright red giants have much shorter lifetimes than subgiants ( and perhaps limited cross - sections , despite larger radii , because of low densities ) it is possible that collisions involving subgiants or faint red giants are much more efficient at producing helium wds than collisions involving brighter red giants . this selection effect would preferentially cause very low - mass ( @xmath40.25 ) wds to form ( since the red giant core mass increases with increasing brightness ) , giving objects with longer cooling times according to webbink ( 1975 ) and sarna et al . ( 1998 ) , and enhancing the number of helium wds seen . we summarize here the results for cvs 14 : ( 1 ) a 4th cv candidate in ngc 6397 has been spectroscopically confirmed , ( 2 ) uv data for cv 1 implies that it has a red disk when compared with field cvs , ( 3 ) the photometry of cg98 combined with kurucz spectra for the estimated secondaries provide strong evidence that cvs 14 all have faint disks and probably low accretion rates ( consistent with faint quiescent dne ) , ( 4 ) the 24686/ line ratios ( together with the faint disks ) imply that cvs 13 may be dq her systems , ( 5 ) the correlations between the 2/ line ratios for cvs 14 and both ( a ) the continuum ratios between and h@xmath2 and ( b ) @xmath23 ( disk ) provide extra evidence that the 6397 cvs are mainly dq her systems . this is consistent with the finding of verbunt et al . ( 1997 ) that cvs 13 could be dq her systems based on their x - ray and optical fluxes . an alternative explanation is that some of the cvs are old novae in hibernating phases between nova eruptions . we conclude that there may be fundamental differences between populations of globular cluster cvs and field / open cluster cvs , perhaps caused by the different formation mechanism of tidal capture and exchange collisions or perhaps because of the different environment in globular clusters . one possible explanation is that interactions cause stars to rotate more quickly , resulting in stronger magnetic fields than in most field stars , as suggested by gc95 . alternatively , vandenberg , larson & de propris ( 1998 ) have suggested that rapidly rotating cores of giant stars in the metal poor globular cluster m30 might reconcile this cluster s luminosity function with stellar evolutionary theory . similar problems exist in understanding the luminosity function of ngc 6397 , although further study of the theory of vandenberg et al . ( 1998 ) is needed . another possibility is that magnetic wds are formed in globulars from differentially rotating cores in blue stragglers ( grindlay 1996 ) . prospects for further work on the cv data include modeling of the disk for cv 1 , power spectrum analysis of both the time series obtained by cg98 and the sub - exposures for the fos spectra , studies of line profile changes with time and detailed spectral modeling incorporating the cluster s low metallicity . a clear test of the hypothesis that most of the 4 cvs are dq her systems is to search for a stable optical ( or x - ray ) period with @xmath69 ( dq hers usually have @xmath70 ; patterson 1994 ) . because the short fos observations of the cvs are inadequate for this purpose , we shall propose to obtain simultaneous spectra of cvs 1 and 2 using moderate time - resolution ( @xmath71 60s ) spectroscopy in the blue ( using stis with a long slit ) , to directly test the magnetic cv hypothesis and place constraints on the hibernating nova scenario . the results for the nf are : ( 1 ) using detailed comparisons with stellar atmospheres from wesemael et al . ( 1980 ) and kurucz ( 1993 ) we have determined log g = 6.25 @xmath0 1.0 , and = 17,500 @xmath0 5,000 k ( consistent with = 22,000 @xmath0 7,000 k using the photometry of cg98 ) , ( 2 ) by using these line parameters and the luminosity of the nf we have shown that the nf spectrum is consistent with a helium wd having a mass of @xmath10.25 and an age between 0.1 and 0.5 gyr ( depending on the models used ) , and ( 3 ) the nf spectrum appears to be significantly doppler shifted from the expected wavelength , suggesting the presence of a dark , massive companion . the low mass of the nf ( and probably similar or lower masses for the others ) suggest that interactions between degenerate stars and subgiants or faint red giants are more efficient at producing helium wds than interactions involving degenerate stars and brighter red giants . although we have not yet made a rigorous attempt to find evidence for velocity variability of the line for the nf , the prospects from subdividing this low s / n spectrum are poor , especially since almost two thirds of the data for the nf was obtained over just one @xmath142 min time segment . clearly , observations over a longer time are needed to confirm that doppler shift evidence presented above and to study radial velocity variations . use of stis with the long slit would enable two nfs to be observed simultaneously , along with many cluster stars providing an ideal radial velocity reference . observations in the blue would also give excellent coverage of balmer absorption lines ( with the exception of h@xmath2 ) , giving considerably more accurate line parameters , and helping determine whether the luminosity difference between the bright nf and the two fainter ones is mainly because of mass or age differences . we thank r. kurucz , l. althaus and o. benvenuto for models , b. hansen , r. di stefano and f. wesemael for discussions and an anonymous referee for helpful comments . this work was partially supported by nasa grants nag5 - 3808 and hst grant go-06742 ( pde and jeg ) , and nasa grant nag5 - 6404 ( cdb ) . & 4687.04 @xmath0 0.59 & 1.26 @xmath0 0.11 & 5.9 @xmath0 0.5 & 21.0 @xmath0 2.0 & 0.0 @xmath0 2.7 h@xmath72 & 4858.09 @xmath0 0.21 & 3.92 @xmath0 0.16 & 18.0 @xmath0 0.8 & 3.8 @xmath0 5.7 & 15.6 @xmath0 1.6 he i & 5872.91 @xmath0 0.45 & 1.51 @xmath0 0.14 & 6.9 @xmath0 0.7 & 21.6 @xmath0 2.2 & 0.0 @xmath0 2.7 h@xmath2 & 6559.19 @xmath0 0.30 & 4.40 @xmath0 0.18 & 21.0 @xmath0 0.9 & 24.0 @xmath0 1.2 & 0.0 @xmath0 1.6 he ii & 4688.37 @xmath0 0.86 & 1.44 @xmath0 0.16 & 11.2 @xmath0 1.3 & 13.9 @xmath0 12.0 & 23.6 @xmath0 7.1 h@xmath72 & 4862.47 @xmath0 0.17 & 4.19 @xmath0 0.15 & 31.9 @xmath0 1.2 & 20.8 @xmath0 2.0 & 11.1 @xmath0 2.1 he i & 5873.98 @xmath0 0.53 & 1.04 @xmath0 0.10 & 7.6 @xmath0 0.8 & 21.7 @xmath0 2.8 & 0.0 @xmath0 3.2 h@xmath2 & 6563.60 @xmath0 0.39 & 3.94 @xmath0 0.16 & 30.2 @xmath0 1.3 & 30.7 @xmath0 1.5 & 2.1 @xmath0 2.1 he ii & 4685.71 @xmath0 0.81 & 0.53 @xmath0 0.05 & 15.2 @xmath0 1.5 & 26.2 @xmath0 5.5 & 10.6 @xmath0 7.3 h@xmath72 & 4860.73 @xmath0 0.12 & 2.07 @xmath0 0.05 & 59.0 @xmath0 1.6 & 13.8 @xmath0 0.9 & 10.5 @xmath0 0.8 he i & 5874.59 @xmath0 0.37 & 0.53 @xmath0 0.03 & 13.6 @xmath0 1.0 & 18.6 @xmath0 1.5 & 0.0 @xmath0 1.6 h@xmath2 & 6562.94 @xmath0 0.24 & 2.81 @xmath0 0.09 & 71.7 @xmath0 2.8 & 3.1 @xmath0 4.1 & 20.2 @xmath0 1.4 he ii & 4685.60 @xmath0 0.67 & 0.20 @xmath0 0.03 & 12.3 @xmath0 1.8 & 21.0 @xmath0 3.6 & 0.0 @xmath0 4.1 h@xmath72 & 4861.84 @xmath0 0.05 & 2.73 @xmath0 0.03 & 158.0 @xmath0 2.7 & 9.4 @xmath0 0.5 & 8.1 @xmath0 0.4 he i & 5874.78 @xmath0 0.18 & 0.79 @xmath0 0.03 & 29.3 @xmath0 1.4 & 15.0 @xmath0 0.8 & 0.0 @xmath0 1.0 h@xmath2 & 6563.27 @xmath0 0.06 & 3.39 @xmath0 0.04 & 106.7 @xmath0 1.7 & 11.9 @xmath0 0.3 & 6.2 @xmath0 0.4 cv 1 & 5.95 & 5.93 & 0.68 & 5.1 & & 8.5 & 7.6 & 8.5 cv 2 & 7.13 & 7.24 & 0.57 & 4.4 & 9.7 & 8.7 & 8.2 & 8.5 cv 3 & 7.84 & 8.18 & 0.49 & 3.8 & 9.3 & 8.9 & 8.8 & 8.7 cv 4 & 8.41 & 8.53 & 0.45 & 3.6 & 10.9 & 9.9 & 9.9 & 9.3

i discuss several physics issues that can be addressed through the present and future program of parity - violating electron scattering measurements . in particular , i focus on strange quark form factors , hadronic effects in electroweak radiative corrections , and physics beyond the standard model . # 1#1 # 1#1 # 1#2 # 1 # 1 # 1#1 # 1^#1 # 1#1 # 1#1 # 1#1 # 1_phys . rev . _ * # 1 * # 1_phys . rev . _ * c#1 * # 1_phys . rev . _ * d#1 * # 1_phys . rev . lett . _ * # 1 * # 1_nucl . phys . _ * a#1 * # 1_nucl . phys . _ * b#1 * # 1_ann . of phys . _ * # 1 * # 1_phys . reports _ * # 1 * # 1_phys . lett . _ * b#1 * # 1_z . fr phys . _ * a#1 * # 1_z . fr phys . _ * c#1 *

nucl-th0302049

parity - violating electron scattering ( pves ) has become an established component of the nuclear physics research program at a variety of accelerators . prior to the late 1990 s , the field was considered somewhat esoteric , with the completion of roughly one experiment per decade : the deep inelastic @xmath1 experiment at slac in the 1970s@xcite ; the quasielastic beryllium experiment at mainz in the 1980s@xcite ; and the elastic carbon measurement culminating in a publication in 1990@xcite . in the last few years , the rate at which new results are being reported has accelerated dramatically compared that of the past 30 years , and in the next several years , we anticipate even more results from the sample collaboration at mit - bates@xcite , the happex@xcite , g0@xcite , pavex@xcite , and q - weak@xcite collaborations at jefferson lab , the a4 experiment at mainz@xcite , and the e158 experiment at slac@xcite . looking further down the road , the prospective up - grade of cebaf may open up additional possibilities for pves studies beyond those currently on the books . suffice it to say , pves has come into its own as an important tool for the study of a broad range of questions in nuclear and particle physics , moving far beyond its earlier incarnation as an almost exotic area of physics ( for a review , see ref . @xcite ) . at the end of the day , what physics do we hope to have accomplished with this tool , and what important questions do we hope the field will have answered ? in this talk , i would like to give my view . at the same time , i will discuss some recent developments that have opened up new directions for study of both hadron structure and particle physics . of course , this theorist s perspective will be weighted in favor of new theory , but i want to emphasize that these theoretical developments would have only academic interest were it not for the substantial progress on the experimental side . in fact , one of the attractive features of the pves community is the close interplay between theory and experiment . the proper interpretation of the very tiny asymmetries measured with pves requires careful theoretical delineation of various contributions . conversely , the judicious choice of target and kinematics , together with hard - nosed experimental study of the systematic effects that could generate false " asymmetries , provide theorists with a clean and unambiguous probe of nucleon structure and the weak interaction . this kind of syngergy between theory and experiment is a hallmark of our field . the present burst of activity in the field has been motivated by a desire to probe the strange quark content of the nucleon . understanding strange quarks is important from a number of perspectives : _ flavor decomposition of nucleon properties_. how much do the different flavors of light quarks contribute to the low - energy properties of the nucleon ? for many years , the standard lore was that up- and down - quark degrees of freedom were sufficient to account for the most familiar aspects of the nucleon , such as its mass , spin , and magnetic moment . analyses of the @xmath2n @xmath3-term suggested that light - quarks were not , in fact , the whole story , and that possibly 20 - 30% of @xmath4 came from @xmath5 pairs@xcite . later , polarized deep inelastic scattering ( dis ) measurements implied that only 30% of the nucleon spin came from quark spin , with the remainder being supplied by quark orbital angular momentum and by gluons@xcite . part of the picture associated with the spin - crisis " was that the fraction of nucleon spin coming from @xmath5 pairs was about 10% , and that these pairs were polarized opposite to the direction of the total spin . in the quark model framework , this idea is a complete mystery . as a result , one would like to know what other nucleon properties are substantially affected by strangeness . hence , the current program of measurements aimed at the strange quark contributions to nucleon electromagnetic properties . the pves strange quark program offers one advantage over other flavor decomposition studies , namely , the theoretically clean character of interpretation . obtaining the @xmath5 contribution to the nucleon mass requires extrapolation of @xmath2n scattering amplitudes into the unphysical regime using a combination of dispersion relations and chiral perturbation theory ( chpt ) . the value for the @xmath6 relative to @xmath7 has varied somewhat over the years , in part due to the ambiguities associated with this extrapolation . the situation regarding the strange quark contribution to the nucleon spin , @xmath8 , is even more ambiguous . the extraction of @xmath8 from polarized dis measurements relies on the assumption of flavor su(3 ) symmetry . the relevant matrix element is not measured directly in the experiments , and information on neutron @xmath9-decay and hyperon semileptonic decays must be incorporated into the analysis using su(3 ) symmetry . for some time , people have worried about the uncertainties associated with su(3)-breaking . recently , my collaborators and i studied these effects in chpt and found them to be potentially serious@xcite . specifically , one has @xmath10 -\frac{5\sqrt{3}}{6 } g_a^8 \\ \nonumber & = & 0.14 - [ 0.12 + 0.25 + 0.10 ] \ \ \ , \end{aligned}\ ] ] where the @xmath11 are the integrals of @xmath12 taken from experiment and @xmath13 is the eighth component of the octet of axial current matrix elements . a value for the latter can only be obtained from experimentally measured axial current matrix elements by invoking su(3 ) symmetry . the chiral expansion of this su(3 ) relationship is shown in the square brackets in the second line of eq . ( [ eq : deltas ] ) ( the first term is the experimental value for the sum of @xmath14 integrals ) . the first , second , and third terms represent the @xmath15 , @xmath16 , and @xmath17 contributions , respectively . it is a manifestly non - converging series . based on the formula , we have no way of estimating the theoretical su(3)-breaking uncertainty in @xmath8 , nor can we really say what the value of @xmath8 is . the advantage of pves is that the interpretation of the results in terms of strange quarks does not suffer from these kinds of ambiguities . once we have sufficiently precise measurements of the proton and neutron electromagnetic and weak neutral current form factors , we know the relative contributions of @xmath18 , @xmath19 , and @xmath20-quarks to the nucleon vector current matrix elements . the relative importance of the strange quark contribution may be either large or small , but either way , the answer will be clear . we will know the flavor decomposition of the nucleon vector current matrix elements and are , thus , negligible@xcite . ] . _ sea quarks and the quark model_. why does the constituent quark model work so well in describing low - energy nucleon properties ? in light of what we know about the presence of sea quarks and gluons in the nucleon , the success of the constituent quark description of the nucleon remains a puzzle . a variety of solutions have been proposed . it may be that sea quarks are simply inert `` when viewed over long - distance scales . a more interesting possibility is that the sea quarks are not inert but that their effects are hidden in the effective parameters of the quark model , such as the constituent quark mass and the string tension '' of the @xmath21 potential . it may also be that nature is more devious than either of these other possibilities suggest and has conspired to conceal the effects of sea quark dynamics through a series of cancellations . the latter view seems to emerge from dispserion relation analyses of isoscalar electromagnetic and strange quark form factors , as i discuss below . in any case , the beauty of the pves measurements is that they probe the @xmath5 sea over distance scales relevant to the quark model description . if either the effective quark model parameters or nature itself has hidden sea quark effects , the measurements will lift the veil of secrecy and allow us to see what the sea is up to at low - energies . _ z - graphs vs. loops_. how do we relate the quark model description of the nucleon with one built out of hadronic degrees of freedom ? in the latter framework , for example , the low - momentum , isovector electromagnetic properties of the nucleon involve a combination of the pion cloud and @xmath22-meson resonance . at the most naive level , one may think of the @xmath22-meson resonance as a z - graph " effect and the pion cloud as involving disconnected quark loops ( see fig . [ fig1 ] ) . in the @xmath23-channel , the @xmath24 coupling involves @xmath25 pair creation , and in order to connect to a nucleon , it must also involve the negative frequency part of the constituent quark propagator ( the z - graph ) . in the same channel , the pion cloud contains a @xmath26 intermediate state . at the quark level , the non - resonant part of this contribution involves the presence of two @xmath27 pairs . with only three constituent quarks in the nucleon , this effect looks like a disconnected quark loop a sea quark effect . on the other hand , in the non - relativistic quark model ( nrqm ) , neither z - graphs or loops are present , and yet one can still reproduce the low - momentum isovector properties of the nucleon . how can this be ? the strange quark contributions to the vector current form factors necessarily involve disconnected quark loops , since there are no valence strange quarks in the constituent quark model nucleon wavefunction . in a hadronic langauge , these loops look , for example , like @xmath28 pairs . dispersion relations also suggest that this kaon cloud is actually dominated by the @xmath29 resonance a z - graph effect that would imply more sizeable strangeness form factors than one gets from loops only . if this picture is confirmed by experiment , then it will force us to think more carefully about how to relate the quark and hadronic descriptions of the nucleon . _ perturbation theory may not apply_. the mass of the strange quark forces us into a theoretical no man s land where we can not easily apply our usual perturbative techniques for making predictions . the strange quark is not heavy enough to use heavy quark effective theory ; since @xmath30 , we have no small expansion parameter as we do for heavier quarks . at the same time , the strange quark may not be light enoughto make su(3 ) chiral perturbation theory applicable ; @xmath31 not a particularly small expansion parameter either . thus , from the theoretical side , the study of strange quarks presents a particularly challenging situation . _ symmetry is impotent_. because the strange quark vector current contains a component that is a flavor su(3 ) singlet , we can not use su(3 ) symmetry to relate the strangeness form factors to any other presently measured quantities . specifically@xcite @xmath32 where @xmath33 is the baryon number , or su(3 ) singlet current , and @xmath34 is the isoscalar electromagnetic ( em ) current . we know quite a bit about the latter , but the pves measurements will , in fact , give us our first information on the form factors associated with @xmath33 . this means that , even if the su(3 ) chiral expansion were well - behaved , there would appear counterterms or low - energy constants ( lec s ) that we can not obtain from other experiments . consequently , we can not in general use chpt to predict @xmath35 and @xmath36 , the strangeness electric and magnetic form factors , respectively . clearly , the study of these form factors will lead to new insights into nucleon structure , and pves provides us with a theoretically clean probe of this physics . the theoretical challenge is to translate the results of pves measurements into a clear understanding of sea quark dynamics at distance scales for which traditional methods of treating nucleon structure run into trouble . generally speaking , theorists have taken one of three approaches to the study of @xmath35 and @xmath36 : hadron models , dispersion relations , and lattice qcd . of these , the latter two represent first principles methods at the hadronic and microscopic levels , respectively . models have their place , but they necessarily involve assumptions and approximations that are not always clearly identified or controlled . for this reason , we have witnessed a substantial number of model predictions for @xmath35 and @xmath36 and a correspondingly wide range of values . should experiment agree with any subset of these calculations , it will be hard to gauge the significance of this agreement . for this reason , my bias favors the first principles treatments , even if they can not yet be implemented in a completely successful way . although as i emphasized above chpt can not provide predictions for @xmath35 and @xmath36 , it nevertheless provides a convenient framework in which to discuss the physics coming out of other non - perturbative methods . for concreteness , i will focus on @xmath36 at small @xmath37 . in this kinematic domain , one has an expansion @xmath38 the strangeness magnetic moment @xmath39 is given by@xcite @xmath40 where @xmath41 is an _ a priori _ unknown low - energy constant that governs the lowest order [ @xmath16 ] contribution and the @xmath43 denote higher order contributions arising from loops and subleading operators . as noted earlier , one can not use existing measurements to predict @xmath39 because @xmath41 contains an su(3 ) singlet component . in contrast , the lowest order [ @xmath17 ] contribution to the magnetic radius arises entirely from chiral loops@xcite : @xmath45 where @xmath46 and @xmath47 are the usual su(3 ) reduced matrix elements and the @xmath43 denote higher order contributions . since @xmath46 and @xmath47 are known from other experiments , to @xmath48 we have a parameter - free prediction for the strange magnetic radius . this result has been used to extrapolate the sample determination of @xmath36 at @xmath49 ( gev@xmath50 to the photon point in order to obtain results for @xmath39@xcite . given the size of the expansion parameter @xmath51 , however , one might worry about the possible importance of higher - order contributions . to that end , the @xmath52 contributions to @xmath53 were recently computed in ref . the loop effects arising at this order nearly cancel those appearing at @xmath17 , and one also encounters a new lec . the resulting expression is @xmath54 \ \ { \rm fm}^2\ \ \ , \ ] ] where the first term in the square brackets gives the loop contribution and @xmath55 is the @xmath52 magnetic radius lec . on general grounds , one would expect the magnitude of @xmath55 to be of order unity with undetermined sign . in principle , then , its contribution could be an order of magnitude more important than the loop contribution . until we are able to determine its value , we really do not know the either the magnitude or the sign of the slope of @xmath36 at the origin . as a result , the determination of @xmath36 at a single kinematic point is not sufficient to determine @xmath39 . this new development has a couple of implications . first , in order to learn how much the strange quarks contribution to the nucleon magnetic moment , we will require an experimental determination of the @xmath37-dependence of @xmath36 at low-@xmath37 . accomplishing this determination will require completion of the the backward angle , low-@xmath37 program as envisioned for the g0 experiment . second , it means that understanding the non - perturbative dynamics of the strange sea as parameterized by @xmath41 and @xmath55 will be key to achieving a theoretical understanding of the strange magnetism . the strangeness electric form factor has a similar chiral expansion . since the nucleon has no net strangeness , the leading @xmath37-dependence is governed by the strangeness electric radius , @xmath56 . in chpt , this term arises at @xmath17 , and both loops and an electric radius lec @xmath57 appear at this order@xcite : @xmath58\ \ { \rm fm}^2\ \ \ .\ ] ] as in the case of the strangeness magnetic moment and radius , we require input from non - perturbative methods in order to obtain predictions for @xmath56 . _ lattice qcd_. in principle , lattice qcd computations provide a means of obtaining values for the constants @xmath41 , @xmath55 , and @xmath57 . the two most recent lattice computations provide conflicting results for these constants . both computations were carried out in the quenched approximation , so that @xmath5 pairs appear only via the operator insertion . even under this quenched approximation , computation of such disconnected loop `` contributions is difficult . for connected insertions , wherein the operator is inserted on a line originating from the hadronic source , one must compute quark propagators from a fixed point in spacetime to all possible points on the lattice . in contrast , disconnected insertions require computation of quark propagators from all points on the lattice to all other points on the lattice . carrying out such a computation is inordinately expensive . to get around this barrier , one employs noise '' methods , in which one computes quark propagators from a random selection of initial points to a random set of final spacetime points@xcite . for a sufficiently large number of noise vectors , operator matrix elements computed using these quark propagators will approach the value one would obtain by computing propagators @xmath59 for all @xmath60 and @xmath61 . the two recent lattice computations adopted different philosophies for treating the noise method . the kentucky - adelaide group employed a relatively small number of gauge field configurations but a large number of noise vectors@xcite . they obtained statistically signficant signals for @xmath35 and @xmath36 at a variety of @xmath37 values , and from these results , one may infer values for the strangeness lec s . in contrast , the computation of ref . @xcite used on the order of 1000 gauge field configurations but a relatively small number of noises . in this calculation , no significant signal was observed for the strangeness vector current matrix elements , but a non - zero value for the scalar density @xmath62 was obtained . resolution of this disagreement will require signficant future work , and the two groups are pursuing new , refined calculations of the strangeness form factors . until a consensus emerges , it is difficult to draw any strong conclusions from lattice computations . for purposes of illustration , however , i will use the results from ref . @xcite below . achieving firm lattice qcd predictions for @xmath63 will ultimately require resolution of a number of issues . clearly , one would like to have in hand an unquenched calculation . in addition , one would like to use sufficiently small values for the light quark masses that one can extrapolate to the physical values using chpt . in this respect , use of chiral fermions either domain wall or overlap is likely to be critical . recent work by the kentucky group suggests that one must use very light quarks in order for the chiral extrapolation to be accurate , and in this respect , overlap fermions may provide some advantage@xcite . _ dispersion relations_. athough the use of dispersion relations is not a microscopic method , it nevertheless provides a first principles approach to computing form factors in a hadronic basis . the inputs required include the analyticity properties of form factors , causality , unitarity , and experimental scattering amplitudes . to be concrete , the dispersion relation for the strangeness magnetic and electric radii is @xmath64 the spectral function @xmath65 can be related to experimental scattering amplitudes by drawing upon some basic ideas in field theory . symbolically , one has @xmath66 where @xmath67 are magnetic and electric projection operators and the states @xmath68 are all states carrying the quantum numbers of @xmath69 . the lighest such state is the @xmath70 state . although pions themselves have no valence strange quarks , three pions can resonate into an @xmath71 and , by virtue of @xmath71-@xmath29 mixing , can give non - zero contributions to the strangeness vector current matrix elements . we have no indication that the @xmath72 and @xmath73 states resonate into an @xmath74 vector meson , so they are usually omitted from the analysis . in order of mass , the next state and the most intuitive contributor is the @xmath75 state . the chpt expressions for loop contributions to @xmath39 and @xmath76 are derived by computing the @xmath75 contribution at one - loop order . however , the work of ref . @xcite implies that the one - loop approximation is physically unrealistic . in order to satisfy the requirement of s - matrix unitarity , one must sum up kaon rescattering corrections to _ all orders_. after doing so , one finds that the kaon cloud contribution to the spectral functions is dominated by the @xmath77 resonance . moreover , the presence of this resonance signficantly enhances the kaon cloud contribution to the form factors . an interesting picture of form factor dynamics then emerges . to illustrate , consider the isoscalar and strangeness magnetic moments . in the case of @xmath78 , the @xmath70 and @xmath75 contributes have nearly equal magnitude but opposite sign , and the resulting cancellation leads to the very small value for the isoscalar moment . for @xmath39 , in contrast , the @xmath70 contribution gets suppressed by the @xmath71-@xmath29 mixing angle ( @xmath79 ) , while the kaon cloud contribution gets enhanced roughly by the inverse of the strange quark s electric charge , or @xmath80 . the resulting value for @xmath39 is roughly @xmath81 sizeable enough to be seen by experiment . in short , nature conspires to hide the effects of resonating @xmath5 pairs in @xmath78 through a cancellation against resonating light quark pairs . far from being inert , the @xmath5 sea is quite active , though hidden from view in purely electromagnetic ( em ) processes . measuring the strangeness form factors directly should allow one to uncover the conspiracy . there is one possible loop hole in this picture . so far , we have only been able to include states up through the @xmath75 state in a rigorous way . the limitation is the absence of strong interaction scattering data involving higher - mass states . if the isovector em form factors are any guide , then the low - mass states ought to suffice to describe the leading strangeness moments . indeed , the non - resonant @xmath26 continuum and @xmath22-meson resonance give nearly all of the low-@xmath37 behavior of the isovector form factors . for the strangeness form factors , however , there is no guarantee that inclusion of the low - mass region is enough . considerations of the large @xmath37-behavior of these form factors along with simple quark counting arguments suggests that higher - mass states ( _ e.g. _ , @xmath82 , _ etc_. ) may be required . unfortunately , our ability to carry out a dispersion theory treatment of the higher - mass region is presently limited by the dearth of data . on the other hand , one could ultimately intepret the results of the @xmath35 and @xmath36 measurements in terms of the relative importance of this higher - mass region . if , for example , the value of @xmath39 differs substantially from @xmath81 as suggested by the sample results , then there would have to be important higher - mass effects in order to overcome the fairly sizeable low - mass contribution . moreover , these higher - mass effects would have to be relatively strong for the strangeness current but relatively small for the isoscalar em current . such a situation would suggest that nature is an even more devious conspirator than we might otherwise think . _ hadronic models_. while i can not review the full breadth of hadron model predictions for @xmath83 , a few comments are warranted . from the standpoint of the dispersion theory treatment described above , many hadronic models are based on the idea of kaon cloud dominance . intuitively , this idea is appealing . it says the strange sea gets polarized because the nucleon fluctuates into a kaon - hyperon pair . many calculations including some i carried out in the past relied on one - loop computations to estimate this effect . the dispersion relation analysis , however , implies that the picture is physically unrealistic . inclusion of rescattering effects , encoded by the lec s @xmath41 , @xmath55 , _ etc . _ , is essential . moreover , @xmath5 pair creation likely plays a more important role than the spatial separation of an @xmath20 and @xmath84 . finally , one may encounter cancellations between various contributions . in this respect , the quark model calculation of geiger and isgur is suggestive@xcite . in that work , a sum over a tower of two hadron intermediate states was considered , with the quark model providing the relevant couplings . as one proceeds up the tower , strong cancellations occur between various contributions , leading to small strangeness form factors in the end . by necessity , however , the calculation was carried out to one - loop level only . whether this pattern of cancellations would persist after rescattering and resonance effects were included is hard to guess . _ some numbers_. while the task of computing strange quark form factors has a long way to go , it is nonetheless useful to discuss some of the numbers coming from the calculations . in table i , i give some of the lattice qcd and dispersion relation values for the lec s @xmath41 , @xmath55 , and @xmath57 . these numbers should be taken with a grain of salt . as noted above , the two most recent lattice qcd computations differ substantially , with one seeing no signal for the strange vector current form factors . the @xmath36 results of the kentucky - adelaide calculation were fit to a dipole form form , and from that a value for @xmath55 was obtained . there is no reason to believe the dipole form is correct , as an insufficient number of low-@xmath37 values were computed to constrain the behavior of the form factor in this region . at the same time , the dispersion theory values result from only the low - mass part of the spectral function , and significant higher - mass contributions are not ruled out . with these caveats in mind , it is interesting to note that the results of the kentucky - adelaide lattice calculation of @xmath39 are consistent with the low - mass dispersion relation value , but rather substantial disagreement occurs for the magnetic and electric radii . even the signs of the two calculations of @xmath76 differ . clearly , there is room for improvement in the theory . on the experimental side , completion of the entire program of pves strange quark measurements is the only way to ensure that defensible theoretical progress is made . .low - energy constants for strange magnetic moment ( @xmath41 ) , strange magnetic radius ( @xmath55 ) , and strange electric radius ( @xmath57 ) . lattice qcd values are obtained from ref . low - mass " dispersion relation contributions obtained from ref . [ cols="^,^,^,^",options="header " , ] while the current program of pves measurements is far from complete , some intriguing puzzles have emerged from the results already reported , particularly in relation to electroweak radiative corrections . theoretically , the challenge is to properly sort out the interplay between electroweak radiative corrections and strong interactions . in this context , a new idea for a future measurement of pv pion photoproduction on the @xmath85 resonance has emerged recently a measurement which may shed new light on an old problem in hadronic weak interactions . it is now well known that the importance of electroweak radiative corrections to the axial vector form factor are enhanced relative to the naively expected @xmath86 scale . this enhancement occurs because ( a ) the tree - level coupling is suppressed by the small vector coupling of the electron to the @xmath87 , @xmath88 , and ( b ) the corrections themselves contain large logarithms . these logarithms only appear in amplitudes involving @xmath89-exchange so they do not affect neutrino probes of the axial vector hadronic current . the large @xmath89-exchange contributions can also be affected by parity - violating quark - quark interactions , which induce a parity - violating @xmath90 coupling . people sometimes refer to this coupling as the nucleon anapole moment coupling . the anapole moment _ per se _ is not a physical observable , since its value depends on the choice of electroweak gauge parameter , @xmath91 . moreover , anapole moment effects can never be distinguished from other electroweak radiative corrections in any experiment . however , we might still use the anapole moment " to refer to the @xmath91-independent contribution of the weak quark - quark interaction to the leading pv @xmath90 coupling ( for a discussion see ref . @xcite ) . it has been known for some time that because the anapole moment ( am ) effects involve hadronic weak interactions , their magnitude is uncertain@xcite . thus , the theoretical prediction for @xmath92 the axial vector form factor measured in pves has a fairly sizeable error bar . were one to try and determine @xmath36 from pv @xmath93 scattering alone , the theoretical @xmath92 uncertainty would add to the error bar for @xmath36 . hence , the sample collaboration has measured both the pv elastic @xmath93 and pv quasielastic @xmath1 asymmetries , which depend on different linear combinations of the two form factors in question and allow an experimental separation of @xmath94 from @xmath36 . the theoretical uncertainty associated with @xmath95 has been estimated to be considerably smaller than for the isovector component , so the isoscalar contribution will be neglected in this discussion could be too small , a possibility that should be investigated . ] . the published sample results@xcite indicate that the radiative corrections to @xmath94 are , indeed , large and have the same sign as given by the calculations of refs . . however , the magnitude may be considerably larger than predicted . if this result is confirmed by future measurements , such as the lower-@xmath96 sample deuterium experiment or g0 , one would like to understand what is behind the enhancement . several ideas have been mentioned : _ nuclear pv effects_. in the extraction of @xmath94 from the deuterium asymmetry , possible contributions to the pv @xmath89-_nucleus _ coupling from pv @xmath97 interactions have not previously been taken into account . in contrast to pv effects in elastic scattering , those entering inelastic scattering can give rise to a @xmath37-independent contribution to the asymmetry : @xmath98 the @xmath99 terms arise from the @xmath87-exchange amplitude and contain the usual electroweak form factors . the @xmath100 term in @xmath101 is generated by @xmath89-exchange , where the @xmath89 couples to the nucleus through a pv transverse electric dipole matrix element . the presence of this term follows directly from an extension of siegert s theorem@xcite , which implies that for elastic scattering , all transverse electric dipole matrix elements are proportional to @xmath37 , while those entering inelastic transitions need not vanish at the photon point . in the case of @xmath100 , pv nucleon - nucleon interactions are responsible for the effect . people ( myself included ) have speculated that for sufficiently small @xmath102 , the @xmath100 term may become quite important relative to the @xmath103 term that contains @xmath92 , and that the larger - than - expected radiative corrections to @xmath92 may actually be a result of neglecting the @xmath100 contribution . to address this question , two groups recently computed the @xmath37-independent contribution generated by nuclear pv @xcite . the results of the two calculations are consistent with each other and indicate that the effect of nuclear pv is far too small to modify the value of @xmath92 extracted from the asymmetry . generally speaking , one has@xcite @xmath104 at @xmath105 ( gev@xmath50 , the ratio is about @xmath106 . thus , if the radiative corrections are truly enhanced , then some other mechanism would have to be the culprit . _ box graphs_. one of the more sizeable contributions to the axial vector radiative corrections arises from the @xmath107-@xmath89 box diagrams ( fig . [ fig2]a ) . the size of this amplitude is @xmath108 \ \ { g_f\over 2\sqrt{2 } } { \alpha\over 4\pi } \left[\ln\frac{m_z}{\lambda } + c_{\gamma z}^a(\lambda)\right]\ \ \ , \ ] ] where @xmath109 , @xmath110 , and @xmath111 . the constants @xmath112 are given by @xmath113 and @xmath114 . note also the presence of the large logarithm . here , @xmath115 is a low - momentum cut - off , usually taken to be about one gev , corresponding to the transition between the perturbative and non - perturbative domains . the constant @xmath116 encodes all the physics of the loop integral associated with momenta @xmath117 . this physics includes hadronic form factors at the hadronic vertices as well as the full tower of hadronic intermediate states in the hadron propagator . to date , no one has attempted to carry out a systematic treatment of all of these effects . only the nucleon intermediate state along with the relevant form factors has been included in model computations@xcite . one might wonder then , whether inclusion of other states , such as the uncorrelated @xmath118 , @xmath85 , @xmath119 , etc , might lead to an enhancement of this contribution . the problem is an open one and remains to be tackled . i like to emphasize that understanding this particular contribution could have implications for the interpretation for other precision electroweak measurements . for example , in neutron @xmath9-decay , the @xmath120 box graph produces a similar structure to that of eq . ( [ eq : zgammabox ] ) . in this case , the @xmath121 amplitude is used to extract the value of @xmath122 and test the unitarity of the ckm matrix . a value for the constant @xmath123 has been estimated , but as in the case of @xmath116 , contributions from the full hadronic green s function have not been included . were @xmath116 large enough to give the enhanced radiative corrections suggested by sample , and were @xmath123 to have a similar magnitude , one would need to change the value of @xmath122 substantially and engender serious problems with ckm unitarity . although such a scenario seems unlikely , one might nevertheless gain new insights into the theoretical uncertainty associated with @xmath123 through a study of @xmath124 . > from a somewhat different perspective , the sample collaboration has also measured the parity - conserving asymmetry associated with transversely polarized electrons@xcite . this asymmetry arises from the @xmath125-exchange box graph , which is also sensitive to the low - energy part of the nucleon green s function . the experimental result for the asymmetry does not agree with theoretical predictions . achieving a better understanding of the @xmath126 box graph may shed new light on this problem , and vice versa . _ @xmath126 mixing_. the largest contribution to the axial vector radiative corrections comes from the @xmath126 mixing tensor , @xmath127 ( fig . [ fig2]b ) . the light quark contribution to @xmath127 can not be computed in perturbation theory , since the quarks have non - perturbative strong interactions . traditionally , this problem is solved by relating the light quark component of @xmath127 to @xmath128 using a dispersion relation and su(3 ) arguments . for purely leptonic scattering , this approach ought to suffice . for @xmath129 scattering , however , one might worry about strong interactions between the quarks in @xmath127 and quarks in the proton . to some extent , these interactions have already been taken into account in anapole moment computations , but the matching of the two calculations onto each other remains to be worked out carefully . again , it seems unlikely that some new effect might emerge when this matching has been achieved , but one should think about it nonetheless . the g0 collaboration plans a study of the @xmath37-dependence of the axial vector @xmath130 transition form factor , @xmath131 . as in the case of @xmath92 , the transition axial vector form factor will be modified by electroweak radiative corrections . the same issues which apply to @xmath92 will also apply to @xmath131 ( for a detailed analysis , see ref . @xcite ) . in principle , these corrections could contain some @xmath37-dependence not associated with the primordial " form factor ( as one might more directly measure with neutrino scattering ) . a more serious issue , however , arises because one is studying an inelastic process . for the same reasons as discussed above , the pv @xmath130 asymmetry will receive a @xmath37-independent term as a consequence of siegert s theorem . in this case , the relevant @xmath132 matrix element is parameterized by an lec called @xmath133 . at @xmath134 one has@xcite @xmath135 where @xmath136 is the leading vector current transition form factor . the presence of this term in the pves asymmetry could complicate the determination of the @xmath37-dependence of @xmath131 , so it would be desirable to know it in some other way . in principle , a study of pv photoproduction could allow one to measure @xmath133 , as discussed by jeff martin elsewhere in these proceedings . a determination of @xmath133 would also be interesting from another standpoint namely , an old problem in nonleptonic weak interactions . in the @xmath137 sector , one has considerable knowledge of hyperon decays . the mesonic decays depend on both s- and p - wave amplitudes . to date , it has not been possible to obtain a successful , simultaneous description of both amplitudes in chpt . another problem appears with the @xmath137 radiative decays : @xmath138 . here , one measures pv asymmetries @xmath139 associated with the direction of the photon relative to the spin of the initially polarized hyperon . these asymmetries arise from the interference of an e1 and an m1 amplitude . in the limit of exact su(3 ) flavor symmetry , the octet of baryons is degenerate and the e1 amplitude vanishes a result known as hara s theorem@xcite . in the real world , su(3 ) symmetry is broken by the @xmath140 vs. @xmath141 mass difference . thus , one would expect the asymmetries to have a size @xmath142 . the experimentally measured @xmath139 are factors of four to five larger than this naive expectation . based on general symmetry considerations , one has no way of explaining what is responsible for these enhancements . it would appear , then , that the application of qcd - based symmetries , such as chiral symmetry and su(3 ) symmetry , to the theoretical analysis of @xmath137 non - leptonic weak interactions fails . one would like to know if this failure is a consequence of having strange quarks involved in the decay processes , or if it is a signature of a more fundamental feature of the interplay of strong and weak interactions . to that end , a study of @xmath133 could provide new insight . the transition induced by the @xmath133 operator is the @xmath143 analog of the @xmath137 e1 amplitudes governing the @xmath139 . if the enhancements of the latter are not due to strange quarks but a more general feature of nonleptonic weak interactions , then one ought to observe an enhanced asymmetry for pv @xmath85 photoproduction , which involves no strange quarks . on the other hand , if strangeness is the key to the breakdown of symmetry considerations , then the @xmath144 asymmetry ought to have its natural size . recently , holstein and borasoy have developed a model that allows us to make this idea quantitative@xcite . the idea is that the @xmath137 hyperon decays are affected by mixing of excited baryons into the ground states due to the hadronic weak interaction . in order to resolve the s - wave / p - wave problem , the mixing matrix elements @xmath145 must have a characteristic size larger than one might have naively expected . after determining what the @xmath145 must be to fix the s - wave / p - wave inconstency , holstein and borasoy used them to predict the @xmath139 via resonance - mixing processes . interestingly , the asymmetries are significantly enhanced by this mechanism , thereby helping to close the gap between the naive symmetry expectation and the experimental results . in principle , the same kind of weak interaction mixing could occur in the @xmath143 sector ( see fig . [ fig3 ] ) . if the mixing matrix elements @xmath146 have the same characteristic size as for the @xmath137 case , then one would expect @xmath133 to be significantly enhanced by factors of 25 or more compared to one s naive expectation . we already know from studies of pv @xmath147 scattering that pv @xmath148 amplitudes ( @xmath149 is a meson ) are of the same size as the @xmath137 @xmath150 amplitudes , so it is not unreasonable to expect the @xmath151 to have the same scale in the @xmath143 and @xmath137 sectors . in this case , @xmath152 would be rather large on the order of ppm and one could measure it with a reasonable amount of beam time . on the other hand , if one looked for an asymmetry of this size and did nt see it , one might raise doubts about the holstein / borasoy proposal and , more generally , conclude that the problem with symmetry truly does hinge on having a strange quark involved . the first few pves measurements were intended to test the standard model ( sm ) . since then , the sm has been so well - tested in a variety of ways that we can , with a high degree of confidence , use the weak @xmath129 interaction as a well - understood probe of hadron and nuclear structure . the recent advances in this field both experimental and theoretical have now afforded us an opportunity to come full circle and again study fundamental aspects of the weak interaction . of course , there is little doubt that the sm is a successful model describing weak interactions far below the planck scale . however , there exist many conceptual reasons for believing that the sm must be embedded in a larger theory . the quest to discover this new " standard model is at the forefront of both particle physics and nuclear physics . unravelling the nature of the neutrino is clearly direction in which nuclear and particle physicists are pursuing this goal . future collider experiments at the large hadron collider and the linear collider will seek direct evidence for new physics . a third , and no less interesting , avenue is to carry out highly precise measurements of electroweak observables . some such observables such as the permanent electric dipole moments of the electron , neutron , and neutral atoms are so small in the sm that the measurement of a nonzero effect would provide smoking gun evidence for new physics . others , such as the pv asymmetries of interest here , have non - zero values in the sm , so one must attempt to probe for tiny deviations from the sm predictions . the presence or absence of such deviations when taken in conjuction with other precision measurements can provide important clues about the structure of the new sm . to illustrate , recall how fits to precision electroweak observables predicted a range for the mass of the top quark prior to its discovery at the tevatron . the subsequent discovery of a top quark with mass lying in this range provided beautiful confirmation of the sm at the level of quantum corrections . the same kind of synergy between indirect searches in precision electroweak measurements and direct collider searches will undoubtedly provide new insights about whatever lies beyond the sm . > from this standpoint , the quantity of interest to pves is @xmath153 , the weak charge of a fermion or system of fermions . at tree level in the sm , it is the coupling of the @xmath107-boson to the fermion @xmath154 . it also governs the strength of the effective @xmath155 interaction at @xmath156 : @xmath157 where , at tree - level in the sm , one has @xmath158 for both the electron and the proton , the magnitude of the weak charge is suppressed : @xmath159 . because of this suppression , @xmath160 are rather transparent to new physics . as we have heard elsewhere in this conference , the goals of the pv m " oller experiment at slac and the pv @xmath93 experiment at jefferson lab are to measure these two weak charges with some hope of finding evidence for new physics . we have also heard that even within the sm , these two measurements are quite interesting . the value of the effective weak mixing angle appearing in eq . ( [ eq : qweak ] ) actually depends on the scale at which one is carrying out the measurement . the sm predicts this scale - dependence . the sm prediction at the z - pole has been confirmed with high precision , but there exist few precise determinations of @xmath161 below the @xmath87-pole . the cesium atomic pv experiment carried out by the boulder group@xcite probed the weak mixing angle via the cesium nucleus weak charge ( @xmath162 ) at a very low scale . although early indications suggested a substantial deviation from the sm , recent developments in atomic theory now imply the cesium atomic pv result is consistent with the sm . since the extraction of @xmath162 from the measured pv transition rate depends on atomic structure calculations , there has always been some question about the theoretical uncertainty associated with this extraction . in the past couple of years , atomic theorists have computed several sub - one percent corrections such as the breit interaction correction@xcite , uehling potential@xcite , and nucleus - enhanced qed radiative corrections@xcite . as of this writing , there appears to be an emerging consensus that all of the important corrections have been properly taken into account and that the value of @xmath163 should be stable at the level of the experimental precision ( about 0.4% ) . a value for @xmath161 at somewhat higher scales has been obtained by the nutev collaboration using deep inelastic neutrino - nucleus scattering@xcite . the result indicates a roughly 3@xmath3 deviation from the value predicted in the sm . as in the case of cesium apv , however , there exists some debate about the theoretical uncertainties associated with this extraction . it has been argued , for example , that small changes in parton distribution functions due to nuclear shadowing and isospin - breaking could substantially change the sm prediction for the neutrino scattering cross sections and reduce the significance of the nutev anomaly@xcite . the collaboration has responded to this suggestion with a vigorous rebuttal@xcite . when the dust finally settles , one will have either learned something about parton distributions in nuclei or have striking evidence for new physics . in the context of these other measurements , the slac and jlab measurements offer several advantages . first , they are being undertaken at values of @xmath37 lying between the cesium atomic pv and nutev scales . second , they are both being performed at roughly the _ same _ scale : @xmath164 ( gev@xmath50 . third , they are theoretically clean , so their intepretation will not be subject to the kinds of controversies that have affected cesium atomic pv and the nutev result . this point is particularly important , so i will spend some time on it below . finally , the two measurements are complementary . one is purely leptonic , while the other is semileptonic . as i will also discuss , a comparsion of the two measurements can provide an interesting diagnostic " tool for probing new physics . in principle , the pv m " oller experiment , being purely leptonic , is the theoretically cleanest observable . there is some theoretical uncertainty in the sm prediction arising from the presence of light quark loops in @xmath127 , these effects are constrained by current conservation . ] . this uncertainty is common to both @xmath165 and @xmath166 . traditionally , the light quark contribution to @xmath127 is obtained by relating it to @xmath128 via a dispersion relation . in order to do so , one must also invoke su(3 ) flavor arguments . the uncertainty associated with this procedure was estimated some time ago by marciano and sirlin@xcite , and it falls below the expected experimental precision for both weak charge measurements . a second source of theoretical uncertainty arises from the extraction of @xmath167 from the measured asymmetries . in the case of the m " oller experiment , one must be able to reliably subtract out contributions from the elastic and inelastic @xmath93 asymmetries . although the number of @xmath93 events is small when compared with the number of m " oller events , the inelastic @xmath93 asymmetries can be relatively large . the basic problem is that we do not know in detail what the inelastic asymmetries are . one strategy for circumventing it is to measure the total asymmetry in a kinematic region where the inelastic @xmath93 asymmetry is dominant , and then extrapolate to the kinematics of the m " oller measurement . to the extent that various components of the inelastic asymmetry evolve with the relevant kinematic variables in a well understood way , this strategy should reduce the corresponding @xmath93 uncertainty to an acceptable level . in the case of @xmath166 , the design of the detector should eliminate any sizeable contribution from inelastic events . however , the elastic @xmath93 asymmetry itself receives hadronic contributions in the guise of nucleon form factors . the associated uncertainty can also be reduced to an acceptable level by drawing on other measurements . specifically , the asymmetry has the form @xmath168\ \ \ , \ ] ] where @xmath169 is a function of nucleon form factors , both electromagnetic and strange . at forward angles and low-@xmath96 , @xmath170 , yielding an overall @xmath171 contribution to the asymmetry . in contrast the @xmath166 term goes as @xmath96 . the strategy , then , is to use the existing program of pv measurements to constrain the @xmath96-dependence of @xmath172 and extrapolate to very low-@xmath96 where the q - weak measurement will be performed and where the @xmath166 term gives the dominant contribution to the asymmetry . this strategy can be implemented in a rigorous way by using chpt to express all the form factors in @xmath172 in terms of their known quark - mass dependence plus lec s . the existing program of experiments , plus the world data set for the electromagnetic form factors , will determine the lec s with sufficient precision to make the uncertainty in the low-@xmath96 asymmetry sufficiently small . no additional theoretical input on nucleon structure will be needed . potentially more problematic considerations arise in arriving at the sm prediction for @xmath166 itself . in particular , the two boson - exchange box graphs depend on hadronic physics via the nucleon green s function in the box . fortunately , we are able to constrain the theoretical uncertainty associated with these effects to be acceptably small . the largest box graph contribution involves the exchange of two @xmath173-bosons . the total correction is about 26% of the tree - level value , largely due to the absence of the @xmath174 factor from the @xmath175 , the ww box amplitude . the latter is dominated by intermediate states of momentum @xmath176 , so the qcd corrections are perturbative . the latter can be computed using a combination of current algebra and the operator product expansion . the result is @xcite @xmath177\ \ \ , \ ] ] where @xmath178 is the fine structure constant evaluated at a scale @xmath179 in the @xmath180 scheme and @xmath181 is the corresponding definition of @xmath182 . inclusion of the @xmath183 contributions generate a @xmath184 correction to @xmath166 , and higher - order corrections in @xmath185 will be at roughly an order of magnitude less important . note that the qcd corrections involve no large logarithms , since the currents involved are conserved ( or partially conserved ) . a similar correction applies to the zz box graphs , but the effect in this case is much smaller roughly @xmath186 in @xmath166 . in contrast , the @xmath187 box graphs receive contributions from both low- and high - momentum scales . this sensitivity to the different scales is reflected in the presence of a large logarithm in the amplitude : @xmath188\ \ \ , \ ] ] where i have omitted contributions suppressed by powers of @xmath189 . as in eq . ( [ eq : zgammabox ] ) , @xmath115 is a scale associated with the transition from the non - perturbative to the perturbative region and is on the order of 1 gev . at present , we can not reliably compute contributions to the loop integral arising from momenta below this scale . instead , we parameterize their effects by the constant @xmath190 . fortunately , the effect of this unknown term is suppressed by the overall @xmath191 prefactor , so we can live with a fairly large uncertainty in @xmath192 without affecting the interpretation of @xmath166 . the presence of this prefactor is , itself , something of a fortunate accident . it arises from the sum of the box and crossed - box graphs . the spacetime structure of this sum dictates that the @xmath193 amplitude arises from the antisymmetric product of the electron electromagnetic current and the vector part of the electron weak neutral current . the latter contains the @xmath191 vector coupling of the electron to the @xmath87 , thereby producing the suppression factor in @xmath194 . to estimate the uncertainty associated with @xmath195 we can follow a philosophy adopted by sirlin in his classic analysis of neutron @xmath9-decay@xcite . in that case , a similar unknown constant arises from the @xmath196 box graph . the size and uncertainty of @xmath197 is constrained because the vector part of the decay amplitude is used to determine @xmath122-decay , though new experiments using ultracold neutrons should provide competitive determinations . ] . in order for @xmath122 to be consistent with the unitarity of the ckm matrix , the uncertainty from @xmath197 can not be larger than about @xmath198 . the hadronic dynamics responsible for @xmath197 and @xmath195 are similar , so one would expect the uncertainty in the two to be comparable . the corresponding uncertainty in @xmath166 is about @xmath199 . a siginficantly larger uncertainty in @xmath197 and @xmath195 would imply unacceptably large deviations from ckm unitarity . of course , one should pursue computations of these quantities from first principles possibly using lattice qcd . however , the fortuitous presence of the @xmath191 suppression factor has given us a sizeable zone of theoretical safety . one other hadronic effect which one might worry about for @xmath166 is the impact of isospin admixtures into the proton wavefunction . on general grounds , one would expect isospin - breaking effects to be of @xmath200 and , therefore , a source of concern . at @xmath156 , however , a diagonal version of the ademollo - gatto theorem@xcite implies that isospin - breaking effects in the matrix element of the hadronic vector current vanish identically@xcite . they only turn on for non - zero @xmath37 . in this case , they will be effectively contained in the measured form factor term @xmath172 and constrained along with all other @xmath37-dependent effects by experiment . given the theoretical interpretability of the two weak charge measurements , one can credibly view them as probes physics beyond the sm . i have been thinking quite a bit about supersymmetry ( susy ) recently , so i will focus my discussion on this brand of new physics . susy remains one of the most strongly - motivated extensions of the standard model for a number of reasons : it gives a solution to the hierarchy problem `` in particle physics ( essentially , the question as to why the electroweak scale is stable ) ; it generates unification of the electroweak and strong couplings close to the planck scale ; it gives a natural candidate for cold dark matter ( the neutralino , @xmath201 ) ; and it contains new sources of cp violation that could help us account for the observed predominance of matter over anti - matter ( the baryon asymmetry of the universe ) . of course , since no one has yet seen any of the superpartners '' of sm particles , they must be heavier than the particles in the sm . consequently , susy must be a broken symmetry . the breaking can not be too large , however ; otherwise , the hierarchy problem re - emerges . future collider experiments at the lhc and lc will go looking for the heavy superpartners . one might also ask about observing their effects indirectly . for example , superpartners can contribute to precision electroweak observables through radiative corrections . perhaps , the most well - known example of such effects are superpartner loop contributions to the muon anomalous magnetic moment . on general grounds , one expects the size of susy radiative corrections to go as @xmath202 , where @xmath149 is a sm mass and @xmath203 is a superpartner mass . for the muon @xmath204 , one has @xmath205 . taking @xmath206 gev , one gets an effect of the order of @xmath207 roughly the size of the current experimental error bar . the possible significant deviation of @xmath208 from the sm prediction would be consistent with susy loop contributions containing superpartners of mass of a few hundred gev . for this reason , the particle physics community has become quite excited by new results for the muon anomaly . for weak interaction processes , such as @xmath9-decay and pves , one has @xmath209 . consequently , the magnitude of susy loop effects , _ relative _ to the sm prediction , is generically of order @xmath106 . however , since @xmath167 are suppressed in the sm , one actually needs a precision of order a few percent rather than a few tenths of a percent to be sensitive to susy radiative corrections . for this reason , it is interesting to study the effect of these radiative corrections on the weak charges . to that end , we need to modify our formula for @xmath153 as follows : @xmath210 here , @xmath211 and @xmath212 are universal , in that they do not depend on the species of fermion being probed with pves . the quantity @xmath213 , in contrast , is species - dependent . at tree - level , one has @xmath214 and @xmath215 . inclusion of loops lead to deviations from these values . in particular , one can think of @xmath216 as the change in the apparent weak mixing angle due to loop effects . the effects of new physics on @xmath212 arise in two ways : first , one has a change in the prediction for @xmath217 due to its defintion in terms of other more precisely known quantities ( @xmath218 , @xmath219 , and @xmath220 ) . this definition gets modified by new physics . second , @xmath216 receives contributions directly from the pv @xmath221 amplitudes . in order to analyze the corrections @xmath222 , @xmath223 , and @xmath224 one needs to compute a large number of loop graphs moreover , one has to take into account the effects of susy - breaking . in general , this task is non - trivial . in the minimal supersymmetric standard model ( mssm ) , susy - breaking is governed by 105 parameters . to simplify the problem and to build a real theory people have invented models of susy - breaking , such as supergravity , which reduce the number of independent parameters to a handful . whether or not these model assumptions are entirely consistent with electroweak data is not entirely clear . we recently completed an analysis of charged current observables and found that the superpartner spectrum implied by susy - breaking models may not be consistent with the precision data@xcite . for this reason , it is desirable to carry out a model - independent analysis of susy radiative corrections , thereby avoiding the simplifying assumptions of susy - breaking models . my collaborators and i have completed such a model - independent analysis of susy radiative corrections to @xmath167 in the mssm@xcite . we found that the effects on the two weak charges are highly correlated , and that the _ relative _ sign of the correction in both cases is positive over nearly all of the available susy - breaking parameter space . this correlation shown in fig . [ fig4 ] arises because the dominant effect occurs via @xmath225 , with some scatter due to the effects of @xmath222 and @xmath224 . moreover , the potential size of the susy radiative corrections could be as large as the projected experimental error bars . what this means is that if both experiments come in consistent with the sm , then one could not say much about the susy - breaking parameters . however , deviations from the sm of the order of @xmath226 or more would depending on the relative sign of the effect start to either favor or exclude parts the parameter space . the situation is helped somewhat by the correlation between the two weak charges in susy . the two measurements together really act like one determination of @xmath225 with better statistics than obtained with either alone . nevertheless , it would be advantageous looking further down the road to pursue even more precise measurements of the weak charges . the effects of susy can also arise at tree - level if one allows for the violation of lepton number ( l ) and/or baryon number ( b ) . if b - l is conserved , then the mssm has another symmetry called r parity " . conservation of r parity is not automatic ; it has to be imposed by hand . there are some phenomenological constraints . the absence of proton decay highly constrains the possibility of b - violating interactions . however , the constraints on l - violating sources of r - parity violation ( rpv ) are much less severe . the existence of any such interactions would , however , sound the death knell for susy dark matter , since the lightest superpartner ( presumably the @xmath201 ) would no longer be stable . thus , one would like to know from experiment whether or not the mssm respects r parity . to that end , a comparison of @xmath165 and @xmath166 measurements could be quite interesting . in fig . [ fig4 ] , i show the possible deviations of these quantities from the sm values due to rpv effects . the interior of the truncated ellipse is the region allowed by other precision measurements at 95% confidence . note that for most of this region , the relative sign of the deviation in @xmath165 and @xmath166 is negative in contrast to the situation for susy loops . in fact , there is nearly no overlap between the regions in this plane associated with the different effects . thus , a comparison of these two measurements could , in principle , serve as a diagnostic " for susy with our without r parity . in fact , the diagnostic potential of these two experiments is even greater if one considers other models of new physics . for example , e@xmath227 grand unified theories allow for the possibility of a relatively low mass , extra neutral gauge boson models are actually a very broad class of models , though they do not encompass all possibilities for a low - mass " @xmath228 . ] . in these models , the effect of a @xmath228 on @xmath165 and @xmath165 would also be correlated ( having the same relative sign ) . however , a sizeable shift in these weak charges would also imply a sizeable deviation of the cesium weak charge from the sm prediction . since @xmath163 currently agrees with the sm , this scenario appears to be disfavored . in contrast , the effects of susy loops on @xmath162 is tiny below the current experimental plus theoretical error due to cancellations between up- and down - quark contributions . thus , one could still have sizeable loop - induced shifts in @xmath167 without being inconsistent with cesium atomic pv . similarly , the allowed rpv region shown in fig . [ fig4 ] already incorporates the constraints from the @xmath163 determination . i am convinced that pves will remain a highly interesting field for some time to come . in addition to providing a definitive answer to the question about strangeness , pves has also opened up possibilities for studying other aspects of hadronic and nuclear structure . these studies rely on our present understanding of the weak neutral current interaction in order to probe novel aspects of low - energy qcd . at the same time , the success of this program has opened the way for new experiments to study the weak interaction itself . looking further into the future , one might anticipate new pv deep inelastic scattering experiments after the cebaf upgrade that may offer new glimpses of higher - twist physics , or even a more precise measurement of the m " oller asymmetry . suffice it to say , however , that completion of the current program of pves measurements is essential to the long - term success of the field . i would like to thank the organizers of this workshop and the support staff for their hospitality during my stay in mainz . i also thank a. kurylov for assistance in preparing this manuscript . this work was supported in part under u.s . department of energy contracts # de - fg03 - 02er41215 and by the national science foundation under award phy00 - 71856 . ramsey - musolf and h .- w . hammer , phys . * 80 * , 2539 ( 1998 ) ; h .- w . hammer , phys . c * 60 * , 045204 ; e-_ibid _ , * 62 * , 049902 ; h .- w . hammer and m.j . ramsey - musolf , phys . c * 60 * , 045205 ; e-_ibid _ , * 62 * , 049903 ; see also r.l . jaffe , phys . b * 229 * , 275 ( 1989 ) .

we describe a monte carlo event generator for the simulation of qcd - instanton induced processes in deep - inelastic scattering ( hera ) . the qcdins package is designed as an `` add - on '' hard process generator interfaced to the general hadronic event simulation package herwig . it incorporates the theoretically predicted production rate for instanton - induced events as well as the essential characteristics that have been derived theoretically for the partonic final state of instanton - induced processes : notably , the flavour democratic and isotropic production of the partonic final state , energy weight factors different for gluons and quarks , and a high average multiplicity of produced partons with a poisson distribution of the gluon multiplicity . while the subsequent perturbative evolution of the generated partons is always handled by the herwig package , the final hadronization step may optionally be performed also by means of the general hadronic event simulation package jetset . and qcd ; instanton ; deep - inelastic scattering ; monte carlo simulation 11.15.kc ; 12.38.lg ; 13.60.hb * program summary * + _ title of program : _ qcdins 2.0 + _ catalogue identifier : _ + _ program obtainable from : _ http://www.desy.de/~t00fri/qcdins/qcdins.html + _ computer for which the program is designed and others on which it has been tested : _ any computer with a fortran 77 compiler + _ operating systems under which the program has been tested : _ linux 2.0.x ; hp - ux 10.2 + _ programming language used : _ fortran 77 + _ memory required to execute with typical data : _ size of executable program is approximately 2.6 mb . the size of the qcdins library itself is about 200 kb ; the required routines from the herwig and jetset libraries constitute the dominant portion of the needed memory . + _ no . of processors used : _ 1 + _ has the code been vectorised or parallelized ? : _ no + _ no . of bytes in distributed program , including test data , etc . : _ 1071106 + _ distribution format : _ ascii + _ cpc program library subprograms used : _ herwig [ 1 ] version 5.9 ; jetset 7.4 [ 2 ] + _ keywords : _ qcd ; instanton ; deep - inelastic scattering ; monte carlo simulation + _ nature of physical problem _ + instantons are a basic aspect of quantum chromodynamics . being non - perturbative fluctuations of the gauge fields , they induce hard processes absent in conventional perturbation theory . deep - inelastic lepton - nucleon scattering at hera offers a unique possible discovery window for such processes induced by qcd - instantons through their characteristic final - state signature and a sizable rate , calculable within instanton - perturbation theory . an experimental discovery of such a novel , non - perturbative manifestation of non - abelian gauge theories would be of fundamental significance . however , instanton - induced events are expected to make up only a small fraction of all deep - inelastic events . therefore , a detailed knowledge of the resulting hadronic final state , along with a multi - observable analysis of experimental data by means of monte carlo techniques , is necessary . _ method of solution _ + the qcdins package is designed as an `` add - on '' hard process generator interfaced to the general hadronic event simulation package herwig . it incorporates the theoretically predicted production rate for instanton - induced events as well as the essential characteristics that have been derived theoretically for the partonic final state of instanton - induced processes : notably , the flavour democratic and isotropic production of the partonic final state , energy weight factors different for gluons and quarks , and a high average multiplicity of produced partons with a poisson distribution of the gluon multiplicity . while the subsequent perturbative evolution of the generated partons is always handled by the herwig package , the final hadronization step may optionally be performed also by means of the general hadronic event simulation package jetset . _ restrictions on the complexity of the problem _ + the default values of the implemented kinematical cuts represent the state of the art limits for the reliability of the generated instanton - induced event rate and event topology . _ typical running time _ + 10 - 100 events per second for a pc with pentium cpu , depending on its clock frequency . on a hp 9000/735 ( 99 mhz ) workstation , 6 events per second are generated . _ unusual features of the program _ + none _ references _ + [ 1 ] g. marchesini et al . , comput . phys . commun . 67 ( 1992 ) 465 . + [ 2 ] t. sjstrand , comput . phys . commun . 82 ( 1994 ) 74 . * long write - up * +

hep-ph9911516

the ground state ( `` vacuum '' ) of non - abelian gauge theories like qcd is known to be very rich . it includes topologically non - trivial fluctuations of the gauge fields , carrying an integer topological charge @xmath1 . the simplest building blocks of topological structure in the vacuum are @xcite _ instantons _ with @xmath2 and _ anti - instantons _ with @xmath3 . instantons represent gluon field configurations that are localized ( `` instantaneous '' ) in euclidean time and space . while they are believed to play an important role in various long - distance aspects of qcd , there are also important short - distance implications . in qcd with @xmath4 ( light ) flavours , instantons induce hard processes violating _ chirality _ in accord @xcite with the selection rule @xmath5 _ chirality _ @xmath6 , due to the general chiral anomaly . while in ordinary perturbative qcd ( @xmath7 ) these processes are forbidden , their experimental discovery would clearly be of basic significance . the deep - inelastic scattering regime is strongly favoured in this respect , since hard instanton - induced processes are both calculable @xcite within instanton - pertubation theory and have good prospects for experimental detection at hera @xcite . qcdins @xcite is a monte carlo package for simulating qcd - instanton induced scattering processes in deep - inelastic scattering ( hera ) . it is designed as an `` add - on '' hard process generator interfaced by default to the monte carlo generator herwig @xcite . it incorporates the theoretically predicted production rate for instanton - induced events as well as the essential characteristics that have been derived theoretically for the partonic final state of instanton - induced processes : notably , the flavour democratic @xcite and isotropic @xcite production of the final state partons , energy weight factors different for gluons and quarks @xcite , and a high average multiplicity @xmath8 of produced partons with a ( approximate ) poisson distribution of the gluon multiplicity @xcite . earlier versions of qcdins have been used already to establish first experimental bounds on the rate of instanton - induced events at hera @xcite and to develop instanton search strategies @xcite . in the present report a comprehensive description is given of the theoretical framework built into the program ( section [ theorie ] ) as well as of the various program components ( section [ qcdins ] ) and of their usage ( section [ usage ] ) . let us briefly summarize in this section the underlying physics picture , some relevant formulae and the main stages involved in qcdins to generate the complete instanton - induced partonic final state . the remaining formulae may be found under the corresponding descriptions of qcdins routines in section [ qcdins ] . ll & l + dis variables : + @xmath9 + @xmath10 + @xmath11 + @xmath12 + @xmath13 + @xmath14 + + variables of instanton - subprocess : + @xmath15 + @xmath16 + @xmath17 + in deep - inelastic @xmath18 scattering , instanton - induced events are predominantly associated @xcite with a process structure as sketched in fig . [ kin - var ] : a photon splitting into a @xmath19-pair fuses with a gluon from the proton in the background of an instanton ( @xmath20 ) or an anti - instanton ( @xmath21 ) . for each ( light ) flavour , @xmath22 , a violation of _ chirality _ is induced , @xmath23=\pm\ , 2 \ \mbox{\ for an } \left\{\begin{array}{l}i\\{{\overline{i}}}\end{array } \right . \,,\ ] ] in agreement with the general chiral anomaly relation @xcite . correspondingly , the partonic final state exhibits `` flavour democracy '' , i.e. @xmath24 ( @xmath25 ) pairs of _ all _ @xmath4 light flavours occur precisely once in case of an instanton ( anti - instanton ) , @xmath26 + n_g\,g \,.\ ] ] as illustrated in fig . [ kin - var ] , one of those partons acts as a current - quark ( jet ) @xmath27 , whereas the other @xmath28 ( anti-)quarks and some number @xmath29 of gluons are directly emitted from the instanton ( anti - instanton ) `` blob '' . instanton - induced processes initiated by a quark from the proton are suppressed by a factor of @xmath30 with respect to the gluon initiated process @xcite . this fact , together with the high gluon density in the relevant kinematical domain at hera , justifies to neglect quark initiated processes . in instanton - perturbation theory , the dominant instanton - induced contribution to the inclusive @xmath31 cross section at cross sections . ] , subject to appropriate kinematical restrictions and ( theoretical ) fiducial cuts , has a convolution - like structure @xcite , @xmath32 \label{evwgt } & & \times \int\limits_{{\rm max}\left(\frac{q^{\prime 2}}{sx^\prime y_{\rm bj\ , max}},\frac{x_{\rm bj\,min}}{x^\prime } \right)}^{z_{\rm max } } \frac{dz}{z}\,f_g ( z ) \int\limits_{x_{\rm bj\,min}}^{x^\prime z - \frac{m_{k}^2}{s } \frac{1}{y_{\rm bj\,max}-\frac{q^{\prime 2}}{sx^\prime z } } } \frac{dx_{\rm bj}}{x_{\rm bj}}\ , \\[2ex ] \nonumber & & \times \int\limits_{{\rm max}\left ( \frac{q^{\prime 2}}{sx^\prime z}+\frac{m_{k}^2}{s } \frac{1}{x^\prime z -x_{\rm bj}},y_{\rm bj\,min } \right)}^{y_{\rm bj\,max } } \frac{dy_{\rm bj}}{y_{\rm bj}}\ , \theta ( s x_{\rm bj } y_{\rm bj } - q^2_{\rm min})\ , \\[2ex ] \nonumber & & \times \frac{1+(1-y_{\rm bj})^2}{y_{\rm bj}}\ p_{q^\prime}^ { { ( i)}}\ .\end{aligned}\ ] ] it involves integrations over the gluon density @xmath33 , the virtual photon flux and the flux of virtual ( anti-)quarks @xmath34 in the instanton - background @xcite , @xmath35 all relevant kinematical variables in eq . ( [ evwgt ] ) are defined in fig . [ kin - var ] , and @xmath36 denotes the electric charge squared of the virtual ( anti-)quark @xmath34 in units of the electric charge squared @xmath37 . [ ht ] in eq . ( [ evwgt ] ) , @xmath38 denotes the instanton - induced total cross section of the @xmath39-subprocess ( c.f . [ kin - var ] ) and contains the essential instanton dynamics . its analytical form @xcite used in qcdins may be found in eq . ( [ qgcross - vxi ] ) . as illustrated in fig . [ isorho ] , @xmath38 is very steeply growing for decreasing values of @xmath40 and @xmath41 , respectively . eventually , our theoretical predictions based on instanton - perturbation theory @xcite will cease to hold . therefore , the following cuts inferred from a high - quality lattice simulation of qcd @xcite are implemented by default in qcdins 2.0 ( table [ floating ] ) , @xmath42 a further cut on the photon virtuality , @xmath43 is applied in order to warrant sufficient suppression of non - planar contributions @xcite , which are hard to calculate and may spoil the validity of eq . ( [ evwgt ] ) . the cross section @xmath38 exhibits a rather weak residual dependence on the renormalization scale @xmath44 . as an `` optimal '' choice , the value @xmath45 , corresponding to the minimum @xcite , @xmath46 , is taken by default ( table [ floating ] ) . however , @xmath38 depends strongly on the qcd scale @xmath47 . since , strictly speaking , the underlying theoretical framework refers to massless quarks , the ( default ) number of flavours is set to @xmath48 ( table [ floating ] ) . the respective value @xmath49 is obtained by a standard 3-loop perturbative flavour reduction ( eq . ( 9.7 ) of ref . @xcite ) from the 1998 world - average of the running qcd coupling at the z - boson mass @xcite , @xmath50 the central values of these parameters are taken as default in qcdins 2.0 ( table [ floating ] ) . this upgrade of @xmath47 , together with the modified cuts ( [ fiducial ] ) and ( [ fiducialq ] ) , represents an improved understanding of the input parameters and a considerable reduction of uncertainties , as compared to the original publication @xcite and earlier versions of qcdins . note that it also implies a significant change in the predicted rate . next , let us summarize the various stages of event generation by means of qcdins . in a first stage , the various bjorken variables @xmath51 of the instanton - induced process ( c.f . [ kin - var ] ) are generated , with a distribution according to the normalized differential cross section from eq . ( [ evwgt ] ) , @xmath52 in the second stage of the event generation , the 4-momenta @xmath53 of the incoming gluon @xmath54 , the virtual photon @xmath55 , the virtual quark @xmath34 and the current quark @xmath27 , respectively , are filled . sudakov decompositions of these momenta are used to incorporate various constraints , e.g. on the momenta squared . the 4-momentum @xmath56 of the outgoing lepton is calculated subsequently . in the third stage , the partonic final state of the instanton - induced @xmath57-subprocess is generated in its centre - of - mass system ( cms ) as follows . the number @xmath29 of produced gluons is generated according to a poisson distribution with mean @xmath58 ( for the cuts ( [ fiducial ] ) ) , as calculated theoretically ( eq . ( [ ng ] ) ) in instanton - perturbation theory @xcite . next , @xmath59 @xmath60$]-``strings '' of partons are set up , each beginning with a quark , followed by a random number @xmath61 of gluons and ending with an anti - quark of randomly chosen flavour . there are @xmath62 gluons in total and , due to the required flavour democracy ( [ flavour - democr ] ) , @xmath19-pairs of all @xmath4 flavours occur precisely once . a quark and a gluon among these @xmath63 partons are ( randomly ) marked as incoming . the momenta @xmath64 of the @xmath65 outgoing partons are then generated in the cms of the instanton subprocess , according to the energy - weighted phase - space @xmath66 these different energy weights for quarks and gluons @xcite , along with the angular isotropy @xcite , are characteristic features of the leading - order partonic final state ( after averaging over colour ) . next , the colour and flavour connections of the partons are set up . the colour flow is obtained simply by connecting the colour lines of adjacent partons within each of the above - mentioned @xmath4 @xmath60$]-``strings '' in a planar manner ( consistent with the leading order @xmath67 expectation ) . this choice is inspired by the leading - order partonic final state ( after averaging over colour ) @xcite , but may well deserve further research . the flavour flow is constructed by connecting the flavour lines of the quark at the beginning of a string with the flavour line of the anti - quark at the end of a string . the hard subprocess generation ends by boosting the momenta of the final state partons to the laboratory frame . while the subsequent perturbative evolution of the generated partons is always handled by the herwig @xcite package , the final hadronization step may optionally be performed also by means of jetset @xcite . this section is devoted to a systematic description of the various routines of the qcdins package that is designed as an `` add - on '' hard process generator , interfaced to the monte carlo generator herwig @xcite . this reference section is organized as follows : while all subroutines and functions of the qcdins package are described in _ alphabetical order _ in section [ routines ] , a _ logical flow - chart _ is provided in form of tables [ flow - chart ] and [ flow - qihgen ] below . they should always be used as the main guide through the description of the package . a routine listed in the n - th column and the m - th row of these tables progressively calls all routines in the ( n+1)-th column starting in the ( m+1)-th row . all routines called at the level of the main hard process generator * qihgen * and below are documented in table [ flow - qihgen ] . a specific application requires the writing of a steering program by the user ( c.f . section [ usage ] and appendix a ) . it must contain the standard herwig - initialization calls as well as the calls to various initialization routines for qcdins . the latter comprise essentially the four routines listed in the second column of table [ flow - chart ] . by calling the last one of these ( * qcloop * ) , the user starts the proper simulation which comprises the chain of internally called qcdins routines as documented in tables [ flow - chart ] and [ flow - qihgen ] . a specific and quite extensive example is provided with the qcdins distribution ( qtesthz.f ) and may be found in the directory qcdtest. it also illustrates the use of the event analysis routine * hwanal * called by herwig after each processed event . meters ( tables [ floating ] , [ flags ] ) & [ -1.5ex]*qiinit * & & & & + flavour reduction : & & & & & + @xmath68 & & [ -1.5ex]*lamnf * & & & + initialize particle " & & & & & + inst in event record & & [ -1.5ex]*qiinih * & & & + print input parame- & & & & & + ters and warnings & [ -1.5ex]*qistat * & & & & + initialize jetset & & & & & + common block data & [ -1.5ex]*gjeini * & & & & + loop over desired & & & & & + number of events & [ -1.5ex]*qcloop * & & & & + generate one event & & * qcdgen * & & & + assign hard process & & & & & + variables ( herwig ) & & & [ -1.5ex]_hwepro _ & & + no action ( modified & & & & & + herwig routine ) & & & & [ -1.5ex]*hwegam * & + call main instanton & & & & & + process generator & & & & [ -1.5ex]*hvhbvi * & + main ( hard ) instanton & & & & & + process generator & & & & & ' '' '' [ -1.5ex ] + generate parton & & & & & + cascades ( herwig ) & & & [ -1.5ex]_hwbgen _ & & + combine jets with & & & & & + correct kinematics & & & & [ -1.5ex]*hwbjco * & + convert herwig to & & & & & + jetset block data & & & [ -1.5ex]*herlund * & & + jetset event record & & & & & + to hepevt common & & & [ -1.5ex]*luhepc * & & + radiation from lepton & [ -1.5ex]*exfrac * & & + check kinematical boundaries & * qicalc & & + generate identity code of current & & & + quark @xmath27 and virtual quark @xmath69 & [ -1.5ex]*qihpar * & & + generate @xmath70 and associated weight & * qihins & & + generate @xmath71 as @xmath72 & & * qirdis & + calculate total cross section of & & & + instanton - induced subprocess @xmath73 & & [ -1.5ex]*q2sig * & + calculate @xmath74-valley action @xmath75 & & & * action + calculate fermionic overlap @xmath76 & & & * omega + calculate lambert w - function & & & * lambertw + calculate saddle - point value of & & & + conformally invariant @xmath74-distance @xmath77 & & & [ -1.5ex]*xi * + * * * * * * ' '' '' calculate inverse running coupling @xmath78 & & & ' '' '' * xqs * + generate number @xmath29 of emitted gluons & * qigmul & & + calculate gluon multiplicity @xmath79 & & * gmult & + * * ' '' '' calculate inverse running coupling @xmath78 & & & ' '' '' * xqs * + calculate @xmath74-valley action @xmath75 & & & * action + calculate lambert w - function & & & * lambertw + calculate flux of virtual quark @xmath69 & * qisplt & & + calculate remaining weight & * qipvwt & & + calculate momentum of incoming gluon & * qikpar & & + generate momentum of virtual photon & * qikgam & & + generate momenta of virtual quark @xmath69 & & & + and current quark @xmath27 & [ -1.5ex]*qikgsp * & & + generate partonic final state & * qistid & & + generate the partons of the instanton & & & + subprocess in form of @xmath80$]-strings & & [ -1.5ex]*qiglst * & + find the incoming partons & & & + in the @xmath80$]-strings & & [ -1.5ex]*qigpar * & + assign masses of outgoing partons & & * qiplst & + generate 4-momenta of outgoing partons & & * qipsgn & + calculate relative energy weight & & & + of outgoing partons & & & [ -1.5ex]*qipswt * + store 4-momenta of outgoing partons & & & + into phep common block of herwig & & [ -1.5ex]*qipsto * & + colour / flavour connections for each string & & * qiccon & + * * * * * * * * * * subroutine * action*(xi4,s , ds , dds ) abcdefghikl calculation of the @xmath74-action as well as its 1st and 2nd derivatives , as function of the conformally invariant @xmath74-distance . + abcdef conformally invariant @xmath74-distance @xmath77 . @xmath74-action @xmath75 , eq . ( [ action ] ) . @xmath81 , @xmath82 the action is calculated according to the exact valley form @xcite , @xmath83 \ln\left [ \frac{1}{2\xi } \bigl ( f(\xi ) + 4\bigr)\right ] \ , , \\[2.4ex ] f(\xi ) & = & \xi^2+\sqrt{\xi^2 - 4}\xi-4 \ , . \end{aligned}\ ] ] subroutine * exfrac*(a ) abcdefghikl optional account of initial state radiation from the lepton . + ab abcdef dummy rescaling factor of the incoming lepton momentum ; double precision output variable . the actual routine has to be provided by the user . subroutine * gjeini * abcdefghikl initialization of the jetset @xcite parameter common blocks ludat1 , ludat2 , ludat3 , ludat4 and ludatr . + * gjeini * has to be called by the user before any other jetset routine . * gjeini * is from ref . @xcite . function * gmult*(xpr , xi_min , xi_max , qlam , kappa , nf , loopfl ) abcdefghikl calculation of the average gluon multiplicity @xmath84 depending on @xmath41 , @xmath85 , @xmath86 , @xmath4 and loop - order . here , @xmath44 and @xmath4 denote the renormalization scale and the number of light flavours , respectively . + abcdefghi @xmath41 @xmath87 ; lower boundary of @xmath77 used for interpolation . @xmath88 ; upper boundary of @xmath77 used for interpolation . @xmath85 @xmath86 @xmath4 ; number of light flavours . abcd 1-loop renormalization group ( rg ) invariance @xcite along with 1-loop form of @xmath89 . 2-loop rg invariance @xcite along with 2-loop form of @xmath90 . ( default ) 2-loop rg invariance along with 3-loop form of @xmath90 . from an analysis based on the generalized ( mueller @xcite ) optical theorem for the @xmath91 forward scattering amplitude and the @xmath74-valley method , one infers @xcite the differential one - gluon inclusive @xmath92 cross section , normalized by the total cross section @xmath38 . the mean gluon multiplicity @xcite is then found by phase space integration , @xmath93 the function * gmult * calculates and returns the average gluon multiplicity ( [ ng ] ) . the stars @xmath94 in eq . ( [ ng ] ) denote the saddle point values of the @xmath74 collective coordinates @xmath95 , @xmath96 and @xmath77 . their computation proceeds as in the descriptions of the functions * q2sig * and * xi*. the required values of @xmath97 are calculated and returned by the function * xqs*. the @xmath74-action @xmath75 and its @xmath77-derivatives are provided by the subroutine * action*. subroutine * herlund * abcdefghikl conversion of the herwig @xcite event record in the hepevt common block to the respective jetset @xcite common block . + * herlund * is a modified @xcite copy from the jetset subroutine luhepc . subroutine * hvhbvi * abcdefghikl call of the main ( hard ) instanton process generator * qihgen*. + replaces a dummy stub in herwig @xcite that was originally used as event generation interface for the monte carlo generator herbvi @xcite for baryon number violating interactions . used in the qcdins package to select qcd - instanton induced processes via mod(iproc/100,100 ) @xmath98 75 . the process code iproc (= 17600 ) has to be set in the user s steering program ( c.f . appendix a ) . furthermore , the qcdins program header is printed . subroutine * hwbjco * abcdefghikl modification of herwig @xcite routine to account for instanton - induced scattering . + the modifications are @xcite : the logical flag ( dispro ) for keeping the lepton momenta fixed in herwig 5.9 is modified to include also instanton - induced dis , ipro = 76 . furthermore , a bug concerning energy - momentum conservation in the original routine from herwig 5.9 has been fixed . subroutine * hwegam*(ihep , zmi , zma , wwa ) abcdefghikl modification to avoid standard generation of the ( virtual ) photon at this stage . + c.f . ref . @xcite this is a modified routine from herwig 5.9 ( c.f . ref . usually , * hwegam * generates an incoming photon from the incoming @xmath99 . within qcdins , however , the photon is generated at a later stage in the subroutine * qikgam*. thus , the herwig routine * hwegam * has been modified to immediately return for instanton - induced processes ( ipro=76 ) . function * lambertw*(x ) abcdefghikl calculation of the principal branch of the lambert w - function @xmath100 for @xmath101 . + x : @xmath102 ; argument of the lambert w - function . @xmath100 is the ( real ) solution of @xmath103 , analytic at @xmath104 . the following simple , but accurate approximation is used and returned by * lambertw * : @xmath105 \ln(x-4)-(1-\frac{1}{\ln(x)})\cdot\ln(\ln(x ) ) ; & { \rm\ for\ } x > 500\ , . \end{array } \right.\ ] ] function * lamnf*(nf , lambda5 ) abcdefghikl calculation of @xmath106 from @xmath107 to order @xmath108 . + abcdefghijklm number of ( light ) flavours , @xmath4 . input value @xmath107 . the flavour reduction of @xmath107 to the desired number of light flavours is performed by using eq . ( 9.7 ) of ref . @xcite . subroutine * luhepc*(mconv ) abcdefghikl conversion of the jetset @xcite event record contents back to the hepevt common block . + mconv = 1 the present routine is a modified @xcite version of the jetset routine * luhepc*. function * omega*(xi4 ) abcdefghikl calculation of the fermionic overlap , as function of the conformally invariant @xmath74-distance . + xi4 : conformally invariant @xmath74-distance @xmath77 . the following simple , but accurate approximation for the fermionic overlap @xcite @xmath109 is used and returned by * omega * : @xmath110 function * q2sig*(xprime , qlam , kappa , loopfl , nf ) abcdefghikl calculation of the total cross section @xmath111\ ] ] for the instanton - induced subprocess , depending on @xmath41 , @xmath85 , @xmath86 , loop - order and @xmath4 . here , @xmath44 and @xmath4 denote the renormalization scale and the number of light flavours , respectively . + abcdefghi @xmath41 @xmath85 @xmath86 abcd 1-loop renormalization group ( rg ) invariance @xcite along with 1-loop form of @xmath89 . 2-loop rg invariance @xcite along with 2-loop form of @xmath90 . ( default ) 2-loop rg invariance along with 3-loop form of @xmath90 . @xmath4 ; number of light flavours . the function * q2sig * calculates and returns the cross section ( [ cs ] ) as derived in ref . @xcite , @xmath112 & & \times \frac{\omega ( \xi_\ast ) ^{2n_f-1}(\xi_\ast -2)^3 v^{\ast\,5 } } { \sqrt{\frac{1}{2}({\tilde{s}}-v_\ast -2 { d(\tilde{s})})^2 + { \tilde{s}}({\tilde{s}}-v_\ast ) { d\left ( \ln \left ( \frac{{d(\tilde{s})}}{\sqrt{\xi_\ast -2 } } \right)\right ) } } } \\[1.6ex ] & & \times \left ( \frac{4\pi } { \alpha_{\overline{\rm ms}}\left(\mu _ r \right)}\right)^{19/2 } \exp \left [ -\frac{4\pi } { \alpha_{\overline{\rm ms}}\left(\mu _ r \right ) } s^{({i\overline{i}})}\left(\xi_\ast \right ) - 2 \left(1-\ln\left(\frac{v_\ast\mu_r}{q^\prime}\right)\right ) \,{\tilde{s}}\right ] \ , . \nonumber\end{aligned}\ ] ] it is expressed entirely in terms of the saddle point values for the @xmath74 collective coordinates , @xmath113 ( conformally invariant distance ) and @xmath114 ( scaled size ) . for given @xmath115 , @xmath85 and ( scaled ) renormalization scale @xmath86 , these are in turn unique solutions of the saddle point equations @xcite @xmath116 \label{eq1 } v_\ast & = & 2\ , d({\tilde{s}}(\xi_\ast))\ , w\left ( \frac{q^\prime}{\mu_r } \frac{\exp\left\{\frac{1}{2 } \left[\frac{4\pi } { \alpha_{\overline{\rm ms}}\left(\mu _ r \right)}\frac{1}{\delta_1\beta_0 } + \frac{{\tilde{s}}(\xi_\ast)}{d({\tilde{s}}(\xi_\ast ) ) } \right ] \right\ } } { 2\,d({\tilde{s}}(\xi_\ast ) ) } \right),\end{aligned}\ ] ] with @xmath117 and @xmath118 the @xmath74-action @xmath75 as well its @xmath77-derivatives , entering the cross section ( [ qgcross - vxi ] ) and eqs . ( [ eq2 ] ) , ( [ eq1 ] ) through @xmath119 are calculated in the subroutine * action*. the fermionic overlap @xmath120 is calculated and returned by the function * omega*. in eq . ( [ eq1 ] ) , @xmath121 denotes the principal branch of the lambert @xmath121-function , i.e. the ( real ) solution of @xmath103 , analytic at @xmath104 . the latter is calculated and returned by the function * lambertw*. the first step in the solution of the saddle - point equations ( [ eq2 ] ) , ( [ eq1 ] ) consists in eliminating @xmath122 in eq . ( [ eq2 ] ) by inserting eq . ( [ eq1 ] ) . next , for given @xmath115 , @xmath123 and @xmath86 , the resulting implicit equation is solved numerically for @xmath113 . this is done by the function * xi * which provides @xmath124 on return . the latter is then inserted into eq . ( [ eq1 ] ) providing @xmath125 . the values of @xmath97 are calculated and returned by the function * xqs*. subroutine * qcdgen * abcdefghikl interface for the generation of one instanton - induced event , including calls to event initialization , hadronization and event termination routines . + * an instanton - induced event is initialized by a call to the herwig @xcite subroutine * hwuine*. * the partonic instanton subprocess is generated by a call to the herwig subroutine * hwepro*. * if the hadronization flag qicont(21 ) is set .true . ( default ) , the event is fully hadronized . else , the event is finalized ( * hwufne * ) immediately after the call of * hwepro*. depending on the control flag for the hadronization model , qicont(18)=.true./.false . , the hadronization is either performed by appropriate herwig @xcite or ( modified @xcite ) jetset @xcite routines , respectively ( see table [ hadroutines ] ) . * furthermore , in this routine energy - momentum conservation in the generated event is checked . .[hadroutines ] depending on the flag qicont(18 ) = .true./.false . , herwig @xcite and jetset @xcite hadronization may be selected . the respective calls to herwig and ( modified ) jetset routines used in * qcdgen * are displayed . [ cols="^,<,<",options="header " , ] subroutine * qiusps(wt ) * abcdefghikl dummy routine for optional modification of relative phase space weight as calculated in * qipswt*. + wt : modified relative phase space weight . called from * qipswt * if qicont(2 ) = .false .. calculation of wt has to be provided by the user . function * xi * ( xpr , xi_min , xi_max , xq , xmu , nf , loopfl ) abcdefghikl the saddle point value of the conformally invariant @xmath74-separation @xmath77 is calculated as function of @xmath115 , @xmath126 , @xmath127 and @xmath4 . + + abcdefghi @xmath115 @xmath87 ; lower boundary of @xmath77 used for interpolation . @xmath88 ; upper boundary of @xmath77 used for interpolation . @xmath126 @xmath127 @xmath4 ; number of light flavours . abcd 1-loop renormalization group ( rg ) invariance @xcite along with 1-loop form of @xmath89 . 2-loop rg invariance @xcite along with 2-loop form of @xmath90 . ( default ) 2-loop rg invariance along with 3-loop form of @xmath90 . first , the saddle point equation ( [ eq2 ] ) is solved analytically for @xmath41 , and the @xmath128-values corresponding to a discrete set of @xmath129 values are calculated as @xmath130 with @xmath131 and @xmath132 from eqs . ( [ eq1])([dadada ] ) inserted : @xmath133 \nonumber & & \hspace{-5ex } w\left ( \frac { \exp\left\ { \frac{1}{2}\frac{{\tilde{s}}}{{d(\tilde{s } ) } } + \frac{1}{2 } \frac{\delta_1 - 1}{\delta_1\beta_0 } \left [ \delta_1 \ln \left ( \frac{\delta_1 x(\mu_r ) } { x(q^\prime ) + ( \delta_1 - 1)x(\mu_r ) } \right ) -1 \right ] x(\mu_r ) + \frac{1}{2}\frac{x(q^\prime ) } { \beta_0 } \right\ } } { 2\,{d(\tilde{s } ) } } \right ) .\end{aligned}\ ] ] in eq . ( [ vastxq ] ) , the explicit dependence of @xmath132 on @xmath134 ( c.f . ( [ eq1 ] ) has been eliminated in favour of an @xmath135 , @xmath136 dependence by means of the standard scale transformation of @xmath90 . here , we have introduced the shorthand @xmath137 the desired continuous inversion @xmath138 is then achieved by means of numerical interpolation based on the above exact supporting points ( @xmath139 ) . function * xqs*(qlam , loopfl , nf ) abcdefghikl calculation of @xmath140 as function of @xmath141 , where @xmath142 is a generic mass scale . + abcdefghik @xmath141 abcd 1-loop form of @xmath89 . 2-loop form of @xmath90 . ( default ) 3-loop form of @xmath90 . @xmath4 ; number of light flavours . the running coupling @xmath90 is calculated according to the explicit formula eq . ( 9.5a ) in ref . @xcite , which is accurate to 3-loop . the quantity @xmath97 is returned . the integer variable loopfl = @xmath143 controls the loop - order at which eq . ( 9.5a ) in ref . @xcite is truncated . qcdins 2.0 should be loaded together with herwig @xcite version 5.9 and jetset @xcite version 7.4 , that are part of the cernlib distribution . the program is a slave system , which the user must call from his own steering program . a very simple example is provided in appendix a. by default , qcdins is compiled in form of a library , libqcdins.a , that may be linked together with libherwig59.a , libjetset74.a and the library libpdflib.a of parton distribution functions to the steering program . an extensive demonstration program , including an interface to the hztool @xcite package , and a detailed installation instruction are included in the distribution . information about the distribution , its update history , an interactive manual , the source code , pictures of typical events etc . can be accessed via the qcdins www site , http://www.desy.de/~t00fri/qcdins/qcdins.html . first of all , we are grateful to m. gibbs who left physics in may 1995 . without his early contributions the qcdins package would presumably not exist . a number of people helped improving the code . in particular , let us mention t. carli , g. grindhammer , h. jung and m. seymour . specifically , we thank h. jung for his efforts to incorporate an interface to the jetset package . moreover , we are grateful to m. seymour for an evaluation of the qcdins package during the desy workshop 1998/99 on monte carlo generators for hera physics . finally , we thank g. ingelman for a careful reading of the manuscript . the following code may serve as the simplest illustration of a steering program for the qcdins / herwig package . note that the two subroutines * hwanal * and * hwaend * must exist . for an example of an interface with the jetset hadronization , we refer to the extensive steering program qtesthz included in the qcdins distribution . .... program qcdins # include " herwig.inc " c force inclusion of block data external hwudat c initialize process number iproc = 17600 c maximal number of events in this run maxev = 1000 c beam particles pbeam1 = 27.5d0 pbeam2 = 820.0d0 part1 = ' e+ ' part2 = ' p ' c initialize common blocks call hwigin c user can reset parameters at this point , otherwise values c set in hwigin will be used . c no vertex information in event printout prvtx=.false . c reset number of shots for initial max weight search ibsh = 5000 lrsud=0 lwsud=77 c seeds nrn(1)=106645412 nrn(2)=135135176 c use laboratory frame usecmf = .false . c compute parameter - dependent constants call hwuinc c number of herwig events to print out maxpr = 1 c call hwusta to make any particle stable call hwusta('pi0 ' ) call hwusta('k_s0 ' ) c initialize default qcdins input parameters call qiinit c print input parameters call qistat c loop over events call qcloop c user 's terminal calculations call hwaend stop end subroutine hwanal return end subroutine hwaend return end .... below , we display the essential output from a test run of the very simple steering program from appendix a. this test run simulates 1000 complete instanton - induced events in deep - inelastic @xmath144 scattering ( hera ) , with @xmath145 gev and @xmath146 gev , in the laboratory frame . all the qcdins parameters correspond to the default values as set in the initialization routine * qiinit * ( c.f . tables [ floating ] and [ flags ] ) . .... = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = qcd instanton monte carlo information version 2.0 = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = parameter value = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = default gluon mass ( gev ) 0.7500e+00 minimum allowed value of x_bj 0.1000e-02 maximum allowed value of z ( proton mom.frac . ) 0.1000e+01 maximum allowed value of y_bj 0.1000e+01 minimum allowed value of y_bj 0.1000e+00 maximum allowed value of x prime 0.9000e+00 minimum allowed value of x prime 0.3500e+00 maximum allowed value of q prime * * 2 ( gev**2 ) 0.1652e+04 minimum allowed value of q prime * * 2 ( gev**2 ) 0.1134e+03 lower cut for me calculation on x prime 0.3500e+00 lowest allowed weight efficiency cut -.1000e-09 minimum instanton invariant mass 0.0000e+00 lambda - ms - bar(nf ) [ gev ] from pdg 1998 0.3459e+00 minimum total k.e . of outgoing partons ( gev ) 0.0000e+00 renormalization point kappa = mu_r / qprime : 0.1500e+00 minimum allowed value of q * * 2 ( gev**2 ) 0.1134e+03 factorization scale ( gev ) 0.1065e+02 = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = i d code assumed for instanton 206 maximum number of mambo iterations per ps wt . 100 maximum number of ps wt . rejections per event 300 number nf of ( light ) flavours 3 maximum average gluon multiplicity in distbn . 10 maximum number of iterations for me generation 20 maximum number of qigmul iterations 40 number of loops in rg - invariance / alpha_s 3 = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = control flag option setting = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = reweight phase space configurations true use default phase space reweighting true use matrix element weight true disregard instanton minimum mass requirement true generate q prime before x prime true enforce mass of current quark in kinematics true enforce limit on maximum number of gluons true ensure mass less than subprocess energy true kill events with insufficient instanton mass true use z generation as dz / z true use x prime * * -n generation for efficiency true use q prime * * -n generation for efficiency true use herwig rather than jetset hadronization true use random azimuth angle for scattered electron true use full hadronization true = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = process code iproc = 17600 random no . seed 1 = 1246579 seed 2 = 8447766 number of shots = 5000 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * qcdins version 2.0 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * .... .... new maximum weight = 2.4520378923431534e-05 new maximum weight = 6.0858601240747906e-02 new maximum weight = 0.1289870599803515 new maximum weight = 0.3662942637479250 new maximum weight = 0.5705336816482935 new maximum weight = 0.8061467165013312 number of events = 0 number of weights = 5000 mean value of wgt = 2.8721e-02 rms spread in wgt = 6.4377e-02 actual max weight = 7.3286e-01 assumed max weight = 8.0615e-01 ihep i d idpdg ist mo1 mo2 da1 da2 p - x p - y p - z energy mass 1 e+ -11 101 0 0 4 0 0.00 0.00 27.50 27.50 0.00 2 p 2212 102 0 0 0 0 0.00 0.00 -820.00 820.00 0.94 3 cmf 0 103 1 2 0 0 0.00 0.00 -792.50 847.50 300.33 ihep i d idpdg ist mo1 mo2 da1 da2 p - x p - y p - z energy mass 5 e+ -11 121 7 9 21 9 0.00 0.00 27.50 27.50 0.00 6 gluon 21 122 7 20 22 19 0.00 0.00 -50.37 50.38 0.75 7 hard 0 120 5 6 9 20 0.53 0.99 -22.98 77.98 74.51 8 inst 0 3 7 0 0 0 0.23 9.33 -43.54 47.01 15.07 9 e+ -11 123 7 5 26 5 8.20 -7.47 3.48 11.63 0.00 10 ubar -2 124 7 14 27 17 -8.43 -1.86 17.19 19.24 0.31 11 dqrk 1 124 7 12 31 13 -0.07 0.52 -7.65 7.67 0.32 12 gluon 21 124 7 13 33 11 0.16 -0.28 -0.64 1.04 0.75 13 dbar -1 124 7 11 35 12 -0.67 -0.12 -2.65 2.75 0.32 14 uqrk 2 124 7 15 37 10 0.01 4.97 -11.73 12.75 0.32 15 gluon 21 124 7 16 39 14 0.53 0.62 -2.17 2.44 0.75 16 gluon 21 124 7 17 41 15 0.43 -0.43 -8.44 8.49 0.75 17 gluon 21 124 7 10 43 16 -0.72 0.09 -2.40 2.62 0.75 18 sqrk 3 124 7 19 45 20 -0.11 2.50 -4.93 5.55 0.50 19 gluon 21 124 7 6 47 18 0.81 0.93 -1.46 2.06 0.75 20 sbar -3 124 7 18 51 6 -0.14 0.52 -1.47 1.64 0.50 21 z0/gama * 23 3 5 7 0 0 -8.20 7.47 24.02 15.87 -21.17 .... .... check of energy - momentum conservation in the event : = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = sum_i(p_i)[gev ] = ( 4.87e-13 -5.49e-13 1.59e-12 -4.20e-12 ) with 60 stable particles in final state contributing . number of events = 1000 number of weights = 27258 mean value of wgt = 2.8837e-02 rms spread in wgt = 6.3371e-02 actual max weight = 7.0199e-01 assumed max weight = 8.0615e-01 9 a. belavin , a. polyakov , a. schwarz and yu . tyupkin , . t hooft , ; ; ( erratum ) ; . s. moch , a. ringwald and f. schrempp , . a. ringwald and f. schrempp , . a. ringwald and f. schrempp , hep - ph/9411217 , in : quarks 94 , eds . d. grigoriev , v. matveev , v. rubakov , d. son and a. tavkhelidze ( world scientific , singapore , 1995 ) p. 170 . m. gibbs , a. ringwald and f. schrempp , hep - ph/9506392 , in : workshop on deep inelastic scattering and qcd ( dis 95 ) , eds . f . laporte and y. sirois ( ecole polytechnique , paris , 1995 ) p. 341 . t. carli , j. gerigk , a. ringwald and f. schrempp , hep - ph/9906441 , in : monte carlo generators for hera physics , eds . a.t . doyle , g. grindhammer , g. ingelman and h. jung , desy - proc-1999 - 02 ( desy , hamburg , 1999 ) p. 329 . g. marchesini et al . , . s. moch , a. ringwald and f. schrempp , hep - ph/9706400 , in : deep inelastic scattering and qcd ( dis 97 ) , eds . j. repond and d. krakauer ( american institute of physics , woodbury , new york , 1997 ) p. 1007 . a. ringwald and f. schrempp , hep - ph/9706399 , in : deep inelastic scattering and qcd ( dis 97 ) , eds . j. repond and d. krakauer ( american institute of physics , woodbury , new york , 1997 ) p. 781 . s. aid et al . , h1 collaboration , . s. aid et al . , h1 collaboration , . t. carli and m. kuhlen , . s. moch , a. ringwald and f. schrempp : in preparation a. ringwald and f. schrempp , . c. caso et al . , particle data group , . h. jung , private communication . t. sjstrand , . v.v . khoze and a. ringwald , . j. verbaarschot , . g. grindhammer , private communication . a. mueller , . m. gibbs and b. webber , . t. carli , private communication . j. owens , . r. kleiss and w. stirling , . m. seymour , private communication . j. bromley et al . , in : future physics at hera , eds . g. ingelman , a. de roeck and r. klanner ( desy , hamburg , 1996 ) vol . 1 ,

we treat the case of a flat stress - free surface ( i.e. , the ground in seismological applications ) separating air from a homogeneous , isotropic , solid substratum overlain by a homogeneous , isotropic , solid layer ( in contact with the ground ) solicited by a sh plane body wave incident in the substratum . the analysis is first carried out in the frequency domain and subsequently in the time domain . the frequency domain response is _ normal _ in that no resonances are excited ( a resonance is here understood to be a situation in which the response is infinite in the absence of dissipation ) . the translation of this in the time domain is that the scattered pulse is of relatively - short duration . the duration of the pulse is shown to be largely governed by radiation damping which shows up in the imaginary parts of the complex eigenfrequencies of the configuration . three methods are elaborated for the computation of the time history and give rise to the same numerical solutions for a large variety of configurations of interest in the geophysical setting under the hypothesis of non - dissipative , dispersionless media . the method appealing to the complex eigenfrequency representation is shown to be the simplest and most physically - explicit way of obtaining the time history ( under the same hypothesis ) . moreover , it is particularly suited for the case in which modes can be excited as occurs when the incident wave is not plane or the boundary condition is not of the stress - free variety for all transverse coordinates on the ground plane .

physics0605117

this work is inspired by the problem of predicting the effects of earthquakes in cities . it is known that the most dangerous effects are produced in cities built on soft underground underlain by a hard substratum . a simple model of the city is considered herein in which the buildings are absent ( i.e. , the ground is flat ) , the soft underground is constituted by a homogeneous , soft layer overlying , and in welded contact with , a homogeneous , hard substratum . this configuration is solicited by a sh plane body wave and the object is to determine the time history of response on the ground , preferably in a numerically - efficient , physically - understandable manner . in the following , we shall be concerned with the determination of the vectorial displacement field @xmath0 on , and underneath , the ground in response to a seismic solicitation . in general , @xmath0 is a function of the spatial coordinates , incarnated in the vector @xmath1 and time @xmath2 , so that @xmath3 . we first carry out our analysis in the frequency domain , and thus search for @xmath4 , with @xmath5 the angular frequency . fourier analysis tells us that @xmath6 and @xmath4 are related by @xmath7 wherein it should be noted that @xmath4 is a generally - complex function , whereas @xmath6 is a real function . the second step will therefore deal with the computation of the integral in ( [ fourier ] ) . since everything is invariant with @xmath8 , the analysis takes place in the @xmath9 ( sagittal ) plane depicted in fig . [ fig5 ] . in this figure : @xmath10 designates the ( trace of the ) interface between the substratum ( half - space domain @xmath11 ) and the soft layer ( laterally - unbounded domain @xmath12 ) , and @xmath13 designates the ( trace of the ) flat ground . the medium @xmath14 above the latter is air , assumed to be the vacumn for the purpose of the analysis . the media in the layer and substratum are the elastic ( or viscoelastic ) solids @xmath15 and @xmath16 respectively . the incident plane wave propagates in @xmath11 toward the the interface @xmath10 and the ground @xmath13 . since the latter is stress - free ( i.e. , the normal and tangential components of traction are nil on the boundary ) , the total displacement field vanishes in the region @xmath17 above the boundary ( see fig . [ fig5 ] ) . [ ptb ] one can always choose the cartesian coordinate system so that the wavevector associated with the incident shear wave ( the subscript @xmath18 will constitute a reminder that we deal with shear = transverse waves in the following ) lies in the @xmath9 plane . this is assumed herein and signifies that the displacement associated with this wave is perpendicular to the @xmath9 plane and therefore lies in a horizontal plane . thus , the incident wave is a shear wave and the associated displacement is horizontal ; i.e. , a shear - horizontal ( sh ) wave . moreover , the motion associated with this wave is , due to the choice of the cartesian reference system , independent of the coordinate @xmath8 . this implies that the resultant total motion induced by this incident wave is independent of @xmath8 , i.e. , the boundary value problem is 2d , so that it is sufficient to look for the displacement field in the @xmath9 plane . actually , since we already know that the total displacement vanishes in the half plane above the boundary we must look for the total displacement field ( hereafter designated by @xmath19 ) only in @xmath11 and @xmath12 . hereafter , we designate the density and lam parameters in @xmath20 by @xmath21 and @xmath22 ( for @xmath23 ) respectively . the mathematical translation of the boundary value problem in the _ space - frequency domain _ is : @xmath24 @xmath25 @xmath26 @xmath27 @xmath28 @xmath29 @xmath30 @xmath31 @xmath32 @xmath33 @xmath34 @xmath35 wherein @xmath36 is the ( unknown ) diffracted field in @xmath37 , @xmath38 the kronecker delta symbol , and : @xmath39 @xmath40 @xmath41 being the angle of incidence with respect to the @xmath42 axis . ( [ sevenhunthirteen ] ) is the space - frequency domain equation(s ) of motion , ( [ sevenhunfourteen])-([sevenhunthirtyeight30 ] ) the boundary condition(s ) , ( [ sevenhunfourtyone ] ) the radiation condition , and ( [ sevenhunfourtytwo])-([sevenhunfourtythree ] ) the description of the incident wave . until further notice , we drop the @xmath43dependence on all field quantities and consider it to be implicit . as in the previous section , and on account of the outgoing wave condition(s ) ( [ sevenhunfourtyone ] ) , we adopt the following field representations : @xmath44dk_{1}~,\ ] ] @xmath45dk_{1}~,\ ] ] @xmath46 @xmath47dk_{1}~,\end{gathered}\ ] ] @xmath48dk_{1}~,\end{gathered}\ ] ] @xmath49 dk_{1}~,\ ] ] wherein @xmath50 @xmath51 with @xmath52 note that these field representations involve nine unknown functions @xmath53 , @xmath54 , @xmath55 . the latter will be obtained by applying the nine boundary conditions embodied in ( [ sevenhunfourteen])-([sevenhunthirtyeight30 ] ) . the use of ( [ sevenhunfourteen ] ) , ( [ sevenhunfourteenb ] ) , ( [ sevenhunthirtyeight25 ] ) , ( [ sevenhunthirtyeight27 ] ) , ( [ sevenhunthirtyeight28 ] ) , and ( [ sevenhunthirtyeight30 ] ) , in ( [ sevenhunfourtyfour])-([sevenhunfiftytwo ] ) gives rise to : @xmath56 the next step consists in using ( [ sevenhunfourteena ] ) in ( [ sevenhunfourtynine ] ) to obtain : @xmath57 \exp(ik_{1}x_{1})dk_{1}=0~~;\forall x_{1}\in\mathbb{r}~.\ ] ] by fourier inversion we get @xmath58 whence @xmath59 wherein @xmath60 by the same token , we can write ( [ sevenhunfourtynine ] ) as @xmath61 so that using ( [ sevenhunfiftyfive ] ) and ( [ sevenhunfiftyseven ] ) together with ( [ sevenhunthirtyeight26 ] ) and ( [ sevenhunthirtyeight29 ] ) leads to : @xmath62 @xmath63 by fourier inversion we find : @xmath64 @xmath65 this system of two equations can be written as the matrix equation @xmath66 it follows that : @xmath67 wherein @xmath68 \exp[-i(k_{2}^{i}+k_{2}^{0})h]~,\ ] ] and @xmath69 wherein @xmath70 \exp(-ik_{2}^{i}h ) ~.\ ] ] the consequence of all this is that : @xmath71 @xmath72 @xmath73 @xmath74 wherein @xmath75 @xmath76 \exp[-2ik_{2}^{0i}h]~,\ ] ] and @xmath77 \exp[-ik_{2}^{0i}h ] ~.\ ] ] these results , together with ( [ sevenhunfourtytwo ] ) , indicate that the diffracted fields in @xmath11 and @xmath12 have the same ( sh ) polarization ( @xmath78 in fig . [ fig5 ] ) as the incident field . the preceding results show that the fields in the two media can be decomposed as follows : @xmath79 wherein @xmath80 } ~,\ ] ] @xmath81}~,\ ] ] and @xmath82 wherein @xmath83}~,\ ] ] @xmath84}~.\ ] ] these results indicate that the diffracted field in @xmath11 reduces to a specularly - reflected wave @xmath85 and the diffracted ( as well as total ) field in @xmath12 reduces to a sum of a refracted - reflected wave @xmath86 and a refracted - transmitted wave @xmath87 . @xmath88 and @xmath85 are plane homogeneous ( body ) waves so that the field in the half - space underneath the layer is composed of two body waves . since we consider only the case of geological interest in which the substratum is harder than the layer , the s - wave phase velocity in the substratum is larger than the s - wave phase velocity in the layer , which means that @xmath89 and consequently @xmath90 is real , so that both @xmath87 and @xmath86 are body waves . thus , the field in the layer is also composed of two body waves . an important corollary of the previous remarks is ( in the case of geological interest ) that an incident ( plane ) body wave in the substratum can only excite ( plane ) body waves in both the substratum and the layer . thus , if we want a surface wave to be excited somewhere underneath the ground , then we have to introduce some sort of modification of either the excitation , the nature of the media , or the nature of the boundary condition on the ground . recall that the frequency domain displacement response in the layer is of the form @xmath91 with @xmath92}~,\ ] ] so that one half of the total frequency domain displacement response on the ground at point @xmath93 is given by @xmath94 wherein : @xmath95 @xmath96 @xmath97 @xmath98^{2}-(s^{i})^{2}}~. % % \ ] ] in the five figures [ aodrickb1 ] , [ aodrickb3 ] , [ aodrickb5 ] , [ aodrickb7 ] , [ aodrickb7a ] we plot the spectrum ( @xmath99 ) of a ricker pulse excitation , the transfer function @xmath100 , and the spectrum of the displacement response @xmath101 . [ ptb ] . @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath107 , corresponding to a case of separated pulses . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] [ ptb ] . @xmath102 , @xmath108 , @xmath104 , @xmath109 , @xmath106 , @xmath110 , corresponding to a case of merged pulses . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] [ ptb ] . @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath110 , corresponding to a case of separated pulses . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] [ ptb ] . @xmath102 , @xmath108 , @xmath104 , @xmath109 , @xmath106 , @xmath111 , corresponding to a case of merged pulses . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] [ ptb ] . @xmath102 , @xmath103 , @xmath112 , @xmath109 , @xmath106 , @xmath107 , corresponding to a case of a so - called `` anomalous '' pulse . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] we note that the spectrum of displacement response contains spikes that are all the sharper and more intense the larger is the contrast between @xmath113 and @xmath114 . as will be shown further on , these sharper spikes lead to larger - duration response in the time domain . we had @xmath115 and due to the fact that @xmath116 is a real function , we must have @xmath117^{*}=u_{3}^{j}(\mathbf{x},-\omega)~,\ ] ] ( wherein the symbol @xmath118 designates the complex conjugate operator ) from which it follows that @xmath119 we shall employ ( [ time3 ] ) or ( [ time1 ] ) to obtain the temporal response from the frequency response function @xmath120 . more precisely , we shall be concerned with the evaluation of @xmath121 \exp(-i\omega t)d\omega~~;~~j=0,1~.\ ] ] for a so - called _ upward - propagating sh - plane wave _ , we have @xmath122 wherein @xmath123 and @xmath41 is the incident angle measured clockwise from the @xmath124 axis . in the above relations , @xmath99 is termed the _ frequency - domain amplitude _ or _ spectrum _ of the incident sh plane wave . the _ time history _ of this incident wave is @xmath125 or @xmath126 the amplitude spectrum @xmath99 of a ricker pulse ( sanchez - sesma 1985 ) is given by @xmath127 wherein @xmath128 , @xmath129 and @xmath130 are real constants ( i.e. , independent of @xmath5 ) . it follows that @xmath131^{*}=-\mathcal{a}\frac{1}{\sqrt{\pi}}\frac{\omega^{2}}{4\alpha^{3 } } \exp\left ( -i\beta\omega-\frac{\omega^{2}}{4\alpha^{2}}\right)=a^{i-}(-\omega)~,\ ] ] so that the temporal history associated with this pulse is @xmath132 more precisely : @xmath133 d\omega= \\ \left ( -\mathcal{a}\frac{1}{4\alpha^{3}\sqrt{\pi}}\right ) \int_{-\infty}^{\infty}\omega^{2 } \exp\left [ -\frac{1}{4\alpha^{2}}\{\omega^{2}+4\alpha^{2}i\omega(t-\beta)\}\right ] d\omega~.\end{gathered}\ ] ] but @xmath134^{2}+4\alpha^{4}(t-\beta)^{2 } ~,\ ] ] so that @xmath135 \int_{-\infty}^{\infty}\omega^{2 } \exp\left [ -\frac{\{\omega+i2\alpha^{2}(t-\beta)\}^{2}}{4\alpha^{2}}\right ] d\omega ~,\ ] ] which , after the change of variables @xmath136 becomes @xmath137 \int_{-\infty}^{\infty}[\varpi - i2\alpha^{2}(t-\beta)]^{2 } \exp\left [ -\frac{\varpi^{2}}{4\alpha^{2}}\right ] d\varpi:= \\ \left ( -\mathcal{a}\frac{1}{4\alpha^{3}\sqrt{\pi}}\right ) \left [ j_{2}(t)-i4\alpha^{2}(t-\beta)j_{1}(t)-4\alpha^{4}(t-\beta)^{2}j_{0}(t)\right ] ~,\end{gathered}\ ] ] wherein : @xmath138 d\varpi=2\int_{0}^{\infty } \exp\left [ -\frac{\varpi^{2}}{4\alpha^{2}}\right ] d\varpi ~,\ ] ] @xmath139 d\varpi=0 ~,\ ] ] @xmath140 d\varpi=2\int_{0}^{\infty}\varpi^{2 } \exp\left [ -\frac{\varpi^{2}}{4\alpha^{2}}\right ] d\varpi ~.\ ] ] employing the following identities ( hodgman 1957 ) : @xmath141 @xmath142 we finally obtain @xmath143 so that @xmath144 \exp[-\alpha^{2}(t-\beta)^{2}]~,\ ] ] or @xmath145 \exp[-\alpha^{2}(t-\beta)^{2 } ] ~.\ ] ] the maxima of @xmath146 are obtained from @xmath147 , i.e. , @xmath148\ } \exp[-\alpha^{2}(t-\beta)^{2}]=0 ~.\ ] ] the solutions of which are @xmath149 it follows that @xmath150 thus , the minimum of @xmath146 is attained at @xmath151 and is equal to @xmath152 . this means that the minimum of @xmath146 is independent of both @xmath129 and @xmath130 . furthermore , @xmath130 is the instant at which the pulse attains its minimum , and this minimum is also the maximum of @xmath153 . @xmath154 when @xmath155 , i.e. , when @xmath156 , so that @xmath157 is an indicator of the width of the main lobe of the pulse , i.e. , the larger is @xmath129 , the smaller is the width of the main lobe of the pulse . the maxima of @xmath146 are attained at @xmath158 and @xmath159 , and their value ( 0.4463 ) is independent of both @xmath129 and @xmath130 . the moduli of the spectra of the ricker pulses for various values of @xmath129 are depicted in fig . [ rickerom ] . [ ptb ] versus @xmath5 for various values of @xmath129 . @xmath102 , @xmath104 . ] note that these spectra do not depend on @xmath130 . the latter only affects the phase of @xmath99 . the time history of the ricker pulses for various values of @xmath129 are depicted in fig . [ rickert ] for the fixed value @xmath104 . this confirms the fact that the larger is @xmath129 , the narrower is the ricker pulse . [ ptb ] versus @xmath2 for various values of @xmath129 . @xmath104 and @xmath102 . ] let @xmath160 be the gaussian function @xmath161~.\ ] ] then @xmath162~,~\frac{d^{2}g(t)}{dt^{2}}=\mathcal{a}[-1 + 2\alpha^{2 } ( t-\beta)^{2}]\exp[-\alpha ^{2}(t-\beta)^{2}]=a^{i-}(t)~,\ ] ] which shows that the ricker pulse is identical to the second time derivative of the gaussian pulse . the wave we are concerned with is actually what was heretofore termed the `` incident plane wave '' . the plane body wave nature of the disturbance is embodied in the frequency domain function @xmath163=a^{i-}(\omega ) \exp\left [ i\omega\left ( \frac{x_{1}}{c_{t}^{0}}s^{i}-\frac{x_{2}}{c_{t}^{0}}\kappa^{0i}\right ) \right ] ~,\ ] ] wherein @xmath164 the ricker pulse nature of the plane wave is embodied in the spectrum function @xmath165 so that the time history of this wave is given by the fourier transform @xmath166d\omega ~,\ ] ] wherein @xmath167 we assume that @xmath16 is non - dispersive , i.e. , @xmath168 and @xmath169 do not depend on @xmath5 , so that @xmath170 is independent of @xmath5 . then @xmath171 is of the same form as @xmath172 ( see ( [ tds135 ] ) ) if we replace therein @xmath2 by @xmath170 , i.e. , @xmath173 so that employing the previous result in ( [ tds148 ] ) : @xmath174 \exp[-\alpha^{2}\{\tau(\mathbf{x},t , s^{i})-\beta\}^{2 } ] ~.\ ] ] _ remark _ the same remarks apply to this time history as to @xmath172 except that @xmath170 replaces @xmath2 . thus , the ricker pulse plane body wave attains its maximum ( in absolute value ) at @xmath175 , and the main lobe of the pulse is all the narrower the larger is @xmath129 . recall that @xmath176 . we encountered previously , in connection with the frequency domain response in @xmath11 and @xmath12 , plane wave functions of the type @xmath177= \\ \mathcal{r}^{j}(s^{i},\omega)a^{i-}(\omega)\exp\left [ i\omega\left ( s^{i}\frac{x_{1}}{c_{t}^{0}}\pm \kappa^{ji}\frac{x_{2}}{c_{t}^{0}}\right ) \right ] ~~;~~j=0,1~,\end{gathered}\ ] ] wherein ( recall that we assumed that @xmath169 does not depend on @xmath5 ) : @xmath178^{2}-(s^{i})^{2}}~,\ ] ] and , recalling that @xmath168 was also assumed not to depend on @xmath5 : @xmath179 e^{\left [ -2i\kappa^{0i}(s^{i})\frac{h}{c_{t}^{0}}\right ] } = \\ r^{0}(\omega,(s^{i})e^{\left [ -2i\kappa^{0i}(s^{i})\frac{h}{c_{t}^{0}}\right ] } ~,\end{gathered}\ ] ] @xmath180 e^{\left [ -i\kappa^{0i}(s^{i})\frac{h}{c_{t}^{0}}\right ] } = \\ r^{1}(\omega,(s^{i})e^{\left [ -i\kappa^{0i}(s^{i})\frac{h}{c_{t}^{0}}\right ] } ~,\end{gathered}\ ] ] wherein @xmath181\ ] ] @xmath182~.\ ] ] thus , we are faced with the problems of evaluating the integrals @xmath183 d\omega~,\ ] ] @xmath184 d\omega~,\ ] ] or @xmath185 d\omega~,\ ] ] @xmath186 d\omega~,\ ] ] wherein : @xmath187 @xmath188 in the following , whenever numerical results are given , they will apply to the field on the ground at point @xmath93 . recall that the time domain displacement response in the layer is of the form @xmath189 so that @xmath190 it is this function , related to the temporal displacement response on the ground , which will be depicted in the graphs that are presented hereafter . the time history of the displacement field in @xmath12 is of the form @xmath192 d\omega= \\ i\int_{-\infty}^{\infty}\frac{a^{i-}(\omega)}{d(\omega , s^{i } ) } \exp\left [ -i\omega\tau^{1\pm}(\mathbf{x},t,\omega , s^{i})\right ] d\omega~,\end{gathered}\ ] ] wherein @xmath193 @xmath194 and the ricker pulse excitation is represented by @xmath195 consequently @xmath196 @xmath197 whence @xmath198\omega\right\ } } { ia\cos(\gamma\omega)-b\sin(\gamma\omega ) } + \\ \frac{\exp\left\ { i[\beta-\tau^{1\pm}(\mathbf{x},t,\omega , s^{i})]\omega\right\ } } { ia\cos(\gamma\omega)+b\sin(\gamma\omega ) } \big ] \exp\left [ -\frac{\omega^{2}}{4\alpha^{2}}\right ] d\omega~.\end{gathered}\ ] ] this formula shows that the integrand is an exponentially - decreasing function of @xmath5 so that the numerical evaluation of the integral should pose no problems . in particular , we replace the upper limit of the integral by @xmath199 , where the latter is such that @xmath200 < < 1 $ ] , so that we are faced with the computation of @xmath201\omega\right\ } } { ia\cos(\gamma(\omega , s^{i})\omega)-b(\omega , s^{i})\sin(\gamma(\omega , s^{i})\omega ) } + \\ \frac{\exp\left\ { i[\beta-\tau^{1\pm}(\mathbf{x},t,\omega , s^{i})]\omega\right\ } } { ia\cos(\gamma(\omega , s^{i})\omega)+b(\omega , s^{i})\sin(\gamma(\omega , s^{i})\omega ) } \big ] \exp\left [ -\frac{\omega^{2}}{4\alpha^{2}}\right ] d\omega~.\end{gathered}\ ] ] to this end , we therefore employ the simplest method : rectangle quadrature . we divide the interval @xmath202 $ ] into @xmath203 equal sub - intervals , of width @xmath204 , and centered at points @xmath205 , so that @xmath206\omega\right\ } } { ia\cos(\gamma(\omega_{l},s^{i})\omega_{l})-b(\omega_{l},s^{i})\sin(\gamma(\omega_{l},s^{i})\omega_{l } ) } + \\ \frac{\exp\left\ { i[\beta-\tau^{1\pm}(\mathbf{x},t,\omega_{l},s^{i})]\omega\right\ } } { ia\cos(\gamma(\omega_{l},s^{i})\omega)+b(\omega_{l},s^{i})\sin(\gamma(\omega_{l},s^{i})\omega ) } \big ] \exp\left [ -\frac{\omega_{l}^{2}}{4\alpha^{2}}\right ] d\omega~.\end{gathered}\ ] ] this result seems to indicate that @xmath191 is complex . however , combining into one the two terms in @xmath207 $ ] , we find @xmath208 \exp\left [ -\frac{\omega_{l}^{2}}{4\alpha^{2}}\right ] d\omega ~,\end{gathered}\ ] ] which shows that @xmath191 is indeed real ( at least for real @xmath209 ) . ( [ rq11 ] ) is the formula we employ for the numerical evaluation of @xmath191 . note that this formula is exact in the limits @xmath210 , @xmath211 . once again , the time history of the displacement field in @xmath12 is of the form @xmath212 d\omega= \\ i\int_{-\infty}^{\infty}\frac{a^{i-}(\omega)}{d(\omega , s^{i } ) } \exp\left [ -i\omega\tau^{1\pm}(\mathbf{x},t,\omega , s^{i})\right ] d\omega~,\end{gathered}\ ] ] wherein @xmath213 @xmath214 and the ricker pulse excitation is represented by @xmath215 from here on , we adopt a strategy that is different from the one in the previous section , notably by writing @xmath216 as @xmath217 ~.\ ] ] the idea is to express @xmath218 as something like @xmath219 and then to employ the power series expansion of @xmath220 to express @xmath221 , but we have to be careful to have @xmath222 in order for this series to converge . thus , we write @xmath223 ~~ ; \\ ~~\left\| \left ( \frac{b(\omega , s^{i})+a}{b(\omega , s^{i})-a}\right ) e^{-2i\gamma(\omega , s^{i})\omega}\right\| < 1~,\end{gathered}\ ] ] @xmath224 ~~ ; \\ ~~\left\| \left ( \frac{b(\omega , s^{i})-a}{b(\omega , s^{i})+a}\right ) e^{2i\gamma(\omega , s^{i})\omega}\right\| < 1~,\end{gathered}\ ] ] so that : @xmath225^{-1}=\left ( -\frac{i}{2}\{b(\omega , s^{i})-a\ } e^{i\gamma(\omega , s^{i})\omega}\right)^{-1}\sum_{m=0}^{\infty } \left ( \frac{b(\omega , s^{i})+a}{b(\omega , s^{i})-a}\right)^{m } e^{-2im\gamma(\omega , s^{i})\omega } ~,\ ] ] @xmath226^{-1}=\left ( \frac{i}{2}\{b(\omega , s^{i})+a\ } e^{-i\gamma(\omega , s^{i})\omega}\right)^{-1}\sum_{m=0}^{\infty } \left ( \frac{b(\omega , s^{i})-a}{b(\omega , s^{i})+a}\right)^{m } e^{2im\gamma(\omega , s^{i})\omega } ~,\ ] ] or @xmath227^{-1}=\sum_{m=0}^{\infty}c_{m}^{-}(\omega , s^{i } ) e^{-i(2m+1)\gamma(\omega , s^{i})\omega } ~,\ ] ] @xmath228^{-1}=\sum_{m=0}^{\infty}c_{m}^{+}(\omega , s^{i } ) e^{i(2m+1)\gamma(\omega , s^{i})\omega } ~,\ ] ] wherein @xmath229 @xmath230 at this point we recall that @xmath231 @xmath232 @xmath233 and _ assume henceforth that @xmath234 and @xmath235 do not depend on @xmath5 _ so that @xmath236 and @xmath237 @xmath238 whence @xmath239 @xmath240 furthermore , we assume that @xmath241 so that @xmath242 is real for all @xmath243 . consequently @xmath244 is real for all @xmath243 , and the condition @xmath245\right\| \gtrless1 $ ] reduces to @xmath246 . it follows that the time history of the field in @xmath12 takes the form @xmath247 d\omega= \\ i\sum_{m=0}^{\infty } c_{m}^{-}(s^{i})\int_{-\infty}^{\infty}a^{i-}(\omega ) \exp\left [ -i\omega\{\tau^{1\pm}(\mathbf{x},t , s^{i})+(2m+1)\gamma(s^{i})\}\right ] d\omega~~ ; \\ ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a } \right ) \right\| < 1~,\end{gathered}\ ] ] @xmath248 d\omega= \\ i\sum_{m=0}^{\infty } c_{m}^{+}(s^{i})\int_{-\infty}^{\infty}a^{i-}(\omega ) \exp\left [ -i\omega\{\tau^{1\pm}(\mathbf{x},t , s^{i})-(2m+1)\gamma(s^{i})\}\right ] d\omega~~ ; \\ ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right ) \right\| > 1~,\end{gathered}\ ] ] or , on account of the ricker pulse nature of the excitation , @xmath249 d\omega~~ ; \\ ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right ) \right\| < 1~,\end{gathered}\ ] ] @xmath250 d\omega~~ ; \\ ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right ) \right\| > 1~.\end{gathered}\ ] ] we recall here the previous result @xmath251 d\omega= \\ \mathcal{a}\left [ -1 + 2\alpha^{2}(t-\beta)^{2}\right ] \exp[-\alpha^{2}(t-\beta)^{2 } ] ~,\end{gathered}\ ] ] so that @xmath252 \times \\ \exp\left [ -\alpha^{2}\left ( \tau^{1\pm}(\mathbf{x},t,\omega , s^{i})-\beta_{m}^{-}(s^{i})\right ) ^{2}\right ] ~~ ; ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right ) \right\| < 1~,\end{gathered}\ ] ] @xmath253 \times \\ \exp\left [ -\alpha^{2}\left ( \tau^{1\pm}(\mathbf{x},t,\omega , s^{i})-\beta_{m}^{+}(s^{i})\right ) ^{2}\right ] ~~ ; ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right ) \right\| > 1~,\end{gathered}\ ] ] wherein @xmath254 although these formulae are exact , they are not suitable for computation due to the presence of the infinite series therein . actually , for practical ( numerical ) purposes , we limit the series to a finite ( @xmath255 ) number of terms ( which is justified by the fact that the terms of the series are exponentially - decreasing with @xmath256 ) so that @xmath257 \times \\ \exp\left [ -\alpha^{2}\left ( \tau^{1\pm}(\mathbf{x},t,\omega , s^{i})-\beta_{m}^{-}(s^{i})\right ) ^{2}\right ] ~~ ; ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right)\right\| < 1~,\end{gathered}\ ] ] @xmath258 \times \\ \exp\left [ -\alpha^{2}\left ( \tau^{1\pm}(\mathbf{x},t,\omega , s^{i})-\beta_{m}^{+}(s^{i})\right ) ^{2}\right ] ~~ ; ~~\left\| \left ( \frac{b(s^{i})+a}{b(s^{i})-a}\right ) \right\| > 1~,\end{gathered}\ ] ] these last two formulae ( which are exact in the limit @xmath259 ) form the basis of what we term the _ power series quadrature method _ for the computation of the time history of the displacement field in @xmath12 . once again , the time history of the displacement field in @xmath12 is of the form @xmath260 d\omega= \\ i\int_{-\infty}^{\infty}\frac{a^{i-}(\omega)}{d(\omega , s^{i } ) } \exp\left [ -i\omega\tau^{1\pm}(\mathbf{x},t,\omega , s^{i})\right ] d\omega~,\end{gathered}\ ] ] wherein @xmath261 @xmath262 and the ricker pulse excitation is represented by @xmath263 before going into details , we recall some general considerations . eqs . ( [ cfprc1])-([cfprc4 ] ) show that the task is to evaluate the fourier integral of a product of two functions @xmath264 and @xmath265 : @xmath266 we make use of the fourier integral representations of @xmath267 and @xmath268 : @xmath269 whose inverses are : @xmath270 we introduce ( [ cfprc6 ] into ( [ cfprc5 ] ) so as to obtain : @xmath271 ~,\ ] ] or , due to the fact that @xmath272 d\omega= 2\pi\delta(\tau - t_{1}-t_{2})~,\ ] ] we find @xmath273 which is a _ convolution _ integral . we can go a step further by expressing everything for positive times : @xmath274 dt_{1 } ~.\ ] ] we now apply this formula to evaluate the time history of response . as usual , we assume that _ the media are dispersionless _ , so that @xmath275 , whence @xmath276 dt_{1 } ~.\ ] ] we choose @xmath277 we recall the result of ( [ tds148 ] ) @xmath278 \exp[-\alpha^{2}(t_{2}-\beta)^{2 } ] ~.\ ] ] and the fourier inverse of @xmath267 is @xmath279 assuming as before that @xmath114 is real , we note immediately that although the denominator of the integral in ( [ cfprc14 ] ) does not vanish for it real @xmath5 , it can vanish for _ complex _ @xmath5 . this suggests that the integral can be evaluated by use of the cauchy theorem by appealing to a suitable integration path in the complex @xmath5 plane . actually @xmath218 vanishes at an infinite number of locations in the complex @xmath280 plane , so that we prefer to proceed as following . in order to stress the fact that the integration variable in ( [ cfprc14 ] ) is real ( i.e. , @xmath281 ) , and for other reasons , we re - write the integral as @xmath282 wherein @xmath283 we assume that @xmath218 vanishes for a denumerable set of _ complex _ frequencies @xmath284 , i.e. , @xmath285 this suggests expanding @xmath216 in a taylor series around @xmath286 : @xmath287 wherein @xmath288 on account of ( [ cfpr14 ] ) we have @xmath289 whence @xmath290 now let us turn to the issue of the actual locations of the zeros of @xmath218 . we search for the complex roots @xmath5 of @xmath291 and assume , as was implicit ( or explicitly stated ) , that @xmath113 , @xmath114 and @xmath292 are real . then @xmath293 + i\cos(\gamma\omega^{'})\left [ a\cosh(\gamma\omega^{"})+b\sinh(\gamma\omega^{"})\right ] = 0 ~,\ ] ] or , owing to the fact that we have a mixture of real and complex quantitites : @xmath294=0 ~,\ ] ] @xmath295 = 0 ~.\ ] ] thus , we have two families of solutions , the first of which correspond to : @xmath296 and the second of which correspond to : @xmath297 the ( so - called _ even _ ) solutions of the first family are : @xmath298 @xmath299 and the ( so - called _ odd _ ) solutions of the second family are : @xmath300 @xmath301 _ remark _ the real parts of @xmath302 are independent of both @xmath113 and @xmath114 . the imaginary parts of the even frequencies are independent of @xmath256 . the imaginary parts of the odd frequencies are independent of @xmath256 . since the @xmath303 functions in these formulae are supposed to be real , the even solutions apply only when @xmath304 and the odd solutions apply only when @xmath305 . owing to the facts that : i ) @xmath306 , ii ) @xmath307 we have ( for @xmath304 ) @xmath308 _ remark _ owing to the facts that : i ) @xmath306 , ii ) @xmath309 we have ( for @xmath310 ) @xmath311 let us now evaluate @xmath312 . we have @xmath313 = \gamma\left [ -ia \sin\left ( \gamma(\omega_{m}^{'}+i\omega_{m}^{"})\right ) + b\cos\left ( \gamma(\omega_{m}^{'}+i\omega_{m}^{"})\right ) \right ] = \\ \gamma\cos(\gamma\omega_{m}^{'})\left [ a\sinh(\gamma\omega_{m}^{"})+b\cosh(\gamma\omega_{m}^{"})\right ] + i\gamma\sin(\gamma\omega_{m}^{'})\left [ -a\cosh(\gamma\omega_{m}^{"})-b\sinh(\gamma\omega_{m}^{"})\right ] ~.\end{gathered}\ ] ] then : @xmath314 + i\gamma\sin(\gamma\omega_{m}^{e'})\left [ -a\cosh(\gamma\omega_{m}^{e"})-b\sinh(\gamma\omega_{m}^{e"})\right ] = \\ \gamma\cos(\gamma\omega_{m}^{e'})\left [ a\sinh(\gamma\omega_{m}^{e"})+b\cosh(\gamma\omega_{m}^{e"})\right ] = -\gamma(-1)^{m}\left ( \frac{b^{2}-a^{2}}{a}\right ) \sinh(\gamma\omega_{m}^{e " } ) ~,\end{gathered}\ ] ] @xmath315 + i\gamma\sin(\gamma\omega_{m}^{o'})\left [ -a\cosh(\gamma\omega_{m}^{o"})-b\sinh(\gamma\omega_{m}^{o"})\right ] = \\ \gamma\sin(\gamma\omega_{m}^{o'})\left [ -a\cosh(\gamma\omega_{m}^{o"})-b\sinh(\gamma\omega_{m}^{o"})\right ] = i\gamma(-1)^{m}\left ( \frac{b^{2}-a^{2}}{a}\right ) \cosh(\gamma\omega_{m}^{o " } ) ~.\end{gathered}\ ] ] the question that arises is whether we have to deal with either even or odd solutions . recall that @xmath316 and , under the ( previous ) assumption of a dispersionless material in the layer , @xmath317 so that @xmath318 \gtrless ( \mu^{0 } ) ^{2}\left [ 1-(s^{i } ) ^{2}\right ] ~,\ ] ] or , recalling that @xmath319 , @xmath320 is equivalent : @xmath321 ~.\ ] ] recall that @xmath322 , so that for @xmath323 ( i.e. , normal incidence ) , @xmath320 ( or @xmath324 ) is equivalent to @xmath325 whereas for @xmath326 ( i.e. , grazing incidence ) , @xmath320 ( or @xmath324 ) is equivalent to @xmath327 in the ( geophysical ) case of interest herein , i.e. , dealing with a soft layer overlying a hard substratum , we have @xmath328 and @xmath329 , so that we are clearly in the situation @xmath330 for all incidence angles . _ this situation is that of odd solutions_. consequently , the time history of displacement in the layer is @xmath331 thus , we are faced with the problem of the evaluation of the integral @xmath332 wherein the important property to note is that @xmath333 which means that the pole of the integrand lies in the lower half part of the complex @xmath5 plane for all @xmath334 . thus , in order to apply cauchy s theorem , we consider the two contour integrals @xmath335 wherein the contours @xmath336 are depicted in figs . [ cplus ] and [ cminus ] . [ ptb ] . ] [ ptb ] . ] then @xmath337 @xmath338 however @xmath339 so that , by virtue of cauchy s theorem @xmath340 @xmath341 however : @xmath342d\theta\big\| ~~ ; \\ ~~t<0~~,~~r>>1 ~,\end{gathered}\ ] ] so that ( due to the fact that @xmath343 for @xmath344 $ ] ) @xmath345 whence ( by virtue of ( [ cfpr46 ] ) @xmath346 similarly @xmath347d\theta\big\| ~~ ; \\ ~~t>0~~,~~r>>1 ~,\end{gathered}\ ] ] so that ( due to the fact that @xmath348 for @xmath349 $ ] ) @xmath350 whence ( by virtue of ( [ cfpr47 ] ) @xmath351 thus , @xmath352 wherein @xmath353 is the heaviside function ( @xmath354 and @xmath355 ) . it follows from ( [ cfprc14 ] ) that @xmath356 due to the fact that @xmath357 , @xmath358 whence @xmath359 or , more explicitly @xmath360\times \\ \exp[-i\omega_{m}^{o}t_{1}-\alpha^{2}(\tau^{1\pm}(\mathbf{x},t , s^{i})-t_{1}-\beta)^{2 } ] dt_{1 } ~.\end{gathered}\ ] ] at this point , we recall that ( with obvious shorthand notation ) : @xmath361 @xmath362 we notice that @xmath363 and @xmath364 are independent of @xmath256 . thus , we can write ( [ cfprc20 ] ) as @xmath365\times \\ \exp[-i\omega_{m}^{o}t_{1}-\alpha^{2}(\tau^{1\pm}(\mathbf{x},t , s^{i})-t_{1}-\beta)^{2 } ] dt_{1 } ~,\end{gathered}\ ] ] wherein @xmath366 ~.\ ] ] thus : @xmath367\times \\ \exp[\omega^{o"}t_{1}-\alpha^{2}(\tau^{1\pm}(\mathbf{x},t , s^{i})-t_{1}-\beta)^{2 } ] dt_{1 } ~,\end{gathered}\ ] ] wherein @xmath368 = \exp\left ( -i\frac{\pi}{2\gamma}t_{1}\right ) \sum_{m=-\infty}^{\infty}\exp\left [ i\frac{m\pi}{\gamma}(\gamma - t_{1})\right ] ~.\ ] ] we make use of the _ poisson sum formula _ ( morse and feshbach , 1953 ) @xmath369 wherein we take @xmath370 and @xmath371 , to obtain @xmath372 = 2\gamma \sum_{m=-\infty}^{\infty}\delta\left ( \gamma - t_{1}+2m\gamma\right ) = 2\gamma \sum_{m=-\infty}^{\infty}\delta\left ( t_{1}-(2m+1)\gamma\right ) ~.\ ] ] but ( recall that @xmath306 ) @xmath373 for @xmath374 , so that @xmath375 = 2\gamma \sum_{m=0}^{\infty}\delta\left ( \gamma - t_{1}+2m\gamma\right ) ~~;~~t_{1}\geq 0 ~,\ ] ] whence @xmath376 or , on account of the properties of the dirac delta distribution , @xmath377 consequently : @xmath378\times \\ \exp[\omega^{o"}t_{1}-\alpha^{2}(\tau^{1\pm}(\mathbf{x},t , s^{i})-t_{1}-\beta)^{2 } ] dt_{1 } ~,\end{gathered}\ ] ] or finally , on account of the sifting property of the dirac delta distribution , @xmath379\times \\ \exp[\omega^{o"}(2m+1)\gamma-\alpha^{2}\{\tau^{1\pm}(\mathbf{x},t , s^{i})-(2m+1)\gamma-\beta\}^{2 } ] ~.\end{gathered}\ ] ] the terms of the series decrease exponentially with @xmath256 so that the series can be approximated by a sum of @xmath255 terms @xmath380\times \\ \exp[\omega^{o"}(2m+1)\gamma-\alpha^{2}\{\tau^{1\pm}(\mathbf{x},t , s^{i})-(2m+1)\gamma-\beta\}^{2 } ] ~,\end{gathered}\ ] ] which is the form adopted in the numerical applications of this method . as shown in the following section , the pole - residue convolution method gives rise to the correct solution in all cases . [ ptb ] computed by the rectangular quadrature method for an `` anomalous '' pulse . @xmath102 , @xmath103 , @xmath112 , @xmath109 , @xmath106 , @xmath107 , @xmath381 , @xmath382 . ] [ ptb ] computed by the power series method for an `` anomalous '' pulse . @xmath102 , @xmath103 , @xmath112 , @xmath109 , @xmath106 , @xmath107 , @xmath383 . ] [ ptb ] computed by the pole - residue convolution method for an `` anomalous '' pulse . @xmath102 , @xmath103 , @xmath112 , @xmath109 , @xmath106 , @xmath107 , @xmath384 . ] [ ptb ] computed by the rectangular quadrature method for separated pulses . @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath110 , @xmath381 , @xmath382 . ] [ ptb ] computed by the power series method for separated pulses . @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath110 , @xmath385 . ] [ ptb ] computed by the pole - residue convolution method for separated pulses . @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath110 , @xmath386 . ] [ ptb ] computed by the rectangular quadrature method for merged pulses . @xmath102 , @xmath108 , @xmath104 , @xmath109 , @xmath106 , @xmath111 , @xmath381 , @xmath382 . ] [ ptb ] computed by the power series method for merged pulses . @xmath102 , @xmath108 , @xmath104 , @xmath109 , @xmath106 , @xmath111 , @xmath387 . ] [ ptb ] computed by the pole - residue convolution method for merged pulses . @xmath102 , @xmath108 , @xmath104 , @xmath109 , @xmath106 , @xmath111 , @xmath388 . ] [ ptb ] . @xmath102 , @xmath108 , @xmath104 , @xmath389 , @xmath106 , @xmath111 , corresponding to the case of a quasi - monochromatic pulse . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] [ ptb ] computed by the rectangular quadrature method for a quasi monochromatic pulse . @xmath102 , @xmath108 , @xmath104 , @xmath389 , @xmath106 , @xmath111 , @xmath381 , @xmath382 . ] [ ptb ] computed by the power series method for a quasi monochromatic pulse . @xmath102 , @xmath108 , @xmath104 , @xmath389 , @xmath106 , @xmath111 , @xmath387 . ] [ ptb ] computed by the pole - residue convolution method for a quasi monochromatic pulse . @xmath102 , @xmath108 , @xmath104 , @xmath389 , @xmath106 , @xmath111 , @xmath388 . ] [ ptb ] . @xmath102 , @xmath108 , @xmath104 , @xmath390 , @xmath106 , @xmath391 , corresponding to the case of a monochromatic pulse . the left - hand curves pertain to moduli , and the right hand curves to phases of the spectra . ] [ ptb ] computed by the rectangular quadrature method for a monochromatic pulse . @xmath102 , @xmath108 , @xmath104 , @xmath390 , @xmath106 , @xmath391 , @xmath381 , @xmath382 . ] [ ptb ] computed by the power series method for a monochromatic pulse . @xmath102 , @xmath108 , @xmath104 , @xmath390 , @xmath106 , @xmath391 , @xmath392 . ] [ ptb ] computed by the pole - residue convolution method for a monochromatic pulse . @xmath102 , @xmath108 , @xmath104 , @xmath390 , @xmath106 , @xmath391 , @xmath393 . ] the three methods are : i ) the rectangle quadrature method , ii ) the power series method , and iii ) the complex frequency pole - residue convolution method . in the set of figures [ rectrickb16 ] , [ resrickb17 ] , [ poisrick18 ] ( for a so - called `` anomalous '' pulse ) , [ rectrickb17 ] , [ resrickb18 ] , [ poisrick19 ] ( for separated pulses ) , and [ rectrickb18 ] , [ resrickb19 ] , [ poisrick20 ] ( for merged pulses ) we exhibit the time history of displacement response at the location @xmath93 on the ground plane . we next choose what seems a typical geophysically - interesting situation ( at least in the frequency domain ) depicted in fig . [ aodrickb10 ] . the corresponding time - domain responses are given in figures [ rectrickb19 ] , [ resrickb20 ] , [ poisrick21 ] . as a last example , we choose a more idealized ( and less realizable due to the very large contrast of physical properties it implies between the layer and the substratum ) geophysical example , the frequency response of which is depicted in fig . [ aodrickb11 ] . the corresponding time - domain responses are given in figures [ rectrickb20 ] , [ resrickb21 ] , [ poisrick22 ] . we notice that all the three methods again give the same results . note that the relatively - long duration and monochromatic nature of the temporal response in figs . [ rectrickb20 ] , [ resrickb21 ] , [ poisrick22 ] are due to the large q single - spike nature of the frequency response , the latter being a result of the large contrast of physical properties and the fact that only one @xmath394 spike is located within the significant part of the spectrum of the ricker pulse . _ remark _ the rectangle quadrature method , embodied in ( [ rq11 ] ) , produces a purely - numerical result which gives no insight as to the physical nature of this response . it was proposed only as a reference solution by which the other two methods could be judged , at least on a numerical basis . this rectangle quadrature method is certainly not optimal , even from the purely - numerical point of view , but obviously one of the simplest to explain and program . by inspection of ( [ psq22 ] ) and comparison with ( [ tds158 ] ) , we see that the power series method gives rise to an expression of the time history response to a ricker pulse that is a sum of displaced ( and increasingly - attenuated ) ricker pulses . this is what one would expect on an intuitive basis for a dispersionless configuration . thus , it would seem that the power series method is the most appropriate one , at least in the situation in which the successive pulses are well - separated . however , in the case in which the successive pulses are not well - separated , intuition is lost ( especially when a long - duration quasi - monochromatic response is produced ) and the power series picture reflects this fact , although it still gives rise to the correct numerical response . however , the power series method can not be applied when @xmath395 as is the case in which love modes are excited . this is the reason why the pole - residue convolution method was proposed . the pole - residue convolution solution in ( [ cfprc34 ] ) expresses the time history of response as a weighted sum of displaced ricker pulses ( the latter would probably be distorted ricker pulses in the presence of dispersion ) . this is close to being intuitive , but what is less intuitive is the fact that the displacements are a function of the real part of the complex zeros of the equation @xmath396 and the weight functions are expressed in terms of the imaginary part of the complex zeros of the equation @xmath396 . the fact that the essential features ( peak values and duration , amongst others ) of the time history are directly - related to the complex eigenvalues of the structure is the essential result we were aiming at in this contribution . the most important parameter is the imaginary part of the eigenvalues since it regulates the height of the succesive ricker pulses and therefore determines the duration of the time domain response . this parameter is a measure of _ radiation damping _ ( which is leakage of energy into the substratum , an attenuation mechanism that exists even in the absence of material dissipation in the layer ) . the pole - residue convolution expression of the time history appears to be similar to the one obtained by the power series method , but the latter method is not applicable when @xmath396 for real eigenvalues in the absence of dissipation ( i.e. , the situation in which it is possible to excite love modes ( groby and wirgin 2005a , b ) ) ; moreover , the power series method does not enable one to predict the duration in an obvious way . some of the numerical results included in this work are rather unexpected . for instance , the time histories given in figs . [ rectrickb18 ] , [ resrickb19 ] , [ poisrick20 ] have quite long durations that one would not expect to occur for a case in which modes can not be excited . actually , this long duration is due to the fact that the only attenuative action in this work is the one due to radiation damping . the duration would be shorter if material dissipation ( i.e. , viscoelasticity ) were taken into account in the layer and/or the contrast between @xmath113 and @xmath114 were smaller . the main result of this contribution is that the three methods give rise to the same solutions for a large variety of scattering configurations . the complex frequency pole - residue convolution method turns out to be the most interesting method since : i ) it is numerically - efficient , ii ) it is explicit as concerns the understanding and quantification of the duration of the time domain response , iii ) it can be employed even in the case in which genuine resonances ( due to mode excitation ) are produced . the part of this study concerning the complex frequency pole residue convolution method constitutes a correction of its counterpart in our previous publications ( groby and wirgin , 2005a ) and ( groby and wirgin 2005b ) . a somewhat similar approach , although applied to a fluid layer in a fluid host , is that of ( conoir , 1987 ) . work remains to be done to take into account dispersion and damping of the material in the layer . the natural follow - up of this study is to elucidate theoretically the nature of the time histories of response not only for the case ( the one treated herein ) in which the configuration is unable to excite ( e.g. , love ) modes , but also in the case in which such modes can be excited ( groby and wirgin 2005a , b ) . * conoir j .- m . , rflexion et transmission par une plaque fluide , in _ la diffusion acoustique _ , gespa n. ( ed . ) , cedocar paris armes , paris , 1987 , 105 - 132 . * groby j .- p . and wirgin a. , 2d ground motion at a soft viscoelastic layer / hard substratum site in response to sh cylindrical seismic waves radiated by near and distant line sources . i. theory , _ geophys.j.int._ , 2005 , 163 , 165 - 191 . * groby j .- wirgin a. , 2d ground motion at a soft viscoelastic layer / hard substratum site in response to sh cylindrical seismic waves radiated by near and distant line sources . ii . computations , _ geophys.j.int._ , 2005 , 163 , 192 - 224 . * hodgman c.d.(ed . ) , _ crc standard mathematical tables _ , chemical rubber publ . co. , cleveland , 1957 , 304 . * morse p.m. and feshbach h. , _ methods of theoretical physics _ , mc graw - hill , new york , 1953 . * sanchez - sesma f.j , diffraction of elastic sh waves by wedges , _ bull.seism.soc.am._ , 1985 , 75 , 1435 - 1446 .

time domain responses of porous media have been studied by some authors , but generally the possible descriptions are given in the frequency domain . the aim of this paper , limited to materials with rigid skeleton considered as equivalent fluids , is to compare in time domain different descriptions by johnson - allard ( ) as well as by pride - lafarge ( ) , with : ) some analytical approximate formulas based upon asymptotic high - frequency expansion ; ) the exact formula by zwikker and kosten for the case of cylindrical pores . the paper starts with a construction analysis of the models ( and ) . then , the green s function in the time domain is defined , written in scaled form , and shown to exhibit interesting properties of materials . in particular , a so - far overlooked decay length describing a high - frequency attenuation - without - distortion effect is identified in terms of brown s tortuosity , johnson s and allard s known characteristic viscous and thermal lengths , and two , unknown in general , characteristic viscous and thermal surfaces . the numerical computation of the green s function is done by fft , with some precautions , because of the importance of the higher frequencies on the response shape : the substraction of the diffusive ( low frequencies ) approximation largely improves the results of the fft . the description is shown to be the best full - frequency general model with remaining small discrepancies due to unsatisfactory account of the mentioned surface - parameters . keywords : pulse propagation ; transient signals ; porous material pacs : 43.55 rv , 43.55 ti , 43.20 gp , 43.20 bi , 43.20 hq

1205.3775

linear wave propagation in homogeneous porous media saturated with a viscothermal fluid such as ambient air , has been the subject of extensive research . traditionally , for the case when the wavelength is large compared to the pore size , it has been described on the basis of the so - called two - scale asymptotic homogenization theory @xcite@xcite . for materials with skeleton sufficiently heavy and/or rigid to be motionless , this leads to an effective medium theory in which the medium permittivities are two linear operators , one representing an effective density and the other an effective compressibility @xcite , @xcite . these operators , while nonlocal in time as a result of delayed responses due to the losses ( temporal dispersion ) , are given as local in space by the homogenization process at the dominant order . in other words , no spatial dispersion arises , so that the response of the material at a given macroscopic point ( physically , an elementary coarse - graining volume ) is determined by the history of the pertinent field variables at the given point but not the neighboring points . as such , this description is a special case not directly applicable to all geometries . it does not describe the situations where structures in the form of helmholtz resonators are presentin this case , if the microgeometry splits in parts with pore sizes sufficiently different to imply different rescalings of the microscopic governing equations in the different parts , a solution can still be written using the principle of asymptotic homogenization @xcite . independently , a general nonlocal theory of propagation along a symmetry axis in macroscopically homogeneous materials having arbitrary microgeometry , has been recently proposed by one of the authors @xcite . finally , different high - frequencies extensions of the idea of two - scale homogenization have also been introduced recently @xcite@xcite . all of these extensions lie outside the scope of the present paper , as defined next . ] . nevertheless , most of the materials used in noise control for sound absorption , do not present in their inner structure very different pores sizes . in this paper we explicitly assume the absence of very different pores sizes , automatically ensuring at long wavelengths the validity of the usual two - scale asymptotic homogenization . it gives us explicit recipes to compute from microgeometry the effective density and compressibility , complex functions of frequency but not of wavenumber , in the above framework . in practice , the computation is not possible in full detail ; it is however not required to be done to arrive at a relatively precise description . in absence of a complete knowledge of the microgeometry , a widely used semi - phenomenological model which depends on a small set of independently measurable geometrical parameters of the structure , is given by the well - known formula of johnson @xmath5 @xmath6 joh for the density , and likewise , the champoux - allard @xcite or the lafarge @xmath5 @xmath6 @xcite formula for the compressibility . these approximate expressions ( denoted @xmath0 ) of the two constitutive functions , essentially are the result of : \1 ) an exact description of the high - frequency limits of the two functions , density and compressibility , in terms of the concepts of ideal - fluid tortuosity @xmath7 and characteristic lengths @xmath8 joh and @xmath9 @xcite , the viscous and thermal relaxation processes being ` frozen ' in this limitin the frozen limit the viscous and thermal relaxation processes have no time to develop ; in the relaxed limit they have enough time to fully develop . ] , \2 ) an exact description of the low - frequency limits , in terms of the concepts of d.c . viscous and thermal permeabilities , @xmath10 @xcite and @xmath11 @xcite , the viscous and thermal relaxation processes being ` relaxed ' in this limit , and finally , \3 ) an assumption that these frozen and relaxed limits 1 ) and 2 ) are connected in the simplest reasonable manner , @xmath12 by means of the simplest ` relaxation ' functions of frequency @xmath13 having their singularities and zeros lying on the imaginary half - axis in the complex @xmath13 plane ( see @xcite appendix a , and @xcite or @xcite appendix c ) . that the singularities are on a half imaginary axis is a mathematical expression of the fact that we actually restrict to the class of materials allowing to apply the direct two - scale homogenization process . the pure causality condition would only require the singularities to be in a half plane . our stronger assumption may be rephrased physically by saying that the fluid velocity pattern at the pore scale is divergence - free for the purpose of the determination of the density , and the pressure pattern is uniform for the purpose of the determination of the compressibility ; or in essence , that the operators are local operators in space , at long wavelengths , in the geometries we consider . the remaining assumption that the functions are the simplest reasonable ` relaxation ' ones , consists in making the additional but adjacent assumption that the considered geometries manifest a relatively narrow not bimodal , for example distribution of pore sizes . then the distribution of poles on the imaginary axis is simple and the whole pattern of response functions on the real axis is strongly determined by the low- and high - frequency behaviors . in this manner , a simple , resp . viscous and thermal , relaxation - transition description of the density and compressibility functions is obtained , that may be thought to be well - verified in a wide class of materials as long as the wavelengths are large compared to the typical dimensions of the coarse - graining averaging volumes . at this point , we may mention a similar relaxation - transition approach , developed by wilson @xcite . in wilson s approach , the emphasis is not made on the low - frequency and high - frequency limits , but directly on the transition . the quality of the description obtainable with wilson s models is comparable to that of model @xmath14 , but the parameters become purely adjustable parameters not precisely defined in terms of the microgeometry , and not clearly obtainable by non - acoustical means . subsequently , the low - frequency relaxed limit 2 ) was made more precise by pride @xmath5 @xmath6 @xcite and lafarge @xcite ( the next viscous and nontrivial thermal terms now being exactly described thanks to the introduction of the additional notions of d.c . viscous and thermal tortuosities @xmath15 and @xmath16 ) , resulting in a slightly improved ( denoted @xmath1 ) description of the relaxation transitions of the two functions . notice that some confusions were present in the original works @xcite and @xcite ( not readily available ) and other subsequent ones ( e.g. @xcite ) , so that the presentation we give later on , of the @xmath1 description , may be worthwhile . in ref . @xcite , fellah @xmath5 @xmath6 concentrated on the time - domain expression of the high - frequencies asymptotics implied by this modela short , non exhaustive list of works concerned with the time - domain description is : @xcite-@xcite . ] . expanding , in the high - frequency limit , the @xmath1 density and compressibility in powers of the inverse stokes number @xmath17 ( defined as the ratio between boundary layer thickness and characteristic pore size ) , and retaining the ( exact ) zero and first order terms and the ( model - dependent ) second order terms , they derived an asymptotic time - domain pressure wave equation . using fractional - derivative and laplace - transform calculus , they were able to solve this equation in elegant manner through the calculation of a corresponding asymptotic green s function of the unbounded medium . recall that a green s function or impulse response , as a function of time , extends and flattens when observed at fixed locations more and more remote from the spatial point where it originates . now , the results of the calculations made in @xcite seem to indicate that the terms of second order yield a significant effect on the green s function for resistive porous materials , and , in addition , that the effect is mainly an effect on the amplitude without noticeable distortion of the time wave pattern ( see figs . 2 and 3 of paper @xcite ) . in the present paper , our purpose is threefold . first , we wish taking advantage of recent clarifications regarding nonlocal ( spatial dispersion ) effects@xcite to propose a lucid review of the above local ( frequency dispersion ) theory , stressing the physical hypotheses behind it . by recalling the construction principle of the models @xmath0 and @xmath18 , we make clear that these models @xmath19 @xmath20 lead to inaccurate descriptions of the second order high - frequency terms @xmath21 . this is highlighted on the simple example of cylindrical circular tubes . this is the matter of section [ be ] . next , we give a very simple analytic derivation of the attenuation - without - distortion finding of ref . @xcite summarized above . indeed , generalizing to arbitrary geometry a work done by polack @xmath5 @xmath6 @xcite for cylindrical circular tubes , we show the following exact property ( for 1d propagation along one axis @xmath22 ) : @xmath23where @xmath24 and @xmath25 are the two first asymptotic green s functions computed by retaining in the wavenumber the terms up to the first and second order on the inverse stokes number @xmath17 respectively . this is done in section [ sw ] , using a scaled - form of the green s function . the first is only determined by @xmath7 , @xmath26 and @xmath9 , and exactly predicted by the models @xmath0 and @xmath1 . the second demonstrates the mentioned attenuation - without - distortion effect through the exponential . the decay length @xmath4 , however , depends in part of the next frozen parameters , two viscous and thermal purely geometrical characteristic surfaces @xmath27 and @xmath28 involved in the above - mentioned terms @xmath21 : @xmath29 ^{-1 } , \label{decay l}\]]where the fluid constants are , @xmath30 the fluid kinematic viscosity , @xmath31 the adiabatic speed of sound , @xmath32 the prandtl number , and @xmath33 the ratio of heat coefficients . now , because the models @xmath0 and @xmath1 ( especially @xmath0 ) give inaccurate predictions for the @xmath21 terms , this decay length @xmath34 will not be accurately captured by the models . indeed , on the simple example of cylindrical circular tubes it can be checked that @xmath34 is completely misrepresented by model @xmath0 , which gives a negative estimate for it , and still largely underestimated by model @xmath1 , which produces only about @xmath35 of its correct value , due to @xmath36 overestimation of surfaces @xmath27 and @xmath28 . both models @xmath37 and @xmath1 do not describe the correct high - frequency attenuation - without - distortion effect . technical details are given in appendix to lighten the main text . finally , our last objective is to show that , in spite of its faulty description of the decay length @xmath4 ( [ decay l ] ) , the model @xmath1 , nevertheless furnishes a relatively precise description of the complete exact green s function , especially when compared to other formulas . the asymptotic analytic green s function @xmath38 provides a reasonable description of the complete green s function with the same number of parameters as @xmath0 , but it uses parameters @xmath27 and @xmath28 which are unknown in general . the merits and drawbacks of the different descriptions are illustrated on the example of circular pores @xmath2 ) computing the exact green s function through fft and the known zwikker and kosten full frequency formulas ; @xmath3 ) computing likewise the model green s functions @xmath0 and @xmath1 through fft ; @xmath39 ) computing the exact asymptotic green s function @xmath38 through ( [ o2]-[decay l ] ) with the known exact values of the involved parameters ( namely @xmath40 , @xmath41 , @xmath42 ) ; and @xmath43 ) computing the ` model asymptotic ' green s functions through ( [ o2]-[decay l ] ) with the model values of the involved parameters ( @xmath44 , @xmath45 , and @xmath46 for @xmath1 and @xmath47 for @xmath0 ) . the results of fft computation are given in section [ fftc ] for the full - frequency models , while the results for asymptotic expansions are given in appendix . we start by recalling the form of the macroscopic equivalent - fluid equations in the frequency domain ( see @xcite):@xmath48where by definition , @xmath49 and @xmath50 are the macroscopic velocity and pressure obtained by coarse - graining ( averaging ) the microscopic fluid velocity and pressure fields , @xmath51 is the time derivative , @xmath52 and @xmath53 are the saturating - fluid density and adiabatic bulk modulus , and @xmath54 and @xmath55 are the dynamic tortuosity and the dynamic compressibility . notice that for simplicity , isotropy or 1d propagation along one principal axis is assumed , so that @xmath54 is a scalar . ( [ 11 ] ) make apparent the frequencies but not the wavenumbers associated to the time- and space - variable fields . this is consistent with the hypothesis that the effects of spatial dispersion are negligible . a necessary but not sufficient condition for this , is that the wavelengths are large . the complementary condition which will ensure that spatial locality is verified , is that no very different pore sizes are present , so that in particular , the presence of structures in the form of helmholtz resonators is excluded . indeed , with resonators , the fluid exchanged to and fro , is always associated by mass conservation with a corresponding spatial inhomogeneity in the wavefield . the resonance can not occur without concomitant spatial inhomogeneity in the macroscopic fields . as such , it is an effect of spatial dispersion , see @xcite p. 360 , no matter how large the wavelengths actually are . now , if we restrict sufficiently the possible geometries , resonances can not occur and the long - wavelength propagation becomes practically devoid of spatial dispersion @xmath12 is given by the traditional homogenization . the application of the asymptotic two - scale homogenization then yields two different microscopic boundary value formal problems to be solved for determining the two constitutive functions @xmath56 and @xmath57 @xcite . the first action - response problem specifies the ( velocity ) response of a viscous incompressible fluid subject to an applied time harmonic , spatially uniform bulk force source term : @xmath58 in the pore space ( with @xmath59 a periodic or stationary random field , in periodic or stationary random geometries ) , and satisfying on the pore surface ( no slip condition ) , @xmath60 in this problem , @xmath61 is a dimensionless unit vector in the direction of the applied bulk force , and @xmath62 is the scaled response velocity field ( dimension of @xmath63 ) . it determines the dynamic permeability @xmath64 and dynamic tortuosity @xmath54 introduced in the landmark paper by johnson @xmath5 @xmath6 @xcite , by the relations : @xmath65where @xmath66 is the coarse - graining averaging operation in the pore space , and @xmath67 is the porosity ( specific connected pore volume ) . the second action - response problem specifies the ( excess temperature ) response of a thermal fluid subject to an applied time harmonic , spatially uniform pressure source term : @xmath68 in the pore space , and satisfying on the pore surface ( no temperature - jump condition ) is assumed sufficiently important to ensure that the specific fluid - solid mass ratio is small ; then the solid is inert thermally and remains at ambient temperature . ] ) , @xmath69here , @xmath70 is a dimensionless unit constant representing the applied pressure , and @xmath71 is the scaled response excess temperature field ( dimension of @xmath63 ) . it determines the functions @xmath72 and @xmath73 , thermal counterparts of functions @xmath64 and @xmath54 introduced by lafarge lafargeth , lafarge , and then , the effective compressibility @xmath74 , by the following relationships : @xmath75where @xmath76 is the ratio of the specific heats . the above action - response problems and response - factor identifications ( d4 ) and ( [ d4 ] ) are written in blind manner by applying the traditional two - scale asymptotic homogenization . as we have insisted , it makes the important assumption that it is possible to neglect spatial dispersion . this is manifested in the first problem by the force source term @xmath61 which is set to a spatial constant ( simultaneously , the velocity field @xmath49 is represented by a divergence - free field ) , and in the second problem by the pressure source term @xmath70 which is also set to a spatial constant ( gradient - free)in fact , what is really meant here is that in this second problem , locally , the pressure field may be viewed as having a uniform gradient ; but the pressure linear variation around the mean may be omitted in a coarse graining volume , as it has no effect on the mean temperature . as soon as spatial dispersion is introduced , it no longer makes sense to represent the source terms by spatial constants . the spatial inhomogeneity of the source terms must be considered . ] . indeed , we could have written these problems and identifications directly without using the homogenization process , on the sole basis of assuming that spatial dispersion effects are absent . now , as a result of the divergence - free and gradient - free nature of velocity and pressure in the given action - response problems , it can be shown that the functions @xmath77 and @xmath78 have purely imaginary singularities @xcite , @xcite , and thus are smooth functions on the real axis this is the point 3 ) mentioned in introduction . this crucial point in the construction of the models will be considered at more length in section @xmath79 . in absence of a complete information on the microgeometry , the two above - mentioned microscopic formal boundary - value problems can not entirely be worked out and their exact detailed solutions ( from which @xmath80 and @xmath55 can in principle be extracted by coarse - graining , see eqs.([d4 ] ) and ( [ d4 ] ) ) are missing . nevertheless , in the limit of high - frequencies and low - frequencies , some general characteristics of the solutions and corresponding functions @xmath54 and @xmath74 may be obtained . that may be sketched as follows . in the limit of high frequencies , the viscous and thermal terms @xmath81 and @xmath82 become negligibly small compared to the other terms . the fluid motions , except for vanishingly small viscous and thermal boundary layers at the pore walls , become close to those of an inviscid nonconducting fluid ( @xmath83 , @xmath84 , with @xmath85 the thermal conductivity @xmath86 . we refer to this limit simply as the ` frozen limit ' ( see footnote [ frozen ] ) . accordingly , the quantities @xmath87 and @xmath88 everywhere tend ( except at the pore walls ) to the ` frozen ' fields @xmath89 and @xmath90 verifying the following equations in the pore space , ( with @xmath91 a periodic or stationary random field , in periodic or stationary random geometries ) , @xmath92and verifying at the pore walls ( @xmath93 is the normal on the latter ) , @xmath94 now assume , following johnson @xmath5 @xmath6 @xcite and allard @xcite , that the pore - surface interface appears locally plane in this asymptotic high - frequency frozen limit . this is in principle an assumption that the viscous and thermal boundary layer thicknesses @xmath95 and @xmath96 eventually become small compared to a characteristic radius of curvature of the pore surface ) become meaningful much more rapidly . ] . then , the functions @xmath54 and @xmath97 expand in integral power series of these thicknesses , which allows us writing @xmath19 @xmath20notice that the case of fractal geometry which modifies the exponent @xmath98 in the first correction terms see @xcite is excluded by the assumption that the pore walls appear locally flat at the scale of the boundary layer thickness ; the presence of sharp edges which modifies the exponent @xmath70 in the second correction terms see @xcite is also excluded by this assumption . ] , @xmath99the geometrical parameters @xmath7 and @xmath100 ( dimensionless ) and @xmath8 and @xmath9 ( dimension of @xmath101 ) must be some pore averages constructed with the frozen fields @xmath89 and @xmath90 . the next order geometrical parameters @xmath27 and @xmath28 ( dimension of @xmath102 ) are dependent on other fields involved in the asymptotic limit and have not been worked out so far . detailed calculations made by johnson and allard @xcite , to which we refer the reader , show that the parameters @xmath103 , @xmath104 , @xmath8 and @xmath105 may be written as follows ( see also , in the most detailed manner for @xmath8 , @xcite ) : @xmath106and , @xmath107where @xmath108 denotes the pore walls and @xmath109 denotes the connected pore volume . parameter @xmath110 is brown s electric formation factor brown ( @xmath89 represents either an inviscid - fluid scaled velocity or acceleration field for the problem of incompressible inviscid fluid flow , accelerating under the action of external bulk force or applied pressure drop , or else , an electric scaled field for the problem of electrical conduction in the bulk fluid @xcite , @xcite ) . parameter @xmath26 is an effective pore radius for dynamically connected pore sizes which was introduced by johnson @xmath5 @xmath6 @xcite for the problem of electrical conduction in the bulk fluid , perturbed by a thin , different conducting layer at the pore walls . parameter @xmath9 is a length characterizing a simpler effective pore radius twice the fluid - volume to fluid - surface ratio sometimes referred to as the kozeny - carman radius ; allard @xcite identified it as the thermal counterpart of parameter @xmath8 . an incomplete reasoning to obtain the parameter @xmath8 , leading to an incomplete expression @xmath111 , is often made , e.g. @xcite @xcite @xcite ; the reasoning inaccuracy outside the viscous boundary layer or the perturbed conducting layer , of a perturbation contribution due to a perturbed ideal - fluid or electrical bulk flow field orthogonal to the leading bulk flow field @xmath89 , and having , contrary to the latter , nonzero normal components at the pore walls . ] was clarified and corrected in @xcite along a line tentatively sketched in @xcite ( appendix d ) . in the opposite relaxed limit of low frequencies , the viscous and thermal terms @xmath112 and @xmath113 eventually become much greater than the inertial terms @xmath114 and @xmath115 and the boundary layers extend to the whole fluid . accordingly , the fields @xmath62 and @xmath71 everywhere tend to the d.c . or ` relaxed ' velocity and excess temperature fields @xmath116 and @xmath117 verifying , in the pore space ( with @xmath118 a periodic or stationary random field , in periodic or stationary random geometries ) , @xmath119and verifying , at the pore walls , @xmath120 in this limit the functions @xmath54 and @xmath121 expand in laurent s series , @xmath122the two first intervening geometrical parameters , @xmath10 ( darcy s viscous permeability ) and @xmath11 ( its thermal counterpart lafargeth , lafarge ) on one hand , both having dimension of @xmath123 , and @xmath15 ( viscous tortuosity ) and @xmath16 ( its thermal counterpart ) on the other hand , both being dimensionless , are pore averages constructed with the relaxed fields @xmath116 and @xmath117 . indeed , simple calculations show that they may be written norr , lafargeth , @xmath124mention that , among other things it is possible to show that @xcite , whatever the geometry , @xmath125@xmath126the equalities being satisfied only for the case of aligned cylindrical pores . as explained before , it is only for sufficiently simple geometries that the assumed long - wavelength nature of the considered fields , automatically imply that the microscopic flow - field can be considered divergence - free for the purpose of determining the density ( eq . ( [ d2 ] ) ) , and likewise , the excess pressure field can be considered gradient - free for the purpose of determining the compressibility . now , johnson @xmath5 @xmath6 have made the important observation that , because the velocity field is locally divergence - free , the singularities of functions @xmath127 and @xmath64 poles , zeros , and branch points necessarily are purely imaginary ( see @xcite appendix a ) . this characteristics of the response functions is explicitly apparent in avellaneda and torquato s solution of principle of the problem ( [ d1]-d3 ) , written in terms of the stokes operator s eigenmodes and relaxation times @xcite . for the functions @xmath128 and @xmath129 , it was similarly shown @xcite that , because the excess pressure field can be considered gradient - freesee footnote [ 5f ] ] , the singularities of functions @xmath130 and @xmath72 also are purely imaginary ( see @xcite appendix c ) . again , this characteristics of the response functions is explicitly apparent in a solution of principle of the problem ( [ d1]-[d3 ] ) , written in terms of the laplace operator s eigenmodes and relaxation times lafargeth , in exactly the same manner as @xcite . the wanted functions must therefore have simple smooth behaviors on the real axis . indeed , the drt ( distribution of relaxation times ) formalism of avellaneda and torquato may be used to explicitly show that in the laplace domain ( @xmath131 ) the hydrodynamic drag function @xmath132 is always a strictly decreasing positive function on the real axis @xmath133 , and the same holds true in the same manner for the corresponding thermal function @xmath134 see lafargeth . alternatively , more recently , the strictly decreasing nature of functions @xmath135 and @xmath136 versus real frequency has been shown in elegant manner by a variational formulation making use of the divergence - free nature ( [ d2 ] ) of the microscopic flow fields @xcite . finally , as observed by johnson @xmath5 @xmath6 @xcite , due to the special location of their singularities and the adjacent hypothesis of the absence of very different pore sizes , the functions @xmath77 and @xmath56 as well as the functions @xmath129 and @xmath128 , may be seeked as the simplest ones satisfying both the frozen and relaxed limits . to proceed , let us introduce a stokes number constructed using johnson s concept of dynamically connected pore size @xmath8 , @xmath137 : @xmath138johnson has proposed the following expression of dynamic tortuosity @xmath139 , @xmath140where @xmath141is a dimensionless shape factor associated to the geometry . this expression is the simplest analytical ansatz that yields the exact first two terms at high frequencies ( eq . ( [ hf ] ) ) and the exact first leading term at low frequencies ( eq . ( [ bf ] ) ) , and automatically satisfies the condition on singularities . as such , and for the general reasons discussed before , it is expected to provide a reasonable description of the exact function @xmath142 . factors of 8 are introduced for convenience in eqs . ( j14-[j15a ] ) , so that @xmath143 for cylindrical circular pores . similarly , to describe the function @xmath73 let us introduce a second stokes number corresponding to thermal effects:@xmath144the ratio @xmath145 @xmath146 is of the order of @xmath147 , not far from unity . proceeding as did johnson , lafarge @xcite proposed to write , @xmath148where @xmath149is the thermal counterpart of shape factor @xmath150 ( @xmath151 for cylindrical circular pores ) , giving in turn a definite model for the dynamic compressibility@xmath152as this model is the elaboration of allard s original attempt to transpose johnson s modelling to thermal effects @xcite , we refer to the combined modeling of functions @xmath153 and @xmath154 eqs . ( [ 12]15 ) as to johnson - allard s ( @xmath0 ) . subsequently , pride @xmath5 @xmath6 @xcite , while studying oscillating viscous flow in convergent - divergent channels , found that the simple formula ( [ j14 ] ) may significantly underestimate the imaginary part of dynamic permeability @xmath77 at low frequencies . to remedy this , they proposed different modified formulas . in essence these are formulas capable to account for the exact value of parameter @xmath15 , which , in such channels , may be significantly increased as compared to johnson s . notwithstanding , this parameter @xmath15 is not singled out in pride . its identification by eq . ( [ lfp2 ] ) ( see @xcite and lafargeth ) shows that it is constructed like the tortuosity @xmath155 , this time for the ` poiseuille'-like velocity pattern . therefore it is a measure of ` disorder ' of the ` poiseuille ' flow , and as such it is increased not only by the convergent - divergent mechanism considered by pride @xmath5 @xmath6 , but also , @xmath156 , by irregularities in the distribution of solid inclusions leading to the existence of privileged flow paths . whatever the cause of the enhanced ` poiseuille disorder ' , a significant increase of factor @xmath157 as compared to johnson s value @xmath158 , will make it more necessary to modify johnson s formula . similar formal considerations hold true also , @xmath159 @xmath160 , for thermal effects and the function @xmath161 . here also , a significant increase of @xmath162 may result when replacing a regular distribution of solid inclusions by an irregular one ( which leaves unchanged the thermal characteristic length @xmath163 ) . now , among the different modifications proposed in @xcite the first was the simplest one , capable to yield the exact first two terms at high and low frequencies , and simultaneously , to automatically fulfil the condition on singularities whatever the values of parameters @xmath67 , @xmath164 , @xmath165 , @xmath8 , and @xmath7 . finally , this formula was expressed by lafarge in terms of the parameter @xmath165 and the same description was then immediately transferable to thermal effects . the corresponding pride - lafarge s ( @xmath1 ) model formulas are:@xmath166where @xmath150 and @xmath167 are as before , and @xmath168 and @xmath169 are the new shape factors given by , @xmath170 these expressions reduce to @xmath171 s by setting @xmath172 and are simple transformations of the latter : they apply the simple group transformation @xmath173 to the basic johnson s square root function @xmath174 . two successive applications of the transformation ( with parameters @xmath168 and @xmath50 ) yield another same transformation ( with parameter @xmath175 ) which preserves both low- and high - frequency limits @xmath176 and @xmath177 . in this way there is some unicity in the @xmath1 modification , which is not extendible in obvious very simple manner . it must be realized that , as a side result of the strong constraints imposed by the special location of singularities and as long as the geometry is relatively simple , the @xmath1 expressions constructed with one more exact term than @xmath0 s at low frequencies , will also describe in a slightly more accurate manner all of the viscous and thermal relaxation description essentially improves the low frequencies if often made in literature . in reality because of the pole in @xmath56 and the way @xmath178 is related to @xmath179 see ( [ be ] ) , the small departures between @xmath18 and @xmath0 are mainly perceptible in the region of intermediate frequencies . ] . nevertheless , it should not be hoped that it is possible to gain , by means of this description , very meaningful information on the frozen parameters @xmath180 and @xmath181 . if the model expressions ( [ 14]-[15b ] ) were exact , the comparison of their high - frequency expansions with the exact ones ( [ hf ] ) would imply @xmath182 hence giving a fixed relation between the set of relaxed and frozen parameters . but the expressions ( [ 14]-[14p ] ) are not exact , and the numbers @xmath168 and @xmath169 in ( [ 15b]-[qpridem ] ) , obtained by low- or high - frequency matching of these non - exact expressions , will not be the same in general . thus we should not hope that the high - frequency expansions obtained with @xmath1 model : @xmath183yield anything precise for the @xmath184 terms . the example of cylindrical circular tubes may serve to illustrate this in quantitative manner . for a material with cylindrical circular pores of identical radius @xmath185 ( say for simplicity , all parallel and aligned along the direction of propagation ) , the two boundary value problems eqs . ( [ d1][d3 ] ) and ( [ d1][d3 ] ) determining @xmath54 and @xmath55 , are easily entirely stated and solved . in effect , these are nothing but the problems considered by zwikker and kosten @xcite in simplifying ( on account of the wide separation between wavelength and tube radius ) the governing equations of the full kirchhoff s theory of sound propagation in a cylindrical circular tube @xcite . we may say , in this respect , that the conventional equivalent - fluid theory neglecting spatial dispersion and expressed by eqs . ( [ 11][d4 ] ) , is the direct generalization to the case of arbitrary geometry , of zwikker and kosten s classic theory . now , zwikker and kosten s result is that @xmath54 , @xmath186 and @xmath55 express via bessel functions as follows : @xmath187where @xmath188 is the following relaxation function , @xmath189(smoothly varying , in a sort ` davidson - cole ' pattern , from relaxed value 1 at low frequencies to frozen value 0 at high frequencies ) . using the known small - arguments and large - arguments series and asymptotic expansions of bessel functions ( or kelvin functions ) , it is simple to derive the following low - frequency and high - frequency exact behaviors : low frequencies : @xmath190 high frequencies : @xmath191then comparing these results with the general low- and high - frequency expansions eqs . ( [ bf ] ) and ( [ hf ] ) , the following parameters values are easily obtained : @xmath192 from eqs . ( [ 15 ] ) and ( [ 15b ] ) , the circular - tube shape factors @xmath150 , @xmath193 and @xmath167 , @xmath169 are identified as:@xmath194whereas for the modified @xmath1 model ( [ qpridem ] ) the latter are : @xmath195 let us now examine how far the models are consistent with the limits ( lfc-[hfc ] ) and collection of parameters ( [ 19 ] ) . when johnson s values @xmath196 are used in the model expressions ( [ 14]-[14p ] ) , no @xmath197 terms appear in the high - frequency limit ( [ hf ] ) : the characteristic surfaces @xmath180 and @xmath198 predicted by model @xmath0 are given infinite values whatever the geometry ( this can be seen also on eqs . ( [ qpridem ] ) ) . simultaneously , in the low - frequency limit , the relaxed parameters @xmath157 , @xmath199 in ( [ bf ] ) are given as @xmath200 , @xmath201 . consider specifically the increments @xmath202 and @xmath203 , @xmath12 the differences @xmath204_{relaxed}-[% \alpha]_{frozen}$ ] produced by the viscous and thermal relaxation processes . for the present case of cylindrical circular tubes , it follows that the relative error @xmath205/ ( ) _ { exact}$ ] made by the model @xmath0 on these increments , is a @xmath206 = -25% error . when the @xmath1 @xmath168 and @xmath169 values ( [ 19b ] ) ( @xmath12 ( [ 15b ] ) ) are used , the latter increments are exactly described , but a 50% overestimation still exists for the characteristic surfaces : putting the values ( [ 19b ] ) in ( [ qpridem ] ) yields @xmath207 . when the modified @xmath1 @xmath168 and @xmath169 values ( [ 19bm ] ) ( @xmath12 ( [ qpridem ] ) ) are used , the characteristic surfaces are exactly described but there remain now a @xmath208 = + 20% error on the viscous and thermal increments . finally , when the values of @xmath168 and @xmath169 are taken as the arithmetic mean of the @xmath1 and modified @xmath1 ones , simultaneous but reduced errors are made : the characteristic surfaces are given with 20% overestimation ( instead of @xmath36 with @xmath1 ) , while the viscous and thermal increments are given with 9% error ( instead of 20% with modified @xmath1 ) . in general , the @xmath0 and @xmath1 models unsatisfactory account of parameters @xmath209 and @xmath181 , will be at the origin of some errors in the description of the propagation of transients at relatively short times or short distances , whereas the modified @xmath1 model description unsatisfactory account of parameters @xmath157 and @xmath199 , will be at the origin of errors in the description of transients at longer times and distances . to study this we compare in what follows different exact ( fully exact or asymptotically exact ) and modelled time green s functions for the case of cylindrical circular tubes . let us first define the green s functions and write the exact asymptotic results that have been described in introduction and are more detailed in appendix . a general method of defining and calculating a green s function in an infinite medium is by means of the effective frequency - dependent wavenumber @xmath210 in this medium . let us define our green s function @xmath211 ( or impulse response ) as a propagated dirac delta impulsive signal @xmath212 imposed at @xmath213 , or more precisely , as the inverse fourier transform of the propagation transfer function @xmath214 : @xmath215 . \label{green}\ ] ] setting to zero the viscosity and thermal conduction coefficients , no frequency dispersion arises . the wavenumber writes @xmath216 where @xmath217 , defined by the first eq . ( [ 1a ] ) , is the frozen speed of sound @xmath218 . the green s function ( [ green ] ) coincides with a dirac delta propagated at this velocity @xmath217 : @xmath219 . setting to nonzero values the viscosity and thermal conduction coefficients the medium wavenumber @xmath220 writes,@xmath221 , \label{23}\]]making apparent a complex function @xmath222 that describes frequency dispersion induced by the viscous and thermal relaxation processes . the green s function ( [ green ] ) writes , @xmath223 \text { . } \label{green2}\ ] ] this green s function @xmath224 now extends and flattens when observed at positions @xmath22 more and more remote from the origin @xmath213 . explicit asymptotic expressions of the green s function defined in this manner will now be obtained by considering high - frequency asymptotic expansions of the wavenumber @xmath220 ( meaning high - frequency asymptotic expansions of the function @xmath225 ) . it will also be convenient to express these functions in scaled form , as functions of a dimensionless delay time @xmath226 and a dimensionless position variable @xmath227 . to take but one example , before being fourier - transformed , the green s function @xmath228 where @xmath220 is say given by the model @xmath1 , may be viewed as a function of : ( _ i _ ) one dimensionless frequency variable , e.g. @xmath229 where @xmath230 is the characteristic viscous relaxation time given by , @xmath231(hence @xmath232 ) , ( _ ii _ ) one dimensionless position variable , e.g. @xmath233(hence @xmath234 can be regarded as a time domain stokes number , when replacing @xmath51 by @xmath235 in eq . ( 12 ) ) . ( _ iii _ ) a number of dimensionless parameters characteristic of the form of the porous space but not of its absolute dimensions ( @xmath236 , @xmath150 , @xmath168 , @xmath167 , @xmath169 , and @xmath237 ) , and , ( _ iv _ ) two dimensionless parameters characteristic of the fluid ( @xmath238 and @xmath239 ) . suppose that the parameters ( _ iii _ ) and ( _ iv _ ) of both the medium and the fluid are held constant . function @xmath228 then reduces to a function of @xmath240 which is parameterized by @xmath227 . there follows that the shape of the corresponding time - domain function will depend , in scaled form , on @xmath227 only , provided the time is counted in a dimensionless manner , @xmath241 for the time elapsed after the first arrival of the signal , @xmath242but function @xmath211 has the dimension of the inverse of time . therefore , @xmath243with @xmath244 a scale - invariant function that depends on the form of the pore space but not on its absolute dimensions:@xmath245 . \label{k2}\]]evidently , function @xmath244 will depend on the description of wavenumber or function @xmath222 : model @xmath0 , model @xmath1 , only some high - frequency terms retained , or complete exact zwikker and kosten s . two successive closed - form analytical exact asymptotic expressions can be derived for the frozen limit , using the series expansion of the wavenumber for the case where the function @xmath222 is represented by its first , and second order terms on @xmath17 . in the first case , using ( eqs . ( [ hf ] ) and ( [ 23 ] ) ) , we write : @xmath246where , @xmath247and obtain , @xmath248where , @xmath249 \text { . } \label{maino1'}\]]in the second case we write @xmath250where , @xmath251and obtain , @xmath252this yields an attenuation without distortion , corresponding to the asymptotic results indicated in eqs . ( [ o2][decay l ] ) . this analytical form of the green s function , exhibiting a superimposed exponential decay , is also a known result of musical acoustics @xcite . concerning the relaxed limit , the purely diffusive green s function @xmath253 can be obtained by assimilating @xmath55 with @xmath76 and @xmath54 with its first leading term @xmath254 in eq . ( [ bf ] ) . in this relaxed d.c . approximation , the wavenumber expands as : @xmath255(we have used the relation ( [ j15a ] ) ) . the corresponding analytical form of the green s function is deduced from the known result given by ref . landau : @xmath256where the notation @xmath257 is used to avoid confusion with eq . ( [ tau ] ) : @xmath258 . \label{diff'}\ ] ] green s functions @xmath211 computed numerically by inverse fft ( fast fourier transform ) of the transfer function @xmath259 , can always be obtained as soon as exact or approximate full - frequency expressions of the wavenumber @xmath220 in eq . ( [ 23 ] ) are known . these fft computations are to be carefully done , however , as it is detailed below . in appendix , the different results , either analytical or numerical , corresponding to the asymptotic green s functions are summarized and compared with the exact results of the fft . for the case of arbitrary porous media , approximate models such as @xmath0 or @xmath260 may be used for @xmath64 , leading to the computation , by fft , of approximate green s functions @xmath211 . for the case of simple workable geometries ( e.g. , cylindrical circular aligned pores of identical radius ) exact expressions are available , leading to the computation , by fft , of exact green s functions @xmath211 . for later use , it will be convenient to distinguish and denote respectively by : @xmath261 the exact ( zwikker and kosten ) green s function ; @xmath262 the full @xmath0 green s function ( @xmath196 ) ; @xmath263 the full @xmath1 green s function ( @xmath264 ) ; @xmath265 the modified full @xmath1 green s function ( defined by @xmath1 s expressions with @xmath266 ) ; @xmath267 the above asymptotic green s function obtained by retaining in the wavenumber the zero and first frozen terms ; @xmath38 the above asymptotic green s function obtained by retaining in the wavenumber the zero , first and second frozen terms . an important point is that the analytical diffusion solution @xmath268 given by eq . ( [ diff ] ) is used in our fft computations to improve the accuracy of calculations . indeed , the numerical computation of the full - frequency exact and model green s functions must be done by fft with some precautions : because of the importance of the higher frequencies on the response shape , subtracting the diffusive ( low frequencies ) approximation largely improves the results . thus , instead of calculating directly the inverse fourier transform @xmath211 of the function @xmath259 , we first compute the inverse fourier transform @xmath269 of the function @xmath270 the latter difference being zero at zero frequency , and increasing smoothly then decreasing rapidly toward zero at high frequency , the function @xmath271 is naturally windowed in the frequency domain . and then we use the relation @xmath272 , with @xmath273 as given by the eqs . ( [ diff]-[diff ] ) above ( see ref . the validity of the fft computation has been checked for first order asymptotic expression @xmath267 , with an accuracy better than 1% . for the computation , the values of the parameters have been chosen close to those of material m1 in ref . the porosity is @xmath274 , the flow resistivity @xmath275 , the permeability @xmath276 , corresponding to a radius @xmath277 @xmath278 . the temperature is @xmath279 , and the prandtl number @xmath280 . the characteristic viscous relaxation time is @xmath281 with respect to time @xmath282 ( both are dimensionless ) . the characteristic viscous relaxation time @xmath283 is defined by eq . ( [ theta ] ) . ( black ) solid line : zwikker and kosten formulae ( eqs . ( [ zw1 ] ) and ( [ zw2 ] ) ) . ( blue ) dotted line : pride - lafarge description ( eqs . ( [ 14 ] ) to ( [ 15b ] ) ) . ( green ) thin solid line : diffusive limit ( eqs . ( diff ) and ( [ diff ] ) ) . for very long times , the diffusive limit is reached.,width=377 ] the chosen length is @xmath284 @xmath285 ( the dimensionless length is @xmath286 ) . [ fig1 ] shows the results for the dimensionless green s function , obtained by fft . the pride - lafarge description is compared with the exact zwikker and kosten formula . as expected , the full @xmath1 description is very satisfactory for short times ( high frequencies ) and long times ( low frequencies ) . for very long times , both descriptions reach the diffusive ( analytical ) limit , @xmath12 the poiseuille behavior is reached . with respect to time @xmath282 ( zoom of figure [ fig1 ] ) . ( black ) , solid line : zwikker and kosten formula . ( blue ) , dotted line : @xmath1 formula . ( red ) , dashed line : @xmath1 modified formula . ( blue ) , mixed line : @xmath0 formula . ( yellow ) , solid , pale line : frozen @xmath287 s green s function @xmath38 see appendix , eqs . ( [ maino1]@xmath288[maino2 ] ) . , width=377 ] in order to emphasize these results , fig . [ fig2 ] shows a zoom of the previous figure , and other approximations have been added . the modified @xmath1 green s function , denoted @xmath265 , can be compared to the @xmath1 green s function . as expected , it is more accurate for short time ( during the signal rise ) than the original @xmath1 function , but it is less accurate for long times , because the choice of the parameters @xmath168 and @xmath169 has been done from the frozen limit instead of the relaxed limit . otherwise the @xmath0 model yields less accurate results than both @xmath1 models . the asymptotic green s function @xmath38 which uses the same number of parameters as @xmath0 performs quite well . on fig . 2 it is observed that the models @xmath1 and @xmath289 are in error mainly in the region of the maximum ; one elementary means to improve the description of the bump shape of the response would be to take the direct mean of the two models green s functions . the response modelled in this manner calculated with values of @xmath168 and @xmath169 equal to the mean of their original and modified @xmath1 values . ] would be very close to the exact one . additional results given later on ( fig.4 ) show however that this is a favourable situation related to the value @xmath290 and that no very significant improvement of model @xmath1 is to be expected in this manner for other @xmath227 values . . ( red ) dashed line : modified green s function @xmath291 ( green ) solid thin line : diffusive green s function @xmath292.(blue ) dotted line : @xmath1 green function @xmath263 at this scale this curve can not be distinguished from the exact result , see fig . [ fig4 ] for details.,width=377 ] the previous results are concerned with a fixed value of the parameter @xmath227 , @xmath12 a fixed value of the thickness of the material layer . in order to compare the different descriptions for several values of @xmath227 , we chose to compare the maximum values of the time responses ( there is a unique maximum ) . for the asymptotic ( frozen ) expressions at the first orders , @xmath293 and @xmath294 , the maximum values are given by an analytical expression , obtained from eq . ( [ maino1]):@xmath295 ^{3/2}e^{-(3/2)}\frac{1}{n_{1}^{2}\xi ^{2}}=\frac{0.2312}{n_{1}^{2}\xi ^{2}}% \text { ; } \label{jkj } \\ max(f_{s2 } ) & = & max(f_{s1})\exp ( -n_{2}\xi ) . \label{jkk}\end{aligned}\ ] ] for both orders the time @xmath296 of the maximum is the same ( this illustrates the attenuation without distortion effect brought by the @xmath184 terms ) . [ fig3 ] shows the result for the exact green s function @xmath261 and the two @xmath1 models , @xmath263 and @xmath265 , as well as the diffusive function @xmath268 . it shows the natural logarithm of the maximum value of @xmath244 with respect to the dimensionless space variable @xmath227 . the two @xmath1 models seem to be very good ; however better insight is found by subtracting the result corresponding to the exact green s function , as shown in fig . as expected , the @xmath1 description is very good for long distances @xmath227 , while the modified @xmath1 description is better for small distances @xmath297 . otherwise both are better than the @xmath0 description . the transition range values of the distance @xmath227 is approximately between @xmath298 and @xmath299 , corresponding to a range for the time domain stokes number @xmath300 . this range is similar to that accepted for the frequency domain stokes number defined by eq . ( [ 12 ] ) ( see e.g. ref . @xcite ) . . ( red ) dashed line : modified @xmath1 @xmath265 . ( blue ) mixed line : @xmath262 . ( yellow ) solid line : frozen @xmath287 s green s function @xmath38 see appendix , eqs . ( [ maino1]@xmath288@xmath288[maino2 ] ) . , width=377 ] a simple analytic formula , eq.(1 ) , can be used to compute the @xmath287 asymptotic green s function in a rigid - framed porous medium . it differs from the @xmath301 asymptotic green s function , by an exponential factor @xmath302 which describes an attenuation - without - distortion effect . we have so far no rigorous statements concerning the geometrical surfaces @xmath209 and @xmath181 that determine the @xmath184 terms in the high - frequency limits ( [ hf ] ) and then also determine the characteristic decay length @xmath4 eq . ( 2 ) . nevertheless these parameters are known for some geometries , such as cylindrical pores . by specializing to this particular geometry , a contrasted situation has been highlighted : while the models fail to give the parameters @xmath180 , @xmath198 , and thus , a correct description of the @xmath287 attenuation - without - distortion effect , they are nevertheless capable to give especially the model @xmath1 a relatively precise description of the complete green s functions . indeed , it is only at very short distances that the asymptotic @xmath287 green s function is close to the complete green s function : its imperfect representation by the models is of no very significant consequence . in connection with this , we note that , when the normalized distance @xmath297 decreases , the maximum error of @xmath1 s model occurs around @xmath303 , precisely when the @xmath287 green s function eventually starts to be valid ( see fig . this suggests that the small remaining errors of @xmath1 s model illustrated in fig . 4 are mainly due to the misrepresentation of parameters @xmath209 , @xmath181 , and that the model would become almost exact if modified to properly account for the latter parameters . the question of the modification to be done remains open . a problem of major interest is the use of the present investigation for the inverse problem , @xmath12 the determination of the parameters of a given material . regarding this , we have illustrated one simple fact : the description of the time domain green s functions is much more precise using the full model expressions than using the asymptotic expressions , as often done in practice . this suggests that there is also an important potential of improvement of the inverse methods of characterization based on recording transmitted and reflected pulses on different thicknesses of a material , provided the full expressions are exploited in the analysis we emphasize , in this respect , the importance of the substraction of the diffusive solution when computing the inverse fft . recall that , since we assume a smooth pore - surface interface , in the frozen limit the product @xmath305 in eq . ( [ 23 ] ) expands in successive integral powers of @xmath17 ( see footnote [ fract ] ) . this has been done with dimensionless variables in section [ ae ] . the two coefficients @xmath306 and @xmath307 have been obtained using the a priori expansions of the functions @xmath54 and @xmath73 ( eqs . ( [ hf]))@xmath308 for the case of cylindrical circular pores , the result for @xmath307 has been given by keefe@xcite , using the zwikker and kosten solution:@xmath309(eq ( [ 45 ] ) also follows by putting the values ( [ 19 ] ) in eq . ( [ n2 ] ) ) . for the general case , @xmath307 requires the missing frozen @xmath184 information @xmath27 and @xmath28 . it will not be given by the asymptotic expansion , eqs . ( [ 16 ] ) and ( [ 18 ] ) , with either @xmath0 or @xmath310 values of @xmath168 and @xmath169 . as explained in section [ ffm ] , in the framework of the @xmath1 model , it is not possible to have a good estimation of the second order term , resulting in an expression for the coefficient @xmath307 which differs from eq . ( [ n2 ] ) : @xmath311 the case of cylindrical circular pores allows checking this . for this case , the latter equation becomes:@xmath312this expression differs from eq . ( [ 45 ] ) . for standard conditions in air , the exact result for @xmath307 is @xmath313 , while the approximated one is @xmath314 , @xmath12 more than three times smaller . notice that @xmath0 model would give a negative estimate of @xmath315 for it . these important discrepancies mean that the models @xmath0 and even @xmath1 will not be able to describe the exact @xmath287 attenuation effect . an alternative to the closed - form green s function obtained for this same @xmath287 asymptotic frozen limit can be obtained using the 1d wave equation that follows from eqs . ( 1 ) , @xmath12 the following helmholtz equation:@xmath316using the high - frequencies asymptotic limits ( [ hf ] ) we get , @xmath317 in the time domain , the corresponding asymptotic wave equation is written as follows @xcite:@xmath318comparison between ( [ 20]-[exalim ] ) and ( [ we ] ) shows that the coefficients are given by:@xmath319 . \label{exaa1c}\end{aligned}\]]the relationship between the coefficients @xmath320 and @xmath306 and @xmath307 is:@xmath321 notice that by using @xmath1 model one would arrive in the asymptotic high - frequency limit to the same asymptotic wave eq . ( [ we ] ) but with the following erroneous value of the index @xmath320 : @xmath322 . \label{a1c}\ ] ] in particular , using @xmath0 model , two important terms disappear as for this case one sets @xmath196 . using this expression for the index @xmath320 in ( [ 1b ] ) corresponds to using the equations ( 14 - 16 ) of fellah @xmath5 @xmath6@xcite , who computed the green s function for an infinite medium described by the above wave equation ( [ we ] ) , by using the laplace transform method notice that in ref . @xcite there was a mistake of a factor 2 in the term under the root in eqs . ( 14 ) and ( [ 14p ] ) , without influence on further equations . moreover the last term @xmath323 in eq . ( [ 18 ] ) was omitted , resulting in a total coefficient of the term in @xmath324 in eq ( [ 18 ] ) equal to @xmath325 . here , consistent with eq . ( [ 18 ] ) , an additional term has been included in the bracket in eq . ( [ a1c ] ) . ] . the fft computations of the exact green s function can be compared with the following expressions : @xmath267 the ( frozen ) @xmath301 green s function ; @xmath38 the ( frozen ) @xmath287 green s function ( eq . ( [ maino2 ] ) with @xmath266 in eq . ( [ 42b ] ) ) ; @xmath326 the @xmath0 @xmath287 green s function ( eq . ( [ maino2 ] ) with @xmath196 in eq . ( [ 42b ] ) ) ; @xmath327 the @xmath1 @xmath287 green s function ( eq . ( [ maino2 ] ) with @xmath264 in eq . ( [ 42b ] ) ) . finally , the fft computations can be compared to the solution of the asymptotic @xmath287 wave equation ( [ we ] ) , for both cases @xmath328 and @xmath264 ( we again choose to compute these solutions using fft , with eqs . ( [ 23 ] ) and ( [ exalim ] ) , without expansion of eq . ( [ 23 ] ) ) : @xmath329 the ( frozen ) @xmath287 green s function ( @xmath266 ) ; @xmath330 the @xmath1 @xmath287 green s function ( @xmath264 ) . these solutions being @xmath287 are expected to be very close to the corresponding solutions obtained using the asymptotic @xmath287 wavenumber . results are plotted on fig . notice that @xmath38 , the ( frozen ) @xmath287 green s function , which is a very simple analytical expression , is the best approximation and leads to interesting results , except at long times . ; @xmath267 ; @xmath331 @xmath327 ; @xmath332 @xmath38 , the latter being also shown in fig . 4.,width=377 ] the @xmath333 description is slightly better than the first order of the frozen asymptotic @xmath267 , but as expected , it is much comparable to the latter , as it severely underestimate the decay length @xmath4 . otherwise , for short distances @xmath227 , the second order of the asymptotic wave equation solution @xmath329 exhibits the expected convergence to the results of the solution @xmath38 based on wavenumber expansion at second order . this convergence is lost for the comparable @xmath1 model estimates @xmath330 and @xmath327 , as a result of using the faulty @xmath1 coefficients ( @xmath334 . for longer distances , the frozen asymptotic @xmath335 solution @xmath329 appears to be less accurate than the frozen asymptotic wavenumber solution @xmath38 : it is not easy to have an interpretation for this result . the second order solution of the wave equation , as presented in ref . @xcite differs ( by definition ) by the third order , with the solution based on wavenumber expansion at second order , the latter being simpler to use in practice looking at the calculation made in the frequency domain , the figure 3 of this paper exhibits a ratio between the @xmath1 description and the @xmath0 one ( @xmath196 ) that is almost independent of frequency : this is the attenuation - without - distortion effect that is described by eq . ( [ maino2 ] ) in our wavenumber - based asymptotic calculation . ] . we wish to thank bruno lombard for fruitful discussions . z.e.a . fellah , c. depollier , m. fellah , w. lauriks , j .- y . chapelon , influence of dynamic tortuosity and compressibility on the propagation of transient waves in porous media , wave motion 41 ( 2005 ) 145161 . fellah , a. wirgin , m. fellah , n. sebaa , c. depollier and w. lauriks , a time - domain model of transient acoustic wave propagation in double - layered porous media , j. acoust . am . 118 ( 2005 ) 661670 . g. lefeuve - mesgouez , a. mesgouez , g. chiavassa , b. lombard , semi - analytical and numerical methods for computing transient waves in 2d acoustic / poroelastic stratified media , wave motion 49 ( 2012 ) , 667 - 680 . r. brown , connection between the formation factor for electrical resistivity and fluid solid coupling factors in biot s equations for acoustic waves in fluid - filled porous media , geophysics 45 ( 1980 ) 12691275 .

the existence of a finite electric dipole moment ( edm ) of the closed - shell atom yb implies parity and time reversal violations involving the nuclear sector . an important effect which can contribute to the yb edm is the tensor - pseudotensor electron - nucleus interaction characterized by the coupling constant . within the standard model ( sm ) of particle physics , as this form of interaction is not allowed . if a finite of yb is observed in experiments , then an estimate of can be obtained by combining with the theoretical calculations . a non - zero implies physics beyond the standard model . in this paper we present the result of our _ ab initio _ calculation of the edm yb using different many - body methods .

physics0111165

discrete symmetry violations in atoms are important phenomena to probe for physics beyond the standard model(sm ) of the particle physics . the electric dipole moment(edm ) of an atom , which is a signature of the simultaneous parity and time - reversal ( p - t ) symmetry violations is one such . there are several possible sources of p - t violation effects within an atom , the tensor pseudo - tensor electron nucleus interaction is an example which is semi - leptonic in nature . the atomic yb , which is a closed - shell atom having a @xmath3 is a very good candidate for atomic edm experiments to probe for the nuclear sector effects . an additional advantage of using a rare earth atom like yb is the closely spaced energy levels . the experimental results when compared with the theoretical results can yield signatures of physics beyond the sm . among the closed - shell atoms , the atom edm of xe@xcite and hg@xcite have been measured . though the hg experiment has set the record of being the most sensitive spectroscopy ever done , the result obtained is null and sets an upper bound to atomic hg edm as @xmath4 . still , it has set bounds on the parameters in particle physics models@xcite . improving the accuracy further is an important challenge in atomic edm experiments . atomic yb offer a possibility of achieving this by using the techniques of laser cooling and trapping . it is also desirable that the edm measurements be done in other closed - shell atoms to verify the physical effects observed . atomic yb also has the advantage of being relatively simple in level structure among the rare earth elements . it has been studied theoretically using a variety of atomic many - body methods and there are ongoing experimental studies . the excitation energies has been studied using relativistic coupled - cluster method@xcite ; the hyperfine structure constants and electric - dipole transition properties have been studied using multireference relativistic many - body perturbation theory@xcite ; and the oscillator strength of some of the important transitions were investigated using multi - configuration relativistic hartree - fock@xcite . the lifetime of some of the crucial levels have been measured experimentally@xcite and atomic yb has been laser cooled and trapped@xcite . it is the candidate for the atomic edm experiments using the methods of laser cooling and trapping of atoms which are in progress@xcite . the p - t violating effects which can be probed using atomic yb are the tensor pseudotensor ( t - pt ) electron - nucleus interaction@xcite and the schiff moment@xcite . an observation of a finite edm of atomic yb could mean nonzero @xmath1 , which is a signature of physics beyond the sm as @xmath1 is zero within sm . hence the theories which allow t - pt electron - nucleus interactions would play a role in understanding nature . the parameter @xmath1 is extracted from the experimental result of atomic edm by combining with the theoretical calculations . accurate atomic theory calculations are needed to obtain precise @xmath1.the atomic edm can also put bounds on the parameters of alternative models in particle physics . the schiff moment which arises due to either nucleon edm or p - t violating interactions between the nucleons can also contribute to the yb edm and can be used to extract the nuclear p - t violation parameters . we have calculated the contribution to atomic yb edm from both these effects but in this paper we will present the result of our calculations for t - pt contribution alone , in a later paper we will present the result of our schiff moment computation . the calculation of yb edm within a limited set of configuration state functions ( csfs ) using diagonalization and bloch equation based many - body perturbation theory ( mbpt ) methods was reported in an earlier paper@xcite . for convenient of reference this paper is called as paper - i hereafter . as a followup , in this paper we present results of calculations using much larger csf space . in addition , we discuss the drawbacks common to the diagonalization and bloch equation based perturbation methods , improvements are discussed and results of using these methods are presented . in this paper we present the method of the calculation . this section is a brief overview of the interaction used in the calculation and bloch equation based mbpt . a detailed description of which are given in the paper - i . the effective t - pt electron - nucleus interaction hamiltonian obtained by treating the nucleus nonrelativistically is @xmath5 where @xmath1 is the t - pt electron - nucleus coupling constant , @xmath6 is the fermi coupling constant , @xmath7 is the nuclear spin , @xmath8 and @xmath9 are the dirac matrices and @xmath10 is the nuclear density . this interaction hamiltonian is effective only within the nuclear region , where @xmath10 is non - zero . and the dependence on @xmath7 implies that it is observable only in odd isotopes of yb which has nonzero @xmath7 . we use fermi nuclear density model for our calculations . the bloch equation of an atom with unperturbed hamiltonian @xmath11 and the residual coulomb interaction @xmath12 as the perturbation is @xmath13 = v_{\rm es}\omega_{\rm es } -\chi_{\rm es}pv_{\rm es}\omega_{\rm es}. $ $ \ ] ] where @xmath14 is the wave - operator , @xmath15 is the correlation operator , @xmath16 and @xmath17 are the projection operators of the model and complementary space . the first and second terms on the right hand side are the principal and renormalization terms respectively . using epstein - nesbet(e - n ) partitioning @xmath11 and @xmath12 can be defined in terms of a set of csfs as @xmath18 where @xmath19 is the dirac - coulomb atomic hamiltonian , and @xmath20 and @xmath21 are the even and odd parity csf spaces respectively . the dirac - coulomb hamiltonian of an atom with @xmath22 electrons in atomic units ( @xmath23 , @xmath24 , and @xmath25 ) is @xmath26 where @xmath27 is the dirac matrix , @xmath28 is the momentum of the electron and @xmath29 is the nuclear potential . introduce the p - t violating interaction hamiltonian @xmath30 as perturbation and define the total perturbation hamiltonian @xmath31 as @xmath32 the interaction hamiltonian @xmath30 can also be expressed in terms of csfs as in @xmath11 and @xmath12 . treating @xmath31 as the perturbation and defining @xmath33 as the total wave operator @xmath34 where @xmath14 remains the same and @xmath35 is the wave - operator which has one order of @xmath30 and all possible orders of @xmath12 before and after @xmath30 . the @xmath30 is treated to first order since it scales as @xmath6 , which is very small . the expression of the atomic edm @xmath0 is @xmath36 where @xmath37 is the effective atomic edm operator . it is the dipole operator dressed with the all order residual coulomb interaction and one order of @xmath30 arranged in all possible sequence . the bloch equation based mbpt has been used for the calculation with a large csfs space as it is more efficient in terms of execution time unlike the direct matrix diagonalization approaches . an atomic many - body theory is size consistent if the properties calculated using it scales linearly as the number of the electrons . the diagonalization and the bloch equation based methods are size consistent within a complete csf space . but it is size inconsistent if the configuration space considered is incomplete . a consequence of incomplete cancellation of the unlinked terms , which scales nonlinearly to the number of electrons . the cancellation is complete when the csf space is complete . to study atomic edm which has important implications and small in magnitude it is preferable to use atomic - many body theory which is size - consistent even with an incomplete csf space . it is also difficult to satisfy the condition of completeness for heavy atoms , which are important candidates for the atomic edm experiments as the number of possible csfs runs into millions for a moderate size orbital space . the wave - operator calculated using the bloch equation within an incomplete csf space is size - consistent if only the linked terms are retained@xcite . the incomplete cancellation of the unlinked terms is then avoided @xmath38p = q\bigg ( { \mbox{$v_{\rm es}$}}\omega_{\rm es}p - \chi_{\rm es}p{\mbox{$v_{\rm es}$}}\omega_{\rm es}p\bigg ) _ { \rm linked } = q\bigg ( { \mbox{$v_{\rm es}$}}\omega_{\rm es}p - \chi_{\rm es}w \bigg ) _ { \rm linked } , \label{eqn2.1}\ ] ] where @xmath39 . redefine the wave - operator in terms of orders of excitation and consider only the single and double excitations . the wave - operator and correlation - operator are @xmath40 where @xmath41 is the order of excitation . the closed - shell single and double excitation wave - operator diagrams are shown in fig[fig1 ] . similarly , the @xmath42 diagrams are shown in fig[fig2 ] . unlike the single particle approach , the diagrams are used as representation of the physical effects and can not be evaluated directly using the usual goldstone rules as e - n partitioning is used . the @xmath43 is the energy and hence a number for closed - shell sytems . from the definitions @xmath44 = q\bigg ( { \mbox{$v_{\rm es}$}}+ { \mbox{$v_{\rm es}$}}\omega_{\rm es } ( 1 ) + { \mbox{$v_{\rm es}$}}\omega_{\rm es}(2 ) - \omega_{\rm es}(m)w\bigg ) _ { m,{\rm linked}}.\ ] ] let @xmath45 be the reference configuration and @xmath46 be the configuration space spanned by singly and doubly excited configurations . the wave - operators can then be expressed as @xmath47 where @xmath48 and @xmath49 are the excitation amplitudes . for closed - shell systems with a single reference , @xmath43 is just a number or closed diagrams . the term @xmath50 is therefore unlinked and does not contribute to the linked bloch equation . the one - body wave - operator equation is @xmath51p\!\ ! = \!\ ! \sum_{ar}\bigg [ { { \big \langle}}\phi_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_0{{\big \rangle}}+ \!\sum_{a'r ' } { { \big \langle}}\phi_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'}^{r'}{{\big \rangle}}x_{a'}^{r'}+\!\!\sum_{a'b'r's'}\!\!{{\big \langle}}\phi_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'b'}^{r's'}{{\big \rangle}}x_{a'b'}^{r 's ' } \bigg ] _ { \rm linked } \!\!\!\!\!\!\!\!\!\!\!\!\!\ ! { \big |}\phi_{a}^{r}{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}\label{eqn2.2}\ ] ] similarly , the two - body wave - operator the equation is @xmath52p\!\!=\!\ ! \sum_{abrs}\bigg [ { { \big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_0{{\big \rangle}}+ \!\sum_{a'r ' } { { \big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'}^{r'}{{\big \rangle}}x_{a'}^{r ' } + \!\ ! \sum_{a'b'r 's ' } \!\!{{\big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'b'}^{r 's ' } { { \big \rangle}}x_{a'b'}^{r's'}\bigg ] _ { \rm linked}\!\!\!\!\!\!\!\!\!\!\!\!\!\!{\big |}\phi_{ab}^{rs}{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}\label{eqn2.3}\ ] ] introduce @xmath30 as perturbation and redefine the perturbation hamiltonian as @xmath53 . the corresponding wave - operators are @xmath54 wave - operators @xmath55 and @xmath56 are same as before but @xmath57 and @xmath58 connect @xmath59 to @xmath60 , where @xmath60 is the configuration space opposite in parity to @xmath45 . within the total configuration space @xmath57 and @xmath58 can be represented as @xmath61 the equations of @xmath57 is @xmath62p & = & \!\!\!\sum_{ar}\bigg [ { { \big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_0{{\big \rangle}}+ \sum_{a'r'}{{\big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'}^{r'}{{\big \rangle}}x_{a'}^{r ' } + \!\!\sum_{a'b'r's'}\!{{\big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'b'}^{r's'}{{\big \rangle}}x_{a'b'}^{r 's ' } \nonumber \\ & & \!\!\!+\sum_{ct}{{\big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_c^t { { \big \rangle}}\overline{x}_c^t + \sum_{cdtu}{{\big \langle}}\overline{\phi}_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_{cd}^{tu}{{\big \rangle}}\overline{x}_{cd}^{tu}\bigg ] _ { \rm linked } \!\!\!\!\!\!\!\!\!\!\!\!\!{\big |}\overline{\phi}_a^r{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}\label{eqn2.4}\end{aligned}\ ] ] the wave - operator defined by equation ( [ eqn2.4 ] ) has only one order of as terms of the form @xmath63 are excluded . the equation of @xmath64 is @xmath65p & = & \sum_{abrs}\bigg [ \sum_{a'r'}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'}^{r'}{{\big \rangle}}x_{a'}^{r ' } + \sum_{a'b'r's'}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'b'}^{r 's ' } { { \big \rangle}}x_{a'b'}^{r 's ' } + \sum_{ct}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_c^t { { \big \rangle}}\overline{x}_c^t \nonumber \\ & & + \sum_{cdtu}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_{cd}^{tu } { { \big \rangle}}\overline{x}_{cd}^{tu}\bigg ] _ { \rm linked } \!\!\!\!\!\!\!\!\!\!\!\!\!\!{\big |}\overline{\phi}_{ab}^{rs}{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}\label{eqn2.5}\end{aligned}\ ] ] the term @xmath66 does not contribute as the one - body interaction hamiltonian can not create double excitations . the equations ( [ eqn2.2])([eqn2.5 ] ) are the required wave - operator equations . consider the term @xmath67 in ( [ eqn2.3 ] ) the diagrams of which are given in fig.[fig3 ] . among the diagrams ( a ) is linked but disconnected and remaining are connected . the disconnected diagrams in the wave - operator can introduce unlinked terms in the next iteration and hence a selection of only linked terms is required , this is difficult in terms of csfs as all the contributions are combined . the other method of separating the wave - operator is in terms of connected cluster operators . it is easier to select connected terms , since it can be done without separating cluster operator into subcomponents . consider the term @xmath68 again , the disconnected contributions has @xmath69 and @xmath70 in @xmath71 and excitation is through the one - body part in @xmath42 , avoiding these terms make the contributions from @xmath68 connected . the third term in ( [ eqn2.5 ] ) still has disconnected terms but using the the same method these can be removed . after these modifications all the terms of the cluster equations are connected . to distinguish from the linked diagram excitation operators define the cluster - operator as @xmath72 , then @xmath73 taking only the linear terms of one and two - particle cluster - operators , the wave - operator is @xmath74 the wave - operator @xmath14 is approximated by the linear cluster terms for the following reasons : 1 . the correlation introduced by @xmath75 is very small compared to the contribution from @xmath76 , which represents a large part of the electron - electron correlation effect . 2 . among the four - body cluster operators @xmath77 is the major contributor but in the present formalism this term can not be included as the csf coupling is not in particle - hole form . 3 . though @xmath78 does not contribute significantly to the electron - electron correlation it is important since and the dipole operators are single - electron operators . the equation of the cluster - operator treating as the perturbation is @xmath79 p = \bigg ( q{\mbox{$v_{\rm es}$}}\omega_{\rm es}p - \chi_{\rm es}wp \bigg ) _ { \rm conn}\ ] ] in closed - shell systems @xmath80 is always disconnected and do not contribute to the cluster equation @xmath79 p = \bigg ( q{\mbox{$v_{\rm es}$}}\omega_{\rm es}p \bigg ) _ { \rm conn}.\ ] ] these are the cepa-0 equations and do not include the epv diagrams . the linked epv diagrams should be avoided but unlinked epv terms should be retained . after suitable transformations the unlinked epv terms can be converted into connected terms@xcite , thus the cluster - operator equation is @xmath79 p = \bigg ( q{\mbox{$v_{\rm es}$}}\omega_{\rm es } p \bigg ) ^{\rm epv } _ { \rm linked}+ \bigg ( q{\mbox{$v_{\rm es}$}}\omega_{\rm es}p \bigg)^{\rm epo}_{\rm conn}.\ ] ] where the first term is epv and second term is non - epv . by rearranging @xmath81 then , the cluster - operator equations are @xmath79 p=\bigg ( q{\mbox{$v_{\rm es}$}}\omega_{\rm es}p \bigg ) ^{\rm epo}_{\rm conn } - \bigg ( q\chi_{\rm es}wp \bigg ) ^{\rm epv}\ ] ] the one - particle cluster operator equation is @xmath82p = \!\!\sum_{ar}\bigg [ { { \big \langle}}\phi_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_0{{\big \rangle}}+ \!\!\sum_{a'r ' } { { \big \langle}}\!\phi_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'}^{r'}{{\big \rangle}}{\cal t}_{a'}^{r ' } + \!\!\!\sum_{a'b'r 's ' } \!\!\!{{\big \langle}}\phi_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'b'}^{r 's ' } { { \big \rangle}}{\cal t}_{a'b'}^{r 's ' } -\!\bigg ( { \cal t}^r_aw\bigg ) ^{\rm epv}\bigg ] { \big |}\phi_{a}^{r}{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}. \label{eqn2.6}\ ] ] the cluster amplitudes are denoted by @xmath83 to distinguish from the one - particle cluster amplitudes represented by @xmath84 . here the calculation is using csfs and @xmath85 is the amplitude of the cluster operator which excites the reference csf @xmath86 to the csf @xmath87 . similarly , the two - particle cluster amplitudes can be defined . @xmath88p = \!\!\sum_{abrs}\bigg [ { { \big \langle}}\phi_{ab}^{rs } { \big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_0{{\big \rangle}}+ \!\!\sum_{a'r ' } \!{{\big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'}^{r ' } { { \big \rangle}}{\cal t}_{a'}^{r ' } + \!\!\!\sum_{a'b'r 's ' } \!\!\!{{\big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'b'}^{r 's ' } { { \big \rangle}}{\cal t}_{a'b'}^{r 's ' } -\!\bigg ( { \cal t}^{rs}_{ab } w \bigg ) ^{\rm epv } \bigg ] { \big |}\phi_{ab}^{rs}{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}. \label{eqn2.7}\ ] ] similarly , the pt - violating cluster - operators @xmath89 are evaluated using the equations @xmath90p\!\!\ ! & = & \!\!\!\sum_{ar}\bigg [ { { \big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_0{{\big \rangle}}+ \sum_{a'r'}{{\big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'}^{r'}{{\big \rangle}}{\cal t}_{a'}^{r ' } + \!\!\sum_{a'b'r's'}\!\!{{\big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'b'}^{r's'}{{\big \rangle}}{\cal t}_{a'b'}^{r 's ' } \nonumber \\ & & \!\!\!\!+\sum_{ct}\!{{\big \langle}}\overline{\phi}_a^r{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_c^t { { \big \rangle}}\overline{\cal t}_c^t + \!\!\sum_{cdtu}{{\big \langle}}\overline{\phi}_{a}^{r}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_{cd}^{tu}{{\big \rangle}}\overline{\cal t}_{cd}^{tu } -\!\ ! \bigg ( \overline{t}^r_a w \!\bigg ) ^{\rm epv } \bigg ] { \big |}\overline{\phi}_a^r{{\big \rangle}}{{\big \langle}}\phi_0{\big |}\label{eqn2.8}\end{aligned}\ ] ] and @xmath91p \!\!&= & \!\!\sum_{abrs}\bigg [ \sum_{a'b'r's'}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$h_{_{\rm ptv}}$}}{\big |}\phi_{a'b'}^{r 's ' } { { \big \rangle}}{\cal t}_{a'b'}^{r 's ' } + \sum_{ct}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_c^t { { \big \rangle}}\overline{t}_c^t \nonumber \\ & & + \,\sum_{cdtu}{{\big \langle}}\overline{\phi}_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\overline{\phi}_{cd}^{tu } { { \big \rangle}}\overline{\cal t}_{cd}^{tu } -\bigg ( \overline{\cal t}^{rs}_{ab } w \bigg ) ^{\rm epv } \bigg ] { \big |}\overline{\phi}_{ab}^{rs}{{\big \rangle}}{{\big \langle}}\phi_0 { \big |}\label{eqn2.9}\end{aligned}\ ] ] using the cluster - operators @xmath92 the atomic edm @xmath0 can be calculated using the operators as @xmath93 where @xmath94 is same as before except that the wave - operator is now in terms of connected clusters . the cluster equations are similar to the cepa-2 equations@xcite , the explicit cepa-0 equations are same as these equations without the epv terms . the cluster based formalism is not an order by order formalism but an iterative scheme where the bloch equation is defined in orders of excitations rather than the orders of perturbation . it is possible to separate the contributions from various many - body effects cleanly using moller - plesset partitioning and an added advantage is the one - to - one correspondence with the diagrams using goldstone evaluation rules . such an approach has been used@xcite to calculate parity non - conservation in atomic cesium to very high accuracy . the e - n partitioning mixes the contributions , but it has the advantage of capturing the static correlation effects very effectively . the cepa-0 equations are exactly the linearized singles and doubles coupled cluster equations , the cepa-2 equations includes a class of non - linear terms . the diagrammatic representation of the principal terms in ( [ eqn2.6 ] ) are shown in fig.[fig4 ] . the contribution from each of the diagrams are as described : 1 . diagram ( a ) contributes to the first term @xmath95 of the cluster equation and is independent of any cluster amplitudes . it is an important term as the iteration proceeds from this term . 2 . the diagrams ( b),(c),(d),(e ) and ( f ) contribute to the second term @xmath96 and the @xmath42 matrix element is coupled with the single excitation cluster amplitude . these start contributing from the second iteration , where the cluster amplitude @xmath97 is just the matrix element @xmath98 in the first iteration . 3 . diagrams ( g),(h),(i ) and ( j ) contribute to the third term @xmath99 and couple the double excitation cluster amplitude with the @xmath42 matrix element . similar to the second term , these diagrams start contributing from the second iteration . consider the diagrams ( c ) and ( e ) , though they resemble the hartree - fock potential scattering diagram these are very different . consider the the bubble part of the diagram ( c ) , it is summed over the occupied orbitals common to both the initial and the final csfs in the matrix element of @xmath42 . an example in yb is if the initial and final csfs are @xmath100 and @xmath101 respectively , the bubble part in ( c ) has all the occupied orbitals except the @xmath102 orbital . this is because both the csfs do not have @xmath102 , where as in the hartree - fock scattering diagram the bubble should have contribution from all the occupied orbitals . a similar description is true of diagrams ( d ) and ( f ) too . the term @xmath103 picks up the effect of the non - linear terms @xmath104 and @xmath105 , which have @xmath106 amplitudes . this implies that the wave - operator assumes the form @xmath107 ^{\rm epv } \label{eqn3.1}\ ] ] terms which are not included in the single excitation cluster amplitude equation are @xmath108^ { \rm epo}.\ ] ] a later section describes the method to choose epv terms from the renormalization part . the diagrams of the principal terms of the double excitation cluster amplitude ( [ eqn2.7 ] ) are shown in fig.[fig5 ] . the first term is similar to that of the single excitation cluster amplitude equation . the second term can be separated as @xmath109 the first term on the right hand side has connected as well as disconnected terms , from which only the connected terms should be retained . the remaining terms are connected and hence linked too since the conditions @xmath110 exclude the disconnected terms . then @xmath111 similarly , the third term can be expanded to @xmath112 \nonumber \\ & & \!\!\!+\!\!\ ! \sum_{a'\ne a , b}\bigg [ { { \big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'b}^{rs } { { \big \rangle}}{\cal t}_{a'b}^{rs } + \!\!\!\sum_{b'\ne a , b}\!\!\!{{\big \langle}}\phi_{ab}^{rs}{\big |}{\mbox{$v_{\rm es}$}}{\big |}\phi_{a'b'}^{rs}{{\big \rangle}}{\cal t}_{a'b'}^{rs}\bigg ] , \nonumber \end{aligned}\ ] ] where all the terms are connected . the triply and quadruply excited terms are excluded . each of the diagrams has an exchange diagram too . terms corresponding to each of the diagrams are : 1 . diagram ( a ) correspond to the first term in the cluster equation and has no dependence on any of the cluster amplitude . 2 . @xmath113 contribute to diagrams ( b ) and ( c ) . the final csf in this term has a hole - particle pair in common with the cluster amplitude . 3 . @xmath114 contribute to diagram ( c ) . though the topology of the diagram is same as that of @xmath113 , it is an epo diagram . the @xmath42 interaction changes the state of the particle and picks up a part of core - virtual correlation effect , which can be identified as core - polarization . 4 . @xmath115 contribute to diagram ( b ) . this also has similar topology with @xmath113 but is again an epo diagram , where there is a change of the hole state and correspond to core - core correlation effect . @xmath116 contribute to diagram ( h ) and ( i ) . these are epo diagrams where a hole - particle change to another hole - particle pair . these contribute to the core - virtual correlation effects . the term @xmath117 contribute to diagrams ( f ) and ( g ) . these are epo diagrams where one of the particle states in @xmath118 is excited to another particle state . this can also contribute to epv diagrams of the first kind , if it is a hole - line epv diagram then it will correspond to ( h ) and ( i ) and if it is particle line epv then diagram ( j ) . 7 . @xmath119 contribute to diagram ( j ) . this is a double excitation where the particle states from the cluster amplitude @xmath120 are excited to different particle states but the hole states remain intact . these terms capture the virtual - virtual correlation effects . @xmath121 contribute to diagram ( d ) and ( e ) . these diagrams correspond to a change of the hole state and are epo diagrams which capture the single - body hole - hole interaction component . this term can also contribute epv diagrams , the hole line epv diagram arising from this term is ( k ) and the particle line epv diagrams are ( h ) and ( i ) . 9 . @xmath122 contribute to ( j ) and is a hole - hole correlation term . thus terms in the cluster equation contribute to different physical effects . so far only the first three terms in the cluster equation have been considered . the last term in the doubly excited cluster amplitude @xmath123 is a renormalization term . it picks up a set of terms non - linear in cluster amplitude @xmath124 these pick up a class of epv terms which are non - linear in cluster amplitudes . consider the expression of @xmath43 @xmath125 the term @xmath126 is picked up through @xmath127 in @xmath43 , which implies that @xmath128 can have one hole(particle ) epv line or a pair of hole - particle epv lines . whereas in the single excitation cluster amplitude equation , the contribution from @xmath126 is captured through the term @xmath129 in @xmath130 . but the number of epv hole - lines or epv particle - lines are the same in both . in general the number of epv hole - lines and particle - lines in @xmath131 is limited by the @xmath42 if @xmath132 and by the cluster amplitudes if @xmath133 or @xmath134 . although @xmath126 is included in the single as well as the double excitation cluster amplitudes the topology of the diagrammatic representations are different . diagrams from @xmath103 has only a pair of hole - particle lines where as @xmath135 has two pairs of hole - particle lines . the terms linear in cluster amplitude can not violate pauli exclusion principal . it is possible only when there are csfs which are epv , which is not possible . the epv diagrams arise from renormalization terms , which are non - linear in cluster amplitude . the diagrams representing the renormalization terms are the cluster diagrams multiplied by the energy diagrams and hence unlinked but these can suitably be rearranged to yield connected diagrams . to select the epv terms all the orbitals are tagged with labels which are prime numbers . the csfs are also assigned a number which is the product of the prime numbers corresponding to the labels of the holes and particles of the csfs . consider a doubly excited csf @xmath71 , the prime numbers @xmath136 be the labels of the hole and particle states and @xmath137 their product , these five numbers identify the csf . however , only three if the csf is singly excited . to maintain consistency the remaining two indices are filled with another prime number not used in labeling the orbitals , let this number be @xmath138 but set the corresponding multiplying factor in @xmath139 as unity . according to this scheme , the ground / reference state of yb is identified by @xmath140 . similarly , labels are also given to the cluster amplitudes . the terms to retain from @xmath103 are those having csfs in @xmath43 identified by @xmath141 which can be divided by one or more of the numbers identifying hole(particle ) of the cluster amplitude @xmath83 . the number of possible division is the number of common hole / particle lines between @xmath83 and @xmath43 . during the selection process division by @xmath142 should be discarded as this does not represent any hole of particle states and this is automatically achieved as the corresponding multiplying factor in @xmath139 is unity . the advantage of this scheme is that it reduces the number of operations required in the selection process . the term @xmath143 of the doubly excited cluster amplitude equation has disconnected components if both the hole and particle states in the initial csf are present in the final csf . these are discarded and only the connected components are chosen . this can be implemented while calculating the matrix elements of @xmath42 . during the matrix element calculation the total number of hole states of the initial and final csfs is calculated . if this is equal to three then these contribute to @xmath143 . in the next step the connected component is chosen by selecting the @xmath42 matrix element which has @xmath70 . then the doubly excited cluster amplitude equation has only connected components . however the discarded components are disconnected but linked . the configuration space is spanned by the csfs constructed from the @xmath144 orbitals . the csfs are generated by single or double excitations from the occupied orbitals to the bound and the continuum virtual orbitals in all possible ways such that it yields the required final angular momentum . for the single reference mbpt , the reference csf of yb is @xmath145 , which be referred as @xmath86 . hence the occupied orbitals are @xmath146-@xmath147,@xmath148-@xmath149 , @xmath148-@xmath150,@xmath151-@xmath152 , @xmath151-@xmath153,@xmath154 and @xmath155 respectively . the csf space generated is not a complete active space but complete for the single and double excitations from the most important outer occupied orbitals within the orbital orbital space . the occupied - orbital shells of the configurations that has been considered are single excitation:=@xmath156 , @xmath157 @xmath158 , @xmath159 @xmath160 and @xmath161 + double excitation:@xmath162 , @xmath163 @xmath164 , @xmath165 , @xmath166 , @xmath167 @xmath168,@xmath169 , + @xmath170 , @xmath171 , @xmath172 , @xmath173 , @xmath174,@xmath175 , @xmath176 , + @xmath177 and @xmath178 . + the remaining electrons are distrubuted among the virtual orbitals in all possible ways . from all the csfs only the @xmath179 even parity and @xmath180 odd parity csfs are chosen . the number of non - relativistic csfs generated are given in table[table1 ] . although not included in the table , csfs with excitations from @xmath181 are also included in the csf space . the total number of odd and even parity csfs with bound virtual orbitals are 9930 and 17087 respectively . the modulus of the en - partitioned energies of the csfs the diagonal hamiltonian matrix elements are as shown in the histograms fig.4 . the two histograms are plotted such the the lowest @xmath182 is shifted to zero and the the range between the lowest and the highest are divided into ten units . the zero on @xmath182 axis are 14064.9531 and 14065.0068 hartrees for the even and odd parity csfs respectively . similarly , the highest @xmath182 are 14067.6720 and 14067.5996 hartrees respectively . from the histogram , the number of csfs with low and high energies are less whereas the number of the configurations that can give the intermediate energy are large . as a result the perturbation series converges fast as only a few of the configurations are quite close to @xmath86 and the energy separation with the rest of the csfs is quite large . number of odd parity csfs in the intermediate energy is more than the even parity as the odd parity configuration space can have many possible intermediate couplings to give @xmath180 , to limit the number of csfs within the memory limitations a selection of csfs is done . the double excitations to @xmath183 and @xmath184 symmetries above the energy of the converged orbitals are not included . another constraint on the choice of configurations is : there should nt be more than four open shells in the non - relativistic notation and eight in the relativistic form , choosing only singly and doubly excited configurations satisfies this condition for a closed - shell atom like yb . this constraint is due to the angular co - efficient computation program . the atomic yb edm is calculated using the bloch equation based mbpt and the size - consistent cepa equations . as a part of the calculation the ground state energy is also computed . the results with different methods are given in the following sections . using the wave - operator @xmath14 the ground state wave - function @xmath186 and energy @xmath185 are @xmath187 the denominator in the expression of @xmath188 is the normalization factor . the first order energy correction of the ground state is zero since the calculations are using the en - partitioned hamiltonian . the even parity csf space of the calculation is spanned by 9930 csfs , of which core part of first 4435 , 4436 - 9094 and remaining csfs are @xmath189 and @xmath190 , and @xmath191 , @xmath192 and @xmath193 , and @xmath194 respectively . using this set of csfs the ground state wave - function is @xmath195 where only the ten important csfs are given explicitly and the normalization constant is @xmath196 . as expected , the most important csfs in @xmath197 are doubly excited except for @xmath198 , which is a singly excited csf and does not interact very strongly with the ground state csf @xmath145 but contributes significantly by correlation through other csfs when the residual interaction is taken to higher orders . values of @xmath185 with increasing size of the even csf space is given in table[table2 ] . csfs are added to the calculation in sequence of excitations from the deeper core orbitals . as the size of the csf space is increased more complicated many - body effects are included in the computation . an important quantity that can be extracted from table[table2 ] is the change in @xmath185 . define the correlation energy @xmath199 as the energy difference between the csf energy of @xmath145 and the energy calculated using the csfs in the even - parity csf space . from the plot of @xmath199 in fig[fig7]a it is evident that the change in the ground state asf energy is not uniform but in steps interrupted by regions of very minimal changes . most significant changes of @xmath199 occur while increasing the csfs from 100 to 500 , from 1000 to 2000 and from 6435 to 7435 csfs respectively . these changes are not the combined effect of all the csfs added but due to a few important ones . the largest change of @xmath199 is while increasing from 1000 to 2000 csfs which corresponds to contribution from the core configuration @xmath200 , that is double excitation from the @xmath102 orbital shell . the remaining two are due to the core configurations @xmath201 and @xmath202 . each contribute @xmath203 , @xmath204 and @xmath205 hartrees respectively , the combined effect adds to 81.20% of the total correlation energy @xmath206 hartrees . from these it can be concluded that , the most important csfs contributing to the correlation energy have core configurations @xmath201 , @xmath200 and @xmath202 . as to be expected the doubly excited csfs are most important to capture the correlation effects and the low lying double excitations from @xmath102 orbital shell has the most significant contribution to correlation energy , it contributes 54.74% of the correlation energy . the plot ( b ) in fig:[fig7 ] indicates the need to include @xmath42 to high orders to capture the correlation effects accurately . from the graph the correlation effect due to two order of @xmath42 is @xmath207 hartrees and decreases in magnitude monotonically till fourth order to @xmath208hartrees but increases in the fourth order to @xmath209hartrees . this trend of oscillation about the final value of @xmath199 continues till convergence . the cycle of the oscillation has a period of four orders , that is in four orders it goes to the same side of the final value of @xmath199 and the amplitude of the oscillation decreases with each cycle . over all , the value of @xmath199 behaves like a damped oscillator with a cycle of four orders . if @xmath185 is calculated by truncating the perturbation to the first few orders where the amplitude of oscillation is quite significant the value of @xmath199 can be erroneous . the fig.[fig8 ] shows the trend of wave - operator convergence . the first graph fig.[fig8]a is the value of the convergence criteria plotted against the order of perturbation and second graph fig.[fig8]b is @xmath210 of the convergence criteria plotted against the order of perturbation . from the first graph it is evident that the convergence criteria begins with a small value but as shown in the second graph in terms of order of magnitude , the convergence is not so fast . the convergence is monotonic with very regular fluctuations . the wave - operator @xmath14 is stored in an order by order sequence . these are accessed as and when required during the calculation of @xmath35 . the wave - operator @xmath35 is calculated introducing the interaction hamiltonian and adding opposite parity csfs to the csf space . once is applied to the wave - operator @xmath14 , it maps onto the odd - parity component of the csf space and can never be mapped back to the even - parity space as is treated to first order only . this is followed by a sequence of residual coulomb interaction @xmath42 , which accounts for the correlation effects within the odd - parity sub - space . in sum total it is a sequence of perturbations applied to the ground state csf @xmath145 , where is sandwiched between all possible arrangements of @xmath42 . after the wave - operator @xmath211 is calculated , the mixed parity ground state wave - function @xmath212 can be written in terms of the wave - operators @xmath14 and @xmath35 as @xmath213 the value of @xmath0 can be calculated using the expression @xmath214 choosing the odd parity csf as mentioned in sec . v the correction to the ground state @xmath197 from the opposite parity sub - space due to is @xmath215 where @xmath216 and only the first ten important csfs are listed . the coefficients are much larger than unity but these should be scaled by the parameter @xmath217 . the product of the coupling constant @xmath1 and nuclear spin @xmath218 is retained as a parameter and @xmath1 can be estimated after combining with the experimental results . the above expression for @xmath219 shows that : 1 . like in the lowest order single - particle calculation , the coefficients of the csfs @xmath220 flip sign for @xmath221 . most of the important csfs are singly excited with respect to the ground state csf @xmath145 , which is to be expected since is a single particle interaction hamiltonian . singly excited configurations like @xmath222 can contribute through three possible many - body routes . first excites @xmath145 to @xmath220 then a sequence of @xmath42 connects it to @xmath222 , second a sequence of @xmath42 connects @xmath145 to @xmath223 and takes it to @xmath222 and third a sequence of @xmath42 takes @xmath145 to a csf @xmath224 which connects to @xmath225 via and another sequence of @xmath42 connects it to @xmath222 . although @xmath222 can not connect directly to the ground state @xmath145 as @xmath226 is close to zero within the nucleus , it is the third most important csf which contributes to @xmath227 this demonstrates the importance of many - body effects . the two most important doubly excited odd - parity csfs for the evaluation of edm are @xmath228 and @xmath229 . more interesting is the second as both the virtual orbitals involved can not contribute to the matrix elements . among the possible many - body routes which can contribute to the co - efficient of @xmath229 one possibility is through the deeper occupied orbitals @xmath230 and @xmath181 , which would contribute to core polarization effects . the value of @xmath0 is calculated using different sets of csfs , where the number of either the even or odd csfs are fixed to the maximum allowed and then increase the number of csfs in the opposite parity csf space . the results of such a sequence of calculations is given in the table [ table3 ] . if the previous sequence of calculation shows the importance of occupied orbitals in the whole csf space , these two sequences demonstrates the significance of the occupied orbitals in csf sub - space of each parity . like the correlation energy @xmath199 there is a significant change of @xmath0 when csfs with double excitations from @xmath102 are included . to appreciate the change better the values of @xmath0 in the two sequence are plotted in fig.[fig8 ] . consider the sequence where the number of even - parity csfs is fixed , @xmath0 increase with the number of the odd - parity csfs , which implies that there are no appreciable cancellations due to the odd parity csfs added . in the second sequence where the number of odd - parity csfs is fixed , @xmath0 decreases as the number of even - parity csfs is increased . in both the sequence there is a significant change in @xmath0 when csfs with double excitation from @xmath102 orbital are added . but , the changes are different in sign , in the even csf space the inclusion of csfs with double excitation from @xmath102 increases the value of @xmath0 where as in the odd - parity csfs it decreases . consider the expression for edm it can be expanded as @xmath231 which can be rewritten in terms of csf coefficients as @xmath232\vec{d } \omega_{\rm es , edm}{\big |}\phi_0 { { \big \rangle}}= 2\sum_j\bigg [ { { \big \langle}}\phi_0 { \big |}+\sum_{i\ne 0}{{\big \langle}}\phi_i { \big |}{\cal c}_i^{\rm ( es)}\bigg ] \vec{d } { \cal c}_j^{\rm ( es , edm)}{\big |}\overline{\phi}_j { { \big \rangle}}. \label{eq5.1}\ ] ] where the definitions of all the quantities are the same as defined in paper - i . within the whole csf space , the contribution from the first term is @xmath233 and the contribution from the second term is @xmath234 , which is just 11.80% of the first term . five most important configurations in the second term from the even - parity sub - space are @xmath235 , @xmath236 , @xmath237 , @xmath238 and @xmath239 and their contributions are @xmath240 , @xmath241 , @xmath242 , @xmath243 and @xmath244 respectively , where @xmath0 is in units of @xmath245 . all these are doubly excited csfs and mixes with the ground state csf significantly but the singly excited csf @xmath198 which is the third important csf of @xmath197 does not contribute strongly . in addition there is shift in the sequence of the important csfs compared to the sequence of contribution to @xmath197 , this is due to difference of dipole and coupling strengths between different csfs . within the whole csf space considered the value of @xmath0 is @xmath246 . in absolute terms this is @xmath247 less compared to the lowest order result of @xmath248 calculated in an earlier section . a major contribution to this difference is the many - body effects , this is because the direct contributions from the csfs added to the configuration space is small . which implies that the contribution from the many - body effects is just 10.81% of the total value and the change is negative . an added advantage of the order by order approach is that the contribution to @xmath0 can be calculated in terms of the order of residual coulomb interaction . earlier while calculating the ground state asf energy it was shown how a truncation in the order of @xmath42 perturbation can give an inaccurate value of @xmath185 . the cepa-0 wave - operator equations of @xmath249 and @xmath250 are identical to the linearized coupled - cluster equations . the atomic yb edm @xmath0 calculated using cepa-0 using the same sequence of csfs as in table [ table3 ] are given in table [ table4 ] and the following can be inferred : 1 . the results from cepa-0 do not differ significantly from the mbpt results for the singly excited csfs with core configurations @xmath251 and @xmath252 . this is because the non - linear terms do not contribute since the configuration space is spanned by singly excited csfs . a significant difference from the mbpt results is expected when doubly excited csfs are included in the csf space . this is observed when the csfs having core configuration @xmath253 are included . the change of @xmath0 calculated using mbpt by fixing the number of odd parity csfs to 17087 is @xmath254 , whereas cepa-0 reduces the change to @xmath255 . 3 . within the whole csf space considered , the value of @xmath0 calculated using cepa-0 is @xmath256 , which is 25.21% larger than the mbpt result . the difference gives a rough estimate of the contribution from the non - linear terms in the cluster equation . it is a rough estimate as the mbpt calculation also includes the contribution from the size - inconsistent terms , which are unphysical and are not included in the cepa formalism . the cepa-0 calculation does not include size - consistent non - linear terms of @xmath249 and @xmath250 . the results using the cepa-2 equations are tabulated in table [ table5 ] . to make comparisons convenient the calculation is done with the same choice of csf sequence as before . comparing with the results from the calculations using mbpt and cepa-0 give the following : 1 . the calculation within the csf space of singly excited even parity csfs with core configurations @xmath251 and @xmath257 and all the odd parity csfs is identical with the result calculated using cepa-0 . this implies that the contribution from the non - linear epv terms with @xmath258 from the even parity subspace is negligible . but , it is different from the mbpt result of the same csf space . and the difference of @xmath259 is due to the epo non - linear terms , which are included in the mbpt . 2 . consider the calculation within the csf space consisting of all the even parity csfs and the singly excited odd parity csfs having core configurations @xmath251 and @xmath257 . the cepa-2 and mbpt results differ by less than 2% . but , the result from cepa-0 differs from both by more than 12% . this shows that the contribution from non - linear terms with @xmath260 is not negligible or the contribution from the size - consistent is significant , with the present results it is difficult to distinguish between the two . 3 . the @xmath0 calculated within the whole csf space considered is @xmath261 and is larger than the mbpt result by 1.4% and less than the cepa-0 result by 31.86% . this implies that the contribution of non - linear terms to @xmath0 is very important . the effect of the epo unlinked terms in @xmath35 can be estimated by calculating @xmath262 with the epo renormalization terms included . the results are given in table [ table6 ] . the final value with the full csf space is @xmath263 , which is suppressed by 5.8% compared to the cepa-2 result . this difference is due to the epo contribution to the renormalization term in @xmath35 . another property which can be compared to gain an insight on the contribution of the size - inconsistent terms is the energy of the ground state @xmath185 . the value of @xmath185 calculated using the wave - operator @xmath14 derived here has no contribution from the size inconsistent terms but it excludes some of the less important size - consistent terms . the difference in the value @xmath185 calculated using the bloch - equation and @xmath14 derived from the cluster equation gives the contribution from the size inconsistent terms . like in @xmath0 the approximation is that the contribution form the epo size consistent terms non - linear in cluster amplitudes is very small , then the difference in the result can in principle be accounted to the size inconsistent terms . comparing the results from different methods , it is clear that a size - consistent theory is preferable for high accuracy computation of atomic edm and the contribution from the non - linear terms in the cluster amplitude is also important . with the present calculation the difference between the mbpt and cepa-2 result can not be accounted distinctly to non - linear terms or the size - inconsistent terms . a better comparison can be made after including all the non - linear terms in the cluster amplitudes . as the orbital space is made larger , the size of the csf space grows very large . this puts a limitation during the calculation as the memory requirement increases , it is manageable if the calculation is done at the level of single particle moller - plesset partitioning of the atomic hamiltonian . at the single particle level the cluster equation with the residual coulomb interaction hamiltonian reduces to the method used used by blundell @xcite and his collaborators . the inclusion of terms non - linear in cluster amplitudes is also relatively easy as compared to the en - partitioning . to check the quality of the wave functions , experimentally known quantities like hyperfine constants and excitation energies needs to be calculated . these will be reported in our later papers . we thank our colleagues rajat choudury , holger merlitz and p. panda for many useful discussions we had and the computer staff for making available to us the r10000 power challenge . 99 t. g. vold , f. j. raab , h. heckel and e. n. fortson , phys . rev . lett . * 52 * , 2229(1984 ) . m. v. romalis , w c griffith and j. p. jacobs and e. n. fortson , phys . lett . * 86 * , 2505 ( 2001 ) e. eliav , u. kaldor , and y. ishikawa , phys . a * 52 * , 291 - 296 ( 1995 ) . s. porsev , yu . g. rakhlina and m. g. kozlov , j. phys . b. * 32 * , 1113(1999 ) . s. g. porsev , yu . g. rakhlina and m. g. kozlov , phys . a * 60 * , 2781(1999 ) . j. migdalek and w. e. baylis phys . rev . a * 33 * , 1417 - 1420 ( 1986 ) c. j. bowers , d. budker , e. d. commins , d. demille , s. j. freedman , a .- nguyen , s .- q . shang , and m. zolotorev phys . a * 53 * , 3103 - 3109 ( 1996 ) . m. watanabe , r. ohmukai , u. tanaka , k. hayasaka , h. imajo and s. urabi , josa b * 13 * , 2377,(1996 ) . k. honda , y. takahashi , t. kuwamoto , m. fujimoto , k. toyoda , k. ishikawa , and t. yabuzaki , phys . a * 59 * , r934-r937,(1999 ) . y. takahashi , m. fujimoto , t. yabuzaki , angom dilip singh , manoj k. samal and b. p. das , in proceedings of cp violation and its origin , edited by k. hagiwara(kek reports , tsukuba , 1997 ) . m. v. romalis and e. n. fortson , phys . a * 59 * , 4547(1999 ) . s. m. barr , int . a * 8*,209,(1993 ) . v. v. flambaum and j. s. m. ginges , arxiv : nucl - th/0108007 angom dilip , bhanu pratap das , warren f perger , m k samal and k p geetha , jour . b * 34 * , 3089(2001 ) . ingvar lindgren and john morrison , _ atomic many - body theory_(springer , new york , 1982 ) . j. goldstone , proc . soc.(london ) * a239 * , 267(1957 ) . c. bloch , nucl . phys . * 6 * , 329(1958 ) . l. m. frantz and r. l. mills * 15 * , 16(1960 ) . w. kutzelnigg , methods in electronic structure theory , ed by h. f. schaefer ( plenum , new york , 1977 ) . s. a. blundell , w. r. johnson and j. sapirstein , phys . rev . a * 43 * , 3407(1991 ) . r. ahlrichs , comp . * 17 * , 31(1979 ) .

the pre - main sequence star v1647 ori has recently undergone two optical / near - infrared ( oir ) outbursts that are associated with dramatic enhancements in the stellar accretion rate . our intensive x - ray monitoring of this object affords the opportunity to investigate whether and how the intense x - ray emission is related to pre - ms accretion activity . our analysis of all fourteen chandra x - ray observatory observations of v1647 ori demonstrate that variations in the x - ray luminosity of v1647 ori are correlated with similar changes in the oir brightness of this source during both ( 20032005 and 2008 ) eruptions , strongly supporting the hypothesis that accretion is the primary generation mechanism for the x - ray outbursts . furthermore , the chandra monitoring demonstrates that the x - ray spectral properties of the second eruption were strikingly similar to those of the 2003 eruption . we find that x - ray spectra obtained immediately following the second outburst during which v1647 ori exhibited high x - ray luminosities , high hardness ratios , and strong x - ray variability are well modeled as a heavily absorbed ( n 4 ) , single - component plasma with characteristic temperatures ( kt 26 kev ) that are consistently too high to be generated via accretion shocks but are in the range expected for plasma heated by magnetic reconnection events . we also find that the x - ray absorbing column has not changed significantly throughout the observing campaign . since the oir and x - ray changes are correlated , we hypothesize that these reconnection events either occur in the accretion stream connecting the circumstellar disk to the star or in accretion - enhanced protostellar coronal activity .

1108.2534

the young stellar object ( yso ) was first noted to be an erupting source when it brightened suddenly in november 2003 , illuminating a new nebula now known as mcneil s nebula . this deeply embedded , low - mass yso is typically considered to be an ex lupi - type object ( an exor " ) though it shares some spectral characteristics with fu orionis @xcite , the prototype of a similar class of erupting pre - main sequence ( pms ) stars ( fuors " ) . exors are observed to brighten irregularly at optical wavelengths up to several times per decade ; these outbursts persist for weeks to months @xcite . fuors are ysos that erupt less often than exors , perhaps only once per century , and fade much more slowly , i.e. , on timescales of years to decades @xcite . the general consensus is that these eruptions are the result of massive accretion events that occur irregularly as young protostars grow ; such accretion episodes may be the primary process through which young stars accrete most of their mass @xcite . the mechanisms underlying exor and fuor outbursts remain very poorly understood ; however , models generally invoke the rapid onset of disk instabilities that lead to sudden inward migration of the inner disk truncation radius and dramatic changes in the star / disk magnetic field configuration ( e.g. , @xcite ) . soft x - ray ( @xmath10.3 kev ) generation via shock - heating of plasma at the bases of pms accretion streams has been proposed for some objects from x - ray grating observations ( see review by * ? ? ? * ) using plasma diagnostics based on helium - like ions , a prime example being tw hydrae @xcite . on the other hand , studies such as the chandra orion ultradeep project ( coup ) indicate that coronal activity , and not accretion , is likely the primary production mechanism for the harder x - rays that are characteristic of most pre - main sequence stars @xcite . debate still lingers as to the primary x - ray generation mechanism in ysos during optical outbursts . since 2003 , v1647 ori has been observed to undergo two distinct outbursts that have been detected in the optical and near - infrared @xcite as well as at x - ray wavelengths @xcite . the magnitudes and timescales for brightening and fading displayed by v1647 ori do not match those of exors . furthermore , the duration of the initial outburst was shorter than those of fuors @xcite . however , spectroscopic data do show signatures of accretion , such as strong h@xmath6 and br@xmath7 emission @xcite , confirming that the eruptions of v1647 ori do resemble those of fuors and exors . v1647 ori has been observed in x - rays before and during both of the eruptions ( as well as after the first of the two eruptions ) detected in the optical / near - infrared since 2003 . although other fuor- or exor - like ysos have been observed in x - rays ( v1118 ori @xcite ; ex lup @xcite ; z cma @xcite ; fu ori @xcite ; v1735 cyg @xcite ) , v1647 ori is the only eruptive yso to undergo such extensive x - ray monitoring . during the 2003 eruption , the sudden rise in flux and subsequent decline in both the optical and near - infrared correlated strongly with a sharp increase and then decline in the x - ray flux , which suggests a common origin @xcite . in this paper , we present and analyze x - ray spectra obtained with the chandra x - ray observatory ( cxo ) during the 20082009 outburst . in [ obs ] , we describe the observations and data reduction . in [ results ] , we examine the trends and patterns seen in the x - ray emission from v1647 ori over the past two outbursts , present modeling results for each of the five recent cxo observations , and compare these results to those obtained from previous ( 20032005 ) cxo and xmm - newton observations . finally , in [ discussion ] we discuss the implications of the results , and we argue that the measured properties of the x - ray emitting plasma are best understood within the framework of the accretion - generated x - ray hypothesis . observations of v1647 ori made with the chandra x - ray observatory ( cxo ) were triggered in 2008 september with a 20 ks observation ( cxo cycle 10 , pi : d. weintraub , obsid 9915 ) , after v1647 ori was reported to have undergone a new optical outburst between early 2008 january and late 2008 august @xcite . subsequent 20 ks observations were initiated in 2009 january and april ( obsids 9916 and 9917 , pi : d. weintraub ) . in addition , in 2008 november , two observations ( obsids 10763 and 8585 ; pi : n. calvet ) were made of the ngc 2068/2071 region that serendipitously included v1647 ori . together , these five cxo pointings yield an extended sequence of observations of this source over a seven - month period immediately following the onset of the optical outburst in 2008 . fortuitously , a field of view that includes v1647 ori was observed in 2002 november , a full year before the start of the 2003 outburst @xcite ; subsequent cxo observations targeting v1647 ori were obtained in 2004 march and april , 2005 august and december , and 2006 may and august @xcite . in addition , targeted xmm - newton observations were obtained in 2004 april @xcite and 2005 march @xcite , and a targeted suzaku observation was obtained in 2008 october @xcite . the 20042006 observations cover a two - year period during which v1647 ori more or less faded steadily from a strong yso x - ray source into a very faint one , albeit with large x - ray flux variability . for direct comparison with our analysis of the 20082009 epoch datasets , we have re - reduced and analyzed all of the 20022006 cxo datasets , using the same techniques and software packages that we have used for the latest ( 20082009 ) datasets . for all chandra observations , the advanced ccd imaging spectrometer ( acis ) was used in one of two imaging configurations . acis detectors have a pixel size of 0.49@xmath8 and each acis - i and acis - s ccd has a field of view of 8.3@xmath9 @xmath2 8.3@xmath9 . cxo / acis has significant sensitivity over the energy range 0.310 kev , with the soft ( @xmath101 kev ) x - ray sensitivity dependent on whether a back - illuminated or front - illuminated ccd serves as the detector ( see below ) . the ciao v4.1 software package and caldb v4.1.04.1.4 calibration files were used to reduce the data and extract pulse - invariant ( pi ) spectra . observation details are given in table [ tab-1 ] . to compare x - ray luminosities from 20082009 with those obtained from previous observations , source spectra were extracted from 2.5@xmath8 radius regions ( making sure the aperture was positioned so as to encompass as many source photons as possible , even when v1647 ori is 4@xmath9 off - axis ) while background spectra were extracted from regions near but beyond 2.5@xmath8 from the target , on the same chip , using 20@xmath8 outer radius extraction apertures . inspection of background light curves reveals background levels to be fairly constant with no evidence of large fluctuations , such as flares , occurring during the observations . spectra were also re - extracted from previous chandra observations using the same apertures to ensure that direct comparisons of results from different observation epochs could be performed reliably . for observations in which the primary target was not v1647 ori ( obsids 2539 , 10763 , and 8585 ) , exposure maps were generated to investigate off - axis position effects . in all three of these observations , the net count and mean count rate corrections due to off - axis source positions were insignificant , on the order of a few percent . the resultant spectral data points of the five recent observations were grouped into energy bins with a minimum of five counts per bin before spectral modeling was done . the count rates were high enough and durations long enough for each of the 20082009 observations that this bin size yielded pi spectra with good statistics . for the 20022006 observations , we employed single - count - minimum binning for very low - count observations and five - count - minimum binning for higher - count observations so as to yield spectra suitable for modeling . because no events with energies less than 0.5 kev and few events with energies greater than 8.0 kev were detected from v1647 ori , we limited our modeling to the 0.58.0 kev energy range . since v1647 ori was imaged with both front- and back - illuminated ccds in the exposure sequence under analysis , we generated synthetic spectra and convolved these spectra with the instrument responses to determine whether mean , broadband ( 0.58.0 kev ) count rates were directly comparable for all observations of v1647 ori . the simulations ( table [ tab - fibi ] ) showed that , for the plasma temperature regimes considered here , the back - illuminated s3 ccd is @xmath110% more sensitive to incoming flux than the front - illuminated acis - i ccds . net counts and mean count rates in the energy range 0.58.0 kev ( table [ tab-1 ] ) were therefore adjusted downward accordingly for the two observations ( obsids 5307 and 5308 ) that used the s3 ccd in order to remove this sensitivity bias . we calculated the median photon energies , mean count rates , and mean hardness ratios for all 14 cxo observations of v1647 ori . these results are reported in table [ tab-1 ] . in calculating the hardness ratios , the hard ( h ) x - rays are defined as those in the range from 2.8 to 8.0 kev and the soft ( s ) x - rays as those with energies from 0.5 to 2.8 kev @xcite . the hardness ratio is then hr = ( h - s)/(h+s ) , such that negative values of hr indicate softer spectra , and positive values indicate harder spectra . the recalculated values and error ranges of the mean count rates and median photon energies for the 20042006 observations differ only marginally from those reported by @xcite . with data from only three epochs in hand , @xcite reported that the x - ray flux brightened and hardened during the outburst ( 2002 november 14 to 2004 march 07 ) and then quickly faded and softened post - outburst ( 2004 march 07 to 2004 march 22 ) . with data from additional epochs available , however , @xcite reinterpreted the late 2004 march data as a short - term downward fluctuation , not as a quick end to the original outburst . they interpret the observed pattern of changes in the median photon energy as evidence that the x - ray spectrum hardened during outburst , remained somewhat hard for at least one year , and then softened after 2005 august 27 as both the optical and x - ray flux from v1647 ori returned to pre - outburst levels . the results obtained here ( for hardness ratio ) confirm this general trend . this hardening and softening ( the mean hardness ratios ( with 1@xmath11 uncertainties ) changed from @xmath120.34 @xmath13 0.31 ( soft ) to + 0.46 @xmath13 0.15 ( hard ) and back to @xmath120.45 @xmath13 0.33 ( very soft ) over the course of three years , from 2002 november through 2005 december ) took place over a timescale of months , with v1647 ori remaining near peak outburst x - ray luminosity and hardness for more than one year - with the notable exception of the 2004 march 22 data . v1647 ori likely remained in its soft x - ray , low x - ray luminosity phase for almost three years , from late 2005 until its latest optical outburst in 2008 august . the five cxo observations covering the interval from 2008 september through 2009 april all show v1647 ori in a hard x - ray , high x - ray luminosity state . as was the case during the 20022006 observations , the source remained near peak x - ray luminosity for at least seven months ( fig . [ overall ] , figure to be discussed in section [ discussion ] ) . as of 2009 december 18 , v1647 ori appeared to still be in an outburst phase with j- , h- , and k - band photometric magnitudes near values measured at the brightest phase of outburst during 2008 september @xcite . thus , it appears that v1647 ori remained near peak luminosity for at least one year . x - ray light curves and hardness ratios as a function of time were extracted for each observation to examine the properties of any possible short - term x - ray variability . in particular , these time series data permit us to investigate whether the trends in x - ray flux and hardness ratio seen in 20022006 were repeated during the recent outburst that began around 2008 august . the extracted x - ray light curves and hardness ratio time series for six of the nine observations from 20022006 are presented in figure [ lcprev ] . x - ray light curves and hardness ratio curves for the five 20082009 chandra observations are presented in figures [ 9915hardness][9917hardness ] , which show ( from top to bottom ) the overall x - ray light curve , the soft ( 0.52.8 kev ) light curve , the hard ( 2.88.0 kev ) light curve , and the hardness ratio time series . energy ranges for the soft and hard x - ray bands follow those defined in @xcite , and plotted uncertainties in mean hardness ratios and count rates are 1@xmath11 . during all five of the most recent ( 20082009 ) observations , v1647 ori experienced very pronounced , short - duration ( @xmath1210 ks ) x - ray flux variations in the form of large rises and subsequent declines in count rate . these fluctuations are evident at 4 ks and 16 ks in figure [ 9915hardness ] , at 8 ks in figure [ 10763hardness ] , at 20 ks in figure [ 8585hardness ] , at @xmath18 ks in figure [ 9916hardness ] , and at the beginning of the observation reported in figure [ 9917hardness ] . the mean count rate of v1647 ori varied by about a factor of @xmath12 for the first two epochs , while in the last three epochs , the mean count rate varied by a factor of @xmath16 . in all five epochs , the bulk of the x - ray variability is contributed by the harder portions of the spectra . the x - ray count rate for v1647 ori was intrinsically lower during most of the 2002 - 2006 observations ( the count rate was @xmath145 ks@xmath15 only on 2004 march 07 ) than in 20082009 ( the count rate was @xmath145 ks@xmath15 during all five observations ) . also , in some cases like obsids 5307 and 5308 ( both from 2004 ) , the observations were on the order of only a few kiloseconds in duration . the low count rates in most of the 20022006 data and the short exposure times in two of these observations make comparisons with the more recent datasets , in terms of the presence or absence of 210 ks timescale variations , fairly difficult . however , observations 5382 and 5383 , obtained in 2005 april and august , respectively more than a year after the initial optical outburst of v1647 ori in 2003 november yielded enough counts that we can compare these two light curves with the more recent observations with some confidence . as seen in these two light curves from 2005 ( fig . [ lcprev ] , lower - left panel , beginning at @xmath110 ks , and lower - middle panel , beginning at @xmath15 ks ) , the x - ray count rate again varied by a factor of @xmath14 on a timescale of 210 ks , a level and duration of short - term variability very similar to that seen in the more recent ( 20082009 ) observations . we conclude that kilosecond - timescale variations are seen in all observations of v1647 ori in which the signal is strong enough and the observing window long enough for us to search for such events , and that the amplitude duty cycle and duration of these few - kilosecond x - ray variations were stable over the five - year period from 2005 to 2009 . such variability appears to be part of the normal x - ray pattern for this object , at least when the count rate is elevated . one therefore needs to interpret the overall flux levels observed during the 2004 march 07 and 2004 march 22 observations , which , respectively , are only 5.5 ks and 4.9 ks in duration , with caution . the highest measured x - ray mean count rate following the latest ( 2008 ) optical outburst of v1647 ori is higher ( 23.6 ks@xmath15 ; one month post - outburst ) than the highest level ( 11.7 ks@xmath15 ; four months post - outburst ) observed during its 20032005 outburst ( table [ tab-1 ] ) . in addition , with the exception of the 2004 march 07 observation , the 20082009 observations ( a period spanning seven months ) show v1647 ori to be consistently brighter than it was during the 20032005 outburst . we find that in 20082009 the x - ray spectrum stays relatively hard , as is shown by the median photon energies , which in 20082009 are consistently about twice the energy ( 3.4 to 3.9 kev ) that was observed before the first eruption ( 2.0 kev in 2002 ) and after v1647 ori had returned to its quiescent state in late 2005 ( 1.7 kev ) . also , while the mean hardness ratios were negative both before ( 2002 november ) and after ( 2005 december ) the first eruption , the mean hardness ratios in 20082009 stay consistently around @xmath10.3 or greater . these results also reveal that v1647 ori exhibited a very interesting sequence of changes of state , beginning in 2004 march . v1647 ori had switched from its elevated state on 2004 march 07 ( 11.7 counts ks@xmath15 , median photon energy of 3.6 kev , mean hardness ratio of + 0.46 ) to what appears to be a quiescent state on 2004 march 22 ( 2.5 cts ks@xmath15 , median photon energy of 2.0 kev , @xmath120.45 mean hardness ratio ) but then was again in an elevated state one year later on 2005 april 11 ( 4.8 cts ks@xmath15 , median photon energy of 3.5 kev , mean hardness ratio of + 0.51 ) . yet by 2005 august 27 , v1647 ori was again returning to quiescence ( 0.9 cts ks@xmath15 , 3.0 kev median photon energy , + 0.01 mean hardness ratio ) and by 2005 december was fully back to quiescence . at least three explanations are plausible for the observed changes we have described in the x - ray emission from v1647 ori that occurred between 2004 march 07 and 2004 march 22 and between 2004 march 22 and 2005 april 11 . one : the outbursts generate both hard and soft x - rays ; most of the time ( 2004 march 07 and 2005 april 11 ) , the hard x - ray plasma is dominant in the x - ray spectrum as observed by cxo , but at certain times ( 2004 march 22 ) the soft x - ray plasma component is seen more clearly . the first major outburst ( 2003 november through early 2004 march ) had ended , at least in the x - ray regime , after only a few months . by 2005 april , the source was again in an elevated state and by 2005 august , four months later , was again returning to quiescence . three : the 2004 march 22 observation caught v1647 ori during a short duration ( 210 ks ) quiescent moment during the extended ( months long ) outburst phase . however , the xmm - newton observation obtained on 2004 april 4 showed a low x - ray flux level at the beginning of the observation that is consistent with the quiescent level observed by cxo that increased in the second part of the observation @xcite . moreover , the supporting evidence from observations made at optical and near - infrared wavelengths for the first outburst ending in 2004 mid - march and for a second outburst beginning later that year ( explanation two " ) is weak , at best . of the remaining two possibilities , we will argue in 4 that explanation one " is more likely . a comparison of the count rates ( figures [ 9915hardness][9917hardness ] , top panels ) and hardness ratios ( figures [ 9915hardness][9917hardness ] , bottom panels ) during the @xmath1210 ks variability periods suggests that the x - ray spectrum of v1647 ori may harden as it brightens and soften as the source fades during these few - kilosecond variations . this appearance of a correlation between the changing count rate and the changing hardness ratio on a few kilosecond timescale during short - term variability , however , is not consistent . in fact , we find a weak anti - correlation between these properties ( correlation coefficient = @xmath120.18 ) on these short timescales ; however , we do find a strong , positive correlation over much longer periods of time between the x - ray luminosity ( calculated in our spectral modeling ; see section [ modeling ] ) and hardness ratio , as shown in figure [ v1647orilumhrtemp ] . in this figure , the top panels show the seven - year x - ray luminosity curve of v1647 ori while the middle panels show the corresponding hardness ratios with a single , average measurement plotted for each observation epoch . a more direct demonstration of this correlation is shown in figure [ longterm_corr ] where we plot the change in mean hardness ratio versus the change in x - ray luminosity between observations of v1647 ori . we derive a correlation coefficient of 0.44 for these two parameters . as seen in the top and middle left - hand panels of figure [ v1647orilumhrtemp ] , from the first chandra observation in 2002 to the first observation of v1647 ori in 2004 , the luminosity increased by @xmath12 orders of magnitude while the mean hardness ratio increased from a relatively soft value of @xmath120.34 to a hard value of 0.46 . from the second to the third observation , the x - ray luminosity dropped by an order of magnitude while the mean hardness ratio softened to a much lower value of @xmath120.45 , much lower than that seen in the 2002 observation of v1647 ori when it was seen to be in quiescence . when the fourth chandra observation occurred in 2005 , the luminosity had risen back to outburst levels again while the mean hardness ratio increased again to + 0.51 , the highest observed to that date . in subsequent chandra observations through early 2006 , the x - ray luminosity of v1647 ori diminished while the hardness ratio dropped . in the top and middle righthand panels of figure [ v1647orilumhrtemp ] , we present the x - ray luminosity and mean hardness ratio versus time for most recent chandra observations of v1647 ori . the same correlation between changes in x - ray luminosity and changes in the mean hardness ratio seen in the 2002 through 2006 observations is evident in all observations from 2008 through 2009 . from 2008 september 18 to 2008 november 27 , there appears to be an anti - correlation with the luminosity decreasing slightly between these observations while the hardness ratio increases . we note , though , that in both of these epochs , the luminosity is high and the spectrum is quite hard . since the short - period light curves of each observation show that there is typically x - ray variability on timescales of kiloseconds , with x - ray flux sometimes increasing by several factors , this apparent anti - correlation might be attributed to short - duration variability . these results suggest that we can identify two states for v1647 ori , as seen in time - averaged data on a 530 ks timescale : * in its _ quiescent _ state , v1647 ori has a low count rate ( @xmath101 ct ks@xmath15 ) , low median photon energy ( @xmath162.5 kev ) , and negative ( soft x - ray dominated ) hardness ratio . * in its _ elevated _ state , the count rate is at least 2550 times greater , the median photon energy doubles , and the hardness ratio becomes strongly positive ( hard x - ray dominated ) . the x - ray evidence shows that when v1647 ori enters a major optical / near - infrared outburst , the x - ray profile switches from quiescent to elevated , and when the outburst phase ends , the x - ray profile switches from elevated to quiescent . modeling of the spectra employed xspec v12.4 . for those spectra with single - count binning , xspec was set to use the cash - statistic instead of the @xmath17@xmath18 statistic to assess the goodness of fit . all models at first assumed a thin , single - temperature plasma ( apec model ) subject to absorption by an intervening column of hydrogen ( wabs component ) . from the 2004 april xmm - newton observations , single - component thermal plasma models of the v1647 x - ray spectrum yield a hydrogen column density , chemical abundance , and plasma temperature of n@xmath0 = 4.1@xmath210@xmath3 @xmath4 , z = 0.8 solar , and kt@xmath5 = 4.2 kev , respectively @xcite , and recent suzaku observations yield similar results @xcite . therefore , initially , n@xmath0 , z , and kt@xmath5 were set approximately to the aforementioned values ( n@xmath0 = 4.0 @xmath2 10@xmath3 @xmath4 , z = 0.8 solar , and kt@xmath5 = 4.0 kev ) . as our initial test indicated that z is poorly constrained by the chandra data , only n@xmath0 and kt@xmath5 were left free to vary during the fitting procedure . for four of the five 20082009 observations , the models converged to best - fit solutions with column density and plasma temperature within a 90% confidence interval of their initial values ; however , the 2008 november 27 spectrum model was unable to converge to a physically - meaningful fit , so the plasma temperature was fixed at kt@xmath5 = 4.0 kev ( table [ tab-3 ] ) . a similar fitting method was used for the 20022006 observations ( table [ tab-2 ] ) . initially , the first five of these spectra spectra were modeled with freely varying hydrogen column densities and plasma temperatures , while the chemical abundance was fixed at 0.8 solar . these spectral models , however , were unable to constrain plasma temperatures or x - ray fluxes and luminosities , so the hydrogen column density was then fixed at n@xmath0 = 4.1 @xmath2 10@xmath3 @xmath4 , and the models were refit . the remaining four spectra , obtained when v1647 ori was reverting to an optical / x - ray quiescent state , were modeled with the hydrogen column density and plasma temperature fixed at n@xmath0 = 4.1 @xmath2 10@xmath3 @xmath4 and kt@xmath5 = 0.86 kev , respectively . given the small number of counts in these four spectra , the associated error ranges for these x - ray fluxes and luminosities were obtained by multiplying the error ranges for the mean count rates of these observations with an appropriate energy conversion factor ( ecf ) , where the ecf for each observation epoch was obtained by dividing the derived x - ray flux by the mean count rate . for the cxo observations of v1647 ori with sufficient total counts ( obsids 9915 and 9917 ) , we performed fits of a two - component thermal plasma model with parameters for hydrogen column density set to 4.0 @xmath2 10@xmath3 @xmath4 , plasma temperatures set to kt@xmath5 = 0.5 kev and 2.0 kev , and chemical abundance fixed at 0.8 solar . visual inspection of the spectral models shows that there is a negligible difference between the best - fit single- and two - component models , f - test results suggest that there is no statistical improvement in the model fits with the addition of a second plasma component ; i.e. , the best - fit parameters for the latter model converge on values such that the contribution of the lower - temperature component is negligible . we conclude , therefore , that all of the 20082009 cxo data are best fit with a single - component model . best - fit models for each of the five recent chandra observations are shown in figure [ bestfive ] . the overall trend of the spectral models is to converge to fits with parameters similar to those of the best - fit single - component model reported by @xcite . intervening hydrogen column densities do vary from model to model but remain in the n@xmath0 @xmath1 46 @xmath2 10@xmath3 @xmath4 range , and plasma temperatures are kt@xmath5 @xmath1 26 kev , slightly higher than but still within the range of uncertainty found by @xcite . determining a robust model for the last three recent epochs was more challenging than for the first two epochs , especially for obsid 9916 ( 2009 january 23 ) . a major concern for us was to assess whether the flux and hardness variability seen in the light curves for these observations ( figs . [ 8585hardness][9917hardness ] ) contributes to the difficulty of fitting a unique model to the spectral data . to assess the effects of variability on the robustness of our spectral analysis , we divided each of these the spectra into two phases , one in which the x - ray flux was relatively lower and one in which the flux was elevated . each of these spectra was then modeled independently . though the hardness ratios suggest that the spectra did harden marginally as v1647 ori went from x - ray - dim to x - ray - bright during these observations , the spectral modeling yields no measurable changes in hydrogen column density or plasma temperature . thus , it appears that the changes in the light curves are likely due only to changes in the plasma emission measure . our initial modeling of the v1647 spectra from 2008 and 2009 suggests the presence of line emission from near - neutral iron at 6.4 kev as well as from the ( unresolved ) helium - like iron k@xmath6-line triplet at 6.64 , 6.67 , and 6.70 kev . the neutral iron line emission at 6.4 kev is often seen in accretion - powered sources and is usually attributed to the fluorescence of cold gas in the presence of a nearby x - ray continuum emission @xcite . thus , the presence of this line in some of the v1647 ori spectra could be attributed to fluorescence of ( neutral ) circumstellar disk material by accretion - generated x - rays . it would not be surprising to detect this emission in the spectra of v1647 ori given that the environment of this yso does appear to contain the necessary ingredients for the formation of 6.4 kev emission , namely a strong , relatively hard x - ray source illuminating cold circumstellar material . observations of other ysos , including roughly a half - dozen sources in the coup survey , with these environmental components have also shown this feature . following @xcite , we added a gaussian component centered at 6.4 kev to the spectral models in order to account for the neutral line component . models of two observations ( 2008 september 18 and 2009 april 21 ) appear to be well fit with the addition of a 6.4 kev line with equivalent widths of @xmath1200 and @xmath1500 ev , respectively , while the 2008 november 27 - 28 and 2009 january 23 are fit well without the additional neutral iron line emission component ( fig . [ irongaussian ] ) . on the other hand , it is possible that the 6.4 kev emission is present during all of the 20082009 observations from cxo but that its spectral signature is muffled by noise in the 2008 november and 2009 january observations . while it appears that we have detected 6.4 kev iron emission in the 2008 september 18 and 2009 april 21 observations , we can not conclude definitively that these detections " are real . we note that the best - fit 6.4 kev equivalent widths are poorly constrained with error ranges extending from zero to roughly twice the equivalent width values . given this range of uncertainty , we can not exclude the possibility that there is no 6.4 kev iron line . we feel confident , however , that these detections are real given that visual and quantitative comparisons of our findings with those detections found by @xcite are very similar . we have compared the intensities / appearances of this line in the various observations of v1647 ori by cxo , xmm - newton , and suzaku . the equivalent widths of this emission feature are very similar to those found in the spectra of the 2004 april 4 xmm observation ( 109 ev ) and in the 2008 october 8 suzaku observation ( @xmath1600 ev ) . we are unable to clearly determine whether there is a correlation between the strength and appearance of the 6.4 kev line and any of the associated plasma characteristics . figure [ overall ] suggests that the overall x - ray flux of v1647 ori is strongly correlated with optical / near - infrared flux . this correlation is revealed more clearly in figure [ longtermhvsx ] , in which we plot the x - ray luminosity versus the i@xmath19-band luminosity . we interpolated i@xmath19-band luminosities for the 20022005 x - ray observation dates . we did not extrapolate i@xmath19-band luminosities for the two 2006 x - ray observation dates due to the highly uncertain flux behavior of v1647 ori . of the 2008 - 2009 x - ray observations , we could only interpolate an i@xmath19-band luminosity for obsid 9916 . with these eight interpolated luminosities , we derived a correlation coefficient between x - ray luminosity and i@xmath19-band luminosity of 0.65 . dramatic increases in optical / near - infrared flux for ysos , such as fu ori , have long been thought to be associated with enhanced accretion @xcite . in such an environment , material is channeled through magnetic funnels from the co - rotation radius of the circumstellar disk down to the photosphere @xcite . hence , when the x - ray flux from a pre - main - sequence star or protostar is elevated and the rapid rise in x - ray emission is directly correlated with large - scale optical outbursts , the correlation itself strongly suggests that accretion is the mechanism responsible for generating the increase in x - ray output @xcite . one way in which accretion - generated x - ray emission could be identified observationally would be through the relatively soft x - rays emitted by the plasma when it plunges onto the stellar surface at free - fall velocities and is shock heated to temperatures of a few million kelvin ( kt@xmath5 @xmath10.3 kev ) @xcite . we can speculate that a second such signature would be much hotter ( kt@xmath5 @xmath1few to tens of kev ) and harder x - rays generated in magnetic reconnection events @xcite within the accretion streams . whether the soft or hard x - ray generating plasma is observable likely depends on the observing geometry . if we have an unobscured line of sight to the footprint of the accretion column , our observations should be sensitive to the cooler plasma ; if the accretion column obscures our view of the accretion - column footprint , our observations should be sensitive only to the hotter plasma in the accretion stream ; and if the obscuration of the accretion footprint is partial , we might detect both plasmas . in between periods of dramatically enhanced accretion , previous studies such as coup @xcite and the xmm - newton extended survey of the taurus molecular cloud ( xest ) @xcite , suggest that the x - ray signature of young stars should be that of normal coronal emission . such emission would be similar to but much fainter than the hot , hard plasma seen from reconnection events in the accretion funnel . during quiescence , the x - ray flux of v1647 ori has a low count rate , low median photon energy , and negative hardness ratio ; in contrast , when the x - ray flux is elevated , the count rate increases by a factor of 25 or greater , the median photon energy doubles , and the hardness ratio becomes strongly positive . most of the v1647 ori light curves reveal that this yso also experiences what appear to be short - term ( few kilosecond ) variability in its x - ray flux . since most of the cxo observations show this short - duration variability when the x - ray count rate is high , the short - term variations are likely part of the normal behavior for v1647 ori . for all 20082009 cxo observations , the x - ray variability is almost entirely seen in hard ( @xmath142.8 kev ) x - rays . our model fits of the v1647 ori spectra indicate that the x - ray signature of the plasma we observe from this source during optical eruptions is predominantly a bright , hot , hard , single - temperature plasma heated to temperatures of 26 kev , which is also consistent with x - rays generated from reconnection events in an accretion funnel @xcite . we must keep in mind , however , that the intervening hydrogen column density will play a significant role in our ability to detect x - ray flux from accretion hotspots at temperatures of a few million kelvin , especially when dealing with column densities as high as those modeled in v1647 ori observations . an accretion - footprint ( shocked ) plasma at a temperature of kt@xmath5 @xmath1 0.3 kev subject to an intervening absorbing column density of a few times 10@xmath3 @xmath4 would have @xmath198% of its x - ray flux extinguished . even given the smallest hydrogen column density found via spectral modeling of the 20042005 v1647 ori observations when the hydrogen column density was allowed to vary freely ( @xmath11 @xmath2 10@xmath3 @xmath4 ) , such a soft component still has @xmath196% of its x - ray flux absorbed . if the hydrogen column density decreased dramatically to n@xmath0 of a few times 10@xmath20 @xmath4 the accretion shock emission could dominate the observed flux . this seems a plausible scenario to explain the softer plasma detected in late 2004 march , especially if the large intervening hydrogen column density inferred at other observing epochs was due mostly to the accretion streams . an alternative explanation is that we observed v1647 ori when it was in the midst of a large accretion episode that pushed the star - disk boundary inward to the point where the accretion became non - magnetospheric @xcite , effectively reducing the amount of hard x - ray flux produced by magnetic reconnection events in the accretion stream . during the 20082009 epoch , when the x - ray luminosity of v1647 ori increased , the spectrum hardened and the emitting plasma increased in temperature ( fig . [ v1647orilumhrtemp ] , right panel ) ; also , when the overall x - ray luminosity decreased , the x - ray spectrum softened and the x - ray generating plasma cooled . these correlations are also seen in the chandra observations following the 2003 eruption ( fig . [ v1647orilumhrtemp ] , left panel ) . however , between 2008 september and 2008 november , the spectrum appeared to harden slightly as the x - ray luminosity decreased slightly . if we are observing x - rays generated predominantly by the @xmath11 kev plasma in the accretion footprint , then a decline in accretion should result in a decline in total x - ray luminosity , particularly in the lower energy flux , resulting in a hardening of the spectrum . therefore , the hardening of the spectrum and decrease in x - ray luminosity observed between 2008 september and november might be a time when cxo was able to see plasma closer to the protostellar surface . in fact , this epoch showed a low temperature ( @xmath12 kev ) plasma , consistent with what could be a mixture of hotter ( 46 kev ) and cooler ( 1 kev ) plasmas . these data may offer evidence that the x - ray flux includes an emission component from the cooler plasma in the footprint of the accretion funnel located in the stellar photosphere . because the cooler plasma suffers greater extinction and because cxo is less sensitive to softer x - ray emission than xmm - newton , cxo would likely only detect the cooler plasma when the viewing geometry is favorable . our modeling work for the x - ray observations obtained in 20042005 and 20082009 , when the mean x - ray count rates were greater than 1 cts ks@xmath15 yield a best fit value for @xmath21 of about 4.1 @xmath2 10@xmath3 @xmath4 , consistent with the results derived by @xcite and a visual extinction of @xmath22 @xmath1 20 @xcite . we were unable to fit @xmath21 in our modeling work for the low count - rate observations during the quiescent period in 20052006 ; however , @xcite obtained a best fit value for @xmath22 of 19 @xmath13 2 , based on their optical , near - infrared , and mid - infrared observations obtained in february 2007 , which was also during the quiescent period ( based on the optical and near - infrared photometry reported by @xcite ) . the observed value of @xmath22 for february 2007 lends strong support to our use of @xmath21 = 4.1 @xmath2 10@xmath3 @xmath4 for our modeling of the quiescent epoch observations . in addition , together these data suggest that @xmath21 and , by implication , @xmath22 , remained essentially unchanged as inferred from the x - ray observations , whether v1647 ori was in the quiescent or elevated x - ray state . on the other hand , as seen at longer wavelengths , the extinction toward v1647 ori has changed . @xcite derive @xmath22 = 13 from near - infrared data obtained in 1998 by 2mass . during the outburst in 20042005 , @xcite found @xmath22 = 810 on 2004 february 18 , @xcite found @xmath23 = 1.26 , @xmath24 = 0.81 , and @xmath25 = 0.5 on 2004 feburary 18 , all of which are consistent with @xmath22 of @xmath15 @xcite , @xcite obtained @xmath22 = 11 from measurements of the 3.1 @xmath26 m water band on 2004 march 9 , and @xcite , who made optical and near - infrared observations from 2004 into very late 2005 , reported @xmath22 of @xmath15 during the 2004 outburst . @xcite then report that @xmath22 increased to @xmath110 by the end of 2005 , when v1647 ori had dramatically faded , and @xcite reported @xmath22 was as high as 19 by early 2007 . clearly , the extinction , as measured via optical and near - infrared measurements , changed first from quiescence to outburst and then from outburst back to quiescence . these apparently discordant results have a straightforward and consistent explanation in the context of an accretion episode . in x - rays , we are essentially detecting v1647 ori along a direct line of sight to the stellar photosphere . our results therefore indicate that the absorption along this direct line of sight which likely includes at least part of a thick circumstellar disk that is tilted about 30 degrees from edge - on @xcite does not change significantly as a function of time , despite the evident changes in x - ray luminosity . the optical and near - infrared photons we observe , however , emerge from the near - photosphere environment of v1647 ori along two paths . one path , along our direct line of sight to the photosphere , produces heavily reddened and extincted light . the second path takes photons nearly perpendicular to our line of sight , through an evacuated polar cavity , where they then scatter into our line of sight @xcite . these photons are bluer and much less heavily extincted than the line - of - sight photons . when v1647 ori is in the quiescent state ( 1998 , late 20052007 ) , we see a faint , reddened , heavily extincted source because the contribution from scattered light is minimal . during the outburst state ( 20042005 ; 20082009 ) , we see a brighter , bluer source because the contribution to the continuum of scattered photons is large . though the plasma temperature of v1647 ori strongly correlates with the x - ray luminosity and hardness ratio ( fig . [ v1647orilumhrtemp ] ) , it is unclear whether we are observing a single - component plasma that increases or decreases in temperature and thus causes the observed changes in x - ray luminosity or if a second , lower - temperature plasma is also present and whose contribution to the total spectrum is overwhelmed by the hotter temperature plasma during outbursts . other sources , such as v710 tau @xcite and ex lupi @xcite have spectra that can be modeled as two - temperature plasmas , with one component fading as the star returns to quiescent levels . @xcite was able to model the xmm - newton observation of v1647 ori in 2004 april with a single - temperature plasma but also found a better fit using a two - component model . however , this observation of v1647 ori was at least twice as long as most of the chandra observations from 20042009 . in addition , xmm - newton is more sensitive to lower - energy photons than cxo / acis . therefore , it may be the case that v1647 ori does have two plasmas that contribute to the x - ray spectra but that acis was unable to consistently detect the lower - temperature component due to the inherit limitations of its design , the shorter observation times of the cxo observations , and the high degree of softer x - ray absorption by the intervening hydrogen column . in order to test whether cxo would be able to detect two distinct plasma temperatures in 5 ks and 20 ks observations , we simulated a two - component plasma ( using _ fakeit _ ) with a normalization ratio and plasma temperatures identical to those found by @xcite and suffering extinction by the same hydrogen column density . the simulated spectra were then convolved with the responses of the front - illuminated acis - i3 and back - illuminated acis - s3 ccds . the simulations incorporated appropriately - scaled normalization parameters so that the resultant count rates were comparable to those of the 2005 and 20082009 cxo observations . each of the simulated spectra was first fit with a single - component model and then with a two - component model . all of the 20 ks observations were found to be fit better using two - component models with each fit yielding values for plasma temperatures and normalization that were close to the original simulation parameters . f - tests also suggested a slight improvement in the model fits if the 75%/25% normalization ratio between the low / high components , as found by @xcite , was forced . when fit with a two - component model , the simulated 5 ks observation of the more sensitive acis - s3 ccd yielded a best fit ( using the 75%/25% normalization ratio ) that converged to a model with two nearly - identical temperature plasmas , i.e. , to a single - temperature plasma , and thus does not lead to a better fit . thus , it appears that cxo should have the sensitivity to detect a two - component plasma such as that found by @xcite if the observations are long enough and the lower - temperature plasma flux is strong enough compared to the higher - temperature plasma flux . the spectral models for each of the five chandra observations obtained in 20082009 consistently converge to fits with a single - temperature plasma that , to within the errors , is @xmath14 kev . as seen with the correlation between x - ray luminosity and the mean hardness ratio , median photon energy , and mean count rate , we also find that when the x - ray luminosity increases or decreases , so does the plasma temperature . this single - component model is consistent with the single - component model of the 2005 april 11 xmm - newton data @xcite . finally , the second ( 2008 ) eruption of v1647 ori has very similar spectral characteristics to that of the first ( 20032005 ) eruption . from tables [ tab-1 ] , [ tab-3 ] , and [ tab-2 ] , the five 20082009 observations of v1647 ori , which span a period of approximately seven months , show * mean hardness ratios that are consistently as hard ( to within the errors ) as the hardest value of any observation following the previous eruption ; * median photon energies at levels that are very similar to the greatest median photon energy of any observation from the previous eruption ; * mean count rates that are 2050 times higher than those observed during x - ray quiescence and that are typically 12 times the highest mean count rate of the 20032006 observations ; * plasma emission measures that are usually 12 times greater than any modeled from the 2003 - 2005 observations ; * and x - ray luminosities that are consistently greater than 1.5 @xmath2 10@xmath27 ergs s@xmath15 , 12 times the x - ray luminosity of v1647 ori during any of the 2003 - 2005 cxo observations . given that the spectral characteristics of v1647 ori are so similar in observations in which the x - ray flux is elevated above the x - ray quiescent level , it appears that the same x - ray generation mechanism was at work during both eruptions and that the plasma characteristics were very similar during the two eruption epochs . the derived x - ray luminosities suggest that the second eruption was more energetic than the first , and given that the modeled emission measures are greater for the observations following the second eruption , it is reasonable to conclude that we observed the same phenomenon in both eruptions but that a larger mass of x - ray emitting plasma was active during the outburst that began in 2008 . our x - ray monitoring demonstrates that the two optical / near - ir outbursts undergone by the enigmatic v1647 ori in less than a decade were accompanied by strong x - ray outbursts . during these two outbursts ( 20032005 and 2008present ) , we see that the x - ray flux rose to peak luminosity over a span of a few weeks and then remained elevated for approximately two years during the first eruption and for at least one year during the second eruption . given that there is very strong evidence that the outbursts observed in the optical and near - infrared regimes are driven by accretion , we conclude that the correlated outbursts in x - rays are also driven by accretion . we find that all of the cxo spectra of v1647 ori are best modeled with a single , moderate - temperature ( 26 kev ) plasma . in almost all cases , the plasma temperature that emerges from models of the cxo spectra is too high to be generated via accretion hotspots on the stellar photosphere but is reasonable for a plasma generated via magnetic reconnection events . however , the x - ray - emitting plasma could also be located in a strongly enhanced stellar corona , or at the inner edge of the circumstellar disk . given that accretion is ongoing , lower - temperature plasma generated by shocks at the accretion footprint is very likely present ; however , during these cxo observations , any such soft component contributed much less flux than the moderate - temperature plasma and so usually did not leave a distinct signature in the x - ray spectra . with the elevated hard x - ray flux lasting the duration of the 20082009 epochs , we conclude that the x - ray flux is not the result of typical coronal flares generated via reconnection events . we believe that since the optical / near - infrared flux remains elevated throughout this observation epoch , we are instead observing the x - rays generated from reconnection events in the accretion stream , with the softer x - ray flux possibly being generated by accretion hotspots at the stellar photosphere . we find no significant change in x - ray absorbing column , indicating that varying optical / ir color measurements , which have previously been interpreted as evidence for variable reddening toward v1647 ori , may instead be indicative of varying contributions from scattered vs. direct photospheric emission from the yso . two of the spectra obtained during the most recent ( 2008 ) outburst appear to show the 6.4 kev neutral iron feature indicating fluorescence from cold ( presumably circumstellar disk ) gas surrounding v1647 ori . with v1647 ori being observed intensely during both outbursts at x - ray , optical , and infrared wavelengths , this objects stands as one of the best characterized systems that exhibits such a close correspondence between x - ray output and accretion rate . as a result of intense monitoring at x - ray , optical , and infrared wavelengths during two successive accretion - driven outbursts , v1647 ori stands as the best characterized yso in terms of the correspondence between x - ray output and accretion rate . we have shown , furthermore , that this yso exhibited strikingly similar x - ray behavior and spectral properties during its recent accretion bursts . these results underscore the need for x - ray monitoring of additional eruptive ysos , so as to evaluate whether the remarkable consistency of v1647 ori is the exception or the norm . we thank nuria calvet for providing early access to the data in cxo obsids 10763 and 8585 . this research was supported via awards numbers go8 - 9016x and go9 - 0006x to vanderbilt university issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 03060 . brahm , p. , kspl , a. , csizmadia , s. , mor , a. , kun , m. , & stringfellow , g. 2004 , , 419 , l39 acosta - pulido , j. a. , et al . 2007 , , 133 , 2020 anthony - twarog , b.j . the h - beta distance scale for b stars - the orion association , , 87 , 1213 - 1222 ( 1982 ) aspin , c. , beck , t. l. , & reipurth , b. 2008 , , 135 , 423 aspin , c. , et al . 2009 , , 692 , l67 audard , m. , briggs , k. r. , grosso , n. , gdel , m. , scelsi , l. , bouvier , j. , & telleschi , a. 2007 , , 468 , 379 audard , m. , et al . 2010 , , 511 , a63 becklin , e. e. , matthews , k. , neugebauer , g. , & willner , s. p. 1978 , , 220 , 831 briceo , c. , et al . 2004 , , 606 , l123 brickhouse , n. s. , cranmer , s. r. , dupree , a. k. , luna , g. j. m. , & wolk , s. 2010 , , 710 , 1835 brittain , s. , rettig , t. , simon , t. , gibb , e. , & liskowsky , j. 2010 , , 708 , 109 fedele , d. , van den ancker , m. e. , petr - gotzens , m. g. , ageorges , n. , & rafanelli , p. 2007 , , 472 , 199 grosso , n. , kastner , j. h. , ozawa , h. , richmond , m. , simon , t. , weintraub , d. a. , hamaguchi , k. , & frank , a. 2005 , , 438 , 159 grosso , n. 2006 , in esa special publication , vol . 604 , the x - ray universe 2005 , ed . a. wilson , 51 - 56 grosso , n. , hamaguchi , k. , kastner , j. h. , richmond , m. , & weintraub , d. a. 2010 , , 522 , a56 gdel , m. & naz , y. 2009 , , 17 , 309 hamaguchi , k. , grosso , n. , kastner , j. h. , weintraub , d. a. , & richmond , m. 2010 , , 714 , l16 hartmann , l. , & kenyon , s. j. 1996 , , 34 , 207 hartmann , l. 1998 accretion processes in star formation ( cambridge , cambridge university press ) herbig , g. h. 1977 , , 217 , 693 herbig , g. h. , aspin , c. , gilmore , a. c. , imhoff , c. l. , & jones , a. f. 2001 , , 113 , 1547 isobe , h. , shibata , k. , yokoyama , t. , & imanishi , k. 2003 , , 55 , 967 itagaki , k. , nakano , s. , & yamaoka , h. 2008 , , 8968 , 2 kastner , j. h. , huenemoerder , d. p. , schulz , n. s. , canizares , c. r. , & weintraub , d. a. 2002 , , 567 , 434 kastner , j. h. , et al . 2004 , , 430 , 429 kastner , j. h. , et al . 2006 , , 648 , l43 knigl , a. , romanova , m. m. , & lovelace , r. v. e. 2011 , arxiv:1106.2356 mcgehee , p. m. , smith , j. a. , hendon , a. a. , richmond , m. w. , knapp , g. r. , finkbeiner , d. p. , ivezi , . , & brinkmann , j. 2004 , , 616 , 1058 mcneil , j. w. , reipurth , b. , & meech , k. 2004 , , 8284 , 1 ojha , d. k. , et al . 2005 , , 57 , 203 ojha , d. k. , et al . 2006 , , 368 , 825 ojha , d. k. , ghosh , s. k. , kaurav , s. s. , bhatt , b. c. , sahu , d. k. , & tej , a. 2008 , , 9006 , 1 porquet , d. , dubau , j. , & grosso , n. 2010 , , 157 , 103 preibisch , t. , et al . 2005 , , 160 , 401 raassen , a. j. j. 2009 , , 505 , 755 reipurth , b. & aspin , c. 2004 , , 606 , l119 sakamoto , t. , wallace , c. a. , donato , d. , gehrels , n. , okajima , t. , & ukwatta , t. n. 2011 , adv . space res . , 47 , 1444 semkov , e. 2004 , inf . bull . variable stars 5578 , 1 semkov , e. 2006 , inf . bull . variable stars 5683 , 1 shibata , k. and yokoyama , t. 2002 , , 577 , 422 shu , f. h. , najita , j. , ostriker , e. , wilkin , f. , ruden , s. , & lizano , s. 1994 , , 429 , 781 shukla , s. j. , weintraub , d. a. , & kastner , j. h. 2008 , , 683 , 893 simon , t. , andrews , s. m. , rayner , j. t. , & drake , s. a. 2004 , , 611 , 940 skinner , s. , briggs , k. , & gdel , m. 2006 , , 643 , 995 skinner , s. , sokal , k. , gdel , m. , & briggs , k. 2009 , , 696 , 766 stassun , k. g. , van den berg , m. , feigelson , e. , & flaccomio , e. 2006 , , 649 , 914 stelzer , b. & schmitt , j. h. 2004 , , 418 , 687 stelzer , b. , hubrig , s. , orlando , s. , micela , g. , mikulek , z. , & schller , m. 2009 , , 499 , 529 tsujimoto , m. , feigelson , e. d. , grosso , n. , micela , g. , tsuboi , y. , favata , f. , shang , h. , & kastner , j. h. 2005 , , 160 , 503 vacca , w. d. , cushing , m. c. , & simon , t. 2004 , , 609 , l29 venkat , v. & anandarao , b. g. 2006 , , 8694 , 2 venkat , v. & anandarao , b. g. 2009 , cbet , 2104 venkat , v. & anandarao , b. g. 2011 , cbet , 2647 vuong , m. h. , montmerle , t. , grosso , n. , feigelson , e. d. , verstraete , l. , & ozawa , h. 2003 , , 408 , 581 zhu , z. , hartmann , l. , gammie , c. , & mckinney , j. c. 2009 , , 701 , 620 llcccccccc 2539 & 2002 nov 14 & 2452593 & s2 & 62.8 & 17 & 0.3 @xmath13 0.1 & 2.0 @xmath13 0.3 & @xmath120.34 @xmath13 0.31 & @xmath120.28 @xmath13 0.28 + 5307 & 2004 mar 07 & 2453072 & s3 & 5.5 & 64 & 11.7 @xmath13 1.5 & 3.6 @xmath13 0.3 & 0.46 @xmath13 0.15 & 0.45 @xmath13 0.14 + 5308 & 2004 mar 22 & 2453087 & s3 & 4.9 & 12 & 2.5 @xmath13 0.8 & 2.0 @xmath13 0.5 & @xmath120.45 @xmath13 0.33 & @xmath120.46 @xmath13 0.33 + 5382 & 2005 apr 11 & 2453472 & i3 & 18.2 & 86 & 4.8 @xmath13 0.5 & 3.5 @xmath13 0.1 & 0.51 @xmath13 0.19 & 0.46 @xmath13 0.12 + 5383 & 2005 aug 27 & 2453610 & i3 & 19.9 & 18 & 0.9 @xmath13 0.2 & 3.0 @xmath13 0.8 & 0.01 @xmath13 0.29 & 0.34 @xmath13 0.25 + 5384 & 2005 dec 09 & 2453714 & i3 & 19.7 & 2 & 0.1 @xmath13 0.1 & 2.2 @xmath13 1.1 & 0.0 @xmath13 0.20 & @xmath120.07 @xmath13 0.92 + 6413 & 2005 dec 14 & 2453719 & i3 & 18.1 & 4 & 0.2 @xmath13 0.1 & 1.7 @xmath13 0.3 & @xmath120.36 @xmath13 0.26 & @xmath121.0 @xmath13 0.75 + 6414 & 2006 may 01 & 2453857 & i3 & 21.6 & 3 & 0.1 @xmath13 0.1 & 1.3 @xmath13 0.3 & 0.09 @xmath13 0.23 & 0.07 @xmath13 0.92 + 6415 & 2006 aug 07 & 2453955 & i3 & 20.5 & 4 & 0.2 @xmath13 0.1 & 2.4 @xmath13 0.4 & @xmath120.27 @xmath13 0.21 & @xmath121.0 @xmath13 0.73 + 9915 & 2008 sep 18 & 2454728 & i3 & 19.9 & 466 & 23.6 @xmath13 1.1 & 3.5 @xmath13 0.1 & 0.36 @xmath13 0.07 & 0.37 @xmath13 0.05 + 10763 & 2008 nov 27 & 2454798 & i2 & 19.7 & 217 & 11.0 @xmath13 0.7 & 3.9 @xmath13 0.1 & 0.65 @xmath13 0.09 & 0.64 @xmath13 0.08 + 8585 & 2008 nov 28 & 2454799 & i2 & 28.5 & 160 & 5.8 @xmath13 0.4 & 3.4 @xmath13 0.1 & 0.40 @xmath13 0.10 & 0.43 @xmath13 0.09 + 9916 & 2009 jan 23 & 2454855 & i3 & 18.4 & 245 & 13.6 @xmath13 0.9 & 3.7 @xmath13 0.2 & 0.41 @xmath13 0.08 & 0.43 @xmath13 0.07 + 9917 & 2009 apr 21 & 2454943 & i3 & 29.8 & 260 & 8.8 @xmath13 0.5 & 3.5 @xmath13 0.2 & 0.28 @xmath13 0.08 & 0.37 @xmath13 0.07 + 9915 & 2008 sep 18 & 0.85 & 77 & 4.1@xmath28 & 3.7@xmath29 & 15.0@xmath30 & 4.2@xmath31 & 8.1@xmath32 + 10763 & 2008 nov 27 & 1.24 & 39 & 4.1 ( fixed ) & 4.0 ( fixed ) & 6.1@xmath33 & 1.8@xmath34 & 3.5@xmath35 + 8585 & 2008 nov 28 & 0.71 & 26 & 5.9@xmath36 & 2.0@xmath37 & 8.2@xmath38 & 1.0@xmath39 & 1.8@xmath40 + 9916 & 2009 jan 23 & 0.79 & 42 & 4.1@xmath41 & 6.0@xmath42 & 6.8@xmath43 & 2.6@xmath44 & 5.0@xmath45 + 9917 & 2009 apr 21 & 0.88 & 44 & 4.1@xmath46 & 3.4@xmath47 & 5.9@xmath48 & 1.5@xmath49 & 2.9@xmath40 + lcccccccc 2539 & 2002 nov 14 & 1.37 & 13 & 0.9@xmath50 & 0.6@xmath51 & 0.03@xmath52 & 0.05@xmath53 + 5307 & 2004 mar 7 & 1.19 & 10 & 4.0 ( f ) & 5.4 @xmath13 1.3 & 1.6 @xmath13 0.5 & 3.1 @xmath13 0.9 + 5308 & 2004 mar 22 & 1.45 & 10 & 1.1@xmath54 & 3.9@xmath55 & 0.2 @xmath13 0.1 & 0.4 @xmath13 0.1 + 5382 & 2005 apr 11 & 0.98 & 14 & 3.4@xmath56 & 3.1@xmath57 & 0.8@xmath58 & 1.5@xmath59 + 5383 & 2005 aug 27 & 1.99 & 16 & 3.2@xmath60 & 0.8@xmath61 & 0.2@xmath62 & 0.4@xmath58 + 5384 & 2005 dec 9 & 1.67 & 1 & 0.86 ( f ) & 0.2@xmath63 & 0.01 @xmath13 0.01 & 0.01 @xmath13 0.01 + 6413 & 2005 dec 14 & 0.87 & 3 & 0.86 ( f ) & 0.9@xmath64 & 0.02 @xmath13 0.01 & 0.05 @xmath13 0.03 + 6414 & 2006 may 1 & 2.83 & 2 & 0.86 ( f ) & 0.3@xmath65 & 0.01 @xmath13 0.01 & 0.02 @xmath13 0.02 + 6415 & 2006 aug 7 & 1.45 & 3 & 0.86 ( f ) & 0.9@xmath65 & 0.01 @xmath13 0.01 & 0.01 @xmath13 0.01 +

we consider kerr spacetimes with parameters and such that , kerr - newman spacetimes with parameters , , and more generally , stationary axisymmetric black hole exterior spacetimes which are sufficiently close to a schwarzschild metric with parameter , with appropriate geometric assumptions on the plane spanned by the killing fields . we show uniform boundedness on the exterior for sufficiently regular solutions to the wave equation , i.e. we show that solutions arising from smooth initial data prescribed on an arbitrary cauchy surface satisfy in the domain of outer communications . in particular , the bound holds up to and including the event horizon . here , is a norm on initial data and depends only on the parameters of the nearby schwarzschild metric . no unphysical restrictions are imposed on the behaviour of near the bifurcation surface of the event horizon . the norm is finite if , and is well - behaved at spatial infinity , in particular , it is sufficient to assume is supported away from spatial infinity . the pointwise estimate derives in fact from the uniform boundedness of a positive definite energy flux . note that in view of the very general assumptions , the separability properties of the wave equation on the kerr background are not used .

0805.4309

the kerr family , discovered in 1963 @xcite , comprises perhaps the most important family of exact solutions to the einstein vacuum equations @xmath18 the governing equations of general relativity . for parameter values @xmath19 ( here @xmath1 denotes the mass and @xmath0 angular momentum per unit mass ) , the kerr solutions represent black hole spacetimes : i.e. asymptotically flat spacetimes which possess a region which can not communicate with future null infinity . the celebrated schwarzschild family sits as the one - parameter subfamily of kerr corresponding to @xmath20 . much of current theoretical astrophysics is based on the hypothesis that isolated systems described by kerr metrics are ubiquitous in the observable universe . despite the centrality of the kerr family to the general relativistic world picture , the most basic questions about the behaviour of linear waves on kerr backgrounds have remained to this day unanswered . this behaviour is in turn intimately connected to the stability properties of the kerr metrics themselves as solutions of @xmath21 , and thus , with the very physical tenability of the notion of black hole . in particular , even the question of the uniform boundedness ( pointwise , or in energy ) of solutions @xmath7 to the linear wave equation @xmath22 in the domain of outer communications has not been previously resolved , except for the schwarzschild subfamily . the main theorems of this paper give the resolution of the boundedness problem for @xmath23 , for the case @xmath2 . solutions to @xmath23 arising from regular initial data remain uniformly bounded in the domain of outer communications . the bound is quantitative , i.e. it is computable in terms of the initial supremum and initial energy - type quantities on initial data . in fact , the results of this paper apply to a much more general setting than the specific kerr metric : boundedness is proven for solutions of @xmath23 on the exterior region of any stationary axisymmetric spacetime sufficiently close to a schwarzschild spacetime with mass @xmath5 . thus , the methods may be of relevance in the ultimate goal of this analysis : understanding the dynamics of the einstein equations @xmath21 in a neighborhood of a kerr metric . we first give a statement of the main results for the special case of kerr and the related kerr - newman family ( this is a family of solutions to the coupled einstein - maxwell system ) . we refer the reader to @xcite . let @xmath4 denote the kerr solution with parameters @xmath24 or more generally the kerr - newman solution with parameters @xmath25 , with @xmath26 and let @xmath27 denote the closure of a domain of outer communications . ( the parameter @xmath28 is known as the charge . ) let @xmath9 be a cauchy hypersurface in @xmath4 crossing the event horizon to the future of the sphere of bifurcation , and such that @xmath29 coincides with a constant-@xmath30 hypersurface , for large @xmath31 , where @xmath30 and @xmath31 denote here the standard boyer - lindquist coordinates on @xmath32 . recall that in such coordinates , the stationary killing field @xmath33 is given by @xmath34 . the kerr - newman solutions are moreover axisymmetric . the penrose diagram , say along the axis of symmetry ( where the axisymmetric killing field vanishes ) , is depicted below : @xmath35 note that @xmath29 is a past cauchy hypersurface for @xmath36 . denotes causal future , not to be confused with currents @xmath37 to be defined later . ] we have that @xmath38 is foliated by @xmath39 for @xmath40 , where @xmath39 is the future translation of @xmath41 by the flow generated by the stationary killing field @xmath42 for time @xmath43 . let @xmath44 denote the unit future normal of @xmath39 . let @xmath45 denote a translation invariant null generator for @xmath46 , and give @xmath47 the induced volume from @xmath48 and @xmath49 . let @xmath50 denote the standard energy momentum tensor associated to a solution @xmath7 of the wave equation @xmath23 @xmath51 define @xmath52 by @xmath53 and @xmath54 by @xmath55 note that the former current is positive definite when contracted with a future - timelike vector field , but is not conserved , whereas the latter current is conserved , but not positive definite when so contracted . [ prwto ] let @xmath4 , @xmath27 , @xmath39 be as above . there exists a universal positive constant @xmath56 , and a constant @xmath12 depending on @xmath1 and the choice of @xmath57 such that if @xmath58 then the following statement holds . let @xmath7 be a solution of @xmath23 on @xmath4 such that @xmath59 . then @xmath60 @xmath61 @xmath62 here the integrals are with respect to the induced volume forms . the integral on the left hand side of @xmath63 can be defined via a limiting procedure . [ deutero ] under the assumptions of the previous theorem , the following holds . let @xmath7 be a solution of the wave equation @xmath23 on @xmath4 such that @xmath64 then @xmath65 in @xmath66 . the hypothesis of theorem [ prwto ] can be re - expressed as the statement that local energy as measured by a local observer be finite , i.e. that @xmath67 , @xmath68 be in @xmath69 , together with the global assumption that @xmath70 the latter in turn is certainly satisfied if @xmath71 vanishes in a neighborhood of @xmath72 . similarly , the hypothesis of theorem [ deutero ] is satisfied for @xmath73 , @xmath68 in @xmath74 , if @xmath71 vanishes in a neighborhood of @xmath72 . finally , note that given an arbitrary cauchy surface @xmath75 for kerr , sufficiently well behaved at @xmath72 , it follows that the right hand side of @xmath76 is bounded by @xmath77 thus the above regularity assumptions could be imposed on an arbitrary cauchy surface . there are no unphysical restrictions on the support of the solution in a neighborhood of @xmath78 . the results of theorems [ prwto ] and [ deutero ] remain true when the kerr or kerr - newman metric is replaced by an arbitrary stationary axisymmetric black hole exterior metric suitably close to schwarzschild , and with suitable assumptions on the geometry of the killing fields . in particular , in addition to smallness , it is required that as in the kerr solution the null generator of the horizon is in the span of the killing fields . the precise assumptions are outlined in section [ ta3n ] . the elusiveness of the results of theorems [ prwto ] and [ deutero ] stems from the well - known phenomenon of _ this is related to the fact that the killing field @xmath33 ( with respect to which the kerr solution is stationary ) is not everywhere - timelike in the domain of outer communications . in particular , there is a region of spacetime where @xmath33 is spacelike , the so - called _ ergoregion_. the boundary of this region is called the _ ergosphere_. the presence of the ergoregion means that the energy current @xmath79 is not positive definite when integrated over spacelike hypersurfaces . thus , the conservation of @xmath79 does not imply _ a priori _ bounds on an @xmath80-based quantity . in particular , the local energy of the solution can be greater than the initial total energy , even if the energy is initially supported where @xmath79 is positive definite . a test - particle version of this fact , where a particle coming in from infinity splits into one of negative energy entering the black hole and one of greater positive energy returning to infinity , is known as the _ penrose process_. the pioneering study by christodoulou @xcite of the `` black hole transformations '' obtainable via a penrose process led to a subject known as `` black hole thermodynamics '' . in the physics literature , where discussion of these issues is inextricably linked to the separability of @xmath23 and decomposition of @xmath7 into modes , the problem of the ergoregion appears as a formidable and perhaps intractable obstacle . it turns out , however , that there are other physical mechanisms at play which have an important role but are not necessarily well reflected from the point of view of separability . in particular , the tendency of waves to eventually disperse ( true in any asymptotically flat spacetime ) coupled with the powerful red - shift effect at the horizon . indeed , these properties , which depend only loosely on the stationarity , tend to make solutions not only stay bounded but decay to a constant in time , even if the local energy increases for a short time . unfortunately , the dispersive properties of waves on black hole backgrounds are severely complicated by the presence of trapped null geodesics . ( the presence of these can easily be inferred by a continuity argument in view of the fact that there exist both null geodesics crossing the horizon and going to null infinity . ) it is only very recently that the role of trapping has been sufficiently well understood in the special case of the schwarzschild family to allow for the first proofs of decay for general solutions of @xmath23 on such backgrounds . see the results described in section [ prevsch ] . in the case of kerr , the techniques introduced for controlling trapping on schwarzschild can not be readily perturbed . this has to do with the fact that these techniques seem to exploit the special property that the trapping concentrates asymptotically on a set of codimension @xmath81 in physical space , the so - called photon sphere . in contrast , in kerr the codimensionality of the space of trapped geodesics can only be properly understood in phase space . this indicates that controlling trapping requires a far more delicate analysis . it would appear from the above that the problem of superradiance could in principle be overcome , but at the expense of a very delicate analysis of trapping . a closer look , however , reveals that the situtation is considerably more favourable . at a heuristic level , the reason for this is the following remark : if one could separate out the `` superradiant '' part of the solution from the `` non - superradiant '' part , then one only has to exploit dispersion for the superradiant part . this latter problem turns out to be much easier than understanding dispersion for the whole solution . to decompose the solution , we must first cut off the solution @xmath7 in the `` time''-interval of interest to obtain @xmath82 and then decompose into two pieces @xmath83 where @xmath84 is to be supported in frequency space ( real frequencies @xmath85 and integer @xmath86 here defined with respect to coordinates @xmath30 and @xmath87 ) only in the range @xmath88 , whereas @xmath89 is to be supported in frequency space only in the range @xmath90 . for spacetimes sufficiently close to schwarzschild , for a suitable choice of the parameter @xmath91 , one can view @xmath89 as essentially non - superradiant , and @xmath84 as the superradiant part . if one can show boundedness for @xmath89 and dispersion for @xmath84 , then one will have proven the uniform boundedness of the sum @xmath7 . for spacetimes sufficiently close to schwarzschild , one can choose @xmath91 sufficiently small so that trapping essentially does not occur for @xmath84 , and the dispersive mechanism of schwarzschild is stable . this relies on the stability of the red - shift effect for considerations close to the horizon . in complete contrast to the standard picture , it is the superradiant part of the solution which would be the better behaved one . in practice , the analysis is of course not as simple as what has been portrayed above , and here again , the stabilising effect of the red - shift acting near the horizon plays an important role . in view of the cutoffs in time , the equations satisfied by @xmath92 and @xmath89 are coupled . moreover , the statement that @xmath89 is non - superradiant while @xmath84 is dispersive must also be understood modulo error terms . it turns out that to control these error terms , one of necessity must have at their disposal an energy quantity which does not degenerate on the horizon , that is to say , the @xmath80-based quantity for which one shows uniform boundedness must be the one of theorem [ prwto ] , and not a quantity analogous to @xmath79 in schwarzschild . in particular , one must understand the red - shift mechanism even for the `` non - superradiant '' part @xmath89 , for which one does not understand dispersion . such stable estimates at the horizon ( corresponding to the energy measured by local observers ) exploiting the red - shift effect were first attained for schwarzschild in our previous @xcite . it is interesting to note , however , that in @xcite , understanding of the red - shift mechanism was always coupled with understanding dispersion , i.e. controlling the trapping phenomenon . in particular , one had to appeal to an understanding of dispersion even to obtain the result of theorems [ prwto ] for schwarzschild . in this paper , we show how understanding the red - shift can be decoupled from understanding dispersion in the non - superradiant case . in addition , we show that the red - shift effect allows us to commute the wave equation with a vector field transverse to the horizon , yielding a new route to higher order estimates and pointwise estimates . an extra side - benefit of our results is thus a new , simpler and more robust proof of theorems [ prwto ] and [ deutero ] even for the case of schwarzschild . see section [ further1 ] . we review in detail previous work on this and related problems . results of the type of theorems [ prwto ] and [ deutero ] for static perturbations of minkowski space pose little difficulty . ( indeed , the analogue of theorem [ prwto ] is immediate , and theorem [ deutero ] can be proven with the help of sobolev inequalities after commuting the equation with the static killing field . ) thus , we shall pass directly to the black hole case . the analogue of theorem [ deutero ] for schwarzschild is a celebrated result of kay and wald @xcite , building on previous work of wald @xcite where the theorem had been proven for the restricted class of data whose support was assumed not to contain the bifurcation sphere @xmath93 . in view of the positive definiteness of @xmath79 in the domain of outer communications , the only essential difficulty is obtaining bounds for @xmath7 up to the horizon ( where @xmath33 becomes null ) , as bounds away from the horizon can be obtained essentially as described immediately above for static perturbations of minkowski space . the arguments of kay and wald to prove the analogue of theorem [ deutero ] relied on the staticity to realize a solution @xmath7 as @xmath94 where @xmath95 is again a solution of @xmath23 constructed by inverting an elliptic operator acting on initial data . in addition , a pretty geometric construction exploiting the discrete symmetries of maximal schwarzschild was used to remove the unphysical restriction on the support near @xmath93 necessary for constructing @xmath96 in the original @xcite . unfortunately , neither of these methods is particularly robust to perturbation . the reason the authors had to resort to such techniques was that theorem [ deutero ] was proven _ without _ proving the analogue of theorem [ prwto ] , rather , using only the conserved flux @xmath79 whose control degenerates as @xmath46 is approached . theorem [ prwto ] for schwarzschild was only proven as part of the decay results of @xcite to be discussed below . turning now to the issue of decay , the first non - quantitative decay result for @xmath23 on schwarzschild is contained in the thesis of twainy @xcite . the first quantitative decay results for solutions of @xmath23 on schwarzschild ( and more generally , reissner - nordstrm ) were proven in @xcite , but were restricted to spherically symmetric solutions , or alternatively , the @xmath97th spherical harmonic @xmath98 of a general solution @xmath7 . ( in fact , this was a byproduct of the main result of @xcite , which concerns decay rates for spherically symmetric solutions to the coupled einstein-(maxwell)-scalar field system . ) quantitative decay results for the whole solution @xmath7 , both pointwise and in energy , were proven in @xcite , in particular , the uniform decay result @xmath99 is an eddington - finkelstein advanced time coordinate and @xmath100 is an appropriate quantity computable on initial data , and @xmath101 denotes say @xmath102 . inequality @xmath103 is sharp as a uniform decay rate in @xmath104 . the results of @xcite exploit both the red - shift effect near the horizon and the dispersive properties . the estimates are derived using a variety of vector field multipliers , in particular , a vector field multiplier @xmath105 such that the flux of @xmath106 gives the local energy at the horizon . the energy identity of @xmath105 quantifies the red - shift effect . weaker decay results were proven independently by blue and sterbenz @xcite for initial data vanishing on @xmath93 , but with control which degenerates on the horizon . in particular , the estimates of @xcite are unstable to perturbation near the horizon . the stability of the estimates of @xcite near the horizon will be of critical importance here . both @xcite and @xcite control trapping effects with the help of vector field multipliers which must be carefully chosen for each spherical harmonic separately . these were inspired by a series of papers by soffer and collaborators , for instance @xcite ; see , however @xcite . the first proof of decay for @xmath7 not relying on spherical harmonic decomposition for the construction of these multipliers is provided by our more recent @xcite . since uniform boundedness is the most basic question which can be asked about @xmath23 on kerr , previous results in this setting are of necessity of a partial nature . in particular , essentially all previous work on @xmath23 is restricted to the projection of @xmath7 to a single azimuthal frequency , or equivalently , to the case where the data are of the form @xmath107 solutions arising from @xmath108 are then of the form @xmath109 . let us call such solutions azimuthal modes . in principle , one could attempt to deduce properties of general @xmath7 by summing relations deduced for each individual azimuthal mode . as we shall see , however , due to the non - quantitative nature of the results described below , in of themselves they unfortunately yield no information about general @xmath7 . nonetheless , even the study of such @xmath110 without regard to uniform control in @xmath86 turns out to be a non - trivial problem . indeed , even for such individual azimuthal modes , the most basic questions had not been previously answered , in particular , the analogue of theorems [ prwto ] or [ deutero ] . this being said , there are interesting partial results concerning @xmath108 that had been previously obtained . in particular , most recently finster _ et al . _ @xcite had been able to show for smooth @xmath110 that for fixed @xmath111 and @xmath112 , @xmath113 under the assumption that the support of @xmath110 does not contain @xmath114 . see however @xcite . in particular , one can deduce @xmath115 for each fixed @xmath111 and @xmath112 , without however a bound on the @xmath116 . the results rest on an explicit integral representation of the solution which is derived using the remarkable ( but all too fragile ) separability properties of the kerr metric . the arguments contain many pretty applications of contour integral methods of classical complex analysis . since these techniques are essentially algebraic , no restriction on the size of @xmath117 need be made provided @xmath118 . in @xcite , under the same assumption on the initial support , the authors deduce that for each @xmath119 , @xmath120 thus , the energy of each mode in the region @xmath121 remains finite but again , no quantitative bound in terms of data can be produced . moreover , from the results of @xcite , one can not deduce that the @xmath116 of @xmath122 and @xmath123 commute with taking @xmath124 or @xmath125 , i.e. @xmath122 is _ a priori _ compatible with @xmath110 blowing up along the horizon : @xmath126 and @xmath123 is compatible with infinite energy concentration near the horizon : @xmath127 as explained before , no statement could be inferred for the general solution @xmath7 from the above statements on individual azimuthal modes , not even a weak statement like @xmath122 or @xmath123 . this is because the @xmath128 and @xmath116 of @xmath129 , @xmath122 and @xmath123 do not _ a priori _ commute with summation over @xmath86 . must be understood as the projection @xmath110 , to agree with what is proven in the body of the paper . ] of course , in view of theorems [ prwto ] and [ deutero ] , one can now infer from @xmath129 corollary [ nqc ] of section [ nqd ] . the somewhat unsatisfying nature of the above previous results deduced with the help of separability are indicative of how difficult it is to obtain quantitative statements about solutions of the wave equation @xmath23 even in the algebraically special case where one has explicit representations of the solution . perhaps this is for the best , however . remarkable though they are , the separability properties of the kerr metric are unstable to perturbation . just as in the case of stability of minkowski space @xcite , understanding the stability properties of the einstein equations near the kerr solution will undoubtedly require robust methods . we hope that the techniques employed here will have further applications in this direction . a related problem to the wave equation is that of the klein - gordon equation @xmath130 with @xmath131 . there is a well - developed scattering theory on schwarzschild for the class of solutions of @xmath132 with finite energy associated to the killing @xmath33 . in particular , an asymptotic completeness statement has been proven in @xcite . this analysis in of itself , however , when specialised to @xmath74 solutions in the geometric sense , only gives very weak information about the solution . in particular , it does not give @xmath80 control of @xmath7 or its angular derivatives on @xmath46 . in the case of kerr , there are again certain partial results for @xmath132 in the direction of scattering for a `` non - superradiant '' subspace of initial data @xcite . these interesting results do not , however , address the characteristic difficulties of superradiance . see also @xcite . finally , we mention that there has been a series of interesting papers concerning the dirac equation on kerr and kerr - newman . see @xcite . for dirac , considerations turn out to be much easier as this equation does not exhibit the phenomenon of superradiance . we shall not comment more about this here but refer the reader to the nice article @xcite . we can not do justice here to the vast work on this subject in the physics literature . see @xcite for a nice survey . the authors thank g. holzegel for comments on the manuscript . the authors also thank the niels bohr institute , copenhagen , for hospitality while this work was being completed . m. d. is support in part by a grant from the european research council . i. r. is support in part by nsf grant dms-0702270 . constants will play an important role in this paper and it is imperative to set the conventions early . in the next section we shall fix a schwarzschild metric with parameter @xmath1 . we shall use the notation @xmath133 and @xmath134 for general positive constants which only depend on the choice of @xmath1 . an inequality true with a constant @xmath133 will be true if @xmath133 is replaced by a larger constant , and similary , for @xmath134 if @xmath134 is replaced by a smaller positive constant . this convention is obviously violated and has to be reinterpreted appropriately . ] we shall use the notation @xmath135 to denote @xmath136 since @xmath133 and @xmath134 denote general constants , we shall apply without comment the obvious algebraic rules @xmath137 , @xmath138 , @xmath139 , etc . we will also require various particular parameters which can be chosen depending only on @xmath1 : @xmath140 the above parameters are not explicitly computed but are determined implicitly by various constraints . before choosing a parameter , say @xmath141 , we shall use notation like @xmath142 , @xmath143 to denote constants depending only on @xmath1 and the as of yet unchosen @xmath141 . it is to be understood that again here , the notation @xmath133 indicates that the constant can always be replaced by a bigger one , and @xmath134 by a smaller one . we shall also use the notation @xmath144 to indicate that the parameter @xmath145 depends on the still unchosen @xmath141 . once @xmath141 is determined , we may replace the expressions @xmath142 , @xmath146 etc . , with @xmath133 , @xmath147 , etc . in this section we shall describe the general class of metrics for which our results will apply . to set the stage , we must first fix some structures associated to a schwarzschild metric . we refer the reader to our previous @xcite for a review of the geometry of schwarzschild . we must first fix a certain subregion of schwarzschild with parameter @xmath5 , relevant coordinates , and a choice of axisymmetric killing field . this will provide the underlying manifold with stratified boundary for the class of metrics to be considered later . let us use the notation @xmath148 to denote the schwarzschild metric . refer to the diagram below : @xmath149 we will denote by @xmath27 the closure of a domain of outer communications in maximal schwarzschild . we have @xmath150 where @xmath46 denotes the _ future event horizon _ and @xmath151 the _ past event horizon_. the intersection @xmath152 is known as the _ bifurcation sphere_. recall the static killing field @xmath33 , timelike on @xmath32 and null on @xmath153 . flow by integral curves of @xmath33 defines a one - parameter family of diffeomorphisms @xmath154 . recall now the area - radius function @xmath31 . on the horizons @xmath153 we have @xmath155 . we will use the notation @xmath156 for the function defined by @xmath157 . associated to schwarzschild will be the constants @xmath158 determined in section [ ysec ] . we may assume say that @xmath159 let @xmath160 be a cutoff function such that @xmath161 for @xmath162 and @xmath163 for @xmath164 . define the hypersurface @xmath165 by @xmath166 this can be extended beyond @xmath46not by the expression @xmath167 , however to a spacelike hypersurface in maximal schwarzschild . let us actually define @xmath165 to include its limit points on the horizon @xmath46 . note of course that in view of the support of @xmath160 , it follows that in the region @xmath164 , @xmath165 coincides with the constant @xmath168 hypersurface . we may define a new coordinate @xmath169 this coordinate is regular on @xmath170 . we have that @xmath171 let us define @xmath172 clearly @xmath173 . we have that @xmath174 for some constants @xmath133 , @xmath134 . recall here the conventions of section [ constants ! ] . for technical reasons , we shall require two auxilliary sets of spacelike hypersurfaces . let @xmath175 be a cutoff function such that @xmath176 for @xmath177 and @xmath178 for @xmath179 . let us define @xmath180 and @xmath181 for an @xmath147 to be determined later with @xmath182 . let us define @xmath183 independently of the choice of @xmath147 , we have that @xmath184 are spacelike , in fact @xmath185 in what follows we shall restrict to @xmath186 the set @xmath187 is again a manifold with stratified boundary ( as was @xmath27 ) , where the boundary is given by @xmath188 . choosing a coordinate atlas consisting of two charts @xmath189 , @xmath190 on @xmath191 , then setting @xmath192 , @xmath193 , it follows that @xmath194 form a coordinate atlas for @xmath187 . we can ensure moreover that the regions of the sphere covered by the charts are restricted so that the metric functions satisfy @xmath195 in these coordinates . note that with respect to both these charts , the vector @xmath196 is the stationary killing field @xmath33 . we will use the above coordinate atlas @xmath197 in formulating our closeness assumptions . a third set of coordinates will be useful for us , namely the coordinates arising from a choice of standard spherical coordinates denotes an azimuthal coordinate . ] @xmath198 on @xmath191 . with respect to @xmath199 coordinates , it follows that @xmath200 is a killing field . let us denote the smooth extension to @xmath27 of this killing field as @xmath201 . note that @xmath201 vanishes precisely at two points on each sphere of symmetry . this corresponds to the locus of points where the @xmath202 coordinates break down . because @xmath203 is not satisfied with respect to these coordinates , they will not be as useful in formulating the closeness assumptions . we will say that schwarzschild is axisymmetric and @xmath201 is a choice of axisymmetric killing field . finally , we shall also at times refer to so - called regge - wheeler coordinates @xmath204 here @xmath30 is the standard schwarzschild time and the coordinate @xmath205 is defined by @xmath206 note that this coordinate is regular in @xmath207 , but sends the boundary to @xmath208 . with respect to these coordinates , we note that @xmath209 extends to a smooth vector field on all of @xmath187 ( i.e. to the event horizon ) , and in fact , in the limit @xmath210 on @xmath211 . this last coordinate system is not useful for formulating closeness assumptions in view of the fact that it breaks down on the horizon . we shall only use regge - wheeler coordinates for making calculations with respect to the schwarzschild metric . finally , a word of caution . since we have several coordinate systems which will be considered , coordinate vectors like @xmath212 will always be referred to in conjunction with a specific coordinate system . we now describe the class of metrics to be allowed . we consider the manifold with stratified boundary @xmath187 defined above . we consider the class of all smooth lorentzian metrics @xmath48 such that : 1 . for @xmath213 sufficiently small , @xmath214 latexmath:[\[\label{locsmal2 } with respect to the atlas @xmath197 . referred to previously . for the relation to kerr - newman , see section [ knse ] . ] 2 . the vector fields @xmath216 and @xmath217 with respect to @xmath218 coordinates are again killing with respect to @xmath48 . 3 . [ lastassump ] there is a function @xmath219 defined on @xmath46 such that @xmath220 is null on the horizon , and @xmath221 above implies that @xmath46 is null with respect to @xmath48 and its null generator lies in the span of @xmath33 and @xmath201 . we may define the _ ergoregion _ to be the region where @xmath33 itself is _ not _ timelike . for sufficiently small @xmath222 , assumptions @xmath223 and @xmath224 imply that @xmath165 is spacelike with respect to @xmath48 , in fact , with our conventions on constants , @xmath225 similarly , we have from @xmath226 that for @xmath222 sufficiently small , @xmath227 and @xmath228 are spacelike , in fact @xmath229 independently of the choice of @xmath147 . note that @xmath230 is again isometric to @xmath165 with respect to @xmath48 , and similarly @xmath231 is isometric to @xmath232 . we will denote by @xmath44 the future normal of @xmath230 : @xmath233 this defines a translation invariant smooth timelike unit vector field on @xmath187 . similarly , we define @xmath234 we will use the notations @xmath235 @xmath236 @xmath237 @xmath238 all integrals without an explicit measure of integration are to be taken with respect to the volume form in the case of a region of spacetime or a spacelike hypersurface , and an induced volume form connected to the choice of a @xmath239-invariant tangential vector field @xmath240 , in the case of @xmath241 . note the following property of the volume integral with respect to the ( almost ) global @xmath242 coordinate system : there exist smooth @xmath243 , @xmath244 such that for all continuous @xmath245 : @xmath246 @xmath247 also let us note that @xmath248 by @xmath249 , it follows that if @xmath135 in the sense @xmath250 , it follows that @xmath251 a similar relation holds with @xmath252 and @xmath184 . let @xmath253 , @xmath254 . then the kerr - newman metric with parameters @xmath28 , @xmath0 satisfies the assumptions of section [ genclass ] . let us sketch how one can implicitly define a kerr - newman metric on @xmath187 in our @xmath255 coordinate system . for convenience , let us do this by defining a new set of coordinates on @xmath207 which are to represent boyer - lindquist coordinates @xmath256 . for this define @xmath257 by @xmath258 @xmath259 by @xmath260 where @xmath261 is defined by @xmath262 , and @xmath263 by @xmath264 where @xmath265 , and @xmath266 by @xmath267 now consider the metric on @xmath207 defined in these new coordinates by @xmath268 writing the metric in @xmath255 coordinates , and then relating this form in turn to the coordinates of @xmath197 one sees immediately that @xmath269 uniformly as @xmath270 , and @xmath271 uniformly as @xmath270 , where @xmath272 denote coordinates of @xmath197 . it follows that given @xmath222 , the assumptions @xmath223 and @xmath273 hold . the remaining assumptions are well - known properties of kerr which are manifest from the boyer - lindquist form . let @xmath274 , @xmath230 be as in section [ genclass ] , and let @xmath16 be an @xmath74 function on @xmath165 , and let @xmath275 be an @xmath69 function on @xmath165 . here the @xmath80 norm is defined naturally with respect to the induced riemannian metric on @xmath165 . by standard theory , there exists a unique solution @xmath7 of the initial value problem @xmath276 with the property that @xmath277 we will in fact require that @xmath278 by density arguments , the main results of this paper would follow if they were proven under the additional restriction that @xmath16 , @xmath275 are in fact smooth , and thus , that @xmath7 is smooth . moreover , we can safely assume that @xmath279 and @xmath275 are supported away from infinity . let us assume this in what follows so as not to have to comment on regularity issues or the _ a priori _ finiteness of certain quantities . it follows in particular from this assumption that @xmath280 moreover , that @xmath71 is supported away from spatial infinity . for a sufficiently regular function @xmath281 , let us define @xmath282 and for @xmath283 a vector field , @xmath284 in addition , let us define the quantity @xmath285 note that this is non - negative . moreover , in the coordinate charts of the atlas @xmath197 , we have @xmath286 by @xmath287 and @xmath288 , we have that , @xmath289 and for all @xmath40 , @xmath290 key to our results will be the uniform boundedness of this quantity . [ kevtriko ] there exist positive constants @xmath222 , @xmath12 depending only on @xmath5 such that the following holds . let @xmath48 , @xmath230 be as in section [ genclass ] and let @xmath16 , @xmath275 , @xmath7 be as in section [ sols ] where @xmath7 satisfies @xmath23 . then , for @xmath291 , @xmath292 inequality @xmath76 of theorem [ prwto ] follows from theorem [ kevtriko ] . ( the universality of the constant @xmath293 in the statement of that theorem follows _ a posteriori _ from a simple scaling argument . ) we note that given @xmath296 , for small enough @xmath297 , the vector field @xmath298 is timelike . for sufficiently regular @xmath281 , let us define @xmath299 note that @xmath300 thus , to prove theorem [ kevtriko ] , it is sufficient to prove @xmath301 with @xmath294 replacing @xmath302 . the significance of the parameter @xmath303 will become clear in the context of the proof . we shall need also a weaker positive definite quantity defined as follows . let @xmath160 denote a cutoff function such that @xmath161 for @xmath164 , and @xmath163 for say @xmath304 . for a sufficiently regular function @xmath281 , define @xmath305 note that we have @xmath306 note also that for @xmath164 , we have @xmath307 and @xmath308 for all @xmath309 , we have @xmath310 @xmath311 for an arbitrary suitably regular function @xmath281 such that @xmath312 is supported away from spatial infinity , recall from @xmath313 and @xmath314 the definitions of @xmath315 and @xmath316 . define also @xmath317 we have @xmath318 where @xmath319 thus , setting @xmath320 we have the identity @xmath321 we will also consider currents modified as follows . given a function @xmath322 , define @xmath323 by @xmath324 @xmath325 @xmath326 identity @xmath327 also holds for @xmath328 as long as appropriate assumptions are made in a neighborhood of spatial infinity . we will always apply @xmath328 to @xmath281 with @xmath329 , and thus , by our assumptions on @xmath312 , such @xmath281 will in fact be supported away from spatial infinity . it will be useful to have a separate notation for currents as defined with respect to the schwarzschild metric . for these we use the notation @xmath330 , @xmath331 , @xmath332 , etc . suppose that @xmath333 is a vector field such that its components @xmath334 are bounded in the atlas @xmath197 . it follows from @xmath223 that @xmath335 the above applies in particular if @xmath336 , i.e. for the case @xmath337 . ( in fact , the @xmath322 term disappears from the difference above . ) note that if the components of @xmath338 are less than @xmath339 we have by the triangle inequality @xmath340 note also that if @xmath341 , @xmath342 , @xmath322 , @xmath343 and @xmath344 are bounded with respect to @xmath197 , where then from @xmath223 , @xmath273 , we obtain @xmath345 if @xmath346 above vanishes , then @xmath347 are examples of _ compatible currents _ in the sense of @xcite . this is a unifying principle for understanding the structure behind much of the analysis for lagrangian equations like @xmath23 . since @xmath33 is killing we have @xmath348 in @xmath164 , @xmath33 is timelike and moreover we have @xmath349 in that region . in all regions we have latexmath:[\[|j_\mu^t(\psi)n_\sigma^\mu|\le b\,j_\mu^{n_\sigma}(\psi ) n^\mu_\sigma , \qquad let @xmath353 denote eddington - finkelstein null coordinates , @xmath354 . ] on @xmath32 and let @xmath205 denote the regge - wheeler coordinate . in the paragraph that follows , coordinate derivatives are with respect to say @xmath355 coordinates , whereas @xmath356 , @xmath357 denote @xmath358 , etc . recall from @xcite that for a vector field @xmath105 of the form : @xmath359 we have @xmath360 let us define @xmath361 , @xmath362 where @xmath363 is a cutoff function such that @xmath364 for @xmath365 , and @xmath366 for @xmath367 , for two parameters @xmath158 , and a small constant @xmath368 . one sees easily that there exist such parameters such that for @xmath365 , @xmath369 @xmath370 @xmath371 let us return now to the coordinate charts of our @xmath197 . we see from the above that the vector field @xmath105 has the property that in @xmath365 , @xmath372 where @xmath373 refer to the coordinate charts of @xmath197 whereas we easily see also that in @xmath374 latexmath:[\[\label{because2 } @xmath367 , @xmath376 . moreover , we note that @xmath105 is a regular vector field , in particular , when expressed with respect to the coordinates of @xmath197 , we have @xmath377 , @xmath378 . because all derivatives appear on the right hand sides of @xmath379 and @xmath380 , these inequalities are stable , i.e. it follows from @xmath381 that for @xmath222 sufficiently small , @xmath382 in @xmath365 , and @xmath383 in @xmath374 , while certainly @xmath384 for @xmath367 . define @xmath385 note that @xmath386 in the region @xmath387 , we have by @xmath388 @xmath389 note the factor of @xmath303 . in the region @xmath365 , we certainly have by @xmath390 @xmath391 for @xmath367 , we have of course @xmath392 in particular , the bound @xmath393 holds in all regions . with the help of @xmath390 and @xmath394 , we obtain easily that @xmath395 holds everywhere , if @xmath303 is sufficiently small . by similar considerations to the above , we see that given @xmath303 , by requiring @xmath297 sufficiently small , we have that @xmath396 is timelike everwhere up to the boundary , and in fact @xmath397 on the other hand , since by assumption [ lastassump ] , @xmath46 is null , @xmath398 controls all tangential derivatives . more precisely , we have @xmath399 @xmath400 on @xmath46 . for the above we have used the full content of assumption [ lastassump ] , as well as the translation invariance of @xmath49 , @xmath44 , @xmath401 , @xmath402 and @xmath396 , which allows us to choose uniform constants @xmath133 . in this section we shall often use regge - wheeler coordinates as many of the computations refer to the schwarzschild metric @xmath405 . in particular , we will consider vector fields of the form @xmath406 . in what follows @xmath407 will denote @xmath408 . in @xmath409 coordinates not the @xmath410 of our chart ! of course , @xmath410 coincides with @xmath30 for @xmath164 . ] we have @xmath411 where @xmath412 denotes the induced metric from @xmath148 on the spheres . we may rewrite the above as @xmath413 let @xmath141 , @xmath414 be parameters to be chosen in what follows . let @xmath415 . given these , we define a function @xmath416 such that @xmath417 ( one can smooth this function , although this is irrelevant . ) we call the resulting vector field @xmath418 . we obtain that in @xmath419 @xmath420 since @xmath421 for @xmath164 , we can rewrite this as @xmath422 where the coordinate derivatives in the last line can now be understood with respect to the atlas @xmath197 . for @xmath222 sufficiently small we obtain from @xmath381 @xmath423 in this region , where we have used @xmath424 . by @xmath394 , it follows that in @xmath425 . @xmath426 for small enough @xmath297 . consider now the region @xmath427 . we have @xmath428 and thus @xmath429 @xmath430 @xmath431 we have thus @xmath432 in this region . thus , by @xmath381 , @xmath433 and @xmath434 we have @xmath435 in this region . it follows now from @xmath436 that @xmath437 in this region , for small enough @xmath297 . in view of @xmath438 , @xmath439 will `` have a sign '' when applied to @xmath440 ( see section [ flest ] ) except for very large values of @xmath31 , namely @xmath441 . to control the behaviour there we will need an additional current . first , let us notice that for the @xmath418 we have selected , the coefficient of @xmath442 is always nonnegative . finally we notice that for @xmath441 , the coefficient of @xmath443 satisfies @xmath444 to choose an additional vector field , let us define @xmath445 where @xmath160 is a smooth cutoff with @xmath163 for @xmath446 and @xmath161 for @xmath447 , and let @xmath448 be the vector @xmath449 finally , define the function @xmath450 and consider the modified current @xmath451 defined by @xmath452 , as well as the associated @xmath453 and @xmath454 . note that for general @xmath245 , we can rewrite @xmath455 from which we see @xmath456 note also the modified error term @xmath457 finally , let us define the currents @xmath458 @xmath459 @xmath460 by our previous remarks , @xmath327 holds for @xmath461 . also , in view of the definition of @xmath322 , identities @xmath462 , @xmath463 and @xmath381 hold for @xmath461 , @xmath464 . let us expand @xmath465 where @xmath466 @xmath467 @xmath468 @xmath469 note that for @xmath447 , we have @xmath470 in particular , for @xmath441 for sufficiently large @xmath144 we have that @xmath471 while in @xmath425 , we have @xmath472 for @xmath473 , let us simply remark that for @xmath474 , we have @xmath475 for @xmath476 , we note first that we have the following asymptotic formula @xmath477 i.e. for @xmath441 for sufficiently big @xmath144 , we have @xmath478 and thus by @xmath479 @xmath480 to consider the behaviour for @xmath481 , let us first note that there exists an @xmath482 depending only on @xmath1i.e . independent of @xmath141 if we require @xmath141 to be sufficiently large such that for @xmath483 we have @xmath484 and thus , in @xmath485 we have @xmath486 for @xmath487 we have @xmath488 and thus , say @xmath489 for @xmath141 sufficiently large . turning to @xmath490 , we note first @xmath491 for large @xmath31 , i.e. we have @xmath492 for @xmath441 for @xmath144 suitably chosen . on the other hand , one sees easily that @xmath482 before could have been chosen such that for all @xmath141 we have @xmath493 for @xmath494 . for @xmath495 , we have @xmath496 we may thus choose @xmath141 large enough so that in this region @xmath497 * let @xmath141 be now chosen . it follows that @xmath498 and @xmath499 can be chosen . these choices thus can be made to depend only on @xmath1 . * let us assume in what follows in this section that @xmath329 . we thus have @xmath500 it follows that @xmath501 thus , in the region @xmath502 , we have @xmath503 note that , in the support of @xmath504 , we have latexmath:[\[(\partial_r^*\psi)^2\sim ( \partial_r\psi)^2 , \qquad we have then by the above bounds and @xmath381 , @xmath433 and @xmath424 that for @xmath506 , @xmath507 for @xmath222 suitably small , whereas for @xmath508 we may write @xmath509 where for the second inequality we require that @xmath222 be sufficiently small . from @xmath510 and the fact that @xmath504 vanishes identically in @xmath365 , we have @xmath511 in the region @xmath365 . to give bounds for the boundary terms , note first that @xmath512 on @xmath46 . it follows that on the horizon , we have @xmath513 one sees easily that for @xmath46 or @xmath230 where @xmath514 or @xmath515 , we have @xmath516 for @xmath222 sufficiently small . in view of the fact that we also have @xmath517 it follows that @xmath518 on @xmath46 or @xmath230 . on the other hand , in view of the the assumption @xmath329 , we have similarly @xmath519 it follows from the above inequalities that @xmath520 on both @xmath230 and @xmath46 . as explained in the introduction , the arguments of this paper hinge on separating the `` superradiant '' part of the solution from the non-``superradiant '' part , and then exploiting dispersion for the former and positive definiteness for the @xmath79 flux through the event horizon for the latter . these two parts will be characterized by their support in frequency space . as we certainly do not know , however , _ a priori _ that @xmath7 is in @xmath521 , we will first need to cut off @xmath7 in @xmath410 . this construction , together with propositions which control the errors that arise , are given in this section . let @xmath175 be a cutoff function such that @xmath176 for @xmath523 and @xmath178 for @xmath524 . given @xmath525 , define @xmath526 we may express this as @xmath527 where @xmath528 and @xmath529 are smooth functions on @xmath187 with @xmath530 @xmath531 @xmath532 and latexmath:[\[|\partial_{(i ) } { } ^{+}\chi^\tau_{\hbox{\rightscissors } } |\le b_q , \qquad with respect to the charts of @xmath197 , for any multi - index @xmath534 of order @xmath535 . moreover , @xmath536 the reader may wonder why the cutoff region is related to @xmath537 , indeed , why @xmath537 have been introduced in the first place . essentially , this is necessary to achieve the propositions of section [ sugkri][revisit ] . we would like to express all errors in terms of the positive definite quantity @xmath538 . this quantity does not contain @xmath7 itself but only derivatives . of course , in view of the fact that , as we shall see , the spherical average @xmath98 does not give rise to errors , this does not generate problems for the region @xmath539 for @xmath540 can be controlled by @xmath538 via a poincar inequality . as @xmath541 , one needs extra negative powers of @xmath31 . our cutoff region diverges from @xmath542 as @xmath543 and this allows us to `` gain '' powers of @xmath31 necessary to control @xmath97th order terms via a poincar inequality in @xmath542 . one can then retrieve estimates all the way to the boundary of the cutoff region using the positive definiteness of @xmath79 for large @xmath31 . let @xmath546 be a smooth cutoff supported in @xmath547 $ ] with the property that @xmath548 in @xmath549 $ ] , and let @xmath550 be a parameter to be determined later . for a smooth function @xmath551 of compact support in @xmath410 , let @xmath552 denote its @xmath86th azimuthal mode . let @xmath553 denote the fourier transform of @xmath281 in @xmath410 . note that @xmath554 . define @xmath555 @xmath556 note of course @xmath557 note in addition that @xmath558 whereas @xmath559 in the application to @xmath560 , we shall write simply @xmath561 and @xmath562 . note finally , that in view of @xmath563 , @xmath564 . note that for @xmath565 , @xmath566 where @xmath567 the kernel @xmath568 is a rescaled copy of a schwarz function of @xmath410 . as a consequence , for any @xmath569 , @xmath570 be a smooth cut - off function supported in @xmath571 such that @xmath572 on @xmath547 $ ] . then , since @xmath573 , we have the reproducing formula @xmath574 where the kernel @xmath575 also satisfies @xmath576 . finally , let @xmath577 be a function smooth away from @xmath578 and with the property that @xmath579 for @xmath580 and @xmath581 for @xmath582 . in particular , the function @xmath583 is smooth and @xmath584 for @xmath580 and @xmath585 for @xmath582 . since @xmath586 , we can write for @xmath565 , @xmath587 where @xmath588 furthermore , @xmath589 where @xmath590 and @xmath591 where @xmath592 the function @xmath593 is equal to one on @xmath594 and @xmath595 for @xmath582 . the kernel @xmath596 satisfies @xmath597 for any @xmath598 . in addition , we have a uniform bound ( coming from @xmath599 decay ) @xmath600 combining we obtain @xmath601 the decomposition of @xmath281 into @xmath544 and @xmath545 is motivated by the desire to compare various @xmath80-type norms of the @xmath401 and @xmath212 derivatives . since this is required at a localised level , however , error terms arise . the precise relations one can make are recorded in this section . the estimates of this section employ standard techniques of elementary fourier analysis . we must be careful , however , to express all `` error terms '' in a form which can be related to our bootstrap assumptions which will be introduced later on . first a lemma . [ flatlem ] let @xmath604 and let @xmath281 be smooth and of compact support in @xmath410 . then @xmath605 recall @xmath606 . note first that by the relations of section [ defflat ] , it follows that for any @xmath598 , we have @xmath607 we have thus @xmath608 it follows that , for @xmath609 , @xmath610 thus , @xmath611 where @xmath612 if @xmath613 $ ] and @xmath97 otherwise . to prove the lemma , in view of the comments in section [ genclass ] on the volume form and plancherel , it would suffice to show that @xmath614 @xmath615 the first term on the right hand side of @xmath616 is bounded by @xmath617 thus , it follows that @xmath618 we have established @xmath619 . the second term on the right hand side of @xmath616 is bounded by @xmath620 we have @xmath621 for @xmath535 chosen sufficiently large . as for @xmath622 , we have @xmath623 the above and @xmath624 give @xmath625 . [ yenilemma ] under the assumptions of the previous lemma , if @xmath626 , then @xmath627 for any @xmath598 , we have latexmath:[\ ] ] the following lemmas are immediate : @xmath899 @xmath900 @xmath901\psi ) ^2\le b \epsilon_{\rm close}^2 \int_{\mathcal{r}(\tau',\tau '' ) } { \bf q}^1(\psi).\ ] ] given @xmath902 , we may choose a @xmath903 ( with @xmath904 as @xmath905 ) such that @xmath906 for convenience , let us require in what follows that @xmath907 . it follows from @xmath327 , the above lemmas and cauchy - schwarz ( applied with a small parameter @xmath908 ) that @xmath909 since @xmath910 , it follows that @xmath908 can be chosen so that @xmath911 from the propositions of section [ elliptprops ] below , we obtain @xmath912 and thus , for small enough @xmath222 , and choosing @xmath913 close enough to @xmath914 ( and thus small enough @xmath903 ) , we obtain @xmath915 ( the choice of @xmath913 having been made , we have written above @xmath916 as @xmath133 following our convention . ) from @xmath917 and @xmath918 , it now follows that @xmath919 it follows immediately that there exists a sequence @xmath920 such that @xmath921 from @xmath887 , we have on the other hand @xmath922 and from @xmath301 @xmath923 from proposition [ elliptprop1 ] it follows that @xmath924 while from proposition [ elliptprop2 ] , it follows that @xmath925 thus in fact , @xmath926 in view of @xmath927 , we obtain now easily @xmath928 we have the following elliptic estimate on spheres : [ elliptprop1 ] let @xmath929 denote a set of constant @xmath31 in a @xmath410 , @xmath31 , @xmath683 , @xmath930 coordinate system . for @xmath7 a solution of @xmath6 , we have @xmath931 note first that @xmath932 let @xmath933 denote the standard laplacian on the unit sphere . in the coordinates of the first paragraph of section [ ysec ] , we may write @xmath934 integrating over @xmath929 endowed with metric of the standard unit sphere , we obtain the elliptic estimate @xmath935 i.e. , in view of the assumptions @xmath223 on the metric , @xmath936 on the other hand , from @xmath223 , @xmath273 we have @xmath937 the above , @xmath938 and @xmath939 yield the proposition , for @xmath222 sufficiently small . in addition , we have the following elliptic estimates away from the event horizon . [ elliptprop2 ] for @xmath7 a solution of @xmath6 , and @xmath940 a parameter with @xmath941 , then , for @xmath222 sufficiently small , we have @xmath942 @xmath943 the proof of this straightforward elliptic estimate is left to the reader . we have the following sobolev - type estimate on schwarzschild let @xmath281 be a smooth function on @xmath944 of compact support . then @xmath945 this in turn follows from the following euclidean space estimate : there exists a constant @xmath946 such that the following holds . let @xmath281 be a smooth function on @xmath947 of compact support . . there exists a positive constant @xmath222 , depending only on @xmath1 , and a positive constant @xmath949 , depending on @xmath1 and @xmath878 , such that the following holds.let @xmath48 , @xmath230 be as in section [ genclass ] and let @xmath16 , @xmath275 , @xmath7 be as in section [ sols ] where @xmath7 satisfies @xmath23 . then , for @xmath291 , @xmath950 where @xmath951 denotes the @xmath878th order spacetime covariant derivative tensor and @xmath952 denotes the induced riemannian norm . theorem [ deutero ] in particular follows from the above . in the schwarzschild case , we may apply the estimates proven here for @xmath89 in sections [ estest ] and [ boubou ] directly to the whole @xmath7 . since no frequency decomposition need be made , no associated error terms arise and the whole argument can be reduced to a few pages . the resulting energy estimate , coupled with the higher order and pointwise estimates of section [ he s ] , yield a new proof for uniform boundedness of solutions to the wave equation on schwarzschild which is in some sense the simplest one yet using neither the discrete isometry exploited by kay - wald @xcite , nor the vector field @xmath953 of our @xcite or @xcite , nor commuting with angular momentum operators . moreover , one shows the uniform boundedness of all derivatives on the event horizon up to all order , whereas previous results could control only tangential derivatives . [ stat ] let @xmath954 be a static spherically symmetric asymptotically flat exterior black hole spacetime bounded by a non - extremal event horizon @xmath46 . then the estimates of theorems [ prwto ] and [ deutero ] hold . our argument is easily adapted to spacetimes which are small perturbations of non - extremal schwarzschild - de sitter , in particular to slowly rotating non - extremal kerr - de sitter , or kerr - newman - de sitter . see @xcite for the setting . one fixes the manifold structure on a subregion @xmath955 where @xmath27 is here the region between a set of black / white hole and cosmological horizons and @xmath57 is a cauchy surface crossing both horizons to the future of the bifurcate spheres . @xmath956 one continues as in the schwarzschild case . the argument is in fact easier at several points . because @xmath31 is bounded in @xmath27 , the zero - order terms pose no difficulty . in particular , one need not introduce the @xmath957 and @xmath958 surfaces , nor must one modify @xmath959 by the addition of @xmath960 . we leave the details for a subsequent paper . p. blue and a. soffer _ errata for `` global existence and scattering for the nonlinear schrodinger equation on schwarzschild manifolds '' , `` semilinear wave equations on the schwarzschild manifold i : local decay estimates '' , and `` the wave equation on the schwarzschild metric ii : local decay for the spin 2 regge wheeler equation '' _ , gr - qc/0608073 , 6 pages

we report moderate resolution 3 - 5 m spectroscopy of the nucleus of ngc 1068 obtained at 0.3 ( 20 pc ) resolution with the spectrograph slit aligned approximately along the ionization cones of the agn . the deconvolved fwhm of the nuclear continuum source in this direction is 0.3 . four coronal lines of widely different excitations were detected ; the intensity of each peaks near radio knot c , approximately 0.3 north of the infrared continuum peak , where the radio jet changes direction . together with the broadened line profiles observed near that location , this suggests that shock - ionization is the dominant excitation mechanism of the coronal lines . the depth of the 3.4 m hydrocarbon absorption is maximum at and just south of the continuum peak , similar to the 10 m silicate absorption . that and the similar and rapid variations of the optical depths of both features across the nucleus suggest that substantial portions of both arise in a dusty environment just in front of the continuum source(s ) . a new and tighter limit is set on the column density of co. although clumpy models of the dust screen might explain the shallowness of the silicate feature , the presence of the 3.4 m feature and the absence of co are strongly reminiscent of galactic diffuse cloud environments and a consistent explanation for them and the observed silicate feature is found if all three phenomena occur in such an environment , existing as close as 10 pc to the central engine .

0906.4885

the unified model of active galactic nuclei ( agns ) explains the most basic observed properties of agns in terms of toroids of dust and gas , whose orientations to our line of sight determine the observed characteristics of these objects . the infrared spectral energy distributions ( seds ) of agns betray the presence of dust in their nuclei . some insight into the properties and distribution of the dust has been gained from observations and modeling of the 9.7 @xmath0 m silicate feature @xcite and the 3.4 @xmath0 m hydrocarbon absorption band @xcite . notable findings include differences in the spectral profile of the 10 @xmath0 m silicate feature between galaxies with differing amounts of obscuration @xcite , and evidence that the toroids are rather small , only a few parsecs in diameter at 10 microns @xcite . at shorter wavelengths many agns are known to have nuclear point sources , whose angular extents are diffraction - limited ( e.g. , at @xmath10.1 resolution as viewed in the @xmath2-band ) . the dust close to an agn is generally expected to be accompanied by large quantities of molecular gas @xcite , so it might be assumed that medium or high resolution infrared spectroscopy would result in the detection of molecular absorption lines there . surprisingly , probably the easiest interstellar infrared molecular band to detect in absorption , the co fundamental ( @xmath3=1 - 0 ) band , whose most prominent lines are located at 4.5 - 5.0 @xmath0 m , has not yet been found toward the nucleus of the prototypical and nearby ( 14.4mpc , 1= 72 pc ; * ? ? ? * ) seyfert 2 galaxy , ngc 1068 , at both 120 km s@xmath4 resolution @xcite and at 20 km s@xmath4 resolution @xcite . emission _ from molecular hydrogen in portions of the central few parsecs @xcite is the only direct evidence for gas phase molecules there . a clumpy torus was first suggested by @xcite to account for dust surviving close to agns and has been invoked by @xcite to explain the behavior of the silicate feature with orientation and the spectral breadth of the far - infrared emission . the model is supported in some specific cases by fits to the sed of an agn ( e.g. , * ? ? ? * ) . in ngc 1068 the presence of prominent dust absorption bands both to longer and shorter wavelengths than the co fundamental band suggest that some co absorption ought to be present . in galactic dense cloud material , a silicate optical depth of 0.4 , the value found by @xcite toward the nucleus of ngc 1068 , implies an h@xmath5 column density of 0.7 @xmath6 10@xmath7 @xmath8 and a co column density of 1.0 @xmath6 10@xmath9 @xmath8 @xcite , which should be easily detectable in the fundamental band . toward the diffuse cloud in front of cyg ob2 no . 12 , which has @xmath10 mag , the 10 @xmath0 m silicate absorption is of comparable strength to that in ngc 1068 and the 3.4 @xmath0 m absorption band , generally assumed to arise from stretching of c - h bonds in solid - state hydrocarbons @xcite , is considerably weaker than that band in ngc 1068 ( e.g. , * ? ? ? * ; * ? ? ? nevertheless several ( narrow ) lines in the co fundamental band have optical depths of 0.3 @xcite . one way for co to be present but difficult to detect is if the co lines are so broad that they blend together , so that individual lines can not be seen . using different methods @xcite and @xcite have each estimated the mass of the nuclear black hole in ngc 1068 to be @xmath12 @xmath6 10@xmath11 m@xmath12 . the dimension of the outer of the torus - like clouds observed by @xcite and recently re - observed by @xcite using mid - infrared interferometry is 2 pc . together these values imply an orbital speed of 225 km s@xmath4 , slightly less than half the spacing of lines in the co fundamental band . thus , if absorbing co is located at the outer edge of the torus , each co transition would produce an absorption 450 km s@xmath4 wide , which is comparable to the co line spacing . in that case and assuming that the co rotational states are thermalized to at least a few hundred kelvins , the @xmath13 band spectrum would show a broad and shallow depression covering most of the band and would not be easily detectable in previous data sets . any co orbiting considerably farther from the nucleus , however , should be readily detectable in individual lines , if present with sufficient column density . for example , gas orbiting at a radius of 10 pc would produce lines with full widths at zero intensity ( fwzis ) of @xmath1200 km s@xmath4 . the high signal - to - noise ratio attainable in the @xmath13 band for ngc 1068 , has allowed us to conduct a new and more sensitive search for co absorption against the @xmath13 band continuum at the nucleus of this galaxy . we also have obtained a 3 - 4 @xmath0 m spectrum of the nucleus in very good seeing and with high enough s / n to begin to discern the spatial distribution of the 3.4 @xmath0 m feature within 0.4 ( 30 pc ) of the nucleus . in the galaxy this band , due to the c - h stretch in ethyl and methyl groups attached to hydrocarbons , is observed only in diffuse clouds @xcite . recent observations of the spatial behavior of the 9.7 @xmath0 m silicate feature toward agns have provided some information on the dust distribution in the central few tens of parsecs @xcite , in several cases revealing extended dusty structures perhaps associated with material fueling the central engine @xcite . @xcite have found stronger evidence for fueling of the ngc 1068 agn , based on high angular resolution observations of h@xmath5 line emission within the central few arc - seconds . more detailed observations of the 3.4 @xmath0 m band and a more sensitive search for co absorption promise a complementary view of the material in the center of ngc 1068 . the 35 @xmath0 m region also includes a number of high - excitation forbidden emission lines . coronal " lines in agns have been a subject of interest during the last few decades for several reasons . first , their high ionization energies ( @xmath14100 ev ) imply that they must be excited by shocks and/or hard photons from the agn , with negligible contribution from hot stars . if the possible sources of excitation can be disentangled , for example through observation of the locations at which various lines arise , the line strengths and ratios may yield valuable information on the nature of the agn continuum in otherwise inaccessible regions of the uv / x - ray spectrum . second , if they are emitted in the inner regions of the narrow line region ( nlr ) , or even associated with the putative torus of the unified model , they can be used indirectly to study those structures . third , the widths , profiles and wavelength shifts of the lines may map extreme conditions such as shocks and outflows in regions close to the central engine , which may be spatially resolvable in nearby agns . optical spectroscopy from _ hubble space telescope _ ( hst ) has provided detailed information on the morphology of the emission line regions and complex velocity fields of the ionized gas associated with radio structures close to the agns ( for ngc 1068 see * ? ? ? although efforts have been made to understand the infrared spectra of agns , because of the lack of high spatial resolution information on a well - defined sample , a consensus on the location and origin of the infrared coronal lines has not yet been reached . in this respect , ngc 1068 is a useful laboratory because of the prominent coronal line spectrum that it displays at optical , near - infrared and mid - infrared wavelengths . whereas the prominent coronal lines in the optical are largely from fe , the 1.02.5 @xmath0 m region includes several other species , notably si , ca and al , and the 3.05.0 @xmath0 m range includes lines from al , si , ar , and na . the new 35 @xmath0 m observations and their reduction are outlined in [ obs ] , and the strengths , distributions and profiles of the emission lines and molecular and solid state absorption features are described in [ emission ] and [ absorption ] , respectively . comparisons with previous observations are mostly found in [ compare ] . the implications of these findings for our understanding of the environments producing the coronal lines and solid state and possible molecular absorptions are discussed separately in [ discuss ] . ccccc 2.93.6 @xmath0 m & 360 & 1400 & 0.41@xmath150.02 & 0.31@xmath150.02 + 3.64.2 @xmath0 m & 360 & 2400 & 0.42@xmath150.02 & 0.30@xmath150.03 + 4.45.3 @xmath0 m & 360 & 2000 & 0.41@xmath150.03 & 0.29@xmath150.03 + m spectra . north is up and east is to the left . edges of the slit ( vertical lines ) are overlayed on fig . 1 of @xcite , a 4 @xmath6 4 field containing the o iii image from @xcite and the 6 cm contours of @xcite . radio source s1 , as designated e.g. in @xcite , is coincident with the near - infrared continuum peak . , width=453 ] spectra of the nuclear region of ngc 1068 and of a nearby calibration star , hr 691 ( f0v , l=4.7 ) , were obtained at the united kingdom infrared telescope ( ukirt ) on mauna kea on 2006 august 10 . the facility 15 @xmath0 m imager / spectrograph uist @xcite was used , with its 0.2 ( 2-pixel wide ) slit centered on the continuum peak and oriented north - south , as shown in fig . [ fig : slit ] . with that slit width uist delivers resolving powers , r , of 1400 - 2400 in various parts of the 35 @xmath0 m region . the orientation roughly corresponds to the position angle of the ionization cones in the central 0.4 @xcite . rows of the array are separated by 0.10 along the slit . table 1 summarizes the observational parameters . the observations were made in superb seeing conditions and through a photometric and relatively dry sky ( 1.5 mm of precipitable h@xmath5o ) . three spectral regions were covered , 2.93.6 @xmath0 m , 3.64.2 @xmath0 m , and 4.45.3 @xmath0 m . total exposure times were six minutes in each of the above spectral intervals , and the telescope was nodded 12north and back , along the slit in an abba pattern which was repeated every two minutes . long - slit spectra and imaging of h i br @xmath16 at 2.17 @xmath0 m @xcite have detected no line emission at the offset location . hr 691 was observed in each spectral interval just prior to ngc 1068 as a flux calibrator and to remove telluric absorption lines from the spectra of ngc 1068 during data reduction . the differences in mean airmass between the pairs of observations of ngc 1068 and calibration star were each one percent or less . the intensity profiles of the continuum from hr 691 , measured along the slit have full widths at half maximum ( fwhms ) of approximately three pixels 0.3 ( equivalent to 22 pc at the distance of ngc 1068 ) in all three wavelength bands observed . assuming that the intensity distribution observed along the slit is circularly symmetric , we estimate that 55 @xmath15 10 percent of the signal from hr 691 fell out of the slit . data reduction employed starlink figaro routines to extract spectra in individual rows of the uist detector array from coadditions of several subtracted pairs of images that had been flat - fielded in the ukirt reduction pipeline . following manual removal of spikes , spectra from pairs of adjacent rows were combined to produce spectra covering various 0.2 @xmath6 0.2 regions along the slit . these were ratioed by the extracted spectrum of the calibration star . wavelength calibration , obtained from telluric lines in the spectrum of the calibration star is better than 0.0001 @xmath0 m . all wavelength scales in this paper are _ in vacuo _ and not corrected for the systemic velocity of ngc 1068 . the flux calibrations of the reduced 2.93.6 @xmath0 m and 3.54.1 @xmath0 m spectral segments agreed to 3.5 percent in the wavelength interval of overlap , attesting to the excellent weather conditions . much of the 4.45.4 @xmath0 m region that lies outside of the 4.55.0 @xmath0 m interval is unusable or of little value because of low telluric transmission and is not shown here . in addition , we do not include the reduced spectrum in several narrow wavelength intervals in the 4.805.00 @xmath0 m region , where deep telluric h@xmath5o lines severely degrade the spectra of both the galaxy and the calibration star . in the 35 @xmath0 m interval the spectrum of the calibration star contains several prominent absorption lines of hydrogen in the brackett , pfund , and humphreys series . of these only br @xmath17 falls in a relatively clean portion of the spectrum and can be removed by interpolation across the observed feature . for the other prominent h i lines we employed a kurucz r=4,000 model spectrum of vega with line strengths somewhat diluted in an attempt to match the h i lines of an f0v star . this technique proved successful in removing the pf @xmath18 and hu 11 - 6 pair at 4.65 - 4.67 @xmath0 m but did not work well for simultaneous removal of the pf @xmath16 and pf @xmath19 lines in the short wavelength 2.93.6 @xmath0 m segment . small wavelength intervals near those lines are omitted in the spectrum presented here . full widths at half maximum ( fwhms ) along the slit were determined for both ngc 1068 and the calibration star from the coadded spectral images the results are shown in table 1 . in all three wavebands the nuclear continuum source in ngc 1068 had a fwhm of approximately four pixels along the slit ( compared to three pixels for the calibration star ) and was thus partially resolved . a simple gaussian deconvolution , using the stellar fwhm as the instrumental value yields intrinsic fwhms for ngc 1068 of 0.270.29 in the ns direction ( corresponding to a characteristic ns radius " for the continuum emission of @xmath110 pc ) . this appears qualitatively consistent with the adaptive optics imaging near these wavelengths by @xcite as well as with the more recent and considerably higher resolution imaging of @xcite . because the central source is partly resolved in the present data , it is possible to crudely investigate the spatial distribution of the 3.4 @xmath0 m absorption feature along the slit , in addition to accurately determining the distributions of the emission lines . m spectrum of the central 0.2 @xmath6 1.0 ( ew x ns ) region of ngc 1068 . wavelengths of atomic lines of interest are shown , as are the the three components of the 3.4 @xmath0 m hydrocarbon feature @xcite and the band center of co , all at the systemic redshift of 0.003793 . the noise level varies but can be estimated at each wavelength from the typical point - to - point fluctuations of the continuum in the vicinity . for simplicity the flux scale assumes that both ngc 1068 and the calibration star are unresolved point sources and that slit losses for them are equal ; this is approximately correct for the continuum from ngc 1068 , but not for the much more extended line emission.,width=453 ] cccc @xmath20alvi@xmath21 $ ] & 153.8 & 3.660 & 0.45@xmath150.10 + h i 8 - 5 & 13.6 & 3.7405 & @xmath220.3 + @xmath20siix@xmath21 $ ] & 303.2 & 3.9357 & 1.5@xmath150.15 + h i 5 - 4 & 13.6 & 4.0522 & @xmath220.8 + @xmath20arvi@xmath21 $ ] & 75.0 & 4.5295 & 2.5@xmath150.7 + @xmath20navii@xmath21 $ ] & 172.1 & 4.6847 & 1.1@xmath150.3 + figure [ fig : spectrum ] shows the 3 - 5 @xmath0 m spectrum extracted from the central ten rows ( i.e. , a 0.2@xmath61.0 region of ngc 1068 ) . the expected locations of hydrogen recombination lines and fine structure lines are indicated for a heliocentric systemic radial velocity of 1137 km s@xmath4 . in addition to the broad 3.4 @xmath0 m hydrocarbon feature @xcite , four high excitation fine structure ( coronal ) lines are seen in emission : [ alvi ] at 3.660 @xmath0 m , [ siix ] at 3.9357 @xmath0 m , [ arvi ] at 4.5295 @xmath0 m and [ navii ] at 4.6847 @xmath0 m ( rest wavelengths ) . of these the [ alvi ] and [ navii ] lines are new detections . a tentative detection of the latter was reported by @xcite . the other lines were previously found by @xcite and @xcite . the coronal lines have fwzis of over @xmath11000 km s@xmath4 and the peak of each line is blue - shifted relative to the systemic velocity of ngc 1068 by @xmath1 300 km s@xmath4 , as has been found previously for other coronal lines in ngc 1068 @xcite . no hydrogen or helium recombination lines are detected in fig . [ fig : spectrum ] . the marginal emission feature near pf @xmath16 has a profile and a velocity centroid that are inconsistent with all of the other lines , leading us to doubt its reality . there is some marginally significant spectral structure in the vicinity of br @xmath17 , near 4.06 @xmath0 m , but no obvious sign of line emission . m continuum , [ siix ] , and [ alvi ] line flux in 0.2 sections and 0.2 steps along the 0.2 wide slit . typical @xmath151@xmath23 uncertainties in line strengths are shown ; upper limits , designated by downward arrows , are 2@xmath23 . the angular resolution ( fwhm = 0.30 ) , derived from the profile of the standard star , is indicated . , width=453 ] the coronal line that is detected at the highest signal - to - noise ratio is the [ siix ] line at 3.95 @xmath0 m . si@xmath24 has an ionization energy of 303 ev , making it one of the most highly excited species detected in ngc 1068 . the next best coronal line for study in these data is the [ alvi ] line at 3.67 @xmath0 m , which is more than three times weaker , but is situated at shorter wavelengths than other lines and in a region of high atmospheric transmittance so that the ambient background is relatively low and the noise slightly lower than at 3.95 @xmath0 m . the ionization energy of al@xmath25 is 154 ev , approximately half that of si@xmath24 . figure [ fig : siix ] shows both the continuum flux density and the fluxes of these two lines , in 0.2@xmath60.2 apertures along the north - south slit . the line fluxes peak approximately 0.3 north of the continuum peak and emission extends northward to 0.8 . little or no line emission is present south of the nucleus . the peaks of the line emission from both [ navii ] ( ionization energy 172 ev ) and [ arvi ] ( ionization energy 75 ev ) also are offset a few tenths of an arc - second to the north of the continuum peak , similar to [ siix ] and [ alvi ] , but their detailed distributions are less certain , because of difficulties in accurately removing telluric absorption features and accurately defining the continuum level in the case of the [ arvi ] lines , and because of the weakness of the emission in the case of the [ navii ] line . line in 0.2 steps along the slit . continuous lines denote spectra at the center and to the north ; dashed lines indicate spectra to the south . numbers in parentheses are continuum offsets . the y - axis is flux density in a 0.2 @xmath6 0.2 aperture and assumes an estimated slit loss factor of 2.2 for the calibration star . the systemic radial velocity and velocity resolution are shown . , width=453 ] figure [ fig : velocity ] shows the velocity profiles of the [ si ix ] line along the north - south slit . the blue - shifted peak seen in the composite spectrum of fig . 1 is present at all positions from the center to 0.8 north . at these locations the velocity of peak line emission is shifted between -200 and -400 km s@xmath4 relative to the systemic velocity of ngc 1068 . to the south no line emission is seen except perhaps 0.4 south where a weak line centered near the systemic velocity may be present . at all of the locations where the line is clearly detected , the velocity profile is asymmetric with an extended blue wing . the wing is most prominent , essentially forming a shoulder , in the central and 0.2 north spectra . overall the line emission appears to extend approximately to -1000 km s@xmath4 and + 400 km s@xmath4 from the systemic velocity . this range is similar to that seen in other coronal lines in the 14 @xmath0 m region , e.g. , by @xcite . the line fluxes within the uist slit given in table 2 were determined in various ways . coronal line fluxes derived solely from fig . [ fig : spectrum ] will be systematically incorrect because the spectrum has been flux - calibrated using a point source , but the line emission is more extended along the slit than is the continuum . the spectra in fig . [ fig : velocity ] and summed fluxes for the [ si ix ] and [ alvi ] lines in the table have been corrected for slit losses , using the spectrum of the calibration star . for [ arvi ] , and [ navii ] , the signal - to - noise ratios are too low to determine fluxes by summing their values at locations along the slit . therefore we have assumed similar distributions for these lines as [ siix ] and [ alvi ] . then their fluxes in the slit can be estimated by scaling the fluxes derived from fig . [ fig : spectrum ] by 0.60 , the ratio of the fluxes of the [ siix ] and [ alvi ] lines in table 2 to the values for them derived from fig . [ fig : spectrum ] . for br @xmath17 and pf @xmath16 upper limits were derived solely from the spectrum in fig . [ fig : spectrum ] . m hydrocarbon absorption feature ( this paper ) and the 9.7 @xmath0 m silicate absorption feature @xcite across the nucleus of ngc 1068 . error bars are @xmath151@xmath23 . , width=453 ] the 3.4 @xmath0 m hydrocarbon absorption feature is the only solid - state feature present in the 35 @xmath0 m spectrum . it extends from 3.35 @xmath0 m to perhaps 3.57 @xmath0 m with a broad absorption maximum near 3.41 @xmath0 m and a shoulder that begins at 3.47 @xmath0 m . assuming that the contribution of the 3.43 @xmath0 m component drops to zero at 3.50 @xmath0 m , the absorption maximum of the shoulder feature occurs close to 3.50 @xmath0 m . after accounting for the redshift of ngc 1068 , the above wavelengths and the overall profile are in good agreement with those observed for the 3.4 @xmath0 m feature in galactic sources @xcite . the optical depth of the 3.4 @xmath0 m feature has been determined by fitting a linear continuum across the 3.103.35 @xmath0 m interval ( excluding the 3.293.33 @xmath0 m region ) and the 3.573.60 @xmath0 m interval . the maximum optical depth of the 3.4 @xmath0 m feature shown in fig . [ fig : spectrum ] is 0.072 @xmath15 0.004 . using the same technique we also have measured the optical depth of the 3.4 @xmath0 m feature in 0.2 steps within @xmath150.4 of the infrared continuum peak . the results are shown in fig . [ fig : chsil ] , together with the measurements of the optical depth of the 9.7 @xmath0 m silicate absorption from @xcite measured in 0.4 steps steps over a wider region . at and south of the continuum peak the optical depth of the 3.4 @xmath0 m feature is approximately 0.08 , while at the northern locations the depth is noticeably less , approximately 0.06 . this pattern is similar to that seen in the silicate feature . the implications of this are discussed in [ discuss ] . the 4.55.0 @xmath0 m portion of the spectrum allows a new search for the presence of the fundamental band of carbon monoxide ( co ) , which should manifest itself as individual absorption lines if the co line width is less than the line spacing ( @xmath1500 km s@xmath4 or as broad depressions on either side of the band center if the co line profiles are comparable to or broader than the spacing . @xcite , observing at a resolution of 20 km s@xmath4 , did not detect narrow co lines . as discussed earlier co absorption lines arising in gas orbiting the agn outside the characteristic radius of the continuum emitting region should be no wider than 200 km s@xmath4 ( fwzi ) . no signs of absorption by individual lines can be seen in the current spectrum , and any broad depression between the band center and the long wavelength can not be deeper than about 1% of the continuum . close examination of the spectrum shows that the continuum in the 4.84.9 @xmath0 m portion of the spectrum , corresponding to p branch transitions with 10@xmath22@xmath26@xmath27 , may be depressed by roughly that amount relative to adjacent portions of the spectrum . this could be interpreted as evidence for absorbing co in dense gas at temperatures of several hundred degrees and with very high velocity dispersion . for an isothermal absorbing slab , lines of co at this temperature with fwhm s of 500 km s@xmath4 and central depths of 1% would produce such a depression and would imply a co column density of @xmath1 1 @xmath610@xmath28 @xmath8 and a gas particle density of order 10@xmath11 @xmath29 . the corresponding portion of the r branch of co is not available for comparison due to telluric absorption and contamination from the [ arvi ] line . because the evidence for this broad co absorption is marginal we regard the above column density as an upper limit . both the offsets of the peaks of the [ siix ] and [ alvi ] lines from the peak of the continuum and the extents of the line emission are somewhat less than those reported by @xcite for the coronal lines that they measured along the ionization cone at position angle 31 degrees . however , their measurements were made at roughly five times lower angular resolution than those here . thus the present determination of the offset should be more accurate . the distributions of the other two coronal lines along the ns slit are not as well known as those of [ siix ] and [ alvi ] , but they appear to be similar . @xcite imaged the 1.96 @xmath0 m coronal [ sivi ] line at 0.2 resolution , finding it to originate mainly to the north - northeast of the the continuum source . the current measurements of the [ siix ] line are largely consistent with @xcite , but failed to detect the fainter line emission that their data show to the south . @xcite imaged the [ sivii ] 2.48 @xmath0 m line emission at 0.10 resolution . their figures show it to be mostly in a clumpy structure centered on the continuum peak . this appears to differ markedly from the distributions of both the [ sivi ] line imaged by @xcite and the [ siix ] line observed spectroscopically by @xcite and by us . the accuracy of the co - registration of their images is not available . in the present data the lines and the continuum were measured simultaneously and the result is clear . the ionization energy of [ sivii ] lies between those of [ alvi ] and [ siix ] ; thus it would be surprising if the intensity distribution of the [ sivii ] line relative to that of the adjacent continuum were not similar to the lines observed here . the new measurements of these lines put significant constraints on the size of the high - ionization line emitting region . the detected [ si ix ] flux in the uist slit is about 3.5 times smaller than that reported by @xcite using the infrared space observatpry ( iso ) with a 14 @xmath6 20 aperture , and three times smaller than that reported by @xcite , who used a 4.4 slit width however , assuming that the [ siix ] line - emitting region through which our observations have sliced is roughly circularly symmetric , the total flux in the [ siix ] region is @xmath15.3 @xmath30 w m@xmath31 , in agreement with the flux measured by iso and , within the uncertainties , with that by marconi et al . using adaptive optics imaging @xcite fund a rather clumpy distribution of the [ sivii ] 2.48 @xmath0 m line emission extended over @xmath11 but one that does not prefer any particular position angle , so the assumption of circular symmetry , while likely incorrect in detail , should allow a reasonable estimate of total line flux . following the same line of reasoning , the [ alvi ] line , which has a very similar flux distribution along the uist slit as [ siix ] ( see figure [ fig : siix ] ) , would have a total flux of 1.6@xmath32 w m@xmath31 . the total fluxes in the [ arvi ] , and [ navii ] lines would be 9 @xmath6 10@xmath33 w m@xmath31 and 4 @xmath6 10@xmath33 w m@xmath31 , respectively . @xcite found a 1.7 times higher flux for the [ arvi ] line , suggesting that the distribution of this ion , which possesses the lowest ionization energy of the four detected , is somewhat more extended than [ siix ] or possesses a large and very faint halo . @xcite and @xcite found that the [ sivi ] and [ sivii ] lines ( with ionization energies of 167 ev and 205 ev , respectively ) each have halos extending to a radius of 70 pc ( 1.0 ) , but they do not provide the relative contributions of the core and halo . the absence of hydrogen 5@xmath344 recombination line emission is another noteworthy aspect of the 3 - 5 @xmath0 m spectrum in figure [ fig : spectrum ] . the br @xmath16 line emission detected by @xcite from the circum - nuclear ring of h ii regions at a radius of 15 - 16 has 12 orders of magnitude lower surface brightness than that seen at the nucleus by @xcite . thus the weakness of the recombination line emission in figure [ fig : spectrum ] is not likely to be due to subtraction of bright line emission in the offset beam , located 12 north of the nucleus , just inside the ring . @xcite have mapped the br @xmath16 line emission in the vicinity of the nucleus at high angular resolution and have kindly made their data available to us . our limit for the br @xmath17 flux in the central 0.2 @xmath61.0of @xmath228 @xmath610@xmath35 w m@xmath31 is 75 times the br @xmath16 line flux in the same aperture . assuming case b conditions this limit corresponds to an upper limit on the reddening of 3.5 mag between the wavelengths of the two lines , and , for a @xmath36@xmath37 dependence for the extinction , a visual extinction of no more than @xmath160 mag to the br @xmath16 line - emitting region . the optical depth of the 3.4 @xmath0 m feature reported here is in good agreement with the values measured by @xcite and @xcite in larger apertures , but disagrees with values about twice as high reported by @xcite and @xcite . as the bulk of the continuum emission comes from the central point source , the differences with @xcite and @xcite are difficult to account for in terms of positional variations . in the case of the former paper , incorrect location of the short wavelength continuum may be responsible for the discrepancy , but this does not seem to be the explanation for the @xcite result . neither we , nor any other of the above observers have detected the nuclear 3.3 @xmath0 m ( pah ) emission feature reported by @xcite . it is possible that some of the features that they see are spurious , perhaps resulting from incorrect cancellation of strong telluric lines present in the 3.253.45 @xmath0 m interval . @xcite and @xcite searched for lines of the co fundamental at resolutions of 250 km s@xmath4 and 120 km s@xmath4 , respectively . the latter authors set a 3@xmath23 upper limit of 7% of the continuum for the depths of individual co lines . in the 4.57 - 4.77 @xmath0 m portion of the current spectrum , obtained at 150 km s@xmath4 resolution and also devoid of co lines , the signal - to - noise ratio on the continuum is about three times higher . thus , the limits both on individual lines separated from one another ( i.e. , with widths less than @xmath1500 km s@xmath4 ) and on a broad depression due to blended co lines are significantly more stringent than those of @xcite . finally , @xcite reported the detection toward the nucleus of ngc 1068 of a narrow absorption feature at 4.69 @xmath0 m , which they attributed to solid co. our spectrum shows no such feature , to a limiting absorption depth of a few percent of the continuum . one of the most important questions concerning coronal lines in agn is the excitation mechanism that drives their emission . here we discuss whether photoionization or shock ionization can account for the locations and strengths of the coronal line emission in ngc 1068 . a summary is provided at the end of this section . @xcite carried out a number of illustrative photoionization simulations in an effort to identify the optimal conditions and locations in which the coronal lines form around agn . assuming ultraviolet radiation from the central engine as the only excitation mechanism and plane - parallel , constant density slabs of gas , they determined the distances from the ionizing source in which the lines are emitted as a function of density . they provide line equivalent widths in the density - distance plane , indicating where the bulk of the emission occurs for each line , and allowing rough comparisons between observed and predicted emission line flux ratios . the [ alvi ] , [ ar vi ] and [ siix ] lines observed here are included in their calculations . their figures assume an ionizing ultraviolet continuum similar to that of a typical seyfert galaxy with @xmath38 ergs@xmath4 , but distances scale as @xmath39 . @xcite and @xcite estimate ionizing luminosities 12 orders of magnitude higher in ngc 1068 . in the discussion below we assume an ionizing luminosity of 10@xmath40 ergs@xmath4 , which implies locations and dimensions of the emission regions are five times larger than those shown by @xcite , if all other parameters remain constant . the model then predicts that the distance and density for peak [ siix ] line emission cover 1.5100 pc and 10@xmath4110@xmath42 @xmath29 , respectively ( with the largest distances corresponding to the lowest densities ) . the observed distance of 20 pc ( 0.3 ) from the nucleus to the peak line emission corresponds to @xmath43@xmath110@xmath44 @xmath29 ) . if the actual peak is closer to the agn than observed but obscured by dust , then higher density gas would be required . for the [ alvi ] and [ arvi ] lines the adjusted @xcite models predict distances to the peak of roughly 15400 pc for the same density range . although the spatial distribution of the [ arvi ] line flux is less precise than the other lines , it appears to be similar to that of the other two lines , with a peak at the same location . clearly the observation that the offset from the agn to the peaks of these lines of with widely varying excitations are the same is incompatible with an ionized medium of uniform density , unless the extinction is highly non - uniform across the nucleus . the current observations can be roughly reconciled with a slab of density several times 10@xmath44 @xmath29 if there is increasing extinction of the coronal line emission closer to agn , of sufficient strength to mask the true peak of the [ siix ] line emission . if the coronal line emission region is symmetrically distributed to the north and south of the agn , the extinction would need increase further to the south of the agn in order to block the line emission there from view . evidence for higher extinction to the south is in fact widespread ( e.g. , * ? ? ? * ; * ? ? ? * and references therein ) . a rough lower limit to the additional extinction required to the south based on the assumption of a symmetric distribution of coronal line emission is two magnitudes at 3.93 @xmath0 m , which corresponds to approximately 70 visual magnitudes . if the highly non - uniform extinction postulated above is not present , only a density distribution of ionized gas strongly peaked 20 pc north of the agn could result in photoionization producing the observed maximum line emission from all of these ions at the same location . the calculations of @xcite also indicate that for these ions the regions of significant coronal line emission can extend to much larger distances ( @xmath1100 pc ) if gas densities decrease gradually with distance from the agn . the near agreement of our estimates of the total flux from the [ siix ] and [ arvi ] lines with measurements made by others in much larger apertures , imply that the ionized gas density must drop steeply at distance greater than @xmath150 pc from the agn . evidence supporting shock ionization in the narrow line regions of agns abounds @xcite . in the case of ngc 1068 , @xcite found that 1.5 north of the nucleus [ fe vii ] is enhanced at the location of the radio jet . they interpreted this as evidence of interaction between the jet and the surrounding medium . the radio jet proceeds outward from the nucleus for an angular distance of several arc - seconds , first nearly due north and then at position angle 35 degrees @xcite . more recently , @xcite and @xcite have studied the influence of the radio jet on the kinematics of [ o iii ] @xmath454959,5007 . they observed an increase in the radial velocity roughly proportional to distance from the nucleus followed by a linear decrease beyond @xmath1100 pc ( 1.4 ) and concluded that the magnitudes of these changes were so large that the neither can the jet be the sole driving force nor can gravity be the sole decelerating mechanism ( frictional forces must be significant ) . however , in the case of that line the deceleration occurs well outside of the region where we observe the higher excitation 35 @xmath0 m coronal lines . the line emission reported here originates in ionized regions much closer to the nucleus than those observed by @xcite and @xcite . nevertheless , evidence for shock excitation of the coronal lines is also readily apparent when one compares the 5 ghz image in fig . 1 of @xcite with our data . assuming , as they and @xcite have inferred , that source s1 is coincident with the agn , and hence with the infrared continuum peak , the peaks of [ siix ] and [ alvi ] line emission at 0.3 north are coincident with the bright radio knot c , where the jet bends to the northeast . this is highly suggestive of an interaction between the jet and gas in the nlr . also suggestive of an interaction at this location is a bright knot of h@xmath46 2.12 @xmath0 m line emission found by @xcite connecting two streamers of line emission , one of which they interpret as feeding the innermost few parsecs . the velocity profiles of the of [ siix ] line along the spatial direction , shown in figure [ fig : velocity ] , also appear to support a jet - nlr interaction . at 0.2@xmath47 north , a prominent blue shoulder is evident in the line profile . that structure also appears to be present 0.4 north , but it weakens significantly at 0.6 . the shoulder at 0.20.4 could be associated with ionized gas accelerated toward us by the interaction of the jet with the ambient gas near knot c. the line profile variations are somewhat reminiscent of the behavior of the [ fevii ] @xmath366087 line reported by @xcite , but the latter is at a much larger distance north of the agn and , also unlike the infrared lines , far from where that line has its peak intensity . if shocks are the dominant excitation mechanism for these coronal lines at radio knot c , then it seems likely that some of the remaining [ siix ] and [ alvi ] line flux outside of that knot originates in other localized regions and is associated with other radio knots . the knot labeled `` ne '' in fig . 1 of @xcite , located 0.4@xmath47 to the ne of knot `` c '' and not included in our slit , is a possible candidate for some of this missing line flux . however , judging from the large extent over which [ sivii ] was observed by @xcite , we do not think that shocks totally dominate over photoionization , and expect that some contribution of the latter is present , perhaps even at radio knot c. the infrared coronal line emission peaks 20 pc to the north of the agn for ions of a wide range of excitation , and extends to 50 pc in the highly excited [ siix ] line and presumably at least as far in the other lines . to the south there is no clear detection of any of the 35 @xmath0 m lines . increased extinction to the south can account for the north - south asymmetry , regardless of the line excitation mechanism . however , photoionization by the agn producing peaks at equal northern distances from the agn for such a large range of excitations requires a highly asymmetric gas distribution and/or a highly asymmetric distribution of obscuring dust to the north . this explanation appears contrived . evidence for collisional excitation of these lines is more compelling ; it consists of the spatial coincidence between radio knot c ( where the jet changes direction ) , a prominent knot of h@xmath46 line emission , and the emission peaks of all of the 35 @xmath0 m coronal lines . the presence at the same location of a localized prominent high velocity blue - shifted shoulder on the [ siix ] line provides further support . higher angular resolution studies of the coronal line emission would allow a more detailed comparison with the morphology of the radio emission . in his subsection we discuss that portion of the interstellar medium producing the hydrocarbon and silicate absorption features at 3.4 @xmath0 m and 9.7 @xmath0 m , but little or no absorption by gaseous co at 4.55.0 @xmath0 m . our discussion makes use of the silicate absorption feature as reported by @xcite ; its similar behavior across the nucleus as the 3.4 @xmath0 m feature provides a strong clue as to the nature of the interstellar medium . the subsection on co examines a number of possible explanations for its absence , and results in a similar conclusion regarding the ism . a summary is provided at the end of this section . it is important to note that in the direction along the ( ns ) slit the continuum source was partially resolved at ukirt and therefore we are assured that it has a characteristic dimension of a few tens of parsecs . thus the bulk of the material producing the 3.4 @xmath0 m absorption feature and not producing co absorption lies 10 pc or more from the central engine . dust and gas in the central few parsecs can only be a minor contributor to the present spectrum and our arguments as to the nature of the interstellar medium therefore apply to the more extended region and not to the core " or the innermost region of the torus observed via interferometry . the spectral profile of the 3.4 @xmath0 m feature is well characterized in the current data . overall the profile is smooth , showing only the two principal sub - features seen in galactic sources , those at 3.420 @xmath0 m and 3.485 @xmath0 m . the profile in ngc 1068 contains no evidence of a distinct 3.385 @xmath0 m sub - feature , which is sometimes present in galactic sources and sometimes blended with the 3.420 @xmath0 m feature @xcite . the data also reveal that the depth of the 3.4 @xmath0 m feature has a similar north - south variation across the nucleus as the 9.7 @xmath0 m silicate absorption ( fig . [ fig : chsil ] ) . both features are at maximum depth on and just to the south of the nucleus . the similarity is also quantitative ; the fractional increase in the depth of each feature at its maximum compared to the depth at adjacent locations is @xmath140 percent . _ this indicates that the carriers of the features are significantly mixed . _ the compactness and nuclear location of the region of enhanced 3.4 @xmath0 m and silicate extinction suggests that the extra absorption to the south occurs quite close to the nucleus . in the silicate feature the weaker off - nuclear absorption extends over 2 to the north and south . because the 3 @xmath0 m continuum source is much more compact , the depth of the 3.4 @xmath0 m feature could not be measured more than 0.4 distant from the nucleus . the lower level and more extended silicate feature largely arises in a circumnuclear region of much larger dimensions than the torus - like structures seen with mid - ir interferometry . it may be associated with material in the disk of the host galaxy , which crosses the line of sight south of the nucleus . the distinction between these two regions is not necessarily sharp @xcite . the ratio of the optical depths of the 3.4 @xmath0 m and silicate features at the infrared continuum peak of ngc 1068 is 0.19 @xmath15 0.02 . in the galaxy the silicate feature is found in both diffuse and dense interstellar clouds , whereas the hydrocarbon feature is found only in diffuse clouds . in diffuse clouds far from the galactic center the ratio of optical depths of the two features can be derived from their optical depths relative to @xmath48 , @xmath48/@xmath49 @xmath50 18.5 @xcite and @xmath48/@xmath51 @xmath50 250 @xcite , yielding @xmath52@xmath53@xmath49 = 0.07 @xmath15 0.01 , a value roughly forty percent of that measured in ngc 1068 . toward the galactic center , where @xmath48 @xmath50 30 mag ) , the value of @xmath52/@xmath49 averaged over sources observed by both @xcite and @xcite is 0.064 @xmath15 0.017 , the same as the above value to within the uncertainties . the ratio _ in just the diffuse gas on the galactic center sightline _ may be estimated as follows . @xcite find @xmath48/@xmath52 = 156 @xmath1516 and @xcite find @xmath49 = 3.6 @xmath15 0.3 . however , one - third of the visual extinction to the galactic center is believed to occur in dense clouds , located mostly in intervening spiral arms @xcite . following @xcite , for values of @xmath48@xmath2212 mag the ratio of visual extinction to silicate optical depth is the same in dense clouds as in diffuse clouds , and thus we associate those 10 mag of dense cloud visual extinction with a silicate optical depth of 0.54 . this gives @xmath52/@xmath49 @xmath50 0.23/3.06 = 0.075 for galactic center diffuse clouds , again far less than the ratio toward the nucleus of ngc 1068 . even if half of the silicate optical depth toward the galactic center arises in the dense clouds in the galactic disk , the above ratio would only be 0.15 , less than the value in ngc 1068 . clearly then , the 3.4 @xmath0 m absorption in ngc 1068 is unusually strong relative to the silicate absorption . this can not be explained away by placing the 3.4 @xmath0 m absorber in a more remote location than the silicate feature , because even in galactic diffuse clouds the optical depth of 0.08 observed in ngc 1068 , corresponds to a silicate optical depth @xmath541 , significantly larger than is observed . the high ratio could be the result of the silicate absorption being significantly filled in , as might arise if thermal emission close to the agn were absorbed in the cooler outer portions of the source . because such filling in would be much less for the 3.4 @xmath0 m feature , the similar variation in optical depths of the features across the nucleus tends to argue that this is not the entire explanation , however . as suggested by @xcite , the high ratio could be explained , at least in part , by the typical continuum surface at 3.4 @xmath0 m lying interior to the typical surface at 9.7 @xmath0 m such that the added pathlength , unobservable at 9.7 @xmath0 m , contains a significant fraction of the observed column density of the 3.4 @xmath0 m absorber . the result also could be a consequence of either unusual elemental abundances or the chemistry of the dust in the vicinity of the agn . the co fundamental band has been searched for previously without success in ngc 1068 at spectral resolutions ranging from 20 km s@xmath4 to 250 km s@xmath4 @xcite . the current data , obtained at 150 km s@xmath4 resolution , significantly tightens the limit on absorption by co in the warm and presumably rapidly moving gas just outside of the 35 @xmath0 m continuum - emitting dust , whose fwhm along the ionization cones is @xmath120 pc . in its lack of detectable co in this band ngc 1068 is similar to the five other type 2 seyferts studied by @xcite , although the limits on co absorption in those galaxies are considerably less strict . we consider three possible explanations for the lack of strong co absorption lines toward the agn in ngc 1068 : ( 1 ) the molecules are distributed among many rotational states ; ( 2 ) the molecules reside in a clumpy obscuring medium ; and ( 3 ) there is a paucity of dense cloud " molecular gas in front of the agn . to explain the non - detection of transitions of co , oh and h@xmath46co lines from low rotational levels towards the highly obscured nucleus of cygnus a at radio wavelengths , it has been pointed out that at the temperatures of the nuclear clouds the lowest rotational levels may contain only a small fraction of the molecules @xcite . for example , in lte at 300 k roughly twenty rotational levels of co have populations within a factor of ten of that of the most populated level , whereas at 25 k only the six lowest levels are so highly populated . it also has been suggested by these authors that under some circumstances a strong and compact nuclear radio source could redistribute the rotational level populations in the surrounding gas , further weakening the strengths of lines from the low rotational levels . @xcite find this explanation unlikely , however , because the 19% of their sources in which high - excitation oh lines are detected also show absorption from the ground state transition . in the case of ngc 1068 , whose 35 @xmath0 m continuum - emitting dust , and presumably the gas associated with this dust , are at temperatures of several hundred kelvins , the lte fractional populations in the low lying levels normally observed at radio and millimeter wavelengths would be much less than in galactic dense clouds . this can not be the explanation for the non - detection of the co fundamental band lines , however , as the @xmath13 band spectrum in fig . [ fig : spectrum ] encompasses lines with a wide range of lower state energies ( corresponding to @xmath26 = 130 in the case of the p branch ) . the second possibility is that , although there is a large amount of molecular material along the line of sight to the nuclear continuum source , it is concentrated into clumps that cover only a small fraction of the source . each individual clump produces deep co lines but as most of the continuum source is unobscured , the net effect is much weaker co absorption . such small cloud volume filling factors have been invoked in order to explain the mid - infrared observational properties of seyfert 1 and seyfert 2 galaxies in the unified model ( in particular the lack of silicate emission in the former and the broad spectral range of the far - infrared continuum ; * ? ? ? small filling factors for toroids surrounding agns are also predicted by the hydrodynamic simulations of @xcite . if the detectability of co absorption is related to the cloud filling factor , then for similar total extinction co should be detected in objects with smooth obscuration ( i.e. , covering the entire nuclear continuum source ) , but not in those with clumpy obscuration . @xcite propose that clumpy and smooth obscuration in agn can be separated on the basis of the strength of the 10 and 18 @xmath0 m silicate features . strong co absorption , in gas that is warm and presumably close to the nuclear source , has been detected in several agn - hosting ultraluminous infrared galaxies ( ulirgs ) @xcite , whose silicate features would classify their dust distribution as smooth , whereas in the single seyfert 2 ulirg whose spectrum would classify it as clumpy , only a weak co band has been tentatively detected @xcite . this may suggest that the strength of the molecular absorption in agn is indeed governed by the clumpiness of the obscuring medium , but a stringent test of this hypothesis will require a larger sample of objects spanning a wide range of dust geometries . nevertheless , the presence in ngc 1068 of the 3.4 @xmath0 m feature , whose spatial behavior suggests that much of it arises in the same gas as much of the silicate feature , close to the agn , raises immediate complications with the clumpy model . that model has the dust and gas local to the agn residing in dense clouds with a small filling factor , so that the silicate absorption , which is strong in each clump , is both diluted and filled in by emission @xcite . the co absorption lines would also be diluted by this arrangement , apparently to undetectability in the case of ngc 1068 . however , the absence of the 3.4 @xmath0 m band in galactic dense molecular clouds suggests that in ngc 1068 the carrier of that band would not reside in such clumps , and thus that a second , more diffuse medium is needed for it ( ignoring , for the moment the evidence that the carriers of the two dust features are mixed ) . the diffuse material could , for example , be a low - density interclump medium . however , the 3.4 @xmath0 m band in galactic diffuse clouds is invariably accompanied by a silicate absorption feature . as pointed out earlier , the previously estimated value of @xmath52/@xmath49 in the diffuse interstellar medium toward the galactic center , when applied to ngc 1068 , implies that most or all of silicate absorption in ngc 1068 must arise in that material carrying the 3.4 @xmath0 m band . the by now almost inevitable solution to these problems is the third explanation : at 313 @xmath0 m essentially all of the absorbing material in the line of sight has the chemical characteristics of galactic diffuse clouds . in this explanation clumpy absorbing material is not needed to account for the lack of detectable co , because in diffuse clouds co is mostly dissociated , comprising only one percent of the carbon . diffuse clouds do have large molecular components , with h@xmath5 accounting for typically half of the hydrogen ; thus h@xmath5 should be available to provide the line emission observed by @xcite . the dust grains in diffuse clouds are comprised of a refractory component responsible for the 3.4 @xmath0 m feature and a silicate component . gas densities in galactic diffuse clouds are 101000 @xmath29 . densities in the inner few tens of parsecs of ngc 1068 might be greater than this . however , at distances of only a few tens of parsecs from a luminous agn it may well be possible that somewhat denser gas also has the above characteristics . as discussed earlier , in the galaxy fairly tight relationships have been found between the visual extinction in the diffuse ism and the depth of the 3.4 @xmath0 m feature @xcite for sightlines to the galactic center and to local diffuse clouds . here we employ the former relationship , @xmath48/@xmath52 = 150 . using it we estimate a typical visual extinction of 12 mag to the 3.4 @xmath0 m continuum - emitting region , implying for a normal gas - to - dust ratio that @xmath55(h ) = 2.3 @xmath6 10@xmath7 @xmath8 . then , assuming that , as in galactic diffuse clouds , one percent of the carbon is in co , and also ( conservatively ) that the typical continuum surface at 3.4 @xmath0 m and 4.7 @xmath0 m are at the same depth , @xmath55(co ) = 6 @xmath6 10@xmath56 @xmath8 , which is consistent with the upper limit of @xmath57 10@xmath28 @xmath8 set by these observations . the above column density of hydrogen is much less than the column density that attenuates our view of the central engine , @xmath55(h ) @xmath14 10@xmath58 @xmath8 @xcite , but much of that gas must be interior to the dust that is emitting the observed 35 @xmath0 m continuum . we have found that the 3.4 @xmath0 m hydrocarbon and 9.7 @xmath0 m silicate bands have spatial variations that are similar across the central 0.2 @xmath6 0.8 arcsecond of ngc 1068 and that only a small fraction of interstellar carbon in this region can be in the form of co. the first finding indicates that the carriers of the 3.4 @xmath0 m and 9.7 @xmath0 m bands are largely mixed and that significant fractions of each are located in material very close to the nucleus . in the galaxy both features are present in diffuse interstellar clouds , but the 3.4 @xmath0 m band is _ only _ observed there . the fraction of co in diffuse clouds is small and our observed upper limit in ngc 1068 is consistent with such a location . measurable co absorption would be present if even a modest fraction of the silicate absorption arose in fully molecular material . the straightforward conclusion is that _ the 3 - 13 @xmath0 m absorption spectrum of ngc 1068 is produced in material chemically similar to the galactic diffuse interstellar medium . _ we reiterate the significance of the lack of detection of co and its incompatibility with models employing clumps of _ dense molecular _ material , even if such material near an agn is capable of producing the 3.4 @xmath0 m feature . if the filling factor of such fully molecular clumps were low enough to dilute " the co band to non - detection , as observed , similar dilutions would occur for the silicate and 3.4 @xmath0 m absorptions , contradicting the observations . the adaptive optics measurements of @xcite show that nuclear thermal infrared continuum arises in part from numerous clumps covering a few tens of parsecs . if the source of the 3 - 13 @xmath0 m continuum is dominated by such clumps , then absorbing material could either be located on surfaces of the clumps or be more continuously distributed in front of them . however , if a significant portion of the continuum is emitted by an extended component , the present observations constrain the degree of clumpiness in the absorbing material . in that case a low , e.g. @xmath220.1 , filling factor for the absorbing clumps probably would be inconsistent with the observed strength of the 3.4 @xmath0 m feature , as it would imply , e.g. , @xmath59(3.4 @xmath0 m ) @xmath14 0.8 in the clumps . the maximum observed optical depth of the feature is 0.5 , in the deeply buried ulirg iras 08572 + 3915 @xcite . the observations reported here bear largely on two important issues : ( 1 ) the ionization mechanism for the high excitation coronal line emission close to the agn ; and ( 2 ) the nature of the dusty environment absorbing the infrared continuum of the agn . regarding the ionized gas , the evidence appears to both favor collisional excitation of the bulk of the coronal line emission and , in view of the spatial coincidences of the various lines , argue against photoionization . regarding the absorbing material , the simplest explanation is that it has much more of the characteristics of diffuse clouds than dense clouds ( clumped or unclumped ) . the strength of the hydrocarbon feature argues against it being distributed in front of the sources of the continuum with a very low filling factor relative to them . because elsewhere it has always been found that silicate dust is always present where a 3.4 @xmath0 m absorption is observed , one can conclude that if the latter is not so distributed then neither is the former . finally , it should be noted that in ngc 1068 the extended source of the 35 @xmath0 m continuum radiation has a characteristic north - south dimension of @xmath120 pc , indicating that the bulk of the continuum emission occurs at distances of @xmath110 pc or more from the central engine . this distance is much greater than that estimated under the assumption that the temperature of the dust is determined principally by the luminosity of the agn and its distance . using an agn luminosity of 1.5 @xmath6 10@xmath60 l@xmath12 @xcite , dust grains heated by isotropic radiation from the agn to the observed 35 @xmath0 m color temperatures of @xmath1500 k would lie only @xmath11 pc from the agn . @xcite has suggested that transient heating of small grains could be responsible for the high temperatures . we speculate that at least in the north - south direction mechanical heating of the dust and gas , perhaps from the agn jets , could also be important . we thank the staff of the united kingdom infrared telescope , which is operated by the joint astronomy centre on behalf of the science and technology facilities council of the u.k . trg s and rem s research is supported by the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the science and technology facilities council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , ministrio da cincia e tecnologia ( brazil ) and secyt ( argentina ) . ara s research was partially supported by the brazilian funding agency cnpq ( 311476/2006 - 6 ) . we are grateful to n. a. levenson for useful discussions and to the anonymous referee for numerous helpful comments . ayani , k. , nakagawa , t. , & nagata , t. 1992 , in `` testing the agn paradigm ; proceedings of the 2nd annual topical astrophysics conference , '' ed . s. holt , s. neff , & c. m. ury , aip conference proceedings , 254 , 528 .

we find a class of non - supersymmetric multi - center solutions of the stu model of five - dimensional ungauged supergravity . the solutions are determined by a system of linear equations defined on a four - dimensional khler manifold with vanishing ricci scalar and a isometry . the most general class of such khler manifolds was studied by lebrun and they have non - trivial-cycles that can support the topological fluxes characteristic of bubbled geometries . after imposing an additional symmetry on the base we find explicit multi - center supergravity solutions . we show that there is an infinite number of regular multi - center solutions with non - trivial topology that are asymptotic to the near - horizon limit of a bmpv black hole . = 10000 * hair in the back of a throat : * * non - supersymmetric multi - center solutions from khler manifolds * * nikolay bobev , ben niehoff and nicholas p. warner + * simons center for geometry and physics + stony brook university + stony brook , ny 11794 , usa + department of physics and astronomy + university of southern california + los angeles , ca 90089 , usa + nbobev@scgp.stonybrook.edu , bniehoff@usc.edu , warner@usc.edu +

1103.0520

the construction of microstate geometries for black holes and black rings is now a fairly well developed subject . the defining idea of such geometries is that they have the same asymptotics at infinity as a given black hole or black ring and yet the solutions are completely smooth and , as a consequence , they resolve the black - hole or black - ring singularity into smooth `` bubbled geometries . '' more generally , it is interesting to see what classes of smooth , multi - centered geometries can be inserted into a black - hole , or black - ring throat since such additional structure may be thought of as `` hair '' on the black object . since the throat geometries are typically asymptotic to an anti de sitter space , such geometric `` hair in the back of a throat '' can be studied quite precisely using holography . one of the primary purposes in finding such backgrounds is not only to elucidate the possible microstate geometries but also to obtain a better semi - classical description of black - hole microstates in general . most of the progress on microstate geometries has been for bps solutions @xcite because such solutions are generically far simpler and are , in fact , governed by a linear system of differential equations @xcite . however , recent work has shown that there are also linear systems of equations that govern very large families of extremal , non - bps solutions @xcite . these families not only include most of the known extremal , non - bps black holes and black rings but also greatly generalize these solutions by including more charges and extending the results to completely new non - bps , multi - centered black - ring solutions . the construction of such solutions is a very important first step in understanding the structure and properties of non - bps solutions in general , particularly since such solutions have played a pivotal role in the analysis of attractor flows @xcite , black - hole bound states , deconstruction @xcite , wall crossing and entropy enigmas @xcite . ultimately one would like to find the non - bps analogs of bubbled solutions in which the singularity of the underlying black object has been resolved and the black - hole , or black - ring , throat is rounded off in some form of smooth geometry . such solutions have come to be known as _ microstate geometries _ and the requisite geometric transition to the bubbled geometries that lies at the heart of such objects has been found and extensively studied for bps solutions @xcite . however , there are very few examples of the corresponding smooth , non - bps geometries @xcite . indeed , there are rather few examples of non - bps solutions with more than one non - trivial bubble in a black - hole background . one of the purposes of this paper is to investigate an interesting new class of bubbled geometries and to find new non - bps bubbled solutions . a crucial first step in finding bubbled geometries is to find smooth , four - dimensional `` base geometries '' with non - trivial homology ( the `` bubbles '' ) that supports the cohomological fluxes that give rise to the charges of the solution . in bps solutions ( and in some non - bps solutions ) these base geometries are necessarily hyper - khler manifolds and are usually chosen to be asymptotic to @xmath4 or @xmath5 so that the final , five - dimensional geometry is asymptotic to five - dimensional , or four - dimensional , minkowski space . there are a rich variety of such admissible base geometries because the metrics are allowed to be ambi - polar , that is , the base metric is allowed to reverse sign from regions with signature @xmath6 to regions with signature @xmath7 . in spite of this apparent peculiarity , the presence of very non - trivial warp factors means that such base metrics can give rise to perfectly viable , smooth , lorentzian five - dimensional geometries . the ambi - polar generalizations of gibbons - hawking ( gh ) metrics @xcite have proven extremely valuable in that they have given rise to extensive and highly computable classes of bps microstate geometries . it was one of the important realizations of @xcite that one could relax the hyper - khler condition on the base geometry and still obtain non - bps solutions from linear systems of equations . in particular , it was shown in @xcite how this could be achieved by starting from a four - dimensional euclidean solution to einstein s equations coupled to an electromagnetic field . this led to some interesting examples based upon israel - wilson metrics but it turns out that many of these naturally arise as spectral flows of more standard non - bps solutions based upon gibbons - hawking metrics . one can also construct a regular non - bps solution with an euclidean kerr - newman base @xcite , however the base has only one topological two - cycle and there is no obvious way to generalize it easily . a natural follow - up question is whether there are any other classes of euclidean electrovac solutions that contain appropriate non - trivial homology and that can be used to generate non - bps microstate geometries . it turns out that there is a relatively straightforward generalization of the gibbons - hawking metrics to khler electrovac solutions and these geometries were studied extensively by lebrun in @xcite . the central result in @xcite is to find the explicit local form of all euclidean , four - dimensional khler metrics that have a @xmath0 isometry and a vanishing ricci scalar . it is then shown in a follow - up paper , @xcite , that these solutions are necessarily electrovac solutions whose electromagnetic field is related to the khler form . this generalizes earlier work on the classification of hyper - khler metrics with @xmath0 isometry @xcite in which it is shown that the metric is determined by a single function that is required to satisfy the affine toda equation . the gibbons - hawking metrics then emerge as precisely the metrics for which the @xmath0 action preserves all three complex structures ( that is , the @xmath0 is tri - holomorphic ) . the `` lebrun metrics '' in @xcite are defined by two functions , one of which must satisfy the affine toda equation and the other of which must essentially be harmonic in a background defined by the affine toda solution . this family of solutions collapses to either the general class of @xmath0-invariant hyper - khler metrics , or to the gh family if one makes an essentially trivial choice for one of the two functions . amongst the general class of lebrun metrics is what we will call the `` lebrun - burns metrics , '' which provide a simple , explicit class of khler metrics on @xmath8 with @xmath9 points blown up s . ] . the end result is rather similar to the gibbons - hawking metrics except that the @xmath10 sections and the harmonic functions on @xmath10 are now replaced by the hyperbolic space , @xmath11 , and its harmonic functions . the lebrun - burns metrics are also asymptotic to @xmath4 and , interestingly enough , the @xmath0 isometry on this @xmath4 does _ not _ act in a manner that matches the tri - holomorphic @xmath0 action on a gibbons - hawking metric that is similarly asymptotic to @xmath4 . thus the solutions obtained from the lebrun - burns metrics will be intrinsically different from those obtained from gh solutions . in this paper we explicitly solve the equations of motion of five - dimensional supergravity on an axisymmetric lebrun - burns base and thus provide an infinite class of non - supersymmetric multi - centered solution . we find that , due to the maxwell flux on the four - dimensional khler base , the five - dimensional backgrounds are not asymptotically flat and have the asymptotics of a warped , rotating @xmath12 space . for certain choices of parameters this becomes the near horizon limit of a bmpv black hole @xcite . we would like to emphasize that although our approach to finding these solutions was inspired by the construction of bps microstate geometries we find the most general axisymmetric solutions within the floating brane ansatz @xcite and with an axisymmetric lebrun - burns base . in particular our solutions include superpositions of non - supersymmetric concentric black rings and other potentially interesting solutions with horizons . in section 2 of this paper we review the process through which one can construct five - dimensional non - bps solutions using the floating - brane ansatz @xcite . in section 3 we introduce the general lebrun metrics and begin solving the linear system in this background , and in section 4 we specialize to the lebrun - burns metrics , discuss their properties and further reduce the linear system and exhibit all the necessary green functions . we will also show , in section 4 , that because the maxwell field on the four - dimensional base involves the khler form , the energy - momentum tensor does not fall off at infinity and hence the lebrun - burns base metrics do not naturally lead to five - dimensional solutions that are asymptotic to flat space . in section 5 we start with the simplest possible lebrun - burns metric , @xmath13 , and use the linear system and green functions of section 4 to construct very simple , explicit examples of non - bps solutions . we find that the natural asymptotic geometries that arise from lebrun - burns base metrics are a warped , rotating @xmath12 space . in section 6 we consider more general lebrun - burns base metrics with non - trivial topology and find the general axisymmetric solution to the system of non - bps equations on such a base . five - dimensional regularity and the absence of closed time - like curves ( ctcs ) puts very stringent conditions on our solutions but in spite of this we find a family of regular solutions with non - trivial bubbles that are asymptotic to the near - horizon region of a bmpv black hole . in section 7 we present our conclusions and suggest some directions for further work . some of the technical details of the construction of the solutions are relegated to the appendix . it is simplest to characterize our solutions in terms of @xmath14 , five - dimensional ungauged supergravity with three @xmath0 gauge fields . this theory may also be thought of as arising from a truncation of eleven - dimensional supergravity on @xmath15 . the bosonic action is given by : @xmath16 where we use the conventions of @xcite . the matrix that defines the kinetic terms can be written as : @xmath17 the scalar fields themselves are not independent and are most conveniently parametrized in terms of three other scalar fields , @xmath18 : @xmath19 which satisfy the constraint @xmath20 . the reason why we use three scalars , @xmath18 , to parametrize two independent scalar fields becomes evident once we write the metric ansatz @xmath21 and define @xmath22 by : @xmath23 this turns out to be an extremely convenient way to express both the scalar fields and the warp factors . the electromagnetic fields are given by the `` floating brane '' ansatz of @xcite , which relates metric coefficient and scalar fields to the electrostatic potentials . in particular , the maxwell potentials are given by : @xmath24 where @xmath25 is a one - form on the base @xmath26 . it is convenient to introduce the magnetic two - from field strengths @xmath27 the four - dimensional base space , @xmath28 , has to be a solution of euclidean einstein - maxwell equations @xmath29 where all quantities are computed in the four - dimensional base metric . the maxwell field is then decomposed as : @xmath30 where @xmath31 is self - dual and @xmath32 is anti - self - dual . the equations of motion @xmath33 imply that @xmath31 and @xmath32 are harmonic . as the notation implies , this defines the magnetic two - from field strength @xmath31 . the linear system that solves the equations of motion can now be written as @xcite : @xmath34 \ , , \qquad \big(\theta^{(2 ) } - * _ 4 \theta^{(2 ) } \big ) ~=~ 2 \ , z_1 \ , \omega^{(3)}_- \ , , \label{zmax1 } \\ \hat \nabla^2 z_2&= & * _ 4 \big [ \theta^{(1 ) } \wedge \theta^{(3 ) } \big ] \ , , \qquad \big(\theta^{(1 ) } - * _ 4 \theta^{(1 ) } \big ) ~=~ 2 \ , z_2 \ , \omega^{(3)}_- \label{zmax2}\,,\end{aligned}\ ] ] and @xmath35 \ , , \label{z3eqn } \\ dk ~+~ * _ 4 dk & = & \frac{1}{2 } \ , \sum_i \ , z_i \ , \big(\theta^{(i ) } + * _ 4 \theta^{(i ) } \big ) \ , , \label{keqn } \end{aligned}\ ] ] where @xmath36 is the laplacian on @xmath28 . indeed , having chosen the electrovac solution that defines the base metric , one uses to read off @xmath31 and @xmath37 . as a result , and are two linear coupled equations for @xmath38 and @xmath39 and @xmath40 and @xmath41 respectively . once these equations are solved , @xmath42 and @xmath43 are solutions to the system of linear equations and . our purpose here is to implement this procedure for the lebrun metrics . the lebrun metric , @xcite , is @xmath44 where @xmath45 and @xmath46 are two functions of @xmath47 which obey the @xmath48 toda equation and its linearized form : @xmath49 the one - form , @xmath50 , satisfies : @xmath51 and the integrability of this differential , @xmath52 , is equivalent to the equation ( [ lineartoda ] ) . the metric is khler , with khler form : @xmath53 it is convenient to introduce frames : @xmath54 and the self - dual forms @xmath55 we will also frequently denote the coordinates by @xmath56 . one can then verify that : @xmath57 is harmonic if and only if @xmath58 satisfies ( [ lineartoda ] ) : @xmath59 note that if one differentiates ( [ toda ] ) with respect to @xmath60 one finds that @xmath61 satisfies ( [ harm1 ] ) and ( [ toda ] ) . define the maxwell field , @xmath62 , by @xmath63 the metric , ( [ lbmet ] ) , is then a solution to the einstein - maxwell equations ( [ electrovacequation ] ) . this fits the form of the linear system obtained in @xcite and so we take : @xmath64 one can now use this in the linear system , , and to solve the equations of motion of supergravity . note that we can absorb the constant @xmath65 by rescaling the coordinate @xmath66 and the function @xmath46 ( which in turn rescales the one - form , @xmath50 ) . throughout the rest of this paper , we set @xmath67 . one finds that the first part of the system is solved by an ansatz : @xmath68 @xmath69 the bianchi identities imply the following relations @xmath70 one can add arbitrary functions of @xmath60 alone to the @xmath71 , but these can be absorbed by a shift @xmath72 , for some function @xmath73 . the functions @xmath74 and @xmath75 can be any solution of ( [ lineartoda ] ) , that is : @xmath76 and given these solutions the functions @xmath77 and @xmath78 are determined by the linear equations : @xmath79 as usual one makes the ansatz @xmath80 where @xmath81 is a one - form on the three - dimensional base . one then finds that the solution may be written as : @xmath82 @xmath83 the functions @xmath84 and @xmath85 must then satisfy the linear equations : @xmath86 \ , . \label{lin3b } \end{aligned}\ ] ] finally , the components of @xmath87 are determined from : @xmath88 @xmath89 the integrability of these equations for @xmath87 follows from the differential equations satisfied by all the other functions . the complete solution is thus obtained by first choosing a background metric , ( [ lbmet ] ) , with functions @xmath45 and @xmath46 that satisfy ( [ toda ] ) and ( [ lineartoda ] ) . the choice of background fixes one of the three electromagnetic fields , via ( [ kform])and ( [ maxf ] ) and , in particular , determines @xmath31 via ( [ max3 ] ) . one then finds @xmath74 and @xmath75 as solutions to the homogeneous equations ( [ leqns1 ] ) and uses these solutions as sources in the linear equations , ( [ lin2a ] ) and ( [ lin2b ] ) , for @xmath77 and @xmath78 . these functions then determine the remaining magnetic fluxes , @xmath90 , and the electrostatic potentials , or warp factors , @xmath91 , @xmath92 from ( [ thetaforms ] ) , ( [ z12form ] ) , ( [ pfns1 ] ) and ( [ pfns2 ] ) . next one solves the linear equation , ( [ lin3a ] ) , for @xmath85 with a source determined by @xmath93 and then solves the linear equation for @xmath84 , ( [ lin3b ] ) , whose source is made from @xmath94 and @xmath85 . the last step is to use all of these functions to solve the linear , first order system , ( [ omx])([omz ] ) for @xmath87 and obtain @xmath43 and the angular momentum vector from ( [ z3form ] ) , ( [ muform ] ) and ( [ kansatz ] ) . while complicated , this is a linear system of equations once one has chosen a solution for the metric function , @xmath45 . before concluding , it is worth noting that taking the function , @xmath45 , to be a constant is a trivial solution to ( [ toda ] ) and then ( [ lineartoda ] ) becomes the harmonic equation on @xmath10 and the metric reduces to the familiar class of gibbons - hawking metrics . similary , if one takes @xmath95 then it also satisfies ( [ lineartoda ] ) , as can be seen by differentiating ( [ toda ] ) with respect to @xmath60 . this yields the general class of hyper - khler metrics with a non - triholomorphic @xmath0 isometry @xcite , which are based upon the affine toda equation . the resulting system of equations for non - bps solutions in five dimensions does not , however , reduce to that considered in the bubbling bps solutions @xcite but is rather more akin to the fluxes considered in @xcite . this is because the flux background is a mixture of self - dual and anti - self - dual fluxes and these break supersymmetry . in particular , the anti - self - dual flux is non - normalizable since it is proportional to the complex structure . thus even the simple gibbons - hawking and toda limits of the lebrun backgrounds extend the class of solutions considered thus far . the lebrun solutions are four - dimensional euclidean einstein - maxwell solutions and it is natural to ask whether some of them preserve supersymmetry . supersymmetric solutions of four - dimensional euclidean einstein - maxwell theory were classified in @xcite . the maximally supersymmetric solutions are @xmath13 or @xmath96 . there are two classes of solutions which preserve half of the supersymmetries - the well - known gibbons - hawking solutions and the euclidean israel - wilson metrics discussed in @xcite . therefore the classification of @xcite also demonstrates that the general lebrun solutions , albeit khler , are non - supersymmetric solutions of einstein - maxwell theory . the lebrun - burns metrics represent a very natural generalization of the gibbons - hawking metrics in that they are four - dimensional khler metrics with @xmath9 2-cycles and associated moduli and they satisfy einstein s equations coupled to a @xmath0 gauge field . they were constructed @xcite ( see also [ 3 ] as cited in @xcite ) as explicit metrics on a decompactification of @xmath97 , the connected sum of @xmath9 @xmath98 s . the simplest class of metrics arises if one takes : @xmath99 and it is then convenient to reparametrize by defining : @xmath100 the lebrun - burns metric can then be written as @xmath101 \label{lbb2}\,.\ ] ] note that the three - dimensional metric is the standard constant - curvature metric on the hyperbolic plane , @xmath102 : @xmath103 the equations ( [ lineartoda ] ) and ( [ adefn ] ) that define the four - dimensional base imply that @xmath104 is a harmonic function on the hyperbolic plane and that @xmath50 is an appropriate one - form on @xmath102 : @xmath105 to avoid a conical singularity at @xmath106 , one must have @xmath107 at this point so that the metric in the @xmath108 direction limits to that of @xmath109 in polar coordinates . thus the metric in the neighborhood of @xmath106 is that of @xmath4 and regularity requires that one restrict the space to @xmath110 . similarly , if one requires @xmath107 at infinity , the space is asymptotic to @xmath111 . note that the circle defined by @xmath66 lies in an @xmath109 plane of the @xmath4 , and the associated isometry therefore only commutes with another @xmath0 factor in the generic @xmath112 holonomy of the base metric . this is quite different from the way in which the isometry associated with the @xmath0 fiber behaves in gh spaces . the green functions of the laplacian on @xmath102 are the functions : @xmath113 where one should remember that @xmath110 on @xmath102 and so this function only has one singularity in the domain of definition . the constant has been added so that @xmath114 vanishes at infinity . given @xmath114 , we can then solve for @xmath50 in . putting @xmath115 , we obtain @xmath116 one can then take : @xmath117 @xmath118 with these choices and @xmath119 , the lebrun - burns metric is a smooth khler metric on @xmath8 blown up at @xmath120 points . it is thus a khler , electrovac generalization of the gibbons - hawking metrics . near @xmath121 , one has @xmath122 @xmath123 and the metric ( [ lbb2 ] ) behaves as : @xmath124 \nonumber \\ & = & c_j\ , q_j \big [ d\rho^2 + \coeff{1}{4 } \rho^2 ( d \theta^2 + \sin^2 \theta \ , d \phi^2 + q_j^{-2 } \ , ( d\tau+\cos \theta \ , d\phi)^2 ) \big ] \ , , \label{metasymp}\end{aligned}\ ] ] where we have introduced spherical polar coordinates about @xmath121 and made a change of variable @xmath125 . thus near the singular points of @xmath104 , the metric is locally @xmath126 , and hence may be viewed as regular in string theory . at infinity one has : @xmath127 @xmath128 and hence @xmath129 and @xmath130 , and the metric is asymptotic to @xmath111 for @xmath131 . exactly as in gibbons - hawking geometries , the lebrun - burns metrics have non - trivial two - cycles defined by the @xmath0 fibers over any curve between the poles of @xmath104 . more specifically , the @xmath0 fiber ( defined by @xmath66 ) taken over a generic line interval in the @xmath102 base describes a cylinder . however , if one runs this interval between two poles of @xmath104 at points , @xmath132 and @xmath133 then the fiber is pinched off at the ends and the result is essentially a topological two - sphere . the asymptotic behavior of the metric at each end of the interval , ( [ metasymp ] ) , means that this two - sphere may , in fact , be modded out by some discrete group that depends upon the values of @xmath134 and @xmath135 . the two - cycles defined in this way will be denoted as @xmath136 and are depicted in fig . [ cycles ] . fiber along a path , in @xmath102 , between any two poles of the potential , @xmath104 . the fiber is pinched off at the poles . here the fibers sweep out a pair of intersecting two - cycles.__,height=113 ] the periods of these cycles are trivial to compute using ( [ kform ] ) : @xmath137 where @xmath138 denote the @xmath60-coordinates of the corresponding poles of @xmath104 . the maxwell fields , @xmath41 , @xmath39 and @xmath31 defined in ( [ thetaforms ] ) and ( [ max3 ] ) have components along the fiber of the form @xmath139 where @xmath77 and @xmath78 satisfy ( [ lin2a ] ) and ( [ lin2b ] ) and @xmath140 from this it follows that these fields have fluxes @xmath141 note , in particular , that for the lebrun - burns metric @xmath142 which vanishes at all the @xmath132 . therefore @xmath31 has _ no non - trivial _ fluxes on the compact two cycles . on the other hand , the complete maxwell field , @xmath62 , ( [ maxf ] ) , does have non - trivial fluxes because it has an anti - self - dual component involving @xmath143 . @xmath144 there is potentially a similar contribution from the `` electric parts '' of the complete gauge fields defined in ( [ aansatz ] ) . specifically , the complete maxwell fields , @xmath145 , have a component along the fiber : @xmath146 this will generically lead to fluxes of the form @xmath147 however , regularity of the metric and the absence of closed time - like curves typically requires that @xmath18 be finite and @xmath148 vanish at the points @xmath132 . thus these fluxes are usually zero . in summary , the bubbled non - bps solutions generically have non - vanishing fluxes for all three maxwell fields but the self - dual part of the third maxwell field , @xmath31 , will have trivial fluxes . the differential operators of interest are : @xmath149 note that @xmath150 is simply the laplacian on @xmath102 . the operator , @xmath151 has been introduced for later convenience because it appears in the equations of motion and it is also useful to note that it has a simple geometric interpretation . observe that the laplacian on @xmath152 may be written as @xmath153 where @xmath154 are cartesians on the first @xmath109 and @xmath155 are polars on the second @xmath109 . thus solving equations that involve @xmath151 may simply be viewed as looking for @xmath156-independent solutions with the flat laplacian on @xmath4 . the equations and solutions that involve @xmath151 are thus extremely familiar from the extensive literature on black rings . in particular , it is useful to note that the following are green functions for @xmath151 : @xmath157 at infinity one has : @xmath158 to solve the linear system , one first solves the homogeneous equations : @xmath159 and then uses these solutions in the equations that define the magnetic fluxes and the angular momentum function , @xmath85 : @xmath160 the last step is to use these solutions in : @xmath161 the physical functions now have the form @xmath162 @xmath163 the equations for @xmath164 reduce to @xmath165 this system of equations has a gauge invariance that leaves the physical solution completely invariant . see appendix [ gauge invariance ] for details . ideally one would like to find solutions that are asymptotically flat and for the metric ( [ metansatz ] ) this means that one must have @xmath166 at infinity . however , this is generically not possible because the maxwell fields , @xmath62 , in ( [ fdecomp ] ) and @xmath167 in ( [ thetaforms ] ) involve the khler form and thus the norm of these fields does not vanish at infinity . to see this , suppose that the @xmath18 go to constants at infinity , then the space - space and time - time components of the full five - dimensional einstein s equations imply that , at infinity , @xmath168 @xmath169 the self - dual parts of the @xmath170 can be made to vanish at infinity , and so can be satisfied . however , the left - hand - side of is positive - definite , and so this equation can not be satisfied because the @xmath167 always have a non - vanishing anti - self - dual part ( given by the khler form @xmath143 ) . this is in contrast to the solutions based on gibbons - hawking metrics for which the @xmath170 vanish at infinity . one can not , therefore , arrange to have @xmath166 at infinity . on the other hand , we will now show that there are solutions that are asymptotic to the near - horizon limit of a bmpv black hole . here we consider solutions in which the base space is completely flat , taking @xmath171 in the lebrun - burns metric . one of the purposes in doing this is to see what kind of asymptotic geometries and black - object geometries can be generated from the lebrun - burns metric using the solution technique of section [ burnssec ] . indeed , we will show that the natural boundary conditions correspond to the near - horizon regions of black holes and black rings . it is also important to note that even though we have set @xmath172 and thus trivialized the metric on the base , the maxwell field , @xmath62 , is still non - zero but is now purely anti - self - dual has vanishing energy - momentum tensor , consistent with the flatness of the base . ] and proportional to the complex structure , @xmath143 . similarly , the other maxwell fields ( [ thetaforms ] ) have both anti - self - dual and self - dual parts on the base . this will generically mean that supersymmetry is completely broken and that the solutions we get will be non - bps . perhaps the simplest non - trivial solution is a spherically symmetric one , whose sources necessarily lie at @xmath173 . in addition we set some of the electric potentials to zero : @xmath174 it is also convenient to introduce polar coordinates in @xmath109 and @xmath4 : we already have @xmath175 and @xmath66 in one copy of @xmath109 and so we define here is not the same as the one in . ] @xmath176 the functions @xmath177 and @xmath85 are then homogeneous solutions to @xmath178 and the spherically symmetric solutions are proportional to @xmath179 ( see ( [ fdefn ] ) ) . we therefore take @xmath180 where @xmath181 and @xmath182 are constant parameters . it is easy to see that one can satisfy ( [ l3eqn1 ] ) by taking : @xmath183 for some function , @xmath184 . the natural choice for @xmath184 is the function @xmath114 in ( [ gdefn ] ) , but this vanishes for @xmath185 , and one must take a limit : @xmath186 one then has @xmath187 the last step is to solve for @xmath87 , for which we can choose the gauge @xmath188 . equations ( [ omx])([omz ] ) then reduce to : @xmath189 for which the solution is : @xmath190 where the homogeneous solutions have been chosen so that @xmath164 goes to zero at infinity . the five - dimensional metric is then : @xmath191 where @xmath192 the conditions for absence of causal pathologies for solutions of our ansatz are discussed in appendix [ causality ] . for the simple solution in this section there is no dirac - misner string in @xmath164 and the condition for absence of ctc s is that all constants @xmath182 , @xmath193 , @xmath194 are non - negative and @xmath195 for generic choice of parameters satisfying the metric has the form of a warped rotating @xmath12 . the general solution has unequal angular momenta in each @xmath109 , and has a distorting warp factor , function , @xmath196 . for the special choice @xmath197 the function @xmath198 becomes a constant and the two angular momenta become equal . the metric then is precisely the near horizon limit of the bmpv black hole @xcite . it is worth emphasizing that the bmpv black hole ( and its near horizon limit ) is a supersymmetric solution of supergravity whereas our solution has anti - self - dual flux that breaks supersymmetry . to get the black ring ( or supertube ) generalization of the foregoing solution , we simply need to use the source functions ( [ gdefn ] ) and ( [ fdefn ] ) with @xmath199 . we can , without loss of generality , take @xmath200 since we are going to consider a single source . we therefore start by taking : @xmath201 for some constants @xmath202 . the supertube will thus be located at @xmath203 . the functions @xmath177 now have a source part and a homogeneous part that is proportional to @xmath58 : @xmath204 and hence @xmath205 where , once again , @xmath193 and @xmath206 are constants . the next step is to solve ( [ meqn1 ] ) and ( [ l3eqn1 ] ) for @xmath85 and @xmath84 , and as in ( [ bhsol3 ] ) it is convenient to shift @xmath84 by @xmath85 . one finds @xmath207 it is straightforward to show that this and ( [ meqn1 ] ) are satisfied by : @xmath208 + \ell_3 \ , g \ , , \label{brsol6 } \\ m & = & - \frac { \ell_1 \ , \ell_2}{2 } \ , \big ( \zeta^{-2 } \ , g^2 ~-~ 4 \ , \rho^2\ , h^2 \label{brsol7 } \big ) ~+~ \mu_0\ , h \ , , \end{aligned}\ ] ] where the constants @xmath209 and @xmath210 multiply homogeneous solutions to the relevant differential equations . combining these results in ( [ z3form ] ) and ( [ muform ] ) one obtains : @xmath211 \ , . \label{brsol9}\end{aligned}\ ] ] the last step is to solve for ( [ omx])([omz ] ) for @xmath164 and one can easily verify that : @xmath212 \ , d \phi \ , , \label{brsol10}\ ] ] where a constant of integration has been adjusted so that @xmath164 goes to zero at infinity . the green functions , @xmath114 and @xmath58 , and the metric that we construct here are , of course , familiar from the standard description of black rings . this is explained further in appendix [ cancoords ] . the first step in checking regularity of the metric is to look for closed time - like curves ( ctc s ) in the @xmath66 and @xmath213 directions near the ring . indeed there is a divergent negative coefficient of @xmath214 unless one requires : @xmath215 which we will use to eliminate @xmath216 . this condition is a familiar regularity requirement for the three - charge , two - dipole charge supertube @xcite . once this term is dealt with there is a sub - leading divergence that then requires @xmath217 since the radius of the supertube is related to @xmath218 we can interpret this relation as a radius formula for the supertube . there is a dirac - misner string in @xmath164 that leads to ctc s for @xmath219 and @xmath220 , and this requires that @xmath221 and so we must fix the homogeneous solutions by taking : @xmath222 with these choices the five - dimensional metric simplifies significantly and it is convenient to introduce two manifestly non - negative functions : @xmath223 and define @xmath224 the angular momentum vector simplifies to : @xmath225 \ , , \label{kang1}\ ] ] and the five dimensional metric may be written as : @xmath226 this form of the metric is reminiscent of the decoupling limit " for supersymmetric black rings / supertubes discussed in @xcite . it would be interesting to dualize our solutions to the d1-d5 duality frame and compare in more detail with the background in @xcite . this metric has no ctc s but is singular at the location of the supertube . this is a generic property of the supertube with two dipole charges and is familiar from the corresponding supersymmetric solutions @xcite . at infinity one has : @xmath227 @xmath228 \ , , \label{asymp2}\ ] ] and hence : @xmath229 the five - dimensional asymptotic metric , ( [ brmet ] ) , now takes the form : @xmath230 which is precisely of the form discussed in section [ flatsol1 ] , _ i.e. _ a warped form of rotating @xmath231 . for the special choice @xmath232 we have @xmath233 and reduces to the near horizon bmpv metric . therefore we can consider the metric as the metric to which all black hole , black ring and multi - center solutions within our ansatz will asymptote for @xmath234 . we will describe this in more detail in the next section . we have been able to solve in complete generality the system of differential equation , , , and on an axisymmetric lebrun - burns base . this provides an infinite class of explicit five - dimensional multi - centered solutions with ( at least ) one time - like and two space - like killing vectors @xmath235 . amongst our solutions are multi - center generalizations of the solutions in section [ flatsol1 ] and [ flatsol2 ] as well as a class of regular bubbled geometries that we discuss in some detail in section [ burnsasymptotics ] below . we will look for solutions on an axisymmetric lebrun - burns base in which the geometry at infinity has the form ( [ bhmet ] ) . the singular points of the harmonic function , @xmath104 , that determines the lebrun - burns base are located along the @xmath175 axis at points @xmath236 : @xmath237 where for convenience we have defined @xmath238 where we will assume that @xmath239 . it is trivial to solve for the functions @xmath74 and @xmath75 @xmath240 solving and for @xmath241 and @xmath85 one finds @xmath242 after a somewhat tedious exercise , @xmath85 , @xmath84 and @xmath243 are collected in appendix [ useidents ] . ] one can also solve equation @xmath244 where we have defined @xmath245 the one form @xmath246 is given by @xmath247 where @xmath248 is a constant which should be fixed so as to avoid ctcs and dirac - misner strings . substituting , , , and in the expressions for @xmath38 , @xmath40 , @xmath43 and @xmath148 , and , one finds the most general non - bps solution on an axisymmetric lebrun - burns base captured by the floating brane ansatz of @xcite . for easy comparison with the solution in section [ flatsol1 ] we have chosen to single out the terms in the solution which have poles at @xmath249 , _ i.e. _ the terms with coefficients involving @xmath193 , @xmath194 , @xmath194 and @xmath182 . in addition to the parameters @xmath193 , @xmath194 , @xmath250 and @xmath182 , the solution in general has @xmath251 parameters : @xmath252 . as we will see in the next subsection imposing regularity and absence of causal pathologies will greatly reduce the number of independent parameters . the solution we construct here will be asymptotic to the metric , which can be viewed as the elementary " solution within our ansatz . these regular solutions on a base with non - trivial topology can be viewed as a non - supersymmetric generalization of the bps bubbled solutions of @xcite . we begin by defining a radial coordinate around each of the poles of the harmonic functions @xmath253 we will be interested in constructing a solution that is regular at the locations of the poles of the harmonic functions , @xmath254 , and is free of ctcs and dirac - misner strings . for @xmath254 we have the following expansion of the harmonic functions @xmath255 since we are looking for a regular bubbled solution in five dimensions we will assume that all functions in the solution have the same singular points ( excluding the point @xmath249 which , as discussed in the previous section , will be treated separately ) . the functions @xmath38 and @xmath40 near a singular point , @xmath254 , diverge as @xmath256 to ensure regularity we should set @xmath257 the function @xmath43 near a singular point , @xmath254 , is @xmath258 the last term in the expression above is divergent and could be made to vanish only for @xmath259 . therefore for a regular @xmath43 one should set @xmath260 it is not hard to show that with this choice of constants the function @xmath148 will limit to a constant near a singular point . the condition for absence of ctc s , where @xmath261 is defined in appendix [ causality ] . ] requires that @xmath148 should vanish at a singular point of @xmath104 and this leads to the constraint : @xmath262 there is one more condition on the constant parameters of the solutions coming from removing a possible dirac - misner string in @xmath164 . however it turns out that after setting @xmath263 the absence of a dirac - misner string in @xmath164 is guaranteed by . to summarize , the conditions for regularity and absence of ctc s and dirac - misner strings near the poles of the harmonic functions requires that we set : @xmath264 note that these conditions are quite different from the regularity and causality constraints for bps bubbled solutions with a gh base @xcite . in particular for the class of bubbled solutions discussed here there is no analogue of the bubble equations " ( or integrability conditions ) familiar from the supersymmetric multi - center solutions @xcite . however we still have an equation that fixes the locations of the poles in the harmonic functions ( but not the distance between them ) in terms of the parameters @xmath265 . our analysis so far does not guarantee the regularity of the supergravity scalars ( _ i.e. _ the khler moduli of the tori in m - theory ) and the absence of causal pathologies at asymptotic infinity . to ensure that we should study the behavior of the solution at @xmath266 . the harmonic functions have the following expansion @xmath267 imposing the regularity and causality constraints at @xmath268 one finds the following constraints on the parameters of the solution : @xmath269 the constraints are easily solved by imposing @xmath270 , however there are in principle other ways to satisfy the relations in , so we will not commit to a specific solution . the asymptotic expansion ( @xmath234 ) of the metric functions in the solution is @xmath271 the constraints together with lead to @xmath272 it is clear that at @xmath266 these regular bubbled solutions are asymptotic to the warped , rotating @xmath12 solution presented in section [ flatsol1 ] . the parameters of the solution can be arranged such that the warp factor in the metric is a constant and the solution is asymptotic to the near horizon bmpv black hole . the axisymmetric multi - center solutions have @xmath273 parameters . the regularity and causality constraints studied in this section impose @xmath274 relations on them , therefore we have a @xmath275-parameter family of regular solutions with non - trivial topology on the base . it should be emphasized that we have only analyzed in detail the condition for absence of ctc s near the singularity of the harmonic functions and at asymptotic infinity . in principle one needs to ensure that there are no ctc s globally and for this one usually has to rely on numerics @xcite . on the other hand , experience with many examples suggests that once one has addressed this at singular points and ensured that the metric coefficients are well - behaved then there are no ctc s globally . it is interesting to note that there is no analog of the bubble equations @xcite for our regular non - bps solutions . bubble equations can be viewed as a form of angular momentum balance that constrains the location of sources and with pure flux solutions , non - trivial bubble equations require non - zero sources for all three fluxes . in our solutions , the magnetic flux of @xmath31 is trivial on the topological two - cycles and the complete maxwell field @xmath62 , has no localized sources . thus one should not be too surprised at the absence of constraints on the location of the remaining flux sources . using the floating brane ansatz of @xcite we have constructed a large class of non - bps multi - centered supergravity solutions . the solutions are determined by a four - dimensional khler base with non - trivial topology and that is a solution of the euclidean einstein - maxwell equations . to find explicit solutions one has to solve a coupled linear system of inhomogeneous differential equations on this base . we managed to construct the most general explicit solution of these equations on the axisymmetric lebrun - burns base . the generic multi - centered solutions will have horizons but we showed explicitly that by a judicious choice of parameters one can make the solutions completely smooth and regular . due to the maxwell flux on the four - dimensional base the five - dimensional solutions are not asymptotically flat but can be arranged to look like a warped , rotating @xmath12 space at asymptotic infinity . for specific choice of parameters the asymptotic metric is exactly the near horizon throat metric of the bmpv black hole . we have thus constructed hair in the back of a throat " . there are a number of possible directions for further work in this area . first , it well - known that bps supertubes with two electric and one magnetic dipole charge are regular in six dimensions in the d1-d5 duality frame @xcite . such solutions can thus potentially provide richer classes of regular geometries . indeed , five - dimensional regularity requires that all the @xmath18 be non - singular but supertubes allow two of the @xmath18 to have poles and the singularities are resolved as kaluza - klein monopoles in six dimensions . the solutions presented in section [ genaxi ] , before five - dimensional regularity was imposed , include solutions that correspond to families of concentric supertubes . removing the singularities as in ( [ zising ] ) required us to set some of the parameters to zero ( see ( [ ellzero1 ] ) ) and while we still found regular solutions with microstate structure , it restricted that family of solutions quite strongly and led us to solutions for which the bubble equations were trivial . we expect that for solutions with supertubes there will be some analog of the familiar radius formula arising from the bubble equations , or integrability conditions . we therefore expect there to be even richer classes of bubbles and `` hair '' if one allows solutions that are regular in six dimensions but not necessarily in five . it is also worth recalling that there are spectral flow methods that map regular , six - dimensional supertube geometries onto five - dimensional , regular bubbled geometries @xcite . for bps solutions , these transformations do not substantially modify the geometry of the four - dimensional base , though they can modify the asymptotics at infinity . on the other hand , for non - bps solutions such spectral flows can completely change the geometry of the base , for example , mapping a hyper - khler geometry onto an israel - wilson electrovac solution @xcite . it would be interesting to see how such spectral flows might modify the solutions considered here , particularly if one first includes supertube configurations . it will almost certainly move one beyond the lebrun class of solutions and perhaps give a richer class of geometries at infinity . there are other natural generalizations of the solutions considered here . our solutions can be uplifted to eleven dimensions where they are sourced by intersecting m2 and m5 branes on @xmath15 @xcite . it is fairly evident that there will also be solutions that can be obtained from intersecting m2 and m5 branes wrapping two - cycles and four - cycles in a more general calabi - yau three - fold . going in the opposite direction , any solution with a lebrun base has a space - like killing vector ( defined by @xmath66-translations ) and so one can perform a dimensional reduction along this direction to find supergravity solutions in four dimensions . these solutions will clearly be non - bps and will represent an infinite class of multi - center four - dimensional solutions that are non - supersymmetric generalizations of the solutions of @xcite . it would be interesting to explore the attractor mechanism for our solutions and make connections with recent discussions on non - bps attractors . the multi - centered solutions of section [ blowingbubbles ] may realize non - bps split attractors . it is interesting to note that in a recent discussion on non - supersymmetric split attractor flows the authors of @xcite also found that there are no bubble equations ( or integrability conditions ) . this fits with our analysis in section [ blowingbubbles ] and it will be very interesting to make this connection more precise . since our solutions are asymptotic to anti - de sitter space one can do holographic analysis of the `` hair '' corresponding to our geometries and understand them as duals to states ( or thermal ensembles ) in the corresponding cft . the solutions presented here have a warped and rotating @xmath276 region and while the @xmath277 correspondence is not understood in such detail as its higher dimensional analogs factor see @xcite . ] there might be some effective approach similar to the one in @xcite . alternatively , one might use a series of dualities and transform the solutions to the d1-d5-p iib duality frame @xcite and study the states in the d1-d5 cft . one might then be able to study the stability of the solutions and make some connection with the recent discussion of hawking radiation from non - supersymmetric solutions of the d1-d5 system @xcite . one would also very much like to find explicit non - supersymmetric solutions that have a throat region that looks like the solutions discussed in this paper but are asymptotically flat at infinity . to achieve this , one will probably have to find a way of breaking the relationship between the background electromagnetic field and the khler form . to achieve this one will probably have to relax some of the simplifying assumptions of the floating brane ansatz @xcite and work with more general ( and complicated ) equations of motion . however , there are almost certainly even broader classes of non - supersymmetric solutions that are determined by linear systems of equations and thus such explicit non - bps solutions may well be within reach . we would like to thank chris beem , iosif bena and clment ruef for useful conversations . nb would like to dedicate this work to the memory of his grandparents kunka and nedelcho . nb is grateful to the usc department of physics and astronomy for warm hospitality while part of this work was completed . this work was supported in part by doe grant de - fg03 - 84er-40168 . [ appendixa ] the general solutions on a lebrun - burns base discussed in section 4 have a gauge invariance " similar to the one present in multi - centered bps solutions with a gh base ( see equation ( 94 ) in @xcite ) ) . it is easy to check that the following transformation leaves and invariant @xmath278 one can also show that the equation for the one - form @xmath164 , , is invariant , therefore the transformation above is a symmetry of the full solution . a supergravity background is causal only if there are no ctcs and dirac - misner strings . to study the constraints imposed by these conditions one should study the five - dimensional metric at a constant time slice : @xmath279 where @xmath280 for absence of ctc s we need to impose the following conditions @xmath281 the last inequality comes from imposing positive definite metric in the six internal directions along @xmath15 upon uplift of our solutions to eleven - dimensional supergavity . the expression for @xmath261 resembles quite closely the one for solutions with a gh base ( see equation ( 102 ) in @xcite ) @xmath282 there is also the possibility of having dirac - misner strings in @xmath164 . to ensure that this does not happen one has to require that @xmath243 vanishes for @xmath283 . here we collect some identities used in section [ blowingbubbles ] . we used the following identities to solve the equations for @xmath241 and @xmath85 @xmath284 the following identities are useful when one solves the equation for @xmath84 @xmath285 @xmath286 @xmath287~. \end{gathered}\ ] ] there are similar identities involving @xmath288 , @xmath289 and @xmath290 that we have used to solve the equation for @xmath243 , however they are pretty lengthy and we refrain from presenting them explicitly . to facilitate comparison of our solutions with the more standard black - ring and supertube solutions , it is useful to recall the canonical separable bipolar coordinates on @xmath4 , @xcite ( we set @xmath200 below ) : @xmath291 in these coordinates the flat metric on @xmath4 takes the form : @xmath292 where @xmath293 . in particular , note that the canonical coordinates , @xmath294 and @xmath295 , are simply related to the green functions that we have been using and thus the solutions of sections 5 and 6 can easily be expressed as rational functions of @xmath294 and @xmath295 . i. bena and n. p. warner , `` bubbling supertubes and foaming black holes , '' phys . d * 74 * , 066001 ( 2006 ) [ arxiv : hep - th/0505166 ] . p. berglund , e. g. gimon and t. s. levi , `` supergravity microstates for bps black holes and black rings , '' jhep * 0606 * , 007 ( 2006 ) [ arxiv : hep - th/0505167 ] . i. bena and n. p. warner , `` black holes , black rings and their microstates , '' lect . notes phys . * 755 * , 1 ( 2008 ) [ arxiv : hep - th/0701216 ] . i. bena and n. p. warner , `` one ring to rule them all ... and in the darkness bind them ? , '' adv . * 9 * , 667 ( 2005 ) [ arxiv : hep - th/0408106 ] . k. goldstein and s. katmadas , `` almost bps black holes , '' jhep * 0905 * , 058 ( 2009 ) [ arxiv:0812.4183 [ hep - th ] ] . i. bena , g. dallagata , s. giusto , c. ruef and n. p. warner , `` non - bps black rings and black holes in taub - nut , '' jhep * 0906 * , 015 ( 2009 ) [ arxiv:0902.4526 [ hep - th ] ] . i. bena , s. giusto , c. ruef and n. p. warner , `` multi - center non - bps black holes - the solution , '' jhep * 0911 * , 032 ( 2009 ) [ arxiv:0908.2121 [ hep - th ] ] . i. bena , s. giusto , c. ruef and n. p. warner , `` supergravity solutions from floating branes , '' jhep * 1003 * , 047 ( 2010 ) [ arxiv:0910.1860 [ hep - th ] ] . i. bena , s. giusto , c. ruef and n. p. warner , `` a ( running ) bolt for new reasons , '' jhep * 0911 * , 089 ( 2009 ) [ arxiv:0909.2559 [ hep - th ] ] . s. ferrara , r. kallosh and a. strominger , `` n=2 extremal black holes , '' phys . d * 52 * , 5412 ( 1995 ) [ arxiv : hep - th/9508072 ] . s. ferrara and r. kallosh , `` supersymmetry and attractors , '' phys . d * 54 * , 1514 ( 1996 ) [ arxiv : hep - th/9602136 ] . a. ceresole and g. dallagata , `` flow equations for non - bps extremal black holes , '' jhep * 0703 * , 110 ( 2007 ) [ arxiv : hep - th/0702088 ] . g. lopes cardoso , a. ceresole , g. dallagata , j. m. oberreuter and j. perz , `` first - order flow equations for extremal black holes in very special geometry , '' jhep * 0710 * , 063 ( 2007 ) [ arxiv:0706.3373 [ hep - th ] ] . s. ferrara , a. gnecchi and a. marrani , `` d=4 attractors , effective horizon radius and fake supergravity , '' phys . d * 78 * , 065003 ( 2008 ) [ arxiv:0806.3196 [ hep - th ] ] . s. bellucci , s. ferrara , a. marrani and a. yeranyan , `` stu black holes unveiled , '' arxiv:0807.3503 [ hep - th ] . f. denef , `` quantum quivers and hall / hole halos , '' jhep * 0210 * , 023 ( 2002 ) [ arxiv : hep - th/0206072 ] . b. bates and f. denef , `` exact solutions for supersymmetric stationary black hole composites , '' arxiv : hep - th/0304094 . f. denef and g. w. moore , `` split states , entropy enigmas , holes and halos , '' arxiv : hep - th/0702146 . g. w. gibbons and s. w. hawking , `` gravitational multi - instantons , '' phys . lett . b * 78 * , 430 ( 1978 ) . c. lebrun , `` explicit self - dual metrics on @xmath296 , '' j. diff . geom . * 34 * ( 1991 ) 223 - 253 . c. lebrun , `` the einstein - maxwell equations , extremal kahler metrics , and seiberg - witten theory , '' arxiv:0803.3734 [ math.dg ] . c. p. boyer and j. d. . finley , `` killing vectors in self - dual , euclidean einstein spaces , '' j. math . * 23 * , 1126 ( 1982 ) . a. das and j. gegenberg , `` stationary riemannian space - times with self - dual curvature , '' gen . * 16 * , ( 1984 ) 817 . j. c. breckenridge , r. c. myers , a. w. peet and c. vafa , `` d - branes and spinning black holes , '' phys . lett . b * 391 * , 93 ( 1997 ) [ arxiv : hep - th/9602065 ] . i. bena , n. bobev and n. p. warner , `` bubbles on manifolds with a u(1 ) isometry , '' jhep * 0708 * , 004 ( 2007 ) [ arxiv:0705.3641 [ hep - th ] ] . j. b. gutowski and w. a. sabra , `` gravitational instantons and euclidean supersymmetry , '' phys . lett . b * 693 * , 498 ( 2010 ) [ arxiv:1007.2421 [ hep - th ] ] . m. dunajski and s. a. hartnoll , `` einstein - maxwell gravitational instantons and five dimensional solitonic strings , '' class . * 24 * , 1841 ( 2007 ) [ arxiv : hep - th/0610261 ] . i. bena and p. kraus , `` three charge supertubes and black hole hair , '' phys . d * 70 * , 046003 ( 2004 ) [ arxiv : hep - th/0402144 ] . i. bena , `` splitting hairs of the three charge black hole , '' phys . d * 70 * , 105018 ( 2004 ) [ arxiv : hep - th/0404073 ] . h. elvang , r. emparan , d. mateos and h. s. reall , `` supersymmetric black rings and three - charge supertubes , '' phys . d * 71 * , 024033 ( 2005 ) [ arxiv : hep - th/0408120 ] . i. bena , c. w. wang and n. p. warner , `` mergers and typical black hole microstates , '' jhep * 0611 * , 042 ( 2006 ) [ arxiv : hep - th/0608217 ] . o. lunin and s. d. mathur , `` ads / cft duality and the black hole information paradox , '' nucl . b * 623 * , 342 ( 2002 ) [ arxiv : hep - th/0109154 ] . o. lunin , j. m. maldacena and l. maoz , `` gravity solutions for the d1-d5 system with angular momentum , '' arxiv : hep - th/0212210 . i. bena , n. bobev , c. ruef and n. p. warner , `` supertubes in bubbling backgrounds : born - infeld meets supergravity , '' jhep * 0907 * , 106 ( 2009 ) [ arxiv:0812.2942 [ hep - th ] ] . i. bena , n. bobev and n. p. warner , `` spectral flow , and the spectrum of multi - center solutions , '' phys . d * 77 * , 125025 ( 2008 ) [ arxiv:0803.1203 [ hep - th ] ] . s. nampuri and m. soroush , `` new perspectives on attractor flows and trees from cft , '' arxiv:1009.5768 [ hep - th ] . m. guica , t. hartman , w. song and a. strominger , `` the kerr / cft correspondence , '' phys . d * 80 * , 124008 ( 2009 ) [ arxiv:0809.4266 [ hep - th ] ] . b. d. chowdhury and s. d. mathur , `` radiation from the non - extremal fuzzball , '' class . grav . * 25 * , 135005 ( 2008 ) [ arxiv:0711.4817 [ hep - th ] ] . s. g. avery , b. d. chowdhury and s. d. mathur , `` emission from the d1d5 cft , '' jhep * 0910 * , 065 ( 2009 ) [ arxiv:0906.2015 [ hep - th ] ] . s. g. avery , b. d. chowdhury and s. d. mathur , `` excitations in the deformed d1d5 cft , '' jhep * 1006 * , 032 ( 2010 ) [ arxiv:1003.2746 [ hep - th ] ] . a. sen , `` state operator correspondence and entanglement in @xmath277 , '' arxiv:1101.4254 [ hep - th ] . r. emparan and h. s. reall , `` a rotating black ring in five dimensions , '' phys . lett . * 88 * , 101101 ( 2002 ) [ arxiv : hep - th/0110260 ] .

we show that the renormalizable yang mills coupled to matter and the higgs field fits all the experimentally observed differential cross sections known in nature . this extended standard model reproduces the experimental gravitational differential cross sections without resorting to the graviton field and instead by exchanging gauge fields . by construction , each so(4 ) generator in quantum gravitodynamics does not commute with the dirac gamma matrices . this produces additional interactions absent to non - abelian gauge fields in the standard model . the contributions from these new terms yield differential cross sections consistent with the newtonian and post newtonian interactions derived from general relativity . dimensional analysis of the lagrangian shows that all its terms have total dimensionality four or less and therefore that all physical quantities in the theory renormalize by finite amounts . these properties make qgd the only renormalizable 4-dimensional theory describing gravitational interactions .

1611.05281

the structure and success of general relativity have lead us to the believe that the spin-2 graviton field mediates quantum gravitational interactions ; but to date this believe has eluded experimental detection . furthermore , the quantum limit of general relativity fails to renormalize properly forcing us to further believe that general relativity comes from a classical projection of a 10 dimensional string . on the other hand , we know that the geometrical object that describes the geodesics in general relativity is the connection and not the metric which enters that construction only after arbitrarily imposing the metric compatibility constraint . then , we can in practice construct a quantum theory of gravity where the fundamental field is the connection as in@xcite or a similar object as we do here . nothing but experimental observation constrains these possibilities . therefore , we should not force the graviton upon ourselves and instead focus on matching the proposed model s differential cross sections to experimental evidence ; more so , when we must yet detect the particle mediating gravitational interactions . here we present quantum gravitodynamics , qgd , the only renormalizable 4-dimensional theory that reproduces the interactions in general relativity to @xmath2 and couples to the standard model through a covariant derivative like the other three interactions in nature . as we show below , all diagrams in qgd match those of it counterpart in general relativity to @xmath2 and the potential that determines a test particle s classical trajectories , are identical for both theories to that same order . as opposed to general relativity which does not renormalize , qgd has a metric void of any dynamics , explicitly lacks a graviton field or any spin-2 state , and instead uses so(4 ) yang mills fields coupled to spinor singlets to mediate the gravitational interactions . furthermore , qgd reproduces all diagrams in general relativity to @xmath2 and therefore all experimental evidence to that same order . however , as opposed to general relativity , the terms in the qgd lagrangian involve products of fields and their derivatives with total dimensionality of at most four and therefore renormalization follows@xcite . thus , qgd is renormalizable , and , like general relativity , constructed in 4 dimensions . because qgd is defined in terms of yang mills fields , it also couples naturally to the standard model to produce a unified model with @xmath3 gauge symmetry that reproduces all the experimentally observed differential cross sections in nature , including those produced by gravitational interactions to @xmath2 . qgd also has the ability to fit those diagrams contributions from gravity that may become available in the future such as higgs - gluon gravitational scattering . in addition , qgd has gravitational su(2 ) yang mills bps states that match the entropy of black holes and type - ii strings . pure yang s gravity@xcite has similar classical equations to those of einstein s relativity and includes schwarzschild - like solutions , whose singularity at the horizon , as opposed to black hole solutions , can not be removed through a coordinate transformation@xcite . qgd and yang s gravity are both constructed in terms of the same connection defined over a fiber bundle , however , yang s gravity is not renormalizable because it couples its yang - mills field directly to the energy - momentum tensor . instead , in qgd , the yang mills fields couple to matter through a @xmath4 covariant derivative , in the same manner as the other forces in nature and which preserves renormalization . therefore , the couplings in qgd ensure that all terms in the lagrangian have dimensionality four or less which guarantees renormalization@xcite . string theory is the only other theory , besides qgd , to incorporate gravitational interactions while preserving renormalization . it has been a successful gravitational theory , because like qgd , scattering processes associated with the @xmath5-function constrain the metric to be a solution of einstein s equation . therefore , like qgd it reproduces all general relativity tests to @xmath2 . like strings , qgd has the necessary bps states to resolve the black hole entropy problem . however , unlike qgd , string theory can not be defined without ambiguity in 4 dimensions and neither can it couple to the standard model which instead must be reformulated along side gravity . the qgd connection defined on the so(4 ) fiber bundle can be projected to the connection defined on the so(1,3 ) tangent bundle . this projection allows the generators of the so(4 ) gauge fields to be defined in terms of dirac gamma matrices @xmath6,\label{so4g}\ ] ] and therefore these generators do not commute with the spacetime dirac @xmath7 matrices . this non - commutativity contrasts qgd from the other gauge theories in the standard model because it introduces terms proportional to @xmath8 $ ] which couple to fermionic matter fields with arbitrary couplings . the map from the so(4 ) gravitational gauge field , @xmath9 , to the spin - connection defined in@xcite motivate calling @xmath9 the connecton . in order to test qgd , we must reproduce the experimental evidence supporting general relativity . as discussed at the end of section 4 , this also suffices to show that qgd will reproduce all effects of general relativity to @xmath2 . the procedure used to fit qed to experimental evidence , see for example@xcite , transfers well to qgd and permits the fitting of all the qgd parameters to experimental observation . the t - matrix calculation in@xcite associated with the tree - level @xmath10 scattering process produces a differential cross section ; comparison of that differential cross section with the one obtained from the schroedinger equation with a coulomb potential allows the fitting of the electron coupling found in qed to the experimentally measured charge used in the coulomb potential . this procedure extends to the quantized but non - renormalizable general relativity theory . in@xcite the tree - level scattering process for two bosons , @xmath11 , produces a differential scattering matrix which fits the schroedinger equation with a newton potential while the 1-loop scattering correction fits the differential cross section produced by the schroedinger equation with the leading classical post newtonian correction to the potential . the strategy implemented in@xcite and@xcite was also successfully implemented to the gravitational sector in string theory @xcite-@xcite . thus , matching the model s differential cross sections to experimental / expected differential cross sections allows us to test if a model correctly describes nature . here we follow the same strategy as that used in@xcite-@xcite to show that qgd coupled to the standard model fits all the known experimental differential cross sections , including those gravitational in nature . section 2 details the symmetries and action derived for a single matter field and the u(1 ) electromagnetic field , both coupled to gravity . in section 3 we calculate the propagators and vertices of the model presented in section 2 . in section 4 we calculate t - matrix elements to produce differential cross sections which are then fitted to the differential cross section of the schroedinger equations with the coulomb , the newtonian and post newtonian potentials as well as the differential cross section for the deflection of light by a point particle in the small angle approximation found in general relativity . section 5 presents the full qgd coupling to the standard model while section 6 fits the differential cross sections to experimentally observed differential cross sections . section 7 moves outside the realm of experimental observation and shows that qgd can include the expected gravitational interactions of the su(2 ) and su(3 ) gauge bosons as well as the higgs doublet . in section 8 we consider the isomorphic @xmath12 yang mills representation of qgd , which when @xmath13 supersymmetrized has bps states whose entropy corresponds to that of a black hole , and that can bridge to type - ii strings@xcite . the conclusions appear in section 9 . the fundamental geometrical object of general relativity , the connection on the tangent bundle , does not require a metric to describe a particle s geodesics . consistent with general relativity and in the spirit of@xcite , qgd chooses a connection over the metric as the fundamental field . qgd does not use a yet to be detected graviton field or spin-2 particle and it uses instead a constant metric void of any dynamics . instead , connections defined on the so(4 ) fiber bundle mediate all gravitational interactions ; these so(4 ) gauge fields alone mediates all the gravitational interactions between particles . these fields suffice to obtain all the differential cross sections expected from general relativity for a single matter - photon system . more importantly , this construction does not have any experimental impediments because we have yet to detect the graviton or any other spin-2 particle . we present the lagrangian @xmath14\psi.\label{lagf}\end{aligned}\ ] ] we use the following conventions . the diagonal and flat space - time metric , @xmath15 , does not carry any dynamics and has signature @xmath16 . @xmath17 represents the levi - civita tensor in 4 dimensions . the vielbeins @xmath18 connect space - time coordinates ( greek indices ) with the so(4 ) fiber bundle coordinates ( latin indices ) . in particular @xmath19 and @xmath20 . the metric @xmath21 , defined on the fiber bundle , has signature @xmath22 . we define the field strengths appearing in ( [ lagb ] ) @xmath23 @xmath24 represents the u(1 ) electromagnetic gauge field and the so(4 ) gauge field @xmath25 , the connecton , mediates of gravitational interactions . the couplings @xmath26 , @xmath27 , @xmath28 and @xmath29 denote coupling constants and @xmath30 . with @xmath31 defined as a copy of a textbook so(n=4 ) gauge field necessarily antisymmetric in indices @xmath32 and @xmath33 , then @xmath34 so that the indispensable property @xmath35 holds . the covariant derivative operator @xmath36 defined in ( [ lagf ] ) acts on a 4 dimensional dirac spinor field @xmath37 @xmath38 and depends on the so(4 ) generators @xmath39 defined in ( [ so4 g ] ) . the u(1 ) local gauge transformation @xmath40 , with @xmath41 a scalar function , acts on the following fields @xmath42 which also transforms the covariant derivative in the standard way @xmath43 but leaves ( [ lagtodo ] ) unchanged . the so(4 ) local gauge transformations with @xmath44 and @xmath45 a 2-tensor , act on the following fields @xmath46 which also transforms the covariant derivative in the standard way @xmath47,\ ] ] but also leaves ( [ lagtodo ] ) unchanged . the so(1,3 ) local gauge transformations acting on the fermions have the same effect as the so(4 ) transformation because @xmath48 however , the so(1,3 ) gauge transformations , can not act on the pure yang - mills lagrangian because the cartan - killing metric for that gauge group , given by the trace over two group elements , is not positive definite which yields states not bounded from below@xcite . this property therefore distinguishes theories defined over the tangent bundle with local so(1,3 ) symmetry from those defined over the fiber bundle with local so(4 ) symmetry . the former can not be used to define yang - mills theories even when the fiber bundles can be projected to the tangent bundle . for the avoidance of doubt , gravitational theories like spin - gravity can be defined over the tangent bundle only because , as a result of the different construction from yang - mills , their spectrum does not become unbounded from below when the cartan - killing metric is not positive definite . the lorenz gauge fixings @xmath49 and property ( [ prop1 ] ) simplify the quadratic component in ( [ lagb ] ) to @xmath50 the structure constant defined in@xcite @xmath51 simplifies the relevant terms of the cubic terms in ( [ lagb ] ) to @xmath52 the terms in ( [ lag13 ] ) take the form @xmath53 and @xmath54 . in the low energy limit the photon momentum , @xmath55 , and the connecton momentum , @xmath56 , satisfy @xmath57 or equivalently @xmath58 this condition accounts for the omission of all other terms . for example , for matter - photon scattering the vertex generated by @xmath59 would only participate at 1-loop level , would necessarily appear twice and thus takes values proportional to the square of the connecton momentum . instead , the vertices produced by the second term in ( [ lag13 ] ) would appear first at tree - level and then at 1-loop level where it would contribute terms proportional to the photon momentum and its square respectively . dimensional analysis of general relativity shows that it does not renormalize properly . however , it follows from ( [ eq1a ] ) that the connecton has the same dimensionality as the u(1 ) gauge field . therefore , the terms in the lagrangian ( [ lag12 ] ) and ( [ lag13 ] ) involve products of fields and their derivatives with total dimensionality four or less . renormalization then follows as shown in@xcite . since @xmath60\ne0 $ ] , we use the relation @xmath61 to express @xmath62\psi&=&g \bar\psi\bar\omega_{\;\;\;\;\mu}^{\mu a}\gamma_a\psi.\label{lf2}\end{aligned}\ ] ] note that ( [ lf2 ] ) does not depend on @xmath63 . this means that gauge invariance constrains ( [ lf1 ] ) but not ( [ lf2 ] ) . thus the rescaling of ( [ lf1 ] ) by a constant redefines the coupling . however , the redefinition guarantees gauge invariance only after a subsequent redefinition of the spinor @xmath37 . on the other hand , a recalling by a constant , even a complex one , of ( [ lf2 ] ) does not affect gauge invariance . this contrasts with su(2 ) of su(3 ) couplings where the dirac matrices commute with their generators and therefore ( [ lf2 ] ) necessarily vanishes . the relations ( [ lf1 ] ) and ( [ lf2 ] ) simplify the lagrangian ( [ lagf ] ) to @xmath64 here the real coupling constants @xmath26 , @xmath27 , @xmath29 and @xmath28 adjust the theory to observation . in particular @xmath26 represents the electric charge and gauge invariance constrains @xmath27 and coincides with the coupling constant found in the purely bosonic sector ( [ lag13 ] ) . instead , gauge invariance does not constrain @xmath28 which can take any value , real , purely imaginary or complex . again , the terms in the lagrangian ( [ lag2 ] ) involve products of fields and their derivatives with total dimensionality four or less . therefore , the terms in the lagrangians ( [ lag12 ] ) , ( [ lag13 ] ) and ( [ lag2 ] ) , and thus lagrangian ( [ lagtodo ] ) , involve products of fields and their derivatives with total dimensionality four or less . renormalization then follows as shown in@xcite . the lagrangian @xmath65 has an complex coupling constant @xmath28 while @xmath66 has a real coupling constant @xmath27 . then , self energy corrections can have imaginary contributions and therefore it would be tempting to assume that the probability density determined by the wavefunction decreases over time and unitarity is lost . this is not the case . we choose a pure real @xmath27 and a pure imaginary @xmath28 and will show when fitting theory to experiment in the subsequent sections that @xmath67 where @xmath68 represents the gravitational constant . the self - energy contributions to @xmath69 for both bosons @xmath24 and @xmath9 come from purely bosonic diagrams and those with fermionic loops . we construct the purely bosonic diagrams with vertex ( c ) and vertex ( d ) defined in section [ pv1 ] . therefore , the diagrams depend only on the real valued coupling constant @xmath27 @xmath70 @xmath71 thus , those diagrams always contribute real amounts to the self energies @xmath72 , @xmath73 , @xmath74 and @xmath75 . as we show below , real contributions to the self energy do not affect unitarity . the diagrams involving fermionic loops always involve an even number of vertices ( a ) or an even number of vertices ( b ) defined in section [ pv1 ] . therefore , corrections to @xmath69 to the self energies of both @xmath24 and @xmath9 always contribute even powers of @xmath28 @xmath76 @xmath77 for some real constants @xmath78 and @xmath79 . even though the coupling @xmath28 is imaginary , all corrections to the self energies for both bosons @xmath24 and @xmath9 contribute real amounts to @xmath69 , even more , because fermions always appear in loops regardless of the order of perturbation , the self energy corrections to the bosons @xmath24 and @xmath9 always remain real . on the other hand , the contributions to the self energies of the fermion contribute real amounts to @xmath80 but complex amounts to @xmath69 . the contribution to the fermion self energy to @xmath80 comes from diagrams constructed from a pair of vertices ( a ) or a pair of vertices ( b ) . therefore @xmath81 for some real constants @xmath82 and @xmath83 . the first imaginary contribution comes from the diagram involving one vertex ( c ) and three vertices ( a ) . therefore @xmath84 therefore , the leading imaginary correction to the fermion self energy contributes at most of @xmath69 . the time evolution of the wavefunction for the fermion @xmath85 where @xmath86 describes the wavefunction at the emitter and @xmath87 represents the wavefunction at the detector , implies that the probability density at time @xmath88 , @xmath89 , decays as @xmath90 when @xmath91 particles are lost as they travel between emitter and detector . this experimentally unobserved reduction in flux prevents large imaginary values of @xmath92 . however , when @xmath93 satisfies @xmath94 then , the self energy @xmath95 can take on imaginary values without loss of unitarity . since the leading imaginary correction to the self energy of the fermion contributes at most @xmath69 , then , @xmath96 whenever @xmath97 years . bearing in mind that the universe has existed less than @xmath98 years and that estimates put the the proton s half life at @xmath99 years , qgd can take imaginary couplings without affecting unitarity . this contrasts with other theories where the coupling constants are much larger than @xmath28 leading to a loss of unitarity if not of gauge invariance as well since @xmath100 . this conclusion is further supported by the absence of diagrams contributing to @xmath69 to the self energies constructed with vertex ( a ) and vertices involving the electromagnetic field . the lagrangian densities ( [ lag12 ] ) and ( [ lag2 ] ) yield the following propagators gluon ( 70,15 ) propagator ( a ) & @xmath101 + fermion ( 70,15 ) propagator ( b ) & @xmath102 + photon ( 70,15 ) propagator ( c ) & @xmath103 + for the avoidance of doubt , in the limit considered here @xmath104 or equivalently @xmath105 . the lagrangian densities ( [ lag13 ] ) and ( [ lag2 ] ) generate the following vertices : |p4.5cm| p8.5cm| vertex1 ( 100,65 ) vertex ( a ) & @xmath106 + vertex2 ( 100,65 ) vertex ( b ) & @xmath107 + vertex3 ( 100,70 ) vertex ( c ) & @xmath108 @xmath109 + vertex4 ( 100,65 ) vertex ( d ) & @xmath110 + in the following section we relate the tree and 1-loop level diagrams to physical processes in order to determine the four parameters in qgd . in this section we calculate 4 scattering amplitudes : the tree - level photon exchange between two particles with equal mass , the tree - level connecton exchange and the 1-loop connecton exchange between two particles with equal mass and the tree - level connecton exchange between a photon and a particle with mass . these scattering amplitudes yield a total of 4 differential cross sections ; we equate the first 3 to the differential cross section for a schroedinger equation with 3 different potentials : the coulomb potential , the newton potential and the post - newtonian potential while the last cross section matches the cross section for gravitational lensing of a photon by a massive point particle . the lagrangian provides 4 coupling constants : @xmath26 , @xmath27 , @xmath28 and @xmath29 . therefore by the end of this section , the 4 coupling constants appearing in the differential cross sections calculated from qgd match those of the schroedinger equation as well as that obtained from gravitational lensing in general relativity . |p5cm| p5cm| dia2 ( 120,80 ) diagram ( a ) & dia1 ( 120,80 ) diagram ( b ) + the contribution from the photon exchange between two mass particles in diagram ( a ) produces the t - matrix @xmath111 with @xmath112 and @xmath113 . the property @xmath114 suppresses the second term in ( [ tree - a ] ) . a further limiting to the spinless schroedinger limit requires summation over the spin of all in - states and out - states so that @xmath115 . thus , in the non - relativistic quantum limit , the third term in ( [ tree - a ] ) vanish exactly reducing ( [ tree - a ] ) to @xmath116 this t - matrix yields the differential cross section @xmath117 while the schroedinger equation with a coulomb potential , @xmath118 with @xmath119 , the measured electric charge , and @xmath120 , the distance between the two particles , produces a differential cross section @xmath121 comparing the differential cross section ( [ cross - a ] ) and ( [ coulomb - a ] ) yields @xmath122 diagram ( b ) contributes to the differential scattering cross section of the connecton exchange between two mass particles . the t - matrix element @xmath123 produces the quantum gravitodynamic differential cross section @xmath124 the schroedinger equation with a newtonian potential , @xmath125 with @xmath68 representing the gravitational constant , @xmath126 representing the measured mass and @xmath120 representing the distance between the two particles , produces a differential cross section @xmath127 comparing the differential cross section ( [ cross - b ] ) and ( [ newton - b ] ) yields @xmath128 and @xmath129 dia3 ( 160,160 ) diagram ( c ) & dia32 ( 160,160 ) diagram ( d ) + 1.0 cm the post - newtonian correction comes from diagrams ( c ) and ( d ) . all other 1-loop diagrams contribute to either analytic terms , or quantum corrections of order @xmath130 and do not contribute to the low energy limit considered here@xcite . in addition , diagrams involving photons and connectons simultaneously do not contribute because the transformation @xmath131 leaves the contribution of the above diagram unchanged while all terms but the first two in ( [ lag13 ] ) remain present in the action . these two diagrams contribute to the quantum gravitodynamic post newtonian t - matrix @xmath132 this result as those below require the following steps . first , we focus only on the electric " form factor which means that @xmath133 equivalently , we disregard terms proportional to @xmath134 $ ] that contribute to the magnetic " form factor . second , the on - shell external momenta imply the relations @xmath135 third , we suppress terms of @xmath136 using the low energy limit property @xmath137 . fourth , we use the approximation @xmath138 and @xmath139 . finally , we use the appendix in@xcite with the expressions for the several feynman integrals . the t - matrix ( [ t - cd ] ) produces the differential scattering cross section @xmath140 the schroedinger equation with a post newtonian correction to the gravitational potential ( see equation ( 39 ) of@xcite ) , @xmath141 where the constant @xmath32 depends on the post - newtonian expansion , produces , after using ( [ tree - b ] ) and ( [ tree - b2 ] ) , the differential scattering cross section @xmath142 comparing the differential cross section ( [ cross - cd ] ) and ( [ einstein - cd ] ) yields @xmath143 therefore , by fixing @xmath26 , @xmath27 and @xmath28 through equations ( [ eq - a ] ) , ( [ tree - b ] ) and ( [ eq - cd ] ) , qgd reproduces the low energy limit of general relativity for the matter sector . therefore , the so(4 ) yang mills theory coupled to matter and an inert metric is a valid renormalizable relativistic quantum field theory to describe gravity s experimental observations . |p5cm| dia6 ( 120,80 ) diagram ( e ) + diagram ( e ) describes the scattering of a fermion @xmath144 with a photon @xmath145 that produces the t - matrix @xmath146 which in turn , in the small angle approximation where @xmath147 , yields the differential cross section @xmath148 where the photon energy , @xmath149 , satisfies @xmath150 derivation of ( [ saa1 ] ) requires the photon on - shell conditions , @xmath151 along with the steps used for the matter - matter scattering above . substitution of ( [ tree - b ] ) and ( [ eq - cd ] ) into ( [ saa1 ] ) yields @xmath152 in the low energy limit , equation ( [ saa2 ] ) describes the rutherford scattering of a mass @xmath153 projectile off of a mass @xmath149 target . in the small angle deflection approximation and after setting the speed of light to 1 , the gravitational lensing of a photon by a point particle of mass @xmath153 in general relativity relates the impact parameter @xmath154 with the deflection angle @xmath155 @xmath156 the relation between the impact parameter , @xmath157 , and the differential cross section , @xmath158 simplifies the latter to @xmath159 or equivalently @xmath160 equation ( [ saa5 ] ) describes the rutherford scattering of mass @xmath149 projectile off of a mass @xmath153 target . the exchange @xmath161 modifies this differential cross section to @xmath162 which then describes the scattering of a mass @xmath153 particle of a photon target with energy @xmath149 . equating ( [ saa2 ] ) with ( [ saa3 ] ) requires that @xmath163 which determines @xmath29 as a function of the known constant @xmath32 . with ( [ tree - light ] ) satisfied , fourier transform in the born approximation of ( [ saa2 ] ) yields the low energy potential @xmath164 which describes the motion of a photon as it had mass equal to its energy in the newtonian approximation . diagram ( e ) also describes the gravitational redshift . when the impact parameter vanishes , a photon starting at @xmath165 approaches the mass target and experiences the potential ( [ potphoton ] ) which accelerates the photon and blueshifts its frequency . thus the energy of the photon at point @xmath166 reads @xmath167 and when @xmath168 @xmath169 as required . diagram ( e ) has the same value as its counterpart diagram in general relativity obtained following the methods of @xcite which produces the exact same potential . renormalization aside , it does so in the frame of reference where the photon scatters off the fermion , and again confirms that we can model photons as particles with mass @xmath170 and ( [ potphoton ] ) . like the gravitational redshift , the shapiro delay also stems from diagram ( e ) and it is just the length of the path traveled by the photon when bending about a mass particle and after reflection from some other body retraces its path back to the source . this feature transpires the quantum limit were diagrams in general relativity match the diagrams for qgd to @xmath2 . in the classical limit , the trajectories of general relativity on a schwarzschild background also follow from consideration of a particle moving about a potential . therefore , since we have shown that both theories have the same potential to @xmath2 , any classical or quantum test that general relativity passes will also be passed by qgd to that same order . the qgd lagrangian ( [ lagtodo ] ) has 4 free parameters @xmath26 , @xmath27 , @xmath28 and @xmath29 . equations ( [ eq - a ] ) , ( [ tree - b ] ) , ( [ eq - cd ] ) and ( [ tree - light ] ) , repeated here : @xmath171 fix these parameters to experimentally observed quantities from qed , and general relativity : the electric charge , the newtonian mass and gravitational constant , the precession of a mass particle about another mass particle and the gravitational lensing of light by mass particles . therefore , qgd suffices to describe the gravitational effects observed in nature without resorting to a spin-2 particle and using instead a quantum connection defined over an so(4 ) fiber bundle . the dimensionality of the connecton is the same as that of the u(1 ) gauge field . therefore , the terms in the lagrangians ( [ lag12 ] ) , ( [ lag13 ] ) and ( [ lag2 ] ) involve products of fields and their derivatives with total dimensionality four or less . then all scattering amplitudes considered in this section renormalize by finite amounts@xcite . the matching of the above 4 differential cross sections with those of general relativity coupled to matter and light are more than simple tests . by matching the differential cross sections we have , at least for two particle interactions , shown that the equations of motion for matter and light coupled to general relativity exactly match those for qgd to @xmath2 . this is manifested by the fact that the schroedinger equation for both theories are the same . therefore , any test that follows from general relativity coupled to matter and light will be reproduced by qgd to order @xmath172 . while it may very well be the case that general relativity and qgd diverge beyond @xmath2 , the reader is reminded that qgd needs to replicate nature and not general relativity . we know that because qgd and general relativity are indistinguishable to @xmath2 , it replicates nature to that order . but general relativity , safe for @xcite , has not been experimentally proven beyond that order , and therefore it may very well be that if qgd and general relativity diverge beyond @xmath2 , it will be up to nature to judge which most closely resembles it , always taking into account that general relativity is at best and effective theory because it is not renormalizable like qgd . it should be further noted that while the differential cross sections pertain two particle interactions and omit single particle and three- , four - particle interactions , these do not impose further constraints . the single particle diagrams involve loops which general relativity can not handle due to lack of renormalization . furthermore , three and four particle interactions are constrained by two particle scattering ; see @xcite for the diagrams used in general relativity . therefore , by considering only the equivalency of two particle scattering processes , we fix the equations of motions for all other scattering processes in both theories . lack of evidence supporting the existence of a spin-2 particle along with the existence of a mapping between an so(4 ) fiber bundle to an so(1,3 ) tangent bundle motivate the formulation of a gravitational theory based on the so(4 ) connecton instead of one based on the graviton . without a graviton field and with a constant diagonal metric void of any dynamics , qgd again incorporates all the necessary gravitational interactions to the standard model using instead @xmath173 so(4 ) gauge fields @xmath174 . the inclusion of the fields @xmath174 suffices to obtain all the experimentally verified differential cross sections , and those expected from general relativity . the experimental evidence for general relativity exists only for the matter and u(1 ) gauge sector . at present , no experimental evidence exists supporting that the su(2 ) , su(3 ) and the higgs sectors of the standard model couple to gravity ; however , here we show that qgd can incorporate such couplings to satisfy the intuitive expectation of general relativity . the lagrangian @xmath175\psi^{(j)}\label{lagfsw}\\ l_{h}&=&\frac{1}{2\beta_{higgs}^2}d_\mu \phi\cdot d^\mu \phi -v(\phi\cdot \phi)\label{laghsw}\end{aligned}\ ] ] remains invariant under so(4 ) transformations as well as @xmath176 transformations . there are @xmath173 covariant derivatives @xmath177 in ( [ lagfsw ] ) . this new artifice allows us to introduce enough parameters to fit all the gravitational scattering amplitudes . the u(1 ) gauge field @xmath178 , the su(2 ) gauge field @xmath179 and the su(3 ) gauge field @xmath180 along with the so(4 ) gauge fields @xmath181 define the strength fields in ( [ lagbsw ] ) @xmath182,\\ w_{\mu\nu}^a&=&\frac{1}{\beta_{su(2)}}\bigg[\partial_\mu w^a_\nu-\partial_\nu w^a_\mu -\frac{g_{su(2)}}{\beta_{su(2)}}c^{abc}w_\mu^b w_\nu^c\bigg ] , \\ g_{\mu\nu}^a&=&\frac{1}{\beta_{su(3 ) } } \bigg[\partial_\mu g^a_\nu-\partial_\nu g^a_\mu -\frac{g_{su(3)}}{\beta_{su(3 ) } } c^{abc}g_\mu^b g_\nu^c\bigg ] , \\ \omega_{\mu\nu}^{(i , m)ab}&=&(\partial_\mu\omega_\nu^{(i , m)ab}+\frac{\alpha_{im } } { \beta_{im}}\epsilon_\nu^{ab\rho}\partial_\mu a^m_\rho)- ( \partial_\nu\omega_\mu^{(i , m)ab}+\frac{\alpha_{im } } { \beta_{im}}\epsilon_\mu^{ab\rho}\partial_\nu a^m_\rho)\nonumber\\ & & -g_{i\,m } c^{ab}_{\;\;\;cdef}(\omega_\mu^{(i , m)cd}+\frac{\alpha_{im } } { \beta_{im}}\epsilon_\mu^{cd\rho } a^m_\rho)(\omega_\nu^{(i , m)ef}+ \frac{\alpha_{im } } { \beta_{im}}\epsilon_\nu^{ef\sigma } a^m_\sigma)\nonumber\\ & & \end{aligned}\ ] ] we define each @xmath183 as a copy of a textbook so(4 ) gauge field necessarily having antisymmetry in indices @xmath32 and @xmath33 , then , @xmath184 so that @xmath185 since @xmath186 where @xmath187 and @xmath188 describes the higgs in the fundamental of so(4 ) , the constant couplings @xmath189 must take the form @xmath190 so that @xmath191 and @xmath192 when representing a quark doublet , the spinors @xmath193 , @xmath194 transform in the irreducible representation of @xmath176 while when representing a lepton doublet , the spinors @xmath193 , @xmath195 transform in the irreducible representation of @xmath196 . then , each of the covariant derivatives @xmath197 , @xmath198,\label{cdsw}\ ] ] acting on a quark doublet @xmath193 transforms covariantly under @xmath3 . the u(1 ) transformation generated by @xmath199 acts on the following fields @xmath200 and leaves ( [ lagtodosw ] ) invariant while each @xmath201 transforms covariantly . since @xmath202=[\tau^a , t^{ab}]=[t^a,\tau^a]=0,\ ] ] su(3 ) transformations generated by @xmath203 act on the following fields @xmath204 where @xmath205 represents the su(3 ) structure constant , to leave ( [ lagtodosw ] ) invariant while each @xmath201 transforms covariantly . similarly , su(2 ) transformations generated by @xmath206 act on the following fields @xmath207 where @xmath205 represents the su(2 ) structure constant , and also leave ( [ lagtodosw ] ) invariant while each @xmath201 transforms covariantly . so(4 ) transformations generated by @xmath208 act on the following fields @xmath209 and also leave ( [ lagtodosw ] ) invariant while each @xmath201 transforms covariantly . under @xmath210 transformations , ( [ cdsw ] ) acting on a lepton doublet @xmath193 transforms covariantly while ( [ lagtodosw ] ) remains invariant . the bosonic component of the lagrangian , ( [ lagbsw ] ) , has quadratic terms which are included in ( [ nov6 ] ) instead . ] @xmath211 the dimensionality of the connecton is the same as that of the u(1 ) gauge field . therefore , the terms in the lagrangian ( [ nov4 ] ) involve products of fields and their derivatives with total dimensionality four . the pure standard model su(2 ) and su(3 ) cubic and quartic bosonic interactions present in ( [ lagbsw ] ) do not mix with the so(4 ) gauge fields and do not merit further treatment here . for the same reasons presented for ( [ lag13 ] ) , the relevant cubic terms in lagrangian ( [ lagbsw ] ) include @xmath212 as in ( [ lag13 ] ) , the terms in ( [ nov5 ] ) take the form @xmath53 and @xmath54 . in the low energy limit the gauge boson momentum , @xmath55 , and the connecton momentum , @xmath56 , satisfy @xmath213 . the terms in the lagrangian ( [ nov5 ] ) also involve products of fields and their derivatives with total dimensionality four . therefore any diagrams constructed from the terms in ( [ nov5 ] ) will renormalize by finite amounts@xcite . after adding the contributions from ( [ lagbsw ] ) dependent on @xmath214 to ( [ laghsw ] ) we obtain the standard model higgs lagrangian along with qgd interactions @xmath215 , \label{nov6}\end{aligned}\ ] ] where @xmath216 describes the higgs in the su(2 ) doublet representation instead of the so(4 ) vector representation and @xmath217 the dual representation of the higgs further strengthens the proposal of the so(4 ) connecton as the carrier of quantum gravitational interactions . the terms in the lagrangian ( [ nov6 ] ) again involve products of fields and their derivatives with total dimensionality four . thus , any corrections involving only these terms will renormalize by finite amounts@xcite . when @xmath193 represents a quark doublet , the lagrangian ( [ lagfsw ] ) simplifies to @xmath218 \psi^{(j ) } , \label{nov8}\end{aligned}\ ] ] where @xmath219 and @xmath220 are the electric charges of each respective quark and @xmath221 @xmath222 similarly , when @xmath193 represents a lepton doublet , the lagrangian ( [ lagfsw ] ) simplifies to @xmath223 \psi^{(j ) } , \label{nov8b}\end{aligned}\ ] ] where @xmath224 and @xmath225 are the electric charges of each respective lepton . in addition to the transformation of the terms in the pure standard model , the electroweak rotation which produces the u(1 ) gauge field @xmath24 only transforms the last term in ( [ nov8 ] ) and ( [ nov8b ] ) @xmath226 where @xmath227 describes the weak mixing angle and takes the same values as in the pure standard model . thus , for a quark doublet @xmath228 \psi^{(j ) } , \nonumber\\ \label{nov8 - 2}\end{aligned}\ ] ] with @xmath229 representing the fermionic sector of the pure standard model after the rotation between @xmath178 and @xmath230 . a similar expression follows for the lepton doublet . the dimensionality of the connecton is the same as that of the u(1 ) gauge field . therefore , the terms in the lagrangian ( [ nov8 - 2 ] ) involve products of fields and their derivatives with total dimensionality of at most four . renormalization then follows@xcite . more importantly , the whole lagrangian , the sum of ( [ nov4 ] ) , ( [ nov5 ] ) , ( [ nov6 ] ) and ( [ nov8 - 2 ] ) , involve products of fields and their derivatives also with total dimensionality at most four and therefore any corrections involving their terms renormalize by finite amounts@xcite . the electroweak rotation transforms @xmath231 into @xmath232 and it transforms @xmath233 into @xmath234 after choosing @xmath235 , @xmath236 and @xmath237 . at this point we can apply the higgs mechanism by selecting the usual vev for the higgs field which will produce the same effects that take place in the standard model . the rotation between the fields does not affect the dimensionality of the fields . therefore , the rotated terms in the lagrangian involve products of fields and their derivatives with total dimensionality at most four . renormalization then follows@xcite . the experimental evidence for gravitational interactions only includes observations for the u(1 ) gauge fields and the fermionic matter fields . thus , while the model can include gravitational interactions with the su(2 ) , su(3 ) and higgs sectors , we limit ourselves in this section to the study of the u(1 ) gauge fields and the fermionic matter fields . this removes all terms in ( [ nov4 ] ) , ( [ nov8 - 2 ] ) , and ( [ nov11 ] ) which carry index @xmath153 as well and allows us to drop altogether the higgs sector ( [ nov6 ] ) . then , lagrangian ( [ nov8 - 2 ] ) expressed in terms of true fermions reduces to @xmath238 \psi^{(j ) } , \nonumber\\ & & \label{nov12}\end{aligned}\ ] ] with @xmath239 running over the number of singlet spinors , @xmath240 and @xmath241 . the lagrangian ( [ nov47 ] ) reduces to @xmath242 and the lagrangian ( [ nov11 ] ) reduces to @xmath243 the dimensionality of the connecton is the same as that of the u(1 ) gauge field . therefore , the terms in the lagrangian ( [ nov12 ] ) , ( [ nov13 ] ) and ( [ nov14 ] ) involve products of fields and their derivatives with total dimensionality four . renormalization then follows as shown in@xcite . in sections 2 to 4 we considered a single particle , like the electron , and successfully reproduced the necessary post - newtonian corrections to fit theory to experiment . expanding the particle spectrum increases the number of scattering cross sections well beyond 4 ; this leads to a lack of solution when a single connecton describes the gravitational interactions . instead we introduced @xmath244 connectons to increase the number of couplings from @xmath245 to @xmath246 and from @xmath247 to @xmath248 . with this increase in the number of connectons the number of parameters in the model also increases sufficiently to fit all these differential cross sections . the lagrangian densities ( [ nov12])-([nov14 ] ) yield the following propagators gluonsm ( 70,15 ) propagator ( d ) & @xmath101 + fermionsm ( 70,15 ) propagator ( e ) & @xmath102 + photon ( 70,15 ) propagator ( f ) & @xmath103 + for the avoidance of doubt , in the low energy limit @xmath104 such that @xmath105 . these same lagrangian densities generate the following vertices : |p5.5cm| p7.5cm| vertex1sm ( 100,70 ) vertex ( e ) & @xmath249 + [ 7ex ] |p5.5cm| p8.5cm| vertex2sm ( 100,70 ) vertex ( f ) & @xmath250 + vertex3sm ( 100,70 ) vertex ( g ) & [ cols="^ " , ] + vertex4sm ( 100,70 ) vertex ( h ) & @xmath251 + [ 7ex ] |p5cm| p5cm| dia2bis ( 120,80 ) diagram ( f ) & dia1bis ( 120,80 ) diagram ( g ) + after assuming as before that the axial current vanishes , the scattering process in diagram ( f ) describing the photon exchange between particle @xmath252 and @xmath56 produces a t - matrix element , @xmath253 which in turn yields a differential cross section contribution , @xmath254 with @xmath255 defined as the reduce mass of particles @xmath252 and @xmath56 . on the other hand , the schrodinger equation for a coulomb potential @xmath256 with @xmath257 and @xmath258 describing the particles measured electric charge and @xmath120 describing the distance between the two particles , yields the differential cross section @xmath259 comparing the differential cross section ( [ sqed ] ) and ( [ coulomb - abis ] ) constrains the parameters @xmath260 and @xmath261 to @xmath262 diagram ( g ) , exchanging a connecton between particle @xmath252 and @xmath56 produces the t - matrix , @xmath263 which in turn produces the differential cross section , @xmath264 on the other side , the differential cross section obtained from the schrodinger equation for a newton potential , @xmath265 where @xmath266 and @xmath267 describe the particles measured mass and @xmath120 the distance between them , produces the differential cross section , @xmath268 comparing the differential cross sections ( [ sqgd ] ) and ( [ newton - abis ] ) imposes the constraint @xmath269 dia3bis ( 160,160 ) diagram ( h ) & dia32bis ( 160,160 ) diagram ( i ) + 1.0 cm diagrams ( h ) and ( i ) produce the 1-loop correction to the connecton exchange between particles @xmath252 and @xmath56 to the t - matrix , @xmath270 after using the same steps used in section 4 , and in turn it produces the differential cross section , @xmath271 the schroedinger equation with the einstein correction to the gravitational potential between two particles , @xmath272 yields the differential scattering cross section @xmath273 comparing the differential cross section ( [ einstein - cdbis ] ) and ( [ solve22 ] ) yields @xmath274 |p5cm| dia6bis ( 120,80 ) diagram ( j ) + the t - matrix contribution from diagram ( j ) scattering a mass particle off of a photon simplifies to @xmath275 after using the small angle approximation , the differential cross section obtained from this t - matrix equates to @xmath276 in the low energy limit , equation ( [ sab1 ] ) describes the rutherford scattering of a mass @xmath277 projectile off of a mass @xmath149 target . for a small angle deflection in general relativity the gravitational lensing of a photon by a point particle of mass @xmath277 , the differential cross section calculation follows in the same manner as in section 4.3 and resulting in @xmath278 the exchange @xmath279 transforms the differential cross section to describe the scattering of a particle of mass @xmath277 from a photon target with energy @xmath149 and as a result takes the form @xmath280 equating ( [ sab1 ] ) with ( [ sab5 ] ) requires that @xmath281 renormalization follows by construction because the structure added to the standard model is just @xmath244 standard yang - mills gauge fields with @xmath1 symmetry . dimensional analysis of the standard model coupled to qgd shows that the connectons @xmath282 have the same dimension as the u(1 ) , su(2 ) and su(3 ) gauge fields . therefore , all integrands in the standard model coupled to qgd , including those involving the connectons @xmath283 have the dimension four or less which guarantees renormalization of all corrections to physical quantities@xcite . if a solution also exists for the system of equations ( [ eqpn11 ] ) , ( [ eqpn12 ] ) and ( [ eqpn13 ] ) , then the renormalizable lagrangian ( [ nov12 ] ) , ( [ nov13 ] ) and ( [ nov14 ] ) correctly describes experimental observation without resorting to the graviton and instead using @xmath1 gauge fields . if a solution exists to the system of equations @xmath284 then , a solution also exists for the system of equations ( [ eqpn11 ] ) , ( [ eqpn12 ] ) and ( [ eqpn13 ] ) . equation ( [ eqqn11 ] ) is a stricter for of ( [ eqpn11 ] ) , ( [ eqqn12 ] ) is a stricter for of ( [ eqpn12 ] ) and ( [ eqqn13 ] ) is a stricter for of ( [ eqpn13 ] ) . while ( [ eqpn11 ] ) , ( [ eqpn12 ] ) and ( [ eqpn13 ] ) are all real , the complex system of equations ( [ eqqn11 ] ) , ( [ eqqn12 ] ) and ( [ eqqn13 ] ) require their imaginary components to vanish . then , the solution to the real components of the system of equations ( [ eqqn11 ] ) , ( [ eqqn12 ] ) and ( [ eqqn13 ] ) also solves the system of equations ( [ eqpn11 ] ) , ( [ eqpn12 ] ) and ( [ eqpn13 ] ) . we require the complex system of equations ( [ eqqn11 ] ) with @xmath285 , @xmath286 instead of ( [ eqpn11 ] ) because solutions to ( [ eqqn11 ] ) ensure that the @xmath80 correction to the @xmath287fermion self energy , @xmath288 does not have an imaginary component , so that @xmath289 does not affect unitarity to @xmath80 . to show existence of a solution to the system ( [ eqqn11 ] ) , ( [ eqqn12 ] ) and ( [ eqqn13 ] ) first note that ( [ eqqn12 ] ) and ( [ eqqn13 ] ) is a complex linear equation system in @xmath248 and then assume that @xmath290 . our model contains 12 fermions with @xmath291 , so that the linear system ( [ eqqn12 ] ) has @xmath292 real equations and @xmath292 imaginary equations . the linear system ( [ eqqn13 ] ) has 12 real and 12 imaginary equations . we can also represent the complex linear system ( [ eqqn12 ] ) and ( [ eqqn13 ] ) instead as a real system of 180 linear equations @xmath293 where @xmath294 and @xmath295 represents a square matrix with dimension n=180 . then , a solution to ( [ lsys1 ] ) exists whenever the @xmath296 or equivalently when the vector rows in @xmath295 are not parallel among themselves . a simple inspection of the coefficients of @xmath297 produced by ( [ eqqn12 ] ) and ( [ eqqn13 ] ) shows that when the complex vectors @xmath298 are not parallel among themselves , the different vector rows in @xmath297 are not parallel . a solution to the complex system of equations ( [ eqqn11 ] ) can be obtained for the set @xmath299 following these steps : 1 ) assign a real random number between 0 and 1 to each @xmath300 and @xmath301 , @xmath302 , 2 ) use the fletcher - reeves - polak - ribiere algorithm to minimize the equation @xmath303 this method produces a minimum of zero for this equation which amounts to solving ( [ eqqn11 ] ) . note that the first squared term in ( [ lsys2 ] ) is the real part of ( [ eqqn11 ] ) while the second squared term is the imaginary part of ( [ eqqn11 ] ) . it is not surprising that a minimum of zero exists for ( [ lsys2 ] ) ; there are @xmath304 variables and only 88 equations . furthermore , the vectors produced by this method are not parallel and explicit calculation of @xmath297 shows that its determinant does not vanish . then , we can choose @xmath305 effectively reducing the number of couplings to matter and the number of connectons by half . ] . therefore , a solution can always be found to the system of 12 fermions and one u(1 ) gauge boson provided there are 180 connectons . this does not preclude the existence of solutions whenever @xmath306 , specially after abandoning the assumption that @xmath290 . the reader should be concerned by the number of parameters necessary to fit qgd coupled to the standard model when compared with general relativity . however , two points should be kept in mind . first , the intention of this section was to show the existence of a solution rather than to find a solution with the smallest possible number of couplings . given the non - linear nature of the system of equations , it is very likely that non - linear solutions with a significantly smaller number of couplings may very well exist ; for example solutions obtained through the gradient method or simulated annealing . second , qgd should only be compared with theories that successfully incorporate gravity and more importantly renormalize . thus , when compared with heterotic strings , which have 496 gauge fields before even considering compactification , the number of gauge fields required by qgd is similar . or , alternatively , like string theory , the gauge fields may represent some compactification of m - theory . nevertheless , we can cast qgd in a more symmetric manner where each of the 25 particles , whether a fermion or a gauge boson , couples to 25 different connectons ; each particle carries charge with respect to 25 non - abelian gauge fields ; each particle connects to a different particle through a unique connecton , including one which self couples . then in this construction a picture emerges where each of the 325 connectons mediates the interaction between two particular particles only ( for example between a higgs and a @xmath307 ) . then , the number , @xmath308 , of differential cross sections to @xmath2 is smaller than the number of couplings which , when omitting the @xmath29 s , counts to @xmath309 : the first term counts the covariant derivative couplings and the second term counts the yang mills couplings . experimental data evidencing the gravitational interactions does not yet exist for the @xmath310 , the @xmath311 and the higgs particles or the gluons . however , we expect for them to become available in the future . nevertheless , qgd can incorporate gravitational interactions for all these particles through the second term in ( [ nov5 ] ) . we present the propagators for the remainder of the bosonic sector gluonb ( 70,15 ) propagator ( g ) & , @xmath312 + wb ( 70,15 ) propagator ( h ) & , @xmath313 + higgsb ( 70,15 ) propagator ( i ) & + where the first propagator describes the gluons , the second describes the electroweak vector bosons @xmath314 and @xmath311 , and the third describes the higgs in the so(4 ) representation . lagrangian ( [ nov5 ] ) produces the following vertices : |p7cm| p7.5cm| vertex5 ( 100,70 ) vertex ( i ) & @xmath315 + [ 7ex ] vertex6 ( 100,70 ) vertex ( j ) & , @xmath316 + [ 7ex ] vertex7 ( 100,70 ) vertex ( k ) & @xmath317 + [ 7ex ] in general we expect that gluon couplings satisfy @xmath318 ; but , due to the mass difference between the w and z particles , we expect @xmath319 and @xmath320 . these additional vertices produce the following tree - level and @xmath80 diagrams : |p6.5cm|p6.5cm| dia6 g ( 120,80 ) diagram ( k ) + dia6w ( 120,80 ) diagram ( l ) + dia6h ( 120,80 ) diagram ( m ) + these diagrams produce the t - matrix elements , @xmath321 which in turn produce the differential cross sections , @xmath322 and @xmath323 after using the low energy limit @xmath324 . although these differential cross sections have little difference among themselves they represent different scattering processes : the gluon does not carry mass while the @xmath325 and higgs do carry mass . therefore , ( [ sg1 ] ) must match the small angle deflection calculated in general relativity which requires @xmath326 and whose derivation follows exactly along the lines in section 6 above . on the other hand the w , z and higgs bosons are all massive and therefore we must compare their differential cross sections to those obtained from the schroedinger equation with a standard newtonian potential . these constraints yield the following conditions @xmath327 |p7cm| p7cm| dia7 ( 160,160 ) diagram ( o ) & dia8 ( 160,160 ) diagram ( p ) + dia7r ( 160,160 ) diagram ( r ) & dia8q ( 160,160 ) diagram ( q ) + finally , diagrams ( o)-(q ) above give the desired correction to @xmath69 for the w , z , and higgs particles interacting with fermions : in the low energy limit , these diagrams produce the t - matrix elements , @xmath328,\\\end{aligned}\ ] ] with @xmath329 . these t - matrices in turn produce the following differential cross sections , @xmath330 comparison with the differential cross sections derived for a schrodinger with a post newtonian potential ( [ einstein - cdbis ] ) imposes @xmath331 inclusion of these differential cross sections further increases the number of differential cross section by @xmath332 for a total of 252 differential cross sections with a consequent increase in the total number of equations to be fitted . however , a solution can be shown to exist following the method in the previous section . we could go on and construct the @xmath80 amplitude for the scattering described by the exchange of a connecton between different bosons as well as the @xmath69 amplitude for the corresponding 1-loop corrections . this requires modifying @xmath333 in ( [ lagtodosw ] ) to read @xmath334 the requirement that the quadratic lagrangian ( [ nov4 ] ) remain invariant requires that the constant real symmetric matrices @xmath335 satisfy @xmath336 the vertices ( h)-(k ) for bosons emitting a connecton will then have a quadratic dependence on the matrices @xmath337 . then , diagrams for fermion - fermion scattering remain unchanged ; diagrams for fermion - boson scattering will produce tree - level t - matrices with quadratic dependence on the matrices @xmath337 ; their 1-loop corrections will produce t - matrices with quadratic and quartic dependence on the matrices @xmath337 ; diagrams for boson - boson scattering will produce tree - level t - matrices with quartic dependence on the matrices @xmath337 ; their 1-loop corrections will produce t - matrices with sextic dependence on the matrices @xmath337 . the coefficients of these matrices will be determined in the same manner as in the previous section : equating the differential cross sections of qgd to those obtained from the schroedinger equation and gravitational lensing . given the lack of experimental evidence , the task of writing each equation which reproduce general relativity expectations is left as a curiosity for the reader . however we note that it may very well turn out that some processes not described by general relativity , like those analogous to s - channel bhabha scattering produced by the term @xmath338 , are confirmed through experiment or that some processes described by general relativity , like those produced by the term @xmath339 , are confirmed to be absent through experiment . we stress that the matrices @xmath337 allow all these possible experimental scenarios . qgd also explains the black hole entropy problem in the same way that strings do . to construct the map between qgd and strings we need first break down so(4 ) into its isomorphic representation @xmath12 . we n=2 supersymmetrize at least one of qgd s su(2 ) gauge groups and consider for example a k3 spacetime manifold . each su(2 ) yang mills theory then has bps states which have the same moduli space as that of a supersymmetric sigma model which in turn can be used to calculate the degeneracy of bps states for large energy states@xcite . the degeneracy thus obtained determines the entropy which coincides with that of certain black hole solutions in string theory @xcite . thus qgd has states whose entropy is that of certain black hole solutions , and therefore , those states represent the quantum description of the black hole solutions in the exact same way they do in string theory . the solution for @xmath340 yang - mills found in@xcite differs from the schwarzschild black hole solution because the singularity of its horizon can not be removed through a coordinate transformation . however , both singularities can be located at @xmath341 . therefore in su(2 ) yang mills , the horizon area and the black hole mass are related in the same way as for the schwarzschild black hole : @xmath342 . this result implies that the su(2 ) yang mills black hole has the same entropy as the schwarzschild black hole which in turn can be expressed in terms of bps solutions . therefore one has a dual representation of the same physical objects in the @xmath13 supersymmetric limit : the macroscopic system is described by black hole solution found in@xcite , while the microscopic system is described by bps states . qgd , as presented here , reproduces the expected newtonian and post newtonian interaction evidenced by experimental measurements for both matter fields and photon fields . in addition to matching the differential cross sections from qgd to those of the schroedinger equation with newtonian and post newtonian corrections as well as the differential cross section for the deflection of light by a point mass particle in general relativity , qgd couples to the standard model in a straight forward and natural manner . thus , qgd does not require us to re - invent the standard model to have a unified theory . this contrasts other attempts at describing quantum gravity which require a recasting of the already successful standard model . the fundamental theory , a straight forward @xmath1 yang - mills , couples to matter in the same manner as su(3 ) and su(2 ) theories describe the other fundamental forces and thus bears a high degree of symmetry with the non - gravitational forces found in nature . dimensional analysis of qgd and qgd coupled to the standard model show that all terms in the lagrangians have dimensionality four or less and therefore renormalization follows@xcite . qgd accommodates the full particle spectrum currently known and yields in the low energy limit the expected relativistic forces provided a sufficiently large number of @xmath1 symmetries exists . the absence in qgd or qgd coupled to the standard model of the graviton should not concern the reader ; after all , we have yet to detect such a particle . the theory here proposed not only reproduces the observed experimental results of general relativity , exhibits renormalization and couples in a straight forward manner to the standard model , it also incorporates yet to be observed , but highly expected , gravitational interactions between matter and the su(3 ) , su(2 ) and higgs bosons as well as those between these bosons . these features make qgd coupled to the standard model with @xmath3 local symmetry the most attractive theory to describes all interactions in nature . i am most grateful to j. lee for providing _ ricci.m _ which was indispensable for all calculations . also indispensable was the encouragement and funding by w. israel during the early attempts to construct this model as well as funding by e. braun in its latter stages . v. n. gribov , j. nyiri , _ quantum electrodynamics : gribov lectures on theoretical physics _ , volume 13 of cambridge monographs on particle physics , nuclear physics and cosmology ( cambridge university press , 2001 ) .

spectral measures provide invariants for braided subfactors via fusion modules . in this paper we study joint spectral measures associated to the rank two lie group , including the mckay graphs for the irreducible representations of and its maximal torus , and fusion modules associated to all known modular invariants .

1404.1863

spectral measures associated to the compact lie group @xmath1 and its maximal torus , nimrep graphs associated to the @xmath2 modular invariants , and the mckay graphs for finite subgroups of @xmath2 were studied in @xcite using information about the generating series of the moments of the spectral measure and the jones series which is related to the poincar series of the subfactor planar algebra associated to the graph ( see e.g. @xcite ) . in @xcite the authors studied spectral measures associated to the compact lie groups @xmath1 and @xmath3 and their maximal tori , nimrep graphs associated to the @xmath2 and @xmath4 modular invariants , and the mckay graphs for finite subgroups of @xmath2 and @xmath4 , using braided subfactor theory . spectral measures associated to the compact rank two lie groups @xmath5 and @xmath6 are studied in @xcite , and for other compact rank two lie groups in @xcite . in this paper and its sequel @xcite we focus on the lie group @xmath7 , the automorphism group of the octonions @xmath8 . it is a simply connected , compact , real rank two lie group of dimension 14 and is the smallest of the exceptional lie groups . it is isomorphic to the subgroup of @xmath9 that fixes any particular vector in its 8-dimensional real spinor representation . we determine spectral measures and joint spectral measures for ( the adjacency matrices of ) various graphs related to the lie group @xmath0 : the mckay ( or representation ) graphs for the irreducible representations of @xmath0 and its maximal torus @xmath10 , nimrep graphs or fusion modules associated to the @xmath0 modular invariants , and in the sequel @xcite , the mckay graphs for finite subgroups of @xmath0 . suppose @xmath11 is a unital @xmath12-algebra with state @xmath13 . if @xmath14 is a self - adjoint operator then there exists a compactly supported probability measure @xmath15 , the spectral measure of @xmath16 , on the spectrum @xmath17 of @xmath16 , uniquely determined by its moments @xmath18 , for all non - negative integers @xmath19 . note that @xmath15 depends on the choice of state @xmath13 on @xmath11 . in the cases we consider , the @xmath12-algebra @xmath11 will be a space of operators which act on the following hilbert space @xmath20 . for the mckay graphs for irreducible representations of @xmath0 we have @xmath21 , whilst for the irreducible representations of its maximal torus @xmath10 we have @xmath22 . for a nimrep graph @xmath23 ( either associated to a @xmath0 modular invariant or the mckay graph for finite subgroups of @xmath0 ) with vertex set @xmath24 , we take @xmath25 . as @xmath0 has rank two , its characters are functions on the maximal torus @xmath10 of @xmath0 . for @xmath2 and @xmath4 it was convenient to determine the spectral measures for the operator @xmath16 given by the adjacency matrices of the various graphs related to these groups ( the mckay graphs for the irreducible representations of the group and its maximal torus , nimrep graphs associated to modular invariants , and the mckay graphs for finite subgroups ) by first determining corresponding measures @xmath26 over the maximal tori of @xmath2 or @xmath4 respectively . this approach is described as a spectral measure blowup in @xcite . in the case of @xmath0 , the maximal torus @xmath10 has dimension one greater than the spectrum @xmath17 , so that there is a loss of dimension when passing from the measure @xmath26 to @xmath15 . this means that there is an infinite family of measures @xmath26 over @xmath10 which correspond to the spectral measure @xmath15 ( details of the relation between @xmath26 and @xmath15 are given in section [ sect : measures - different_domains ] ) . in order to remove this ambiguity , we also consider measures over the joint spectrum @xmath27 of commuting self - adjoint operators @xmath16 and @xmath28 . the abelian @xmath12-algebra @xmath29 generated by @xmath16 , @xmath28 and the identity 1 is isomorphic to @xmath30 , where @xmath31 is the spectrum of @xmath29 . then the joint spectrum is defined as @xmath32 . in fact , one can identify the spectrum @xmath31 with its image @xmath33 in @xmath34 , since the map @xmath35 is continuous and injective , and hence a homeomorphism since @xmath31 is compact @xcite . in the case where the operators @xmath16 , @xmath28 act on a finite - dimensional hilbert space , this is the set of all pairs of real numbers @xmath36 for which there exists a non - zero vector @xmath37 such that @xmath38 , @xmath39 . then there exists a compactly supported probability measure @xmath40 on @xmath33 , which is uniquely determined by its cross moments @xmath41 for all non - negative integers @xmath19 , @xmath42 . the spectral measure for @xmath16 is then given by the pushforward @xmath43 of the joint spectral measure @xmath40 under the orthogonal projection @xmath44 onto the spectrum @xmath45 . to study the spectral measures for the nimrep graphs associated to the @xmath0 modular invariants , we use the theory of braided subfactors and @xmath46-induction , which we now briefly review . for a fuller discussion on braided subfactors and @xmath46-induction see @xcite . the verlinde algebra of @xmath0 at level @xmath47 is represented by a non - degenerately braided system of endomorphisms @xmath48 on a type @xmath49 factor @xmath50 . its fusion rules @xmath51 reproduce exactly those of the positive energy representations of the loop group of @xmath0 at level @xmath47 , @xmath52 . furthermore its statistics generators @xmath53 , @xmath54 obtained from the braided tensor category @xmath48 match exactly those of the kac - peterson modular @xmath53 , @xmath54 matrices which perform the conformal character transformations ( see footnote 2 in @xcite ) . from the verlinde formula we see that this family @xmath55 of commuting normal matrices can be simultaneously diagonalised , i.e. @xmath56 , where the summation is over each @xmath57 and @xmath58 is the trivial representation . the intriguing aspect being that the eigenvalues @xmath59 and eigenvectors @xmath60 are described by the modular @xmath53 matrix . a braided subfactor is an inclusion @xmath61 where the dual canonical endomorphism decomposes as a finite combination of elements of the verlinde algebra , i.e. a finite combination of endomorphisms in @xmath48 . such subfactors yield modular invariants through the procedure of @xmath46-induction which allows two extensions of @xmath62 on @xmath50 , depending on the use of the braiding or its opposite , to endomorphisms @xmath63 of @xmath64 , so that the matrix @xmath65 is a modular invariant @xcite . the systems @xmath66 are called the chiral systems , whilst the intersection @xmath67 is called the neutral system . then @xmath68 , where @xmath69 denotes a system of endomorphisms consisting of a choice of representative endomorphisms of each irreducible subsector of sectors of the from @xmath70 $ ] , @xmath71 , where @xmath72 is the inclusion map . although @xmath48 is assumed to be braided , the systems @xmath66 or @xmath73 are not braided in general . the action of each @xmath50-@xmath50 sector @xmath71 on the @xmath64-@xmath50 sectors @xmath74 produces a nimrep ( non - negative integer matrix representation of the original verlinde algebra ) @xmath75 , i.e. @xmath76 . the spectrum of ( each ) @xmath75 reproduces exactly the diagonal part of the modular invariant @xcite . since the nimreps are a family of commuting matrices , they can be simultaneously diagonalised and thus the eigenvectors @xmath77 are the same for the nimrep graphs @xmath75 for all @xmath71 . we have @xmath78 , where the summation is over each @xmath57 with multiplicity given by the modular invariant , i.e. the spectrum of @xmath75 is given by @xmath79 with multiplicity @xmath80 . the set of @xmath81 with multiplicity @xmath82 is called the set of exponents of @xmath83 . along with the identity invariants for @xmath0 for all levels @xmath47 , there are two exceptional invariants due to conformal embeddings at levels 3 , 4 @xcite and another exceptional invariant at level 4 @xcite . these are all the known @xmath0 modular invariants . since the centre of @xmath0 is trivial , there are no orbifold modular invariants . this list was shown to be complete for all prime heights @xmath84 such that @xmath85 @xcite , and for all other @xmath86 @xcite . the paper is organised as follows . in section [ sect : rep_theoryg2t2 ] we describe the representation theory of @xmath0 and its maximal torus @xmath10 , and in particular focus on the fundamental representations of @xmath0 . in section [ sect : measures - different_domains ] we discuss spectral measures for @xmath0 over different domains , showing how measures over a region in the complex plane yields a unique @xmath87-invariant measure over @xmath10 , where @xmath87 is the weyl group of @xmath0 . in section [ sect : measure : a12infty(g2 ) ] we determine the ( joint ) spectral measures associated to the ( adjacency matrices of the ) mckay graphs given by the action of the irreducible characters of @xmath0 on its maximal torus @xmath10 , and in section [ sect : measureainftyg2 ] the ( joint ) spectral measures associated to the ( adjacency matrices of the ) mckay graphs of @xmath0 itself . in both cases we focus on the fundamental representations of @xmath0 , and determine these ( joint ) spectral measures over @xmath10 and the ( joint ) spectrum of these adjacency matrices . finally in section [ sect : measures_nimrepg2 ] we determine joint spectral measures over @xmath10 for nimrep graphs arising from @xmath0 braided subfactors . the irreducible representations @xmath88 of @xmath0 are indexed by pairs @xmath89 such that @xmath90 . we denote by @xmath91 the fundamental representation of @xmath0 of dimension 7 , where @xmath92 . the maximal torus of @xmath9 is @xmath93 , for @xmath94 , where @xmath95 , and the maximal torus of @xmath0 is the subset of @xmath54 such that @xmath96 , which is isomorphic to @xmath10 . then the restriction of @xmath97 to @xmath10 is given by the @xmath98 block - diagonal matrix @xmath99 for @xmath100 . we also denote by @xmath101 the second fundamental representation of @xmath0 , the adjoint representation which has dimension 14 . the lie group @xmath4 is a subgroup of @xmath0 . the generating function for the @xmath102 branching rules was determined in @xcite . in particular , the fundamental representations @xmath97 , @xmath103 of @xmath0 branch into the following irreducible @xmath4 representations : @xmath104 the representations @xmath105 , @xmath106 are conjugate to one another and are the fundamental three - dimensional representations of @xmath4 . the eight - dimensional representation @xmath107 is the adjoint representation of @xmath4 and is obtained from the product of the fundamental representations by removing one copy of the trivial representation @xmath108 , since @xmath109 . then from ( [ eqn : branch_rulesg2-su(3 ) ] ) the restriction of @xmath103 to @xmath10 is given by the @xmath110 block - diagonal matrix @xmath111 for @xmath100 . let @xmath112 , @xmath113 be the irreducible characters of @xmath0 , @xmath10 respectively , where @xmath114 . the characters @xmath115 of @xmath0 are self - conjugate and thus are maps from the torus @xmath10 to an interval @xmath116 . for @xmath94 , @xmath117 , the characters of @xmath10 are given by @xmath118 , and satisfy @xmath119 . if @xmath120 , @xmath121 is the restriction of @xmath122 , @xmath123 respectively to @xmath10 , we have from ( [ eqn : restrict_rho1g2_to_t2 ] ) and ( [ eqn : restrict_rho2g2_to_t2 ] ) that @xmath124 then @xmath125 for any @xmath126 . the mckay graph for an irreducible representation @xmath62 of a group @xmath83 is the graph whose vertices are labelled by the irreducible representations of @xmath83 , and which , for any two irreducible representations @xmath127 of @xmath83 , has @xmath128 ( directed ) edges from @xmath129 to @xmath130 , where the decomposition of @xmath131 into irreducible contains @xmath128 copies of @xmath130 . the mckay graph of @xmath10 for the first fundamental representation @xmath97 is identified with the infinite graph @xmath132 , illustrated in figure [ fig - a_12infty(g2)1 ] , whose vertices may be labeled by pairs @xmath133 such that there is an edge from @xmath134 to @xmath134 , @xmath135 , @xmath136 , @xmath137 , @xmath138 , @xmath139 and @xmath140 . , width=132 ] , width=207 ] , width=283 ] , width=283 ] similarly , the mckay graph of @xmath10 for the second fundamental representation @xmath103 is identified with the infinite graph @xmath141 , illustrated in figure [ fig - a_12infty(g2)2 ] , where multiplication by @xmath103 corresponds to the edges illustrated in figure [ fig - mult_rho2g2 ] . these graphs @xmath142 are essentially @xmath87-unfolded versions of the graphs @xmath143 , where @xmath87 denotes the weyl group @xmath144 of @xmath0 . by ( * @xmath1453.5 ) , @xmath146 and @xmath147 . here @xmath148 is the path algebra of the graph @xmath23 , where @xmath149 is the algebra generated by pairs @xmath150 of paths from the distinguished vertex @xmath151 such that @xmath152 and @xmath153 , with multiplication defined by @xmath154 . we now consider instead the fixed point algebra of @xmath155 , @xmath156 under the action of the group @xmath0 given by the fundamental representations @xmath97 , @xmath103 respectively , where @xmath0 acts by conjugation on each factor in the infinite tensor product . the characters @xmath112 of @xmath0 satisfy @xmath157 whilst for @xmath158 , @xmath159 where @xmath160 if @xmath161 or @xmath162 . thus the mckay graph of @xmath0 for the first fundamental representation @xmath97 is identified with the infinite graph @xmath163 , illustrated in figure [ fig - a_infty(g2)1 ] , where we have made a change of labeling to the dynkin labels @xmath164 . this labeling is more convenient in order to be able to define self - adjoint operators @xmath165 , @xmath166 in @xmath167 below . the dashed lines in figure [ fig - a_infty(g2)1 ] indicate edges that are removed when one restricts to the graph @xmath168 at finite level @xmath47 ( here @xmath169 ) , c.f . section [ sect : measures_akg2 ] . , width=264 ] , width=264 ] similarly , the mckay graph of @xmath0 for the second fundamental representation @xmath103 is identified with the infinite graph @xmath170 , illustrated in figure [ fig - a_infty(g2)2 ] , again using the dynkin labels @xmath164 , and where the dashed lines again indicate edges that are removed when one restricts to the graph @xmath168 at finite level @xmath47 ( here @xmath169 ) . by ( * @xmath1453.5 ) we have @xmath171 and @xmath172 . the weyl group @xmath87 of @xmath0 is the dihedral group @xmath144 of order 12 . if we consider @xmath144 as the subgroup of @xmath173 generated by the matrices @xmath174 , @xmath175 , of orders 2 , 6 respectively , given by @xmath176 then the action of @xmath144 on @xmath10 given by @xmath177 , for @xmath178 , leaves @xmath179 invariant , for any @xmath180 . any @xmath144-invariant measure @xmath181 on @xmath10 yields a pushforward probability measure @xmath182 on @xmath183 by @xmath184 for any continuous function @xmath185 , where @xmath186 for all @xmath187 . there is a loss of dimension here , in the sense that the integral on the right hand side is over the two - dimensional torus @xmath10 , whereas on the right hand side it is over the interval @xmath188 . thus the preimage @xmath189 $ ] of any point @xmath190 in the interior of @xmath188 is infinite , and there is an infinite family of pullback measures @xmath181 over @xmath10 for any measure @xmath191 on @xmath188 , that is , any @xmath181 such that @xmath192 ) = \nu_{\mu}(x)$ ] for all @xmath193 will yield the probability measure @xmath191 on @xmath188 as a pushforward measure by ( [ eqn : measures - t2-ij_g2 ] ) . we introduce below an intermediate probability measure @xmath194 which lives over a ( two - dimensional ) subregion @xmath195 , for @xmath196 , for which there is a unique @xmath144-invariant measure @xmath197 on @xmath10 . this measure @xmath194 specializes to the spectral measures @xmath198 , @xmath191 of @xmath62 , @xmath81 respectively . the permutation group @xmath199 appears as the subgroup generated by @xmath174 and the matrix @xmath200 of order 3 ( c.f . * equation ( 37 ) ) ) . then @xmath144 is generated by @xmath199 and the @xmath201 matrix @xmath202 which sends @xmath203 , @xmath204 . a fundamental domain of @xmath205 is thus given by a quotient of the fundamental domain of @xmath206 , illustrated in figure [ fig : fund_domain - a2int2 ] ( see @xcite ) , by the @xmath207-action given by the matrix @xmath202 . a fundamental domain @xmath208 of @xmath10 under the action of the dihedral group @xmath144 is illustrated in figure [ fig : fund_domain - g2int2 ] , where the axes are labelled by the parameters @xmath209 , @xmath210 in @xmath211 . in figure [ fig : fund_domain - g2int2 ] , the lines @xmath212 and @xmath213 are also boundaries of copies of the fundamental domain @xmath208 under the action of @xmath144 , whereas they are not boundaries of copies of the fundamental domain under the action of @xmath199 in figure [ fig : fund_domain - a2int2 ] . the torus @xmath10 contains 12 copies of @xmath208 , so that @xmath214 for any @xmath144-invariant function @xmath215 . the fixed points of @xmath10 under the action of @xmath199 are the points @xmath216 , @xmath217 and @xmath218 , but only the point @xmath216 is fixed under the action of the whole of @xmath144 . under @xmath219 the point @xmath216 maps to 7 , 14 in the intervals @xmath220 , @xmath221 respectively , whilst the points @xmath217 , @xmath218 both map to -2 ( in both intervals ) . , title="fig:",width=207 ] + , title="fig:",width=207 ] + let @xmath222 and let @xmath223 be the map @xmath224 . we denote by @xmath225 the image of @xmath226 in @xmath34 . the joint spectral measure @xmath194 is the measure on @xmath225 uniquely determined by its cross - moments as in ( [ eqn : cross_moments_sa_operators ] ) . then there is a unique @xmath144-invariant pullback measure @xmath197 on @xmath10 such that @xmath227 for any continuous function @xmath228 . any probability measure on @xmath225 yields a probability measure on the interval @xmath229 , given by the pushforward @xmath230 of the joint spectral measure @xmath194 under the orthogonal projection @xmath231 onto the spectrum @xmath232 . see ( * ? ? ? * section 2.5 ) for more details . the following @xmath199-invariant measures on @xmath10 , defined in ( * definition 1 ) , will be useful later . note that these measures are also invariant under @xmath144 . [ def:4measures ] let @xmath233 , @xmath234 . we define the following measures on @xmath10 : 1 . @xmath235 , the uniform dirac measure on the @xmath199-orbit of the points @xmath236 , @xmath237 , @xmath238 , for @xmath239 , @xmath240 . @xmath241 , the uniform dirac measure on the @xmath199-orbit of the points @xmath242 , @xmath243 , @xmath244 , @xmath245 , @xmath246 , @xmath247 , for @xmath248 , @xmath249 , @xmath250 . the sets @xmath251 , @xmath252 are illustrated in ( * ? ? ? * figures 4 , 5 ) . for @xmath249 and @xmath253 , @xmath254 , whilst @xmath255 . the cardinalities of the other sets are @xmath256 for @xmath249 , and @xmath257 . some relations between these measures are given in ( * ? ? ? * section 2 ) . for the remainder of the paper we focus on the fundamental representations @xmath97 and @xmath103 of @xmath0 . we first consider their restrictions to @xmath10 . as discussed in section [ sect : rep_theoryg2t2 ] , their corresponding mckay graphs are @xmath259 , illustrated in figures [ fig - a_12infty(g2)1 ] , [ fig - a_12infty(g2)2 ] for @xmath260 respectively . we define commuting self - adjoint operators which may be identified with the adjacency matrix of @xmath259 . we define operators @xmath261 , @xmath262 on @xmath263 by @xmath264 where the unitary @xmath265 is the bilateral shift on @xmath266 . let @xmath267 denote the vector @xmath268 . then @xmath269 is identified with the adjacency matrix of @xmath259 , @xmath260 , where we regard the vector @xmath270 as corresponding to the vertex @xmath271 of @xmath259 , and the operators of the form @xmath272 which appear as terms in @xmath269 as corresponding to the edges on @xmath259 . then @xmath273 corresponds to the vertex @xmath274 of @xmath259 for any @xmath275 , and applying @xmath276 to @xmath270 gives a vector @xmath277 in @xmath278 , where @xmath279 gives the number of paths of length @xmath19 on @xmath259 from @xmath271 to the vertex @xmath274 . we define a state @xmath13 on @xmath280 by @xmath281 . we use the notation @xmath282 to denote the multinomial coefficient @xmath283 . then we have cross moments @xmath284 where @xmath285 if @xmath286 then @xmath287 for all @xmath288 , and we get a non - zero contribution when @xmath289 and @xmath290 . so we obtain @xmath291 where the summation is over all integers @xmath292 such that @xmath293 . if @xmath294 then @xmath295 for all @xmath288 , and we get a non - zero contribution when @xmath296 and @xmath297 . so we obtain @xmath298 where @xmath299 , @xmath300 , @xmath301 and the summation is over all integers @xmath302 such that @xmath303 . the ranges of the restrictions ( [ eqn : restriction - chi1 ] ) , ( [ eqn : restriction - chi2 ] ) of the characters @xmath304 , @xmath260 , of the fundamental representations of @xmath0 to @xmath10 are given by @xmath305 , where @xmath306 $ ] and @xmath307 $ ] . since the spectrum @xmath308 of @xmath265 is @xmath309 , the spectrum @xmath310 of @xmath261 is @xmath311 $ ] , and the spectrum @xmath312 of @xmath262 is @xmath313 $ ] . we now determine the @xmath144-invariant spectral measure @xmath314 on @xmath10 for the graphs @xmath259 , @xmath260 . [ thm : measureg2-t2 ] the joint spectral measure @xmath315 ( on @xmath10 ) for the graphs @xmath259 , @xmath260 , is given by the uniform lebesgue measure @xmath316 . _ proof : _ the @xmath317 cross moment is given by @xmath318 where @xmath319 , @xmath320 are as in ( [ eqn : r1-g2 ] ) , ( [ eqn : r2-g2 ] ) , since @xmath321 . this is equal to the cross moments @xmath322 given in ( [ eqn : crossmomentsa_12infty(g2 ) ] ) . @xmath323 in fact , the measure @xmath315 given above is the joint spectral measure over @xmath10 for the pair of mckay graphs ( @xmath324 , @xmath325 ) for any pair @xmath326 of irreducible representations of @xmath0 , by a similar proof . thus the spectral measure over @xmath10 is independent of the choice of irreducible representations used to construct the mckay graphs . let @xmath328 and @xmath329 , or explicitly , @xmath330 and denote by @xmath331 the map @xmath332 . for @xmath0.,title="fig:",width=226 ] + we now describe @xmath333 , illustrated in figure [ fig : domaind - g2 ] , which is the joint spectrum @xmath334 of the commuting self - adjoint operators @xmath261 , @xmath262 . the boundary of @xmath208 given by @xmath335 yields the curves @xmath336 , @xmath337 for @xmath338 $ ] , which are both given by the parametric equations @xmath339 the boundary of @xmath208 given by @xmath340 yields the curve @xmath341 given by the parametric equations @xmath342 where @xmath343 $ ] . finally , the boundary @xmath344 of @xmath208 yields the curve @xmath345 given by the parametric equations @xmath346 where @xmath347 $ ] . as functions of @xmath190 , the boundaries of @xmath327 are obtained by writing @xmath348 in terms of @xmath190 in the above parametric equations , which are at worst quadratic in @xmath348 . the boundaries of @xmath327 are thus given by the curves @xcite @xmath349 , \label{eqn : boundaryd - y - g2 - 1 } \\ & c_2 : & \hspace{-10 mm } y & = -5(x+1)+2(x+2)^{3/2 } , \qquad x \in [ 7/9,7 ] , \label{eqn : boundaryd - y - g2 - 1a } \\ & c_3 : & \hspace{-10 mm } y & = -5(x+1)-2(x+2)^{3/2 } , \qquad x \in [ -2,-1 ] , \label{eqn : boundaryd - y - g2 - 2}\\ & c_4 : & \hspace{-10 mm } 4y & = x^2 + 2x-7 , \hspace{30 mm } x \in [ -1,7 ] . \label{eqn : boundaryd - y - g2 - 3}\end{aligned}\ ] ] on the other hand , writing these curves as function of @xmath350 involves cubic equations in @xmath348 , and the boundaries of @xmath327 are given by the curves @xmath351 , \label{eqn : boundaryd - x - g2 - 1 } \\ & c_2 : & \hspace{-10 mm } x & = -1 + 4p_3(y)+4p_3(y)^2 , \qquad y \in [ 10/27,5 ] \label{eqn : boundaryd - x - g2 - 1a } \\ & c_3 : & \hspace{-10 mm } x & = -1 + 4p_1(y)+4p_1(y)^2 , \qquad y \in [ -2,5 ] , \label{eqn : boundaryd - x - g2 - 2 } \\ & c_4 : & \hspace{-10 mm } x & = -1 + 2(y+2)^{1/2 } , \hspace{18 mm } y \in [ -2,14 ] , \label{eqn : boundaryd - x - g2 - 3}\end{aligned}\ ] ] where @xmath352 is given by @xmath353 , for @xmath354 and @xmath355 , and we take the positive square root in equation ( [ eqn : boundaryd - x - g2 - 3 ] ) . under the change of variables ( [ eqn : x - g2 ] ) , ( [ eqn : y - g2 ] ) , the jacobian @xmath356 is given by @xmath357-g2 } j ( \theta_1,\theta_2 ) & = 8 \pi^2 ( \cos(2 \pi ( 2\theta_1 + \theta_2 ) ) + \cos(2 \pi ( \theta_1 - 3\theta_2 ) ) + \cos(2 \pi ( 3\theta_1 - 2\theta_2 ) ) \nonumber \\ & \qquad - \cos(2 \pi ( \theta_1 + 2\theta_2 ) ) - \cos(2 \pi ( 3\theta_1 - \theta_2 ) ) - \cos(2 \pi ( 2\theta_1 - 3\theta_2 ) ) ) . \qquad\end{aligned}\ ] ] the jacobian is real and is illustrated in figures [ fig : j_overt2 - 3dplot - g2 ] , [ fig : j_overt2-contours - g2 ] , where its values are plotted over the torus @xmath10 . over @xmath10.,title="fig:",width=302 ] + over @xmath10.,title="fig:",width=207 ] + with @xmath358 , @xmath260 , the jacobian is given in terms of @xmath359 by @xmath360-g2}\end{aligned}\ ] ] the jacobian @xmath361 is invariant under @xmath362 , whilst @xmath363 . thus @xmath364 is invariant under the action of @xmath144 , and we seek an expression for @xmath364 in terms of the @xmath144-invariant variables @xmath190 , @xmath350 , which may be obtained as a product of the roots appearing as the equations of the boundary of @xmath327 in ( [ eqn : boundaryd - y - g2 - 1])-([eqn : boundaryd - y - g2 - 3 ] ) , and is given ( up to a factor of @xmath365 ) as ( see also @xcite ) @xmath366 for @xmath367 , which can easily be checked by substituting for @xmath190 , @xmath350 as in ( [ eqn : x - g2 ] ) , ( [ eqn : y - g2 ] ) . note that the jacobian ( [ j2-g2 ] ) is a cubic in @xmath350 , with the three roots appearing as the equations of the boundary of @xmath327 in ( [ eqn : boundaryd - y - g2 - 1])-([eqn : boundaryd - y - g2 - 3 ] ) . however , although the jacobian ( [ j2-g2 ] ) is a quintic in @xmath190 , only four of the roots appear as the equations of the boundary of @xmath327 in ( [ eqn : boundaryd - x - g2 - 1])-([eqn : boundaryd - x - g2 - 3 ] ) . the fifth root @xmath368 only intersects with @xmath327 at the point @xmath369 . the factorization of @xmath361 in ( [ j2-g2 ] ) and the equations for the boundaries of @xmath327 given in ( [ eqn : boundaryd - y - g2 - 1])-([eqn : boundaryd - x - g2 - 3 ] ) will be used in sections [ sect : measurea12inftyg2-r ] , [ sect : measureainftyg2-r ] to determine explicit expressions for the weights which appear in the spectral measures @xmath370 over @xmath305 in terms of elliptic integrals . from ( [ j2-g2 ] ) we see that the jacobian vanishes only on the boundary of @xmath327 , or over @xmath10 only on the boundaries of the images of the fundamental domain @xmath208 under @xmath144 . since @xmath361 is real , @xmath371 , and we have the following expressions for the jacobian @xmath361 : @xmath372 , @xmath373 and @xmath367 . note that the expression under the square root is real and non - negative since @xmath364 is . then @xmath374 and from theorem [ thm : measureg2-t2 ] and ( [ eqn : measuret2=12c ] ) we obtain the joint spectral measure @xmath375 ( over @xmath327 ) for the graphs @xmath259 , @xmath260 , is @xmath376 we now compute the spectral measure @xmath378 over @xmath305 , which is determined by its moments @xmath379 for all @xmath380 , where @xmath381 for @xmath260 respectively . for @xmath382 we set @xmath383 in ( [ eqn : integral_overdj - g2 ] ) and integrate with respect to @xmath350 . similarly , setting @xmath384 in ( [ eqn : integral_overdj - g2 ] ) , the measure @xmath385 is obtained by integrating with respect to @xmath190 . more explicitly , using the expressions for the boundaries of @xmath327 given in ( [ eqn : boundaryd - y - g2 - 1])-([eqn : boundaryd - x - g2 - 3 ] ) , the spectral measure @xmath382 ( over @xmath386 $ ] ) for the graph @xmath132 is @xmath387 , where @xmath388 is given by @xmath389 , \\ \displaystyle{12 \int_{(x^2 + 2x-7)/4}^{-5(x+1)+2(x+2)^{3/2 } } |j(x , y)|^{-1 } \ , \mathrm{d}y } & \textrm { for } x \in [ -1,7 ] . \end{array } \right.\ ] ] the weight @xmath388 is an integral of the reciprocal of the square root of a cubic in @xmath350 , and thus can be written in terms of the complete elliptic integral @xmath390 of the first kind , @xmath391 . using ( * ? ? ? 235.00 ) , @xmath392 for @xmath393 $ ] , where @xmath394 , whilst for @xmath395 $ ] , @xmath396 the weight @xmath388 is illustrated in figure [ fig - jx - t2-g2 ] , up to a factor @xmath397 . the spectral measure @xmath385 ( over @xmath398 $ ] ) for the graph @xmath141 is @xmath399 , where @xmath400 is given by @xmath401 , \\ & \displaystyle{12 \int_{-1 + 4p_1(y)+4p_1(y)^2}^{-1 + 4p_3(y)+4p_3(y)^2 } |j(x , y)|^{-1 } \ , \mathrm{d}x + 12 \int_{-1 + 4p_2(y)+4p_2(y)^2}^{-1 + 2(y+2)^{1/2 } } |j(x , y)|^{-1 } \ , \mathrm{d}x } & \textrm { for } y \in [ 10/27,5 ] , \\ & \displaystyle{12 \int_{-1 + 4p_2(y)+4p_2(y)^2}^{-1 + 2(y+2)^{1/2 } } |j(x , y)|^{-1 } \ , \mathrm{d}x } & \textrm { for } y \in [ 5,14].\end{aligned}\ ] ] a numerical plot of the weight @xmath400 is illustrated in figure [ fig - jy - t2-g2 ] , again up to a factor @xmath397 . , width=192 ] , width=340 ] we now consider the fundamental representations @xmath403 , @xmath260 , of @xmath0 . as discussed in section [ sect : rep_theoryg2t2 ] , their corresponding mckay graphs are @xmath404 , @xmath260 . this section follows the same arguments as ( * ? ? ? * @xmath1456.2 ) . the new feature here is the presence of terms such as @xmath405 for @xmath406 , where @xmath407 is the unilateral shift to the right on @xmath408 , which correspond to the fact that certain edges on the graphs @xmath409 , @xmath260 , only appear when far enough away from the boundary of the graph . we define self - adjoint operators @xmath165 , @xmath166 on @xmath167 by @xmath410 let @xmath267 denote the vector @xmath268 . then @xmath411 is identified with the adjacency matrix of @xmath404 , @xmath260 , where we regard the vector @xmath270 as corresponding to the vertex @xmath271 of @xmath404 , and the operators of the form @xmath412 which appear as terms in @xmath411 as corresponding to the edges on @xmath404 . then @xmath413 corresponds to the vertex @xmath274 of @xmath404 for any @xmath414 , and applying @xmath415 to @xmath270 gives a vector @xmath277 in @xmath416 , where @xmath279 gives the number of paths of length @xmath19 on @xmath404 from @xmath271 to the vertex @xmath274 . a term of the form @xmath417 , @xmath418 , will be zero if @xmath419 . thus for example , @xmath420 , which corresponds to the fact that there is no self - loop at the point @xmath421 for any @xmath422 , whereas @xmath423 for all @xmath424 , corresponding to the fact that there is a self - loop at these points . it is not immediately obvious that these two operators commute . however this can be easily seen from the fact that @xmath404 are the multiplication graphs for the characters @xmath304 of the fundamental representations @xmath403 , @xmath260 , of @xmath0 , where the vertices of @xmath404 are labeled by the characters of the irreducible representations of @xmath0 . any vector @xmath425 can be written as a linear combination of elements of the form @xmath426 . this is not obvious from the definition of the operators @xmath411 given in ( [ eqn : vn1 ] ) , ( [ eqn : vn2 ] ) . however this also can be seen from the fact that @xmath404 are the multiplication graphs for the characters of the fundamental representations of @xmath0 , where @xmath270 corresponds to the character @xmath427 of the trivial representation . the characters of all other irreducible representations can be written as a linear combination of products of the form @xmath428 , which follows from ( * ? ? ? * proposition 1 ) where @xmath429 and @xmath430 . thus the vector @xmath270 is cyclic in @xmath167 , and we have @xmath431 . we define a state @xmath13 on @xmath432 by @xmath281 . since @xmath432 is abelian and @xmath270 is cyclic , we have that @xmath13 is a faithful state on @xmath432 . then by ( * ? ? ? * remark 2.3.2 ) the support of @xmath433 is equal to the joint spectrum @xmath434 of @xmath165 , @xmath166 . recall the decompositions of @xmath435 and @xmath436 for the characters of @xmath0 given in section [ sect : rep_theoryg2t2 ] . these can be written as @xmath437 , where @xmath438 is the adjacency matrix of @xmath409 , the mckay graph for the fundamental representation @xmath403 of @xmath0 . this equation can be interpreted as meaning that @xmath411 ( identified with the adjacency matrix @xmath438 of @xmath409 ) has eigenvector @xmath439 for eigenvalue @xmath440 , @xmath441 ^ 2 $ ] . thus the spectrum of @xmath411 is given by @xmath442 , and the joint spectrum @xmath434 is @xmath327 . the moments @xmath443 count the number of closed paths of length @xmath19 on the graph @xmath404 which start and end at the apex vertex @xmath271 . we prove in section [ sect : measures_akg2 ] that the measure given by @xmath444 is the joint spectral measure over @xmath10 of @xmath411 , @xmath260 , where @xmath445 is the uniform lebesgue measure on @xmath309 , @xmath204 . in fact , the measure @xmath315 is the joint spectral measure over @xmath10 for the pair of mckay graphs ( @xmath446,@xmath447 ) for any pair @xmath326 of irreducible representations of @xmath0 . we see that @xmath314 is ( up to some scalar ) the reduced haar measure of @xmath0 ( c.f . * @xmath1456.3 ) ) . we now determine the spectral measure @xmath448 over @xmath305 . from ( [ eqn : measuret2=12c ] ) and ( [ eqn : integral_overdj - g2 ] ) , with the measure given in section [ sect : measureainftyg2-t2 ] , we have that @xmath449 where @xmath327 is as in section [ sect : measurea12inftyg2-d ] , and @xmath450 for @xmath260 respectively . thus the joint spectral measure over @xmath327 is @xmath451 , which is the reduced haar measure on @xmath0 ( * ? ? ? * @xmath1456.3 ) . the measure @xmath452 over @xmath453 is obtained by integrating with respect to @xmath350 in ( [ eqn : integral_overdj - g2 - 2 ] ) , whilst the measure @xmath454 over @xmath455 is obtained by integrating with respect to @xmath190 in ( [ eqn : integral_overdj - g2 - 2 ] ) . more explicitly , using the expressions for the boundaries of @xmath327 given in ( [ eqn : boundaryd - y - g2 - 1])-([eqn : boundaryd - x - g2 - 3 ] ) , the spectral measure @xmath452 ( over @xmath386 $ ] ) for the graph @xmath163 is @xmath456 , where @xmath457 is given by @xmath458 , \\ \displaystyle{\int_{(x^2 + 2x-7)/4}^{-5(x+1)+2(x+2)^{3/2 } } |j(x , y)| \ , \mathrm{d}y } & \textrm { for } x \in [ -1,7 ] . \end{array } \right.\ ] ] the weight @xmath457 is the integral of the square root of a cubic in @xmath350 , and thus can be written in terms of the complete elliptic integrals @xmath390 , @xmath459 of the first , second kind respectively , where @xmath391 and @xmath460 . using ( * ? ? ? * equation 235.14 ) , @xmath457 is given by @xmath461,\end{aligned}\ ] ] for @xmath393 $ ] , where @xmath394 , whilst for @xmath395 $ ] , @xmath457 is given by @xmath462.\end{aligned}\ ] ] the weight @xmath457 is illustrated in figure [ fig - jx - g2 ] , up to a factor @xmath397 . the spectral measure @xmath454 ( over @xmath398 $ ] ) for the graph @xmath170 is @xmath463 , where @xmath464 is given by @xmath465 , \\ & \displaystyle{\int_{-1 + 4p_1(y)+4p_1(y)^2}^{-1 + 4p_3(y)+4p_3(y)^2 } |j(x , y)| \ , \mathrm{d}x + \int_{-1 + 4p_2(y)+4p_2(y)^2}^{-1 + 2(y+2)^{1/2 } } |j(x , y)| \ , \mathrm{d}x } & \textrm { for } y \in [ 10/27,5 ] , \\ & \displaystyle{\int_{-1 + 4p_2(y)+4p_2(y)^2}^{-1 + 2(y+2)^{1/2 } } |j(x , y)| \ , \mathrm{d}x } & \textrm { for } y \in [ 5,14].\end{aligned}\ ] ] a numerical plot of the weight @xmath464 is illustrated in figure [ fig - jy - g2 ] , again up to a factor @xmath397 . , width=192 ] , width=340 ] suppose @xmath83 is the nimrep associated to a @xmath0 braided subfactor at some finite level @xmath47 with vertex set @xmath466 . we define a state @xmath13 on @xmath467 by @xmath468 , where @xmath267 is the basis vector in @xmath467 corresponding to a distinguished vertex @xmath151 . note that the state @xmath13 ( and thus the spectral measure ) depends on the choice of distinguished vertex @xmath151 . we choose the distinguished vertex @xmath151 to be the vertex with lowest perron - frobenius weight . consider the nimrep graph @xmath75 . the eigenvalues @xmath469 of @xmath75 are given by the ratio @xmath470 , where @xmath81 belongs to the set @xmath471 of exponents of @xmath83 , @xmath472 ( note that we are now using the dynkin labels ) , and the @xmath53-matrix at level @xmath47 is given by @xcite : @xmath473,\end{aligned}\ ] ] where @xmath474 , @xmath475 , @xmath476 , and @xmath477 , @xmath478 for @xmath479 . letting @xmath480 , @xmath481 , we obtain @xmath482,\end{aligned}\ ] ] which is ( up to a scalar factor ) nothing but the @xmath53-function @xmath483 ( see ( * ? ? ? * section 2.3 ) for a discussion on orbit functions ) . then we see that @xmath484 where @xmath485 , and hence the spectrum @xmath486 of @xmath75 is contained in @xmath229 . note that here we are using the dynkin labels whereas in section [ sect : rep_theoryg2t2 ] we used labels @xmath487 . consider now the pair of nimrep graphs @xmath75 , @xmath488 , which have joint spectrum @xmath489 . the @xmath317 cross moment @xmath490 , where @xmath491 , @xmath492 , is given by @xmath493 . let @xmath494 be the eigenvalues of @xmath75 with corresponding eigenvectors @xmath495 , normalized so that each @xmath495 has norm 1 . as the nimreps are a family of commuting matrices they can be simultaneously diagonalised , and thus the eigenvectors of @xmath75 are the same for all @xmath62 ) . then @xmath496 , where @xmath497 is the diagonal matrix with the eigenvalues @xmath494 on the diagonal , and @xmath498 is the unitary matrix whose columns are given by the eigenvectors @xmath495 , so that @xmath499 where @xmath500 is the entry of the eigenvector @xmath495 corresponding to the distinguished vertex @xmath151 . thus there is a @xmath144-invariant measure @xmath314 over @xmath10 such that @xmath501 for all @xmath62 , @xmath81 . note from ( [ eqn : evalues - nimrep - g2 ] ) , ( [ eqn : moments - nimrep - g2 ] ) that the measure @xmath314 is a discrete measure which has weight @xmath502 at the points @xmath503 for @xmath187 , @xmath504 , and zero everywhere else . thus the measure @xmath314 does not depend on the choice of @xmath62 , @xmath81 , so that the measure over @xmath10 is the same for any pair @xmath505 , even though the corresponding measures over @xmath506 , and indeed the subsets @xmath225 themselves , are different for each such pair . we will now determine this @xmath144-invariant measure @xmath314 over @xmath10 for all the known @xmath0 modular invariants , where we will focus in particular on the nimrep graphs for the fundamental generators @xmath403 , @xmath260,which have quantum dimensions @xmath507[7][12]/[4][6]$ ] , @xmath508[8][15]/[3][4][5]$ ] respectively , where @xmath509 $ ] denotes the quantum integer @xmath509 = ( q^m - q^{-m})/(q - q^{-1})$ ] for @xmath510 . the nimrep graphs @xmath511 were found in @xcite , whilst @xmath512 for the conformal embeddings at levels 3 , 4 were found in @xcite . the realisation of modular invariants for @xmath0 by braided subfactors is parallel to the realisation of @xmath2 and @xmath4 modular invariants by @xmath46-induction for a suitable braided subfactors @xcite , @xcite respectively . the realisation of modular invariants for @xmath6 is also under way @xcite . the graphs @xmath514 , @xmath260 , are associated with the trivial @xmath0 modular invariant at level @xmath47 . they are illustrated in figures [ fig - a_infty(g2)1 ] , [ fig - a_infty(g2)2 ] respectively , where the set of vertices is now given by @xmath515 . the set of edges is given by the edges between these vertices , except for certain self - loops at the cut - off which are indicated by dashed lines ( in figures [ fig - a_infty(g2)1 ] , [ fig - a_infty(g2)2 ] the dashed lines indicate the edges to be removed when @xmath169 ) . the eigenvalues @xmath516 of @xmath514 , @xmath260 , are given by the ratio @xmath517 with corresponding eigenvectors @xmath518 , where @xmath519 . then with @xmath520 , we obtain @xmath521 \qquad \label{eqn : pf - evector1-g2}\end{aligned}\ ] ] and hence we see that @xmath522 where @xmath523 is given by ( [ eqn : j[theta]-g2 ] ) and in ( [ eqn : pf - evector = j - g2 ] ) we have @xmath524 so that @xmath525 and @xmath526 . as a consequence of the identification ( [ eqn : pf - evector = j - g2 ] ) between the perron - frobenius eigenvector and the jacobian we can obtain another expression for the jacobian @xmath361 . recall that the perron - frobenius eigenvector for @xmath168 can also be written in the kac - weyl factorized form @xcite : @xmath527 now @xmath528 whilst @xmath529/(k+4)\sqrt{3 } = 64\sin(\xi ) \sin(3\xi ) \sin(4\xi ) \sin(5\xi ) \sin(6\xi ) \sin(9\xi)/(k+4)\sqrt{3}$ ] . thus we see that @xmath530 . then from ( [ eqn : pf - evector = j - g2 ] ) we have @xmath531 so that the jacobian @xmath523 can also be written as a product of sine functions . we now compute the spectral measure for @xmath514 . now summing over all @xmath535 corresponds to summing over all @xmath536 , or equivalently , since such points satisfy @xmath537 , to summing over all @xmath538 such that @xmath539 since @xmath540 and @xmath541 give the same points in @xmath10 for @xmath542 , the last three conditions are equivalent to @xmath543 if we let @xmath208 be the limit of @xmath544 as @xmath547 , then @xmath208 is a fundamental domain of @xmath10 under the action of the group @xmath144 , illustrated in figure [ fig : fund_domain - g2int2 ] . since @xmath548 along the boundary of @xmath208 , which is mapped to the boundary of @xmath327 under @xmath549 , we can include points on the boundary of @xmath208 in the summation in ( [ eqn : sum_for_measure - a(g2 ) ] ) . since @xmath364 is invariant under the action of @xmath144 , we have @xmath550 where @xmath551 is the image of @xmath544 under the action of the weyl group @xmath552 and some additional points which lie on the boundaries of the fundamental domains ( i.e. where @xmath548 ) . we illustrate the points @xmath553 such that @xmath554 in figure [ fig : c2w - g2 ] . the points in the interior of the fundamental domain @xmath208 , those enclosed by the dashed line , correspond to the vertices of the graph @xmath555 . we can now easily deduce the spectral measure ( over @xmath10 ) for @xmath402 claimed in section [ sect : measureainftyg2-t2 ] . letting @xmath547 , the measure @xmath562 becomes the uniform lebesgue measure @xmath563 on @xmath10 . thus : then the spectral measure for @xmath514 over @xmath327 or @xmath305 , @xmath260 , has the same weights as the spectral measure for the infinite graph @xmath404 given in section [ sect : measureainftyg2-r ] , but the measure here is a discrete measure . the graph @xmath568 , illustrated in figure [ fig - graph_e3_g2 ] , is the nimrep graph associated to the conformal embedding @xmath567 , and is one of two nimrep graphs associated to the modular invariant @xmath569 which is at level 3 and has exponents @xmath570 . the other nimrep graph associated to this modular invariant is @xmath571 considered in the next section . following ( * ? ? ? * @xmath1456 ) we can compute the principal graph and dual principal graph of the inclusion @xmath567 . the chiral induced sector bases @xmath572 and full induced sector basis @xmath573 , the sector bases given by all irreducible subsectors of @xmath574 $ ] and @xmath575 $ ] respectively , for @xmath576 , are given by @xmath577 , [ \alpha_{(1,0)}^{\pm } ] , [ \alpha_{(2,0)}^{(1 ) } ] , [ \alpha_{(2,0)}^{(2 ) } ] \},\\ { } _ m \mathcal{x}_m & = \ { [ \alpha_{(0,0 ) } ] , [ \alpha_{(1,0)}^+ ] , [ \alpha_{(1,0)}^- ] , [ \alpha_{(2,0)}^{(1 ) } ] , [ \alpha_{(2,0)}^{(2 ) } ] , [ \eta_1 ] , [ \eta_2 ] , [ \eta_3 ] \},\end{aligned}\ ] ] where @xmath578 = [ \alpha_{(1,0)}^{\pm } ] \oplus [ \alpha_{(2,0)}^{(1 ) } ] \oplus [ \alpha_{(2,0)}^{(2)}]$ ] , @xmath579 = [ \eta_1 ] \oplus [ \eta_2 ] \oplus [ \eta_3]$ ] , and @xmath580 . the fusion graphs of @xmath581 $ ] ( solid lines ) and @xmath582 $ ] ( dashed lines ) are given in figure [ fig - full_e3_g2 ] , see also ( * ? ? ? * figure 17(a ) ) . the marked vertices corresponding to sectors in @xmath67 have been circled . note that multiplication by @xmath581 $ ] ( or @xmath582 $ ] ) does not give two copies of the nimrep graph @xmath566 as one might expect , but rather one copy each of @xmath566 and @xmath571 . this is similar to the situation for the @xmath4 conformal embedding @xmath583 ( * ? ? ? * @xmath1455.2 ) . let @xmath584 denote the injection map @xmath585 , @xmath586 and @xmath587 its conjugate . the dual canonical endomorphism @xmath588 for the conformal embedding can be read from the vacuum block of the modular invariant : @xmath589 = [ \lambda_{(0,0 ) } ] \oplus [ \lambda_{(1,1)}]$ ] . by ( * corollary 3.19 ) and the fact that @xmath590 , the canonical endomorphism @xmath591 is given by @xmath592 = [ \alpha_{(0,0 ) } ] \oplus [ \eta_1].\ ] ] then by ( * ? ? ? * theorem 4.2 ) , the principal graph of the inclusion @xmath567 of index @xmath593 is given by the connected component of @xmath594 \in { } _ n \mathcal{x}_n$ ] of the induction - restriction graph , and the dual principal graph is given by the connected component of @xmath595 \in { } _ m \mathcal{x}_m$ ] of the @xmath596-multiplication graph . the principal graph and dual principal graph are illustrated in figures [ fig - ghj_graph_e3_g2 ] and [ fig - ghj_graph_e3_g2_dual ] respectively . these principal graphs are sometimes referred to as `` haagerup with legs '' ( * ? ? ? * @xmath1454.2.4 ) . the principal graph in figure [ fig - ghj_graph_e3_g2 ] appears as the intertwiner for the quantum subgroup @xmath566 in ( * ? ? ? * @xmath1454.4 ) . one can also construct a subfactor @xmath597 with index @xmath598 , where @xmath64 is a type iii factor . this subfactor has already appeared in @xcite ( see also the appendix in @xcite , @xcite and @xcite ) . the chiral systems @xmath66 are near group @xmath12-category of type @xmath599 , where @xmath600 , generated by a self - conjugate irreducible endomorphism @xmath601 of @xmath64 and an outer action @xmath46 of @xmath83 on @xmath64 , such that @xmath602[\rho]=[\rho]=[\rho][\alpha_i]$ ] and @xmath603=\bigoplus_{i=1}^3 [ \alpha_i ] \oplus 3[\rho]$ ] , for @xmath604 . here @xmath605 , @xmath606 and @xmath607 . the index @xmath608 of @xmath609 is thus @xmath610 , where @xmath611 is the statistical dimension of @xmath612 . its principal graph is illustrated in figure [ fig - principal_graph_e3_g2 ] and is the bipartite unfolded version of the graph @xmath566 . the dual principal graph is isomorphic to the principal graph as abstract graphs ( * ? ? ? * corollary 3.7 ) . a spectral measure of @xmath619 , @xmath620 ( over @xmath10 ) , with distinguished vertex @xmath621 in figure [ fig - graph_e4star_g2 - 1 ] , is @xmath622 where @xmath241 is as in definition [ def:4measures ] and @xmath623 is the dirac measure at the point @xmath190 . j. bckenhauer and d. e. evans , modular invariants and subfactors , in _ mathematical physics in mathematics and physics ( siena , 2000 ) _ , fields inst . commun . * 30 * , 1137 , amer . soc . , providence , ri , 2001 . j. bckenhauer and d. e. evans , modular invariants from subfactors , in _ quantum symmetries in theoretical physics and mathematics ( bariloche , 2000 ) _ , contemp . math . * 294 * , 95131 , amer soc . , providence , ri , 2002 . p.f . byrd and m.d . friedman , _ handbook of elliptic integrals for engineers and scientists _ , die grundlehren der mathematischen wissenschaften , band 67 , second edition , revised . springer - verlag , new york , 1971 . a. ocneanu , paths on coxeter diagrams : from platonic solids and singularities to minimal models and subfactors . ( notes recorded by s. goto ) , in _ lectures on operator theory _ , ( ed . b. v. rajarama bhat et al . ) , the fields institute monographs , 243323 , amer . soc . , providence , r.i . , 2000 . a. ocneanu , the classification of subgroups of quantum @xmath626 , in _ quantum symmetries in theoretical physics and mathematics ( bariloche , 2000 ) _ , contemp . math . * 294 * , 133159 , amer . soc . , providence , ri , 2002 .

we explore two bilarge neutrino mixing anztze within the context of abelian flavor symmetry theories : ( ) , , , and ( ) , , . the first pattern is proposed by two of us and is favored if the atmospheric mixing angle lies in the first octant , while the second one is preferred for the second octant of . in order to reproduce the second texture , we find that the flavor symmetry should be , while for the first pattern the flavor symmetry should be extended to with and of different parity . explicit models for both mixing patterns are constructed based on the flavor symmetries and . the models are extended to the quark sector within the framework of grand unified theory in order to give a successful description of quark and lepton masses and mixing simultaneously . phenomenological implications are discussed . 2.5 cm * bilarge neutrino mixing and abelian flavor symmetry * 0.2 cm 0.5 cm gui - jun ding , s. morisi , and j. w. f. valle + .2 cm_department of modern physics , _ + _ university of science and technology of china , hefei , anhui 230026 , china _ .3 cm _ ahep group , instituto de fsica corpuscular c.s.i.c./universitat de valncia + edificio de institutos de paterna , apartado 22085 , e46071 valncia , spain _ .3 cm .3 cm _ institut fr theoretische physik und astrophysik , + universitt wrzburg , 97074 wrzburg , germany _ 0.7 cm 2truecm

1211.6506

our knowledge of the neutrino oscillation parameters has enormously improved in recent years . in particular the daya bay @xcite , reno @xcite and double chooz @xcite collaborations have established that the reactor mixing angle @xmath20 at about @xmath21 confidence level , confirming the early hints for a nonzero @xmath22 @xcite . recent global analyses @xcite of neutrino oscillation parameters , including the data released at the neutrino-2012 conference , find that @xmath22 is non - zero at about @xmath23 , and non - maximal atmospheric mixing angle @xmath7 is preferred . however , it still is nt clear which octant @xmath7 lies in . the global fit of ref . @xcite , prefers @xmath7 in the second octant with the best fit value @xmath24 for normal ( inverted ) neutrino mass hierarchy , although this hint is quite marginal and first octant values of @xmath7 are well inside the @xmath25 range for normal hierarchy and at @xmath26 for inverted spectrum . while the independent phenomenological analyses of atmospheric neutrino data in ref . @xcite obtain a preference for @xmath7 in the first octant for both mass hierarchies and exclude maximal mixing at the @xmath27 level , the best fit value is found to be @xmath28 for normal ( inverted ) neutrino spectrum . alternative recent global fits claim both the first and second @xmath7 octants are possible @xcite . as for the mass - squared difference , the best fit values of @xmath29 and @xmath30 are @xmath31 and @xmath32 respectively , which lead to @xmath33 . here the values shown in parentheses correspond to the inverted neutrino mass hierarchy . note that the three groups give almost the same @xmath34 ranges for the lepton mixing parameters . from the theoretical or model - building point of view , one implication of this significant experimental progress is that it excludes the tri - bimaximal mixing ansatz for neutrino mixing @xcite , unless the underlying theory is capable of providing sufficiently large corrections . so far many suggestions have been advanced to explain the new data , in particular the largish @xmath22 @xcite . instead of seeking for new mass - independent lepton mixing matrices to replace the tri - bimaximal pattern @xcite , which may be derived from certain discrete flavor symmetries , ref . @xcite proposed a novel wolfenstein - like _ ansatz _ for the neutrino mixing matrix . in this scheme , all three lepton mixing angles are assumed to be of the same order to first approximation @xmath35 where @xmath36 is the cabibbo angle , and the symbol @xmath37 " implies that the above relations contain unknown factors of order one , the freedom in these factors can be used to obtain an adequate description of the neutrino mixing . inspecting the global data fitting @xcite , we see that @xmath38 and @xmath39 , which is proposed in the so - called tri - bimaximal - cabibbo mixing @xcite and also appeared in the context of quark - lepton complementarity @xcite . such bilarge mixing pattern @xcite would clearly provide a good leading order approximation for the current neutrino mixing pattern , if the atmospheric neutrino mixing angle @xmath7 turns out to lie in the first octant . however , the second octant of @xmath7 can not be ruled out and is supported by the analyses in refs . @xcite . in this case the texture @xmath40 could be taken as a viable model - building standard . we shall refer to two mixing patterns as @xmath1 and @xmath5 textures respectively . the difference between @xmath1 and @xmath5 mixing lies in the order of magnitude of the atmospheric mixing angle @xmath7 ; the @xmath1 mixing pattern would be favored if future experiments establish that @xmath7 belongs to the first octant and the deviation from maximal mixing is somewhat large ; otherwise , @xmath5 mixing is preferred . it is well - known that the observed hierarchies of masses and flavor mixing in the quarks and charged leptons sectors can be conveniently characterized by the cabibbo angle . as a result the @xmath1 and @xmath5 parametrization may have deep implications for the theoretical formulation of the ultimate unified theory of flavor . a lot of work in the literature has demonstrated that the smallness and hierarchy of the quark masses and mixing angles can be naturally generated in theories which , at low energy , are described effectively by an abelian horizontal symmetry , which is explicitly broken by a small parameter @xcite . it certainly follows a natural path to try and apply these ideas on abelian family symmetries developed for the quarks to the lepton sector . in this work , we shall investigate whether and how the @xmath41 and @xmath5 textures can be reproduced naturally from the abelian horizontal flavor symmetry . for generality we assume that the light neutrino masses arise from lepton - number - violating effective weinberg - like operators . the paper is organized as follows . in section [ sec : th ] , we present the effective low energy theory for the abelian @xmath42 flavor symmetry and its extension to @xmath9 . we find that , in order to produce the @xmath1 texture without fine - tuning , the family symmetry should be @xmath9 with @xmath10 and @xmath11 of opposite parity . models for the @xmath1 and @xmath5 schemes are constructed in section [ sec : bl1 ] and section [ sec : bl2 ] respectively . these models are extended to include quarks within the @xmath14 grand unified theory ( gut ) , the observed patterns of both quark and lepton masses and flavor mixings are reproduced , and the general phenomenological predictions of the models are discussed . finally , our conclusions are summarized in section [ sec : conclusion ] . our theoretical framework is defined as follows . for definiteness we consider a low energy effective theory with the same particle content as the supersymmetric standard model ( sm ) . in addition to supersymmetry and the sm gauge symmetry , we introduce a horizontal @xmath42 symmetry and a sm singlet chiral superfield @xmath43 which is charged under the @xmath42 family symmetry ; without loss of generality , we normalize its charge to @xmath44 . the effective yukawa couplings of the quarks and leptons are generated from nonrenormalizable superpotential terms of the form @xmath45 where @xmath46 are higgs doublets , @xmath47 and @xmath48 are the left - handed quark and lepton doublets respectively , @xmath49 , @xmath50 and @xmath51 are the right - handed up - type quark , down - type quark and charged lepton superfields respectively , and @xmath52 , @xmath53 are generation indices . the parameter @xmath54 is the cutoff scale of the @xmath42 symmetry , and @xmath55 denotes the @xmath42 charge of the field @xmath56 . note that @xmath57 and @xmath58 do not appear in the exponents since one can always set the horizontal charges of the higgs doublet @xmath59 and @xmath59 to zero by redefinition of the @xmath42 charges . the last term of eq.([eq : u(1 ) ] ) is the high - dimensional version of the effective lepton - number - violating weinberg operator . for the froggatt - nielsen flavon field @xmath43 , the supersymmetric action contains a fayet - iliopoulos term and the associated d - term in the scalar potential provides a large vacuum expectation value ( vev ) for the scalar component of @xmath43 . the d - term in the potential is given by @xmath60 where @xmath61 is the fayet - iliopoulos term . the vanishing of @xmath62 requires @xmath63 we note that this flavor symmetry breaking mechanism is also frequently exploited in discrete flavor symmetry model building @xcite . once the horizontal symmetry is broken by the vev @xmath64 , one obtains the quark and lepton mass matrices whose elements are suppressed by powers of the small parameter @xmath65 , which for simplicity is usually assumed to be characterized by the cabibbo angle , i.e. , @xmath66 , then we have @xmath67 where @xmath68 is the electroweak scale vev of the higgs doublet @xmath46 . the factors @xmath69 , @xmath70 , @xmath71 and @xmath72 are not constrained by the flavor symmetry and are usually assumed to be of order one , the freedom in these factors is used in order to obtain a quantitative description of the fermion masses and flavor mixings . since the holomorphicity of the superpotential forbids nonrenormalizable terms with a negative power of the superfield @xmath43 , one has @xmath73 if @xmath74 . similarly @xmath75 if @xmath76 , @xmath77 if @xmath78 , and @xmath79 if @xmath80 . in our framework , the light neutrino masses are generated by the high - dimensional effective weinberg operators shown in the last term of eq.([eq : u(1 ) ] ) , consequently , the light neutrinos are majorana particles and its mass matrix @xmath81 is symmetric with @xmath82 to generate light neutrino mass via type i seesaw mechanism , the structure of the light neutrino mass matrix @xmath81 is independent of the @xmath83 charge assignments @xcite , unless there are holomorphic zeros in neutrino dirac mass matrix @xmath84 or in majorana mass matrix @xmath85 for the heavy fields @xmath86 . ] furthermore , if all the horizontal charges are positive , the hierarchial structure of the mass matrices shown in eq.([eq : mass ] ) allows a simple order of magnitude estimate for the various mass ratios and mixing angles : @xmath87},\quad\quad \frac{m_{\ell_i}}{m_{\ell_j}}\sim\lambda^{f(l_i)-f(l_j)+f(e^c_i)-f(e^c_j)},\quad\quad \sin\theta_{ij}\sim\lambda^{f(l_i)-f(l_j)}\,\end{aligned}\ ] ] where @xmath88 is the light neutrino mass , and @xmath89 denotes the element of the quark cabibbo - kobayashi - maskawa quark - mixing matrix ( ckm ) mixing matrix . we note that the sign @xmath90 implies that there is an unknown order one coefficient in each relation , so that the actual value of the mass ratios and mixing angles may slightly depart from the naive `` power counting '' estimate . moreover , if some fields carry negative @xmath91 charges , then holomorphy plays an important role and the estimates ( [ eq : estimate ] ) could be violated as well . for the @xmath41 mixing pattern , both @xmath92 and @xmath93 are of order @xmath94 , then we should require @xmath95 this implies @xmath96 , as a result , we have @xmath97 . therefore we conclude that the @xmath1 mixing pattern can not be naturally produced from a pure @xmath42 flavor symmetry . turning to the @xmath5 mixing pattern given by @xmath98 , @xmath2 and @xmath3 , one should choose @xmath99 10.mm then we have the @xmath100 and @xmath101 @xmath102 entries of the charged lepton mass matrix are of the same order , hence the diagonalization of the charged lepton mass matrix leads to large 2 - 3 mixing . in addition , we obtain @xmath103[mnuiii ] clearly the ( 2 - 3 ) sector of the light neutrino mass matrix has a democratic structure , thus large mixing in this ( 2 - 3 ) sector is naturally obtained . however , barring the presence of special cancellations , the masses of the second and the third light neutrinos are typically expected to be of the same order in this case . as a result , the three light neutrinos are quasi - degenerate and strong parameter fine - tuning is required in order to account for the hierarchy between the measured mass squared differences @xmath104 and @xmath105 . 10.mm in order to avoid this kind of fine - tuning in obtaining an acceptable pattern of neutrino oscillation parameters , we must go beyond the pure @xmath42 flavor symmetry case considered above . let us now move to the extended flavor symmetry @xmath106 . this kind of abelian symmetry is somewhat complex and not yet fully discussed , as far as we know , since most of the previous work concentrated on @xmath42 or @xmath107 flavor symmetry . we now consider @xcite three sm singlet superfields @xmath108 , @xmath109 and @xmath110 with the horizontal charges @xmath111 in exactly the same way as the single @xmath42 case , the three flavons @xmath108 , @xmath109 and @xmath110 could get non - vanishing vevs determined by corresponding the d - terms . in general the vevs @xmath112 , @xmath113 and @xmath114 are different @xcite . for simplicity , we take in what follows : @xmath115 , @xmath116 and @xmath117 . the effective yukawa couplings are given by extending eq.([eq : u(1 ) ] ) with new flavons @xmath108 , @xmath109 and @xmath110 as follows : @xmath118}\left(\frac{\theta_3}{\lambda}\right)^{\left[z_n(q_i)+z_n(u^c_j)\right]}\\ \nonumber&&\qquad + ( y_d)_{ij}q_id^c_jh_d\left(\frac{\theta_1}{\lambda}\right)^{f(q_i)+f(d^c_j)}\left(\frac{\theta_2}{\lambda}\right)^{\left[z_m(q_i)+z_m(d^c_j)\right]}\left(\frac{\theta_3}{\lambda}\right)^{\left[z_n(q_i)+z_n(d^c_j)\right]}\\ \nonumber&&\qquad + ( y_e)_{ij}l_ie^c_jh_d\left(\frac{\theta_1}{\lambda}\right)^{f(l_i)+f(e^c_j)}\left(\frac{\theta_2}{\lambda}\right)^{\left[z_m(l_i)+z_m(e^c_j)\right]}\left(\frac{\theta_3}{\lambda}\right)^{\left[z_n(l_i)+z_n(e^c_j)\right]}\\ \label{eq : u(1)_extended}&&\qquad + ( y_{\nu})_{ij}\frac{1}{\lambda}l_{i}l_{j}h_{u}h_{u}\left(\frac{\theta_1}{\lambda}\right)^{f(l_i)+f(l_j)}\left(\frac{\theta_2}{\lambda}\right)^{\left[z_m(l_i)+z_m(l^{}_j)\right]}\left(\frac{\theta_3}{\lambda}\right)^{\left[z_n(l_i)+z_n(l^{}_j)\right]}\end{aligned}\ ] ] where @xmath119 is the @xmath120 charge of the field @xmath56 , and the brackets @xmath121 $ ] around the exponents denote that we are modding out by @xmath10 ( @xmath11 ) according to the @xmath122 ( @xmath123 ) addition rule , namely , @xmath124 = \left\ { \begin{array}{lll } r & \mbox{if } & r < m\\ r - m & \mbox{if } & r\ge m\\ \end{array}\right . \ ] ] where @xmath125 . we note that the charge assignments of the higgs doublets @xmath59 and @xmath126 have been set to @xmath127 by redefining the flavor symmetry charges of the fields . thus , the fermion mass matrix can be expressed in term of the horizontal charges as @xmath128+\left[z_n(q_i)+z_n(u^c_j)\right]}v_u\ , , \\ \nonumber&&(m_d)_{ij}=(y_{d})_{ij}\lambda^{f(q_i)+f(d^c_j)+\left[z_m(q_i)+z_m(d^c_j)\right]+\left[z_n(q_i)+z_n(d^c_j)\right]}v_d\ , , \\ \nonumber&&(m_{e})_{ij}=(y_{e})_{ij}\lambda^{f(l_i)+f(e^c_j)+\left[z_m(l_i)+z_m(e^c_j)\right]+\left[z_n(l_i)+z_n(e^c_j)\right]}v_d\ , , \\ & & ( m_{\nu})_{ij}=(y_{\nu})_{ij}\lambda^{f(l_i)+f(l_j)+\left[z_m(l_i)+z_m(l^{}_j)\right]+\left[z_n(l_i)+z_n(l^{}_j)\right]}\frac{v^2_u}{\lambda}\,.\end{aligned}\ ] ] consider the quark sector , the flavor mixing angles there are given by @xmath129-\big[z_{m(n)}(q_j)+z_{m(n)}(u^c_j)\big]}\,,\\ v_{ij}^{d}&\sim&\lambda^{(f(q_i)+f(d^c_j))-(f(q_j)+f(d^c_j))+\big[z_{m(n)}(q_i)+z_{m(n)}(d^c_j)\big]-\big[z_{m(n)}(q_j)+z_{m(n)}(d^c_j)\big]}\,.\end{aligned}\ ] ] for @xmath130 the ckm matrix elements describing the charged current weak interaction of quarks behave approximatively as @xmath131 and therefore the ckm mixing @xmath132 is expected to scale as @xmath133 . in order to compare with the pure @xmath42 horizontal symmetry case , we can define an effective flavor charge in the general case @xmath134 as @xmath135 then it is clear that @xmath136 where @xmath137 and we have used eq.([pmm ] ) and the fact that @xmath138-\big[z_m(q_j)+z_m(u^c_j)\big]=z_m(q_i)-z_m(q_j)\pm \alpha m \\ & & \big[z_m(q_i)+z_m(d^c_j)\big]-\big[z_m(q_j)+z_m(d^c_j)\big]=z_m(q_i)-z_m(q_j)\pm \alpha m\end{aligned}\ ] ] where @xmath139 . the condition for the value @xmath140 follows similarly . likewise for the lepton sector , one obtains @xmath141 therefore , the masses and mixing angles can be enhanced or suppressed by @xmath142 relative to the scaling predictions obtained when the family symmetry is the continuous flavor symmetry @xmath143 because of the discrete nature of @xmath144 . note that in the case where the light neutrino masses are generated by the type i seesaw mechanism and all fermion charges are positive , the neutrino masses and mixing angles still do not depend on the details of the right - handed neutrino sector , except for the possible enhancement or suppression associated to the @xmath144 flavor symmetry . 10.mm furthermore , when the flavor symmetry is reduced to @xmath8 by taking @xmath145 , all the above results remain valid . it is remarkable that we can employ the @xmath8 flavor symmetry to maintain the @xmath5 mixing while achieving very different neutrino masses without fine - tuning . we shall restrict our attention to the case of a @xmath146 symmetry which is the minimal nontrivial @xmath122 group ( see , for example , the explicit model construction given in sec . [ sec : bl2 ] below ) . in this case just the @xmath122 symmetry can reproduce a hierarchy in neutrino masses of order @xmath147 consistent with the observed ratio of solar - to - atmospheric splittings . in contrast , note that since the reactor neutrino mixing is necessarily of order @xmath148 the @xmath8 flavor symmetry can not produce the @xmath41 mixing pattern . indeed for such @xmath1 texture one has @xmath2 and @xmath4 , which is in conflict with the required linear behavior of the reactor mixing angle @xmath3 . note parenthetically that the @xmath149 group consists of only the identity element , so the group @xmath150 is isomorphic to @xmath42 , and the @xmath149 charge of field is 0 , hence the flavor symmetry @xmath150 produces a wrong scaling behavior @xmath97 . we now turn to the realistic case of the @xmath9 flavor symmetry . if both solar and atmospheric neutrino mixing angles are of order @xmath94 then the reactor angle would be constrained to be of order @xmath151 . as a result , one can have @xmath3 if the parity of @xmath10 and @xmath11 is opposite . this is an interesting observation of the present work . in section [ sec : bl1 ] , a concrete model for the @xmath1 mixing pattern is presented based on the flavor symmetry @xmath12 . since an abelian flavor symmetry can not predict the exact value of the @xmath152 coefficients in front of each invariant operator , we must content ourselves with explaining the orders of magnitude of fermion masses and flavor mixing parameters . to identify the phenomenologically acceptable mass matrices , we will estimate the various mass ratios and mixing angles as approximate powers of the small parameter @xmath94 . the hierarchies in the quark mixing angles are clearly displayed in wolfenstein s truncated form @xcite of the parametrization of the ckm matrix @xcite : @xmath153 where the quantities @xmath154 , @xmath155 and @xmath156 are experimentally determined to be of order one . therefore the order of magnitude of the three mixing angles is given in terms of the @xmath94 as @xmath157 the charged fermion mass ratios at the grand unified theory ( gut ) scale should satisfy @xcite @xmath158 as well as @xmath159 for the intrafamily hierarchy . the first identity is the well - known @xmath160 unification relation . for the neutrinos , we required that the lepton mixing is of @xmath1 or @xmath5 type depending on the octant of @xmath7 . for the quark sector , all the explicit models are properly constructed to meet the requirement @xmath161 and @xmath162 as has been shown in the previous section , one can reproduce the @xmath1 texture within the framework of @xmath163 family symmetry , where @xmath10 and @xmath11 should have different parity . for concreteness , we shall use @xmath164 and @xmath165 for our model . for such symmetry choice the possible model realization of the @xmath1 texture is not unique . as a concrete example , here the horizontal charges of the lepton fields are taken to be @xmath166 one immediately obtains the charged lepton mass matrix @xmath167 which yields the mass ratios @xmath168 that are consistent with the experimental requirements . for the charged assignments in eq.([eq : ass1 ] ) , the light neutrino mass matrix is given by @xmath169 it predicts the light neutrino mass eigenvalues as follows : @xmath170 the neutrino mass spectrum is normal hierarchy , this is confirmed by subsequent numerical analysis . it is remarkable that this model gives rise to @xmath171 and @xmath172 , which is in excellent agreement with the experimental data . in conventional @xmath42 or @xmath8 flavor symmetries , if any ratio between neutrino masses is an odd power of the small breaking parameter , generally the mixing angle between the two neutrinos will vanish @xcite . the crucial point is that the element @xmath173 , which would have been @xmath174 under the continuous @xmath143 symmetry , is enhanced to @xmath175 due to the discrete symmetry @xmath176 . diagonalizing the mass matrices in eq.([eq : cmm ] ) and eq.([eq : nmm ] ) by the standard perturbative techniques described in refs . @xcite , we get the three lepton flavor mixing angles @xmath177 hence the @xmath1 pattern is produced automatically . note that the solar neutrino mixing @xmath92 arises from order @xmath94 contributions from the diagonalization of both @xmath178 and @xmath81 , while at leading order the reactor and the atmospheric neutrino mixing angles receive contribution only from the neutrino mass matrix @xmath81 . the off - diagonal elements @xmath179 and @xmath180 are enhanced by @xmath181 and @xmath182 respectively , hence we have @xmath3 and @xmath4 instead of the naive expectations @xmath183 and @xmath184 characteristic of the continuous flavor symmetry case . in the following , we shall extend the model to encompass also quark sector . since gut relates quarks and leptons , the transformation properties of quark fields can be determined from those of leptons . in order to give a successful description of the observed fermion mass hierarchies and mixings simultaneously under the same flavor symmetry acting on quarks and leptons we work in the framework of @xmath14 , for definiteness . another motivation of considering @xmath14 unification is the anomaly cancellation . if the @xmath42 flavor symmetry is gauged then a general assignment of flavor charges to the fields will be anomalous . one can imagine the anomaly to be canceled via the green - schwarz mechanism @xcite , however , one must check whether the correct relations are satisfied @xcite . a convenient way to ensure that the flavor charges are amenable to cancellation is to have the flavor symmetry to commute with the @xmath14 group @xcite . here we propose a model with the quark and lepton matter assignments manifestly compatible with potential unification within @xmath14 . a complete study of a realistic grand unified model model addressing the well - known problems such as the doublet - triplet splitting , the proton lifetime and gauge coupling unification , is beyond the scope of the present paper and will be studied elsewhere . in the conventional @xmath14 grand unified theory , the fields @xmath185 and @xmath48 of the same generation are assigned to a @xmath186 multiplet , the fields @xmath47 , @xmath187 and @xmath188 are unified in the @xmath189 representation . since the flavor symmetry is required to commute with the gauge symmetry , this means that the fields in each gauge multiplet transform in the same way under the flavor symmetry . consequently , the quantum numbers of the quark fields under the flavor symmetry @xmath12 are as follows : @xmath190 we note that although there are many possible assignments to produce the @xmath1 texture in the neutrino sector , only a few of them can satisfy the quark sector phenomenological constraints within @xmath14 . it is well - known that the minimal @xmath14 grand unified theory predicts that the down - type quark mass matrix is the transpose of the charged lepton mass matrix , therefore the down - type quarks and charged lepton masses are closely related : @xmath191 , @xmath192 and @xmath193 , which are in gross disagreement with the measured fermion masses and must be corrected @xcite . this can be done through the contribution of renormalizable @xcite or non - renormalizable @xcite operators to the yukawa matrices . following ref . @xcite , we introduce an additional @xmath194 singlet superfield @xmath195 transforming as a @xmath196 of @xmath14 , which has non - renormalizable couplings to fermions of the form @xmath197 . the yukawa couplings of the down - type quark and charged leptons then arise from the two @xmath198 invariant superpotential terms could in principle also give a contribution in the up sector . however , following ref . @xcite we neglect such a term since it is not needed to reproduce the up - type quark masses . ] . @xmath199}\\ \label{eq : dm}&&\qquad\times\left(\frac{\theta_3}{\lambda}\right)^{\left[z^{}_4(\mathbf{10}_i)+z_4(\bar{\mathbf{5}}_j)\right ] } , \end{aligned}\ ] ] which , after the scalar components of @xmath195 acquires a vev , lead to : @xmath200+\left[z_4(q_i)+z_4(d^c_j)\right ] } \,,\\ \label{eq : um}&&(\mathbf{y}_e)_{ij}=\big((c_1)_{ij}-3 \kappa\,(c_2)_{ij } \big)\lambda^{f(q_j)+f(d^c_i)\left[z_3(q_j)+z_3(d^c_i)\right]+\left[z_4(q_j)+z_4(d^c_i)\right]}\,,\end{aligned}\ ] ] where @xmath201 , which breaks the transposition relation between @xmath202 and @xmath203 and can explain the difference between down - type quarks and charged lepton masses . in our numerical fits , we take @xmath204 for illustration and find that realistic values for down - type quarks and charged lepton masses can be reproduced . the superpotential for the up - type quark mass is @xmath205}\left(\frac{\theta_3}{\lambda}\right)^{\left[z^{}_4(\mathbf{10}_i)+z_4(\mathbf{10}_j)\right]}\,,\ ] ] where one has @xmath206 due to the constraint of the @xmath14 gauge symmetry . then one can express the effective yukawa couplings for the up - type quark in terms of the flavor symmetry charges as @xmath207+\left[z^{}_4(q_i)+z_4(q_j)\right]}\ ] ] with the assignments dictated by eq.([eq : qs ] ) , one has the following patterns for the up- and down - type quark mass matrices , @xmath208 which yield latexmath:[\[\begin{aligned } we note that both the up and down quark sector contribute @xmath147 to the mixing element @xmath210 , therefore an accidental enhancement of @xmath211 among the undetermined order one coefficients @xmath212 , @xmath213 and @xmath214 is required in order to describe the correct cabibbo angle . the remaining ckm mixing angles @xmath215 and @xmath216 arise solely from the diagonalization of the down - type quark mass matrix @xmath217 . in addition , the pattern given by eq.([eq : quark_mm ] ) leads to the following quark mass scalings : @xmath218 which describe the experimental data satisfactorily . note that the second term in eq.([eq : dm ] ) accounts for the mass difference between the down - type quarks and charged leptons , allowing for an acceptable charged fermion mass pattern . in order to see in a quantitative way how well the model describes the observed values of the fermion masses and mixings , we perform a numerical analysis , within three independent different seeding methods : namely flat , gaussian and exponential distributions . the modulus of the undetermined order one coefficients are taken to be random numbers with flat , gaussian and exponential distributions in turn , the corresponding phases are varied between 0 and @xmath219 . the probability density function @xmath220 of the three distributions is well known @xmath221 for flat distribution , we take @xmath222 and @xmath223 for illustration in the present work . in the case of gaussian distribution , @xmath224 we set the mean @xmath225 and the standard deviation @xmath226 in our numerical calculation . the probability density function for the exponential distribution is @xmath227 its statistic mean is @xmath228 , and @xmath94 is taken to be 1 as a typical value for numerical simulation . to the extent that our results are independent of the choice of seeding method , they are robust and not simply an artifact of the choice of the seed function . the coefficients @xmath212 , @xmath213 , @xmath214 and @xmath72 are treated as random complex numbers with arbitrary phases and absolute value in the interval of @xmath229 $ ] . then we calculate the quark and lepton masses as well as the ckm and lepton mixing matrix entries which are required to lie in the experimentally allowed ranges . the numerical results are found to be nicely consistent with the above theoretical estimates and qualitative discussions . since the flavor parameters of the quark sector are precisely measured , here we focus on the neutrino sector . as an example the predicted distributions for the light neutrino masses and atmospheric mixing parameter are shown in fig . [ fig : mixing_parameter_bl1 ] . the light neutrino masses follow the normal hierarchy pattern and , for all the points produced , though all non - vanishing , they are rather tiny , with most of the expected @xmath230 values below @xmath231 . as to the mixing angles , no specific values of @xmath232 and @xmath22 are favored within @xmath34 , and hence they are not shown in the figure , @xmath233 and @xmath234 , hence , as before , these are not shown . ] . in contrast , however , the atmospheric neutrino mixing angle @xmath7 obeys @xmath235 , which means that non - maximal @xmath7 values are preferred , as indicated by current neutrino oscillation global analyses post - neutrino 2012 @xcite , with a preference for the first octant . this has been one of our motivations for introducing @xmath1 mixing pattern , which leads to @xmath93 of order @xmath94 at leading order . [ cols="^,^,^ " , ] the recent neutrino oscillation experimental highlights : ( i ) rather large value of reactor mixing angle @xmath22 and ( ii ) indication of significant deviation of the atmospheric neutrino mixing angle @xmath7 from maximality may change our theoretical approach for constructing neutrino mass models . in this paper , we study the wolfenstein - like mixing schemes : @xmath1 mixing in which @xmath2 , @xmath3 , @xmath4 , and @xmath5 mixing , in which @xmath2 , @xmath3 , @xmath236 . the largish @xmath22 can be naturally accommodated in both of them , the two mixing patterns differ in the order of magnitude of @xmath93 , the @xmath1 texture is favored for @xmath7 in the first octant , while @xmath5 is preferred for the second octant @xmath7 . in order to produce the @xmath1 mixing without invoking unnatural cancellation , the abelian flavor symmetry should be @xmath9 with the parity of @xmath10 and @xmath11 being opposite . a concrete model based on @xmath12 family symmetry is constructed , where the light neutrino mass hierarchy @xmath171 is realized due to the discrete nature of @xmath176 . the ratio @xmath237 is expected to be of order @xmath147 in this model , which is in good agreement with experimental data in contrast with conventional @xmath42 or @xmath238 flavor symmetry constructions . furthermore , the model is embedded into the @xmath14 grand unified theory to describe the quark masses and mixing simultaneously . as for the @xmath5 mixing , it can be reproduced within the framework of pure @xmath42 flavor symmetry . however , the light neutrino mass spectrum is expected to be quasi - degenerate , hence fine - tuning of the neutrino mass parameters is needed in order to achieve the observed mass - squared differences . to improve upon this situation , the family symmetry is enlarged to @xmath13 , which gives rise to both large atmospheric neutrino mixing @xmath98 and hierarchical neutrino masses . the model is extended to @xmath14 grand unified theory as well . we show that both models can give a successful description of the observed quark and lepton masses and mixing angles , and the numerical results are nicely in agreement with the theoretical estimates and the qualitative discussions . the light neutrinos are normal mass hierarchy in both models , quasi - degenerate spectrum is strongly disfavored . if the next generation high precision neutrino oscillation experiments determine that the neutrino mass spectrum is inverted hierarchy , both our constructions will be ruled out . the present framework can not predict the cp violating phases @xmath239 , @xmath233 and @xmath234 . the @xmath240-decay effective mass @xmath241 is predicted to be rather small in both constructions , substantial part of the data are below the sensitivity of future experiments except for a region of the @xmath5 model indicated in fig . [ fig : onu2b_bl2 ] . therefore future @xmath240-decay experiments such as cuore , majorana and gerda iii will provide another important test of the present models . this work was supported by the national natural science foundation of china under grant no 10905053 , chinese academy kjcx2-yw - n29 , dfg grant wi 2639/4 - 1 and the 973 project with grant no . 2009cb825200 ; by the spanish mineco under grants fpa2011 - 22975 and multidark csd2009 - 00064 ( consolider - ingenio 2010 programme ) , by prometeo/2009/091 ( generalitat valenciana ) , and by the eu itn unilhc pitn - ga-2009 - 237920 . is supported by a juan de la cierva grant . an _ et al . _ [ daya - bay collaboration ] , phys . lett . * 108 * , 171803 ( 2012 ) [ arxiv:1203.1669 [ hep - ex ] ] . j. k. ahn _ et al . _ [ reno collaboration ] , phys . rev . lett . * 108 * , 191802 ( 2012 ) [ arxiv:1204.0626 [ hep - ex ] ] . y. abe _ et al . _ [ double - chooz collaboration ] , phys . lett . * 108 * , 131801 ( 2012 ) [ arxiv:1112.6353 [ hep - ex ] ] . k. abe _ et al . _ [ t2k collaboration ] , phys . lett . * 107 * , 041801 ( 2011 ) [ arxiv:1106.2822 [ hep - ex ] ] . p. adamson _ et al . _ [ minos collaboration ] , phys . lett . * 107 * , 181802 ( 2011 ) [ arxiv:1108.0015 [ hep - ex ] ] . d. v. forero , m. tortola and j. w. f. valle , phys . d * 86 * , 073012 ( 2012 ) [ arxiv:1205.4018 [ hep - ph ] ] . g. l. fogli , e. lisi , a. marrone , d. montanino , a. palazzo and a. m. rotunno , phys . d * 86 * , 013012 ( 2012 ) [ arxiv:1205.5254 [ hep - ph ] ] . m. c. gonzalez - garcia , m. maltoni , j. salvado and t. schwetz , arxiv:1209.3023 [ hep - ph ] . p. f. harrison , d. h. perkins and w. g. scott , phys . b * 530 * ( 2002 ) 167 [ hep - ph/0202074 ] ; p. f. harrison and w. g. scott , phys . b * 535 * ( 2002 ) 163 [ hep - ph/0203209 ] ; z. -z . xing , phys . b * 533 * ( 2002 ) 85 [ hep - ph/0204049 ] ; x. g. he and a. zee , phys . b * 560 * , 87 ( 2003 ) [ hep - ph/0301092 ] . h. -j . he and f. -r . yin , phys . rev . d * 84 * , 033009 ( 2011 ) [ arxiv:1104.2654 [ hep - ph ] ] ; y. shimizu , m. tanimoto and a. watanabe , prog . * 126 * , 81 ( 2011 ) [ arxiv:1105.2929 [ hep - ph ] ] ; j. -m . chen , b. wang and x. -q . li , phys . d * 84 * , 073002 ( 2011 ) [ arxiv:1106.3133 [ hep - ph ] ] ; z. -z . xing , [ arxiv:1106.3244 [ hep - ph ] ] ; e. ma and d. wegman , phys . lett . * 107 * , 061803 ( 2011 ) [ arxiv:1106.4269 [ hep - ph ] ] ; y. -j . zheng and b. -q . ma , [ arxiv:1106.4040 [ hep - ph ] ] ; x. -g . he and a. zee , phys . d * 84 * , 053004 ( 2011 ) [ arxiv:1106.4359 [ hep - ph ] ] ; s. zhou , phys . b * 704 * , 291 ( 2011 ) [ arxiv:1106.4808 [ hep - ph ] ] ; n. haba and r. takahashi , phys . lett . b * 702 * , 388 ( 2011 ) [ arxiv:1106.5926 [ hep - ph ] ] ; d. meloni , jhep * 1110 * , 010 ( 2011 ) [ arxiv:1107.0221 [ hep - ph ] ] ; s. morisi , k. m. patel and e. peinado , phys . rev . d * 84 * , 053002 ( 2011 ) [ arxiv:1107.0696 [ hep - ph ] ] ; w. chao and y. -j . zheng , arxiv:1107.0738 [ hep - ph ] ; h. zhang and s. zhou , phys . b * 704 * , 296 ( 2011 ) [ arxiv:1107.1097 [ hep - ph ] ] ; x. chu , m. dhen and t. hambye , jhep * 1111 * , 106 ( 2011 ) [ arxiv:1107.1589 [ hep - ph ] ] ; p. s. bhupal dev , r. n. mohapatra and m. severson , phys . rev . d * 84 * , 053005 ( 2011 ) [ arxiv:1107.2378 [ hep - ph ] ] ; s. antusch and v. maurer , phys . d * 84 * , 117301 ( 2011 ) [ arxiv:1107.3728 [ hep - ph ] ] ; w. rodejohann , h. zhang and s. zhou , nucl . phys . b * 855 * , 592 ( 2012 ) [ arxiv:1107.3970 [ hep - ph ] ] ; y. h. ahn , h. -y . cheng and s. oh , [ arxiv:1107.4549 [ hep - ph ] ] ; s. f. king and c. luhn , jhep * 1109 * , 042 ( 2011 ) [ arxiv:1107.5332 [ hep - ph ] ] ; q. -h . cao , s. khalil , e. ma and h. okada , phys . d * 84 * , 071302 ( 2011 ) [ arxiv:1108.0570 [ hep - ph ] ] ; d. marzocca , s. t. petcov , a. romanino and m. spinrath , jhep * 1111 * , 009 ( 2011 ) [ arxiv:1108.0614 [ hep - ph ] ] ; s. -f . ge , d. a. dicus and w. w. repko , [ arxiv:1108.0964 [ hep - ph ] ] ; riazuddin , arxiv:1108.1469 [ hep - ph ] ; f. bazzocchi , [ arxiv:1108.2497 [ hep - ph ] ] ; t. araki and c. -q . geng , jhep * 1109 * , 139 ( 2011 ) [ arxiv:1108.3175 [ hep - ph ] ] ; s. antusch , s. f. king , c. luhn and m. spinrath , nucl . b * 856 * , 328 ( 2012 ) [ arxiv:1108.4278 [ hep - ph ] ] ; h. fritzsch , z. -z . xing and s. zhou , jhep * 1109 * , 083 ( 2011 ) [ arxiv:1108.4534 [ hep - ph ] ] ; r. n. mohapatra and m. k. parida , phys . rev . d * 84 * , 095021 ( 2011 ) [ arxiv:1109.2188 [ hep - ph ] ] ; a. rashed and a. datta , [ arxiv:1109.2320 [ hep - ph ] ] ; p. o. ludl , s. morisi and e. peinado , [ arxiv:1109.3393 [ hep - ph ] ] ; n. okada and q. shafi , arxiv:1109.4963 [ hep - ph ] ; a. aranda , c. bonilla and a. d. rojas , [ arxiv:1110.1182 [ hep - ph ] ] ; g. -j . ding , l. l. everett and a. j. stuart , nucl . phys . b * 857 * , 219 ( 2012 ) [ arxiv:1110.1688 [ hep - ph ] ] ; d. meloni , arxiv:1110.5210 [ hep - ph ] ; s. dev , s. gupta , r. r. gautam and l. singh , phys . b * 706 * , 168 ( 2011 ) [ arxiv:1111.1300 [ hep - ph ] ] ; k. n. deepthi , s. gollu and r. mohanta , arxiv:1111.2781 [ hep - ph ] ; a. rashed , arxiv:1111.3072 [ hep - ph ] ; w. buchmuller , v. domcke and k. schmitz , arxiv:1111.3872 [ hep - ph ] ; i. d. m. varzielas , arxiv:1111.3952 [ hep - ph ] ; r. de adelhart toorop , f. feruglio and c. hagedorn , [ arxiv:1112.1340 [ hep - ph ] ] ; s. f. king and c. luhn , arxiv:1112.1959 [ hep - ph ] ; s. gupta , a. s. joshipura and k. m. patel , arxiv:1112.6113 [ hep - ph ] . h. ishimori and t. kobayashi , phys . d * 85 * , 125004 ( 2012 ) [ arxiv:1201.3429 [ hep - ph ] ] ; s. dev , r. r. gautam and l. singh , phys . b * 708 * , 284 ( 2012 ) [ arxiv:1201.3755 [ hep - ph ] ] . bhupal dev , b. dutta , r. n. mohapatra and m. severson , phys . d * 86 * , 035002 ( 2012 ) [ arxiv:1202.4012 [ hep - ph ] ] ; x. zhang and b. -q . ma , phys . lett . b * 710 * , 630 ( 2012 ) [ arxiv:1202.4258 [ hep - ph ] ] ; i. k. cooper , s. f. king and c. luhn , jhep * 1206 * , 130 ( 2012 ) [ arxiv:1203.1324 [ hep - ph ] ] ; z. -z . xing , chin . c * 36 * , 281 ( 2012 ) [ arxiv:1203.1672 [ hep - ph ] ] ; y. -l . wu , phys . b * 714 * , 286 ( 2012 ) [ arxiv:1203.2382 [ hep - ph ] ] ; y. bentov and a. zee , phys . b * 714 * , 80 ( 2012 ) [ arxiv:1203.2671 [ hep - ph ] ] ; h. -j . he and x. -j . xu , arxiv:1203.2908 [ hep - ph ] ; d. meloni , jhep * 1205 * , 124 ( 2012 ) [ arxiv:1203.3126 [ hep - ph ] ] . y. h. ahn and s. k. kang , arxiv:1203.4185 [ hep - ph ] ; i. d. m. varzielas and g. g. ross , arxiv:1203.6636 [ hep - ph ] ; d. hernandez and a. y. .smirnov , arxiv:1204.0445 [ hep - ph ] ; c. hagedorn and d. meloni , nucl . phys . b * 862 * , 691 ( 2012 ) [ arxiv:1204.0715 [ hep - ph ] ] ; m. fukugita , y. shimizu , m. tanimoto and t. t. yanagida , phys . b * 716 * , 294 ( 2012 ) [ arxiv:1204.2389 [ hep - ph ] ] ; s. zhou , arxiv:1205.0761 [ hep - ph ] ; c. hagedorn , s. f. king and c. luhn , arxiv:1205.3114 [ hep - ph ] ; g. altarelli , f. feruglio , l. merlo and e. stamou , jhep * 1208 * , 021 ( 2012 ) [ arxiv:1205.4670 [ hep - ph ] ] ; g. altarelli , f. feruglio and l. merlo , arxiv:1205.5133 [ hep - ph ] ; a. meroni , s. t. petcov and m. spinrath , arxiv:1205.5241 [ hep - ph ] ; m. j. baker , j. bordes , h. m. chan and s. t. tsou , arxiv:1206.0199 [ hep - ph ] ; x. zhang and b. -q . ma , arxiv:1206.0519 [ hep - ph ] ; e. ma , a. natale and a. rashed , int . j. mod . a * 27 * , 1250134 ( 2012 ) [ arxiv:1206.1570 [ hep - ph ] ] ; n. haba and r. takahashi , arxiv:1206.2793 [ hep - ph ] ; a. g. dias , a. c. b. machado and c. c. nishi , arxiv:1206.6362 [ hep - ph ] ; p. m. ferreira , w. grimus , l. lavoura and p. o. ludl , arxiv:1206.7072 [ hep - ph ] ; g. altarelli , f. feruglio , i. masina and l. merlo , arxiv:1207.0587 [ hep - ph ] ; y. bentov , x. -g . he and a. zee , arxiv:1208.1062 [ hep - ph ] ; w. rodejohann and h. zhang , arxiv:1207.1225 [ hep - ph ] ; r. n. mohapatra and c. c. nishi , arxiv:1208.2875 [ hep - ph ] . g. -j . ding , nucl . b * 862 * , 1 ( 2012 ) [ arxiv:1201.3279[hep - ph ] ] . s. f. king , c. luhn and a. j. stuart , arxiv:1207.5741 [ hep - ph ] . s. m. boucenna , s. morisi , m. tortola and j. w. f. valle , phys . d * 86 * , 051301 ( 2012 ) [ arxiv:1206.2555 [ hep - ph ] ] . s. f. king , arxiv:1205.0506 [ hep - ph ] ; s. antusch , c. gross , v. maurer and c. sluka , arxiv:1205.1051 [ hep - ph ] ; h. minakata and a. y. .smirnov , phys . d * 70 * , 073009 ( 2004 ) [ hep - ph/0405088 ] ; n. li and b. -q . ma , eur . j. c * 42 * , 17 ( 2005 ) [ hep - ph/0504161 ] ; n. qin and b. q. ma , phys . b * 702 * , 143 ( 2011 ) [ arxiv:1106.3284 [ hep - ph ] ] ; y. h. ahn , h. -y . cheng and s. oh , phys . rev . d * 83 * , 076012 ( 2011 ) [ arxiv:1102.0879 [ hep - ph ] ] . e. dudas , c. grojean , s. pokorski and c. a. savoy , nucl . b * 481 * , 85 ( 1996 ) [ hep - ph/9606383 ] ; k. choi , e. j. chun and h. d. kim , phys . b * 394 * , 89 ( 1997 ) [ hep - ph/9611293 ] ; n. irges , s. lavignac and p. ramond , phys . d * 58 * , 035003 ( 1998 ) [ hep - ph/9802334 ] ; j. k. elwood , n. irges and p. ramond , phys . lett . * 81 * , 5064 ( 1998 ) [ hep - ph/9807228 ] ; j. r. ellis , g. k. leontaris and j. rizos , jhep * 0005 * , 001 ( 2000 ) [ hep - ph/0002263 ] ; h. k. dreiner and m. thormeier , phys . d * 69 * , 053002 ( 2004 ) [ hep - ph/0305270 ] ; h. k. dreiner , h. murayama and m. thormeier , nucl . b * 729 * , 278 ( 2005 ) [ hep - ph/0312012 ] ; m. -c . chen , d. r. t. jones , a. rajaraman and h. -b . yu , phys . d * 78 * , 015019 ( 2008 ) [ arxiv:0801.0248 [ hep - ph ] ] ; l. f. duque , d. a. gutierrez , e. nardi and j. norena , phys . d * 78 * , 035003 ( 2008 ) [ arxiv:0804.2865 [ hep - ph ] ] . l. wolfenstein , phys . lett . * 51 * , 1945 ( 1983 ) . j. schechter and j. w. f. valle , phys . d * 22 * ( 1980 ) 2227 ; w. rodejohann and j. w. f. valle , phys . rev . d * 84 * , 073011 ( 2011 ) [ arxiv:1108.3484 [ hep - ph ] ] . h. fusaoka and y. koide , phys . d * 57 * , 3986 ( 1998 ) [ hep - ph/9712201 ] ; g. ross and m. serna , phys . b * 664 * , 97 ( 2008 ) [ arxiv:0704.1248 [ hep - ph ] ] ; z. -z . xing , h. zhang and s. zhou , phys . d * 77 * , 113016 ( 2008 ) [ arxiv:0712.1419 [ hep - ph ] ] . y. grossman and y. nir , nucl . b * 448 * , 30 ( 1995 ) [ hep - ph/9502418 ] ; y. grossman , y. nir and y. shadmi , jhep * 9810 * , 007 ( 1998 ) [ hep - ph/9808355 ] ; y. nir and y. shadmi , jhep * 9905 * , 023 ( 1999 ) [ hep - ph/9902293 ] . m. b. green and j. h. schwarz , phys . b * 149 * , 117 ( 1984 ) . l. e. ibanez and g. g. ross , phys . b * 332 * , 100 ( 1994 ) [ hep - ph/9403338 ] ; p. binetruy and p. ramond , phys . b * 350 * , 49 ( 1995 ) [ hep - ph/9412385 ] ; v. jain and r. shrock , phys . b * 352 * , 83 ( 1995 ) [ hep - ph/9412367 ] ; e. dudas , s. pokorski and c. a. savoy , phys . b * 356 * , 45 ( 1995 ) [ hep - ph/9504292 ] ; p. binetruy , s. lavignac and p. ramond , nucl . b * 477 * , 353 ( 1996 ) [ hep - ph/9601243 ] ; p. h. chankowski , k. kowalska , s. lavignac and s. pokorski , phys . d * 71 * , 055004 ( 2005 ) [ hep - ph/0501071 ] . a. e. nelson and d. wright , phys . d * 56 * , 1598 ( 1997 ) [ hep - ph/9702359 ] . h. georgi and c. jarlskog , phys . b * 86 * , 297 ( 1979 ) . j. r. ellis and m. k. gaillard , phys . b * 88 * , 315 ( 1979 ) . g. altarelli , f. feruglio and i. masina , jhep * 0011 * , 040 ( 2000 ) [ hep - ph/0007254 ] . j. schechter and j. w. f. valle , phys . d * 25 * , 2951 ( 1982 ) . a. osipowicz _ et al . _ [ katrin collaboration ] , arxiv : hep - ex/0109033 ; see also : http://www-ik.fzk.de/ katrin / index.html c. arnaboldi _ et al . _ [ cuore collaboration ] , nucl . instrum . a * 518 * , 775 ( 2004 ) [ hep - ex/0212053 ] ; r. ardito , c. arnaboldi , d. r. artusa , f. t. avignone , iii , m. balata , i. bandac , m. barucci and j. w. beeman _ et al . _ , hep - ex/0501010 . s. r. elliott [ majorana collaboration ] , j. phys . ser . * 173 * , 012007 ( 2009 ) [ arxiv:0807.1741 [ nucl - ex ] ] ; v. e. guiseppe _ et al . _ [ majorana collaboration ] , ieee nucl . 2008 * , 1793 ( 2008 ) [ arxiv:0811.2446 [ nucl - ex ] ] . i. abt , m. f. altmann , a. bakalyarov , i. barabanov , c. bauer , e. bellotti , s. t. belyaev and l. b. bezrukov _ et al . _ , hep - ex/0404039 ; s. schonert ( gerda collaboration ) , j. phys . ser . * 203 * , 012014 ( 2010 ) . p. binetruy , s. lavignac , s. t. petcov and p. ramond , nucl . b * 496 * , 3 ( 1997 ) [ hep - ph/9610481 ] ; s. lola and g. g. ross , nucl . b * 553 * , 81 ( 1999 ) [ hep - ph/9902283 ] ; m. tanimoto , phys . b * 456 * , 220 ( 1999 ) [ hep - ph/9901210 ] ; m. s. berger and k. siyeon , phys . d * 62 * , 033004 ( 2000 ) [ hep - ph/0003121 ] ; m. s. berger and k. siyeon , phys . d * 64 * , 053006 ( 2001 ) [ hep - ph/0005249 ] ; j. sato and k. tobe , phys . rev . d * 63 * , 116010 ( 2001 ) [ hep - ph/0012333 ] .

chlorine abundances are reported in 15 evolved giants and one m dwarf in the solar neighborhood . the cl abundance was measured using the vibration - rotation 1 - 0 p8 line of hcl at 3.69851 m . the high resolution l - band spectra were observed using the phoenix infrared spectrometer on the kitt peak mayall 4 m telescope . the average [cl / fe ] abundance in stars with 0.72[fe / h].20 is [cl / fe]=(0.10.15 ) dex . the mean difference between the [cl / fe ] ratios measured in our stars and chemical evolution model values is ( 0.16.15 ) dex . the [cl / ca ] ratio has an offset of.35 dex above model predictions suggesting chemical evolution models are under producing cl at the high metallicity range . abundances of c , n , o , si , and ca were also measured in our spectral region and are consistent with f and g dwarfs . the cl versus o abundances from our sample match cl abundances measured in planetary nebula and regions . in one star where both hcl and hcl could be measured , acl/cl isotope ratio of 2.2.4 was found , consistent with values found in the galactic ism and predicted chemical evolution models .

1609.01626

a full knowledge of stellar abundance patterns in a variety of stellar sources is useful to understand the production of each element and the chemical enrichment history of stellar populations . galactic chemical evolution models rely on theoretical yields which are tested through observations of stellar abundances . multiple elements have been well studied in galactic evolution , particularly the alpha elements and iron peak elements in galactic populations . some odd atomic number light elements are difficult to measure due to their low abundances and spectral features that lie outside optical wavelengths . thus the light , odd elemental abundance patterns are poorly constrained @xcite . this work focuses on chlorine , an odd z element with two stable isotopes , @xmath0cl and @xmath5cl . both chlorine isotopes are formed during hydrostatic and explosive oxygen burning phases @xcite , although the explosive oxygen burning phase during supernova produces a higher yield than hydrostatic burning @xcite . a model for a 25 m@xmath6 supernova explosion shows the @xmath0cl abundance generated from hydrostatic burning is 4.82 x 10@xmath7 m@xmath6 while explosive oxygen burning generates an abundance of 7 x 10@xmath7 m@xmath6 @xcite . @xmath0cl is thought to be primarily produced when @xmath8s captures a proton . the other isotope , @xmath5cl is thought to be primarily produced from radioactive decay of @xmath5ar , and can be created during neon burning @xcite . models of chlorine production in both core collapse supernova ( ccsne ) @xcite and type ia supernova @xcite show chlorine yields vary as a function of mass and metallicity of the progenitor star . these yields have led galactic enrichment models to predict constant [ cl / fe ] @xmath9 log(n@xmath10/n@xmath11)@xmath12 - log(n@xmath10/n@xmath11)@xmath6 ] ratios over a range of metallicities @xcite . measurement of the chlorine abundances in stellar atmospheres is extremely difficult and little empirical data are available to compare with cl evolution models . the solar cl abundance in particular highlights the difficulties in cl abundance measurements ; no chlorine abundance measurement is possible from the quiet photosphere . an early attempt using near infrared measurements of weak atomic cl lines provided inconclusive abundance measurements and gave an upper limit on the chlorine abundance , a(cl ) @xmath14 5.5 @xcite . chlorine can also be measured in molecular form . hydrogen chloride ( hcl ) molecular vibrational - rotational lines are found in the l - band spectrum of stars . however , the low dissociation potential of hcl limits the molecule to lower temperature stellar atmospheres . hcl features in the spectral range 3.633 @xmath1 m and 4.166 @xmath1 m have been measured in solar sunspot umbrae spectra and resulted in a solar cl abundance of 5.5@xmath30.3 @xcite . chlorine features at x - ray wavelengths can also be present during solar flares ; an abundance of a(cl)=5.75@xmath30.26 has been measured using the line in solar flare spectra @xcite . due to the difficulties in measuring the cl abundance in stellar atmospheres , @xcite suggested that a cl abundance of a(cl)=5.32@xmath15 derived from nearby regions from @xcite may be a suitable proxy for the solar abundance . finally , the the meteoric value for cl of 5.25@xmath30.06 @xcite may also be a proxy for the cl abundance in the sun . while no direct chlorine measurements have been reported in the photospheres of stars , forbidden cl lines , such as features at @xmath45500@xmath16 in the optical regime , provide chlorine measurements in planetary nebulae and regions . more recent cl abundance measurements @xcite in regions were derived by measuring multiple ions of cl without using ion correction factors . these improved measurements have lowered the cl abundance measured in nearby regions and allowed the measurement of the galactic radial cl abundance gradient . the radial slope finds a cl abundance at the solar galactocentric radius of a(cl)=5.05 @xcite . both and planetary nebulae surveys have shown that the gradients of cl and o with galactocentric distance are nearly identical , @xcite for observations and @xcite for pn observations , implying cl and o production are highly correlated . other planetary nebula studies used cl as a proxy for the iron abundance , consistent with current models of cl production @xcite . millimeter and submillimeter hcl emission features have been measured in the interstellar medium and provide tests of the @xmath0cl/@xmath5cl ratio in both galactic and extragalactic sources . the solar system @xmath0cl/@xmath5cl ratio of 3.13 was found from meteoric cl isotope abundances @xcite . measurements of proto - planetary cores have found a @xmath0cl/@xmath5cl ratio of 3.2 @xmath30.1 @xcite matching solar values . a survey of 27 star forming regions , molecular clouds , and the circumstellar envelops of evolved stars , observed with the caltech submillimeter observatory , found @xmath0cl/@xmath17cl ratios that varied between 1 and 5 ( uncertainties ranged from 0.3 to 1.0 ) with the majority of sources having a ratio between 1.1 and 2.5 @xcite . chlorine has also been detected in extragalactic sources ; recently , h@xmath18cl@xmath19 has been detected in a lensed blazar with a measured @xmath0cl/@xmath5cl ratio of 3.1@xmath20 at a redshift of 0.89 @xcite . we report here our cl abundances measured in stars using hcl molecular lines at infrared wavelengths . previous infrared spectroscopic studies found hcl molecular lines in the 4 @xmath1 m wavelength regime in the atmospheres of cool agb stars @xcite . this paper describes the observations , source selection , and data reduction in section [ sec::reduction ] . section [ sec::abund ] describes how the line - list was constructed , the spectral synthesis implementation , and the uncertainties in our abundance measurements . section [ sec::discussion ] discusses the atmospheric parameters necessary for hcl to form in a solar atmosphere , the cl isotopic abundance , and comparisons to theoretical cl enrichment models . the hcl vibration - rotation fundamental is located in the 3 @xmath1 m region where the telluric spectrum can be strong . due to the large rotational constant ( b@xmath21 @xmath2210.6 @xmath23 ) for the @xmath24 hcl vibration - rotation transition , the spectral lines are widely spaced . to select optimum regions to observe we examined both telluric and arcturus spectra @xcite at the wavelength of each h@xmath0cl line . hcl does not appear in the spectrum of arcturus so this process provided a list of the regions with the fewest blends . the 1 - 0 p8 h@xmath0cl transition at 3.69851 @xmath1 m was selected for observation . the 3.7 @xmath1 m spectral region is show in figure [ fig::loggffits ] for the arcturus , solar , and sunspot umbrae spectra . these spectra were all retrieved from noao atlases ; the solar spectrum was taken from @xcite , the arcturus spectrum from @xcite , and the the sunspot umbrae spectra from @xcite . the h@xmath0cl molecular feature is only observable in low temperature stars . in the figure [ fig::loggffits ] h@xmath0cl is present in the cool sunspot spectrum but not in the arcturus atlas or solar atlas . the program stars were observed with the phoenix infrared spectrometer @xcite at the f/16 focus of the kpno mayall 4 meter telescope in 2014 december and 2015 june . the 0.7 arcsecond four pixel slit was used resulting in a spectral resolution of @xmath450,000 . on 2014 december 7 and 2014 december 8 and 2015 june 7 and 8 the 3.6940 @xmath1 m to 3.7110 @xmath1 m region of echelle order 15 was selected with a narrow band order sorting filter . on 2014 december 9 and 10 the spectral range was slightly shifted , 3.6900 @xmath1 m to 3.7070 @xmath1 m . l l l l c c bd+68 946 & - & 7:31 2015 june 7 & m3.5 v & 4.60 & 70 + @xmath25 and & 6860 & 5:05 2014 december 10 & m6 iii & 1.85 & 100 + rz ari & 18191 & 3:31 2014 december 9 & m0 iii & 0.87 & 180 + @xmath26 cet & 18884 & 5:08 2014 december 8 & m1.5 iii & 1.82 & 130 + omicron ori & 30959 & 6:38 2014 december 10 & m3.2 iii & 0.66 & 120 + v1261 ori & 35155 & 7:50 2014 december 10 & s4 & 2.14 & 130 + @xmath1 gem & 44478 & 8:35 2014 december 10 & m3 iii & 1.86 & 130 + v613 mon & 49368 & 9:27 2014 december 9 & s5 & 2.46 & 140 + 27 cnc & 71250 & 10:39 2014 december 10 & m3 iii & 0.58 & 130 + @xmath27 vir & 112300 & 3:35 2015 june 8 & m3 iii & 1.19 & 160 + 83 uma & 119228 & 4:16 2015 june 7 & m2 iii , ba 0.5 & 0.34 & 160 + aw cvn & 120933 & 4:37 2015 june 7 & k5 iii & 0.01 & 170 + 52 boo & 138481 & 6:06 2015 june 7 & k5 iii , ba 0.5 & 1.21 & 200 + & 147923 & 6:38 2015 june 7 & m2 [ s ? ] & 3.46 & 130 + @xmath28 aqr & 216386 & 1:36 2014 december 10 & m2 iii & 0.67 & 130 + 30 psc ( yy psc ) & 224935 & 1:44 2014 december 10 & m3 iii & 0.40 & 130 + @xmath29 psc & 10380 & 4:09 2014 december 7 & k3 iii & 1.36 & 110 + & 10824 & 4:27 2014 december 7 & k4 iii & 1.74 & 120 + 14 tri & 15656 & 4:46 2014 december 7 & k5 iii & 1.63 & 80 + & 20468 & 5:36 2014 december 7 & k2 ii & 1.46 & 110 + & 20644 & 5:46 2014 december 7 & k4 iii & 0.88 & 130 + 23 eri & 23413 & 6:05 2014 december 7 & k4 iii & 2.40 & 100 + & 29065 & 6:22 2014 december 7 & k4 iii & 1.82 & 100 + & 52960 & 9:31 2014 december 7 & k3 iii & 2.04 & 140 + @xmath30 cmi & 58972 & 9:44 2014 december 7 & k3 iii & 0.99 & 100 + @xmath31 gem & 60522 & 9:53 2014 december 7 & m0 iii & 0.23 & 130 + g gem & 62721 & 10:03 2014 december 7 & k4 iii & 1.24 & 110 + & 88230 & 12:15 2014 december 7 & k8 v & 2.96 & 130 + & 218792 & 2:20 2014 december 7 & k4 iii & 2.66 & 100 + 22 psc & 223719 & 2:06 2014 december 7 & k4 iii & 2.04 & 130 + 3 cet & 225212 & 2:32 2014 december 7 & k3 iab & 1.40 & 130 + there is considerable thermal background emission at 3.7 @xmath1 m . to remain within the linear response of the detector single integrations were limited to @xmath4150s . depending on the magnitude of the individual target , exposure times were between 1 to 150 seconds . the noise contributed by thermal emission from background radiation limited the stellar sample to bright infrared sources . all the stars observed had k@xmath25 . spectral type k and m , population i , stars were targeted . a full list of stars that showed chlorine feature are displayed in table [ table::obslog ] . the second portion of this table shows a list of stars that were analyzed but did not present hcl features in their spectra . they are included to extend our analysis over large temperature ranges . stars with effective temperatures between @xmath43300 k and @xmath4 4300 k were chosen as candidate stars for the observations . the spectral region covered does not contain enough spectral features to determine atmospheric parameters for the sample stars . while numerous stellar oh lines are present in this wavelength range these lines do not cover a large enough range of excitation potential suitable to set atmospheric parameters for the sample stars . thus we selected stars with atmospheric parameters available within the literature . standard observing procedures were used with the star nodded between two slit positions . each program star was observed close to zenith with an airmass less than 1.4 , most between 1.0 and 1.1 . additional standard stars to be used for telluric line removal were observed with air masses similar to target stars . flat and dark calibration images were taken at the beginning of each night . data reduction was accomplished using the iraf software suite . the standard near - ir reduction process discussed by @xcite was followed . after each image was trimmed , the dark images were combined using a median filter with a sigclip rejection algorithm . the flats were median combined with an avsigclip rejection algorithm . the combined dark image was subtracted from the combined flat image . each set of stellar observations had been taken in groups of four with two different slit positions using the sequence `` abba '' . using the iraf task imarith , adjacent observations were subtracted from one another to remove sky , dark , and common detector blemishes . each image was then divided by the dark subtracted , normalized flat field image . the resulting images were extracted using the iraf task apall and multiple observations were combined using the iraf task scombine . the images from the december 2014 observing runs were either median or sum combined depending on the cosmic ray and bad pixel contamination . after the combination process , each spectrum was normalize . telluric lines were removed using the iraf task telluric. this correction was done using standard stars observed each night . the telluric contamination in the spectral region is minimal , shown in figure [ fig::exspec2 ] . each telluric correction was done with a standard star that most closely matched the air mass of the observed star . finally , the exposures were wavelength calibrated using stellar lines identified in the arcturus atlas and in each of the stars . the line list , shown in table [ table : : linelist ] , was derived from multiple sources sources , the kurucz atomic line database , @xcite for oh lineschris / lab.html ] , @xcite for nh lines , and the hitran database @xcite . the @xcite oh log gf values for lines in our spectral range are consistent with those in the kurucz molecular line data base . the atomic line parameters were adopted from the kurucz database and ch values were from @xcite ( values listed in the kurucz database ) . the oscillator strength ( f - value ) for the hcl features was calculated from the einstein a coefficient recorded in the hitran database , using equation [ eq::loggf ] ( * ? ? ? * pg.153 ) . @xmath32 the log gf values were tested by fitting high resolution solar photospheric spectrum and arcturus spectrum using synthetic spectra . the atmospheric parameters and abundances for each star can be found in table [ table::params ] . the solar abundances are adopted from @xcite and the arcturus atmospheric parameters and abundances were adopted from @xcite with the exception of the nitrogen abundance which was adopted from @xcite . plane parallel and spherical atmosphere models for the sun and arcturus respectively were generated from the marcs grid @xcite and were interpolated using the program available on the marcs site @xcite . dissociation potentials for each molecular species were found the literature and are listed in table [ table : : linelist ] ; sources include @xcite for oh , @xcite for hcl , @xcite for nh , and @xcite for ch . synthetic spectra were generated using the 2014 version of the moog spectral analysis software @xcite . the synthetic spectra were convolved with a gaussian to match the line profiles in arcturus and solar spectra . the best fits to the spectra are shown in figure [ fig::loggffits ] . the astrophysical log gf values found for the molecular lines are consistent with the theoretical values with the exception of two oh lines at 37050.462 @xmath16 and 37059.961 @xmath16 ; their log gf values were increased by 0.1 to better fit the arcturus and solar spectra . the log gf values of the atomic features were on average 0.4 larger than the atomic features in the kurucz database . fe i features in the spectra could not be fit simultaneously in both the arcturus spectra and the solar spectrum , possibly due either to blending weak , unidentified molecular lines or other features , or due to errors in the excitation potential . our best estimates for the log gf values for the fe i lines are included in table [ table : : linelist ] . l c c c c c l nh & 36921.287 & 2707.70646 & 1.655 & -3.085 & 3.3693 & 5 - 4 r3e10 + nh & 36921.494 & 2707.69132 & 1.655 & -3.012 & 3.3693 & 5 - 4 r1e12 + nh & 36921.506 & 2707.69041 & 1.655 & -3.048 & 3.3693 & 5 - 4 r2f11 + oh & 36942.692 & 2706.10656 & 1.247 & 3.859 & 4.4130 & 3 - 2 p2f12.5 + oh & 36954.896 & 2705.21496 & 1.248 & 3.859 & 4.4130 & 3 - 2 p2e12.5 + ch & 36957.174 & 2705.04229 & 0.3580 & 3.570 & 3.4948 & 2 - 1 r1e3.5 + ca i & 36963.164 & 2704.605 & 5.5637 & 1.15 & & 3s@xmath33f@xmath33f@xmath34 - 3s@xmath33g@xmath35g@xmath36 + ca i & 36967.510 & 2704.286 & 5.5637 & 1.25 & & 3s@xmath33f@xmath33f@xmath36 - 3s@xmath33g@xmath35g@xmath37 + ca i & 36967.866 & 2704.260 & 5.5637 & 0.10 & & 3s@xmath33f@xmath33f@xmath36 - 3s@xmath33g@xmath35g@xmath36 + nh & 36969.559 & 2704.17095 & 0.264 & -4.723 & 3.3693 & 1 - 0 p1e12 + nh & 36970.709 & 2704.08687 & 0.264 & -4.763 & 3.3693 & 1 - 0 p2f11 + nh & 36971.829 & 2704.00489 & 0.264 & -4.803 & 3.3693 & 1 - 0 p3e10 + ca i & 36973.485 & 2703.849 & 5.5637 & 1.34 & & 3s@xmath33f@xmath33f@xmath37 - 3s@xmath33g@xmath35g@xmath38 + ca i & 36973.909 & 2703.818 & 5.5637 & 0.10 & & 3s@xmath33f@xmath33f@xmath37 - 3s@xmath33g@xmath35g@xmath37 + si i & 36975.824 & 2703.678 & 7.0692 & 0.77 & & 3p@xmath39s@xmath40p@xmath41 - 3p@xmath39p@xmath40d@xmath42 + oh & 36979.19 & 2703.43514 & 1.727 & 4.514 & 4.413 & 5 - 4 p2f6.5 + oh & 36981.688 & 2703.25474 & 1.243 & 3.825 & 4.413 & 3 - 2 p1e13.5 + h@xmath0cl & 36985.103 & 2703.0074 & 0.0929 & 4.178 & 4.432 & 1 - 0 p8 + oh & 36998.379 & 2702.03348 & 1.244 & 3.825 & 4.413 & 3 - 2 p2f13.5 + h@xmath5cl & 37010.023 & 2701.1874 & 0.0929 & 4.178 & 4.432 & 1 - 0 p8 + oh & 37034.102 & 2699.42489 & 1.030 & 3.782 & 4.413 & 2 - 1 p2f15.5 + nh & 37040.866 & 2698.96521 & 1.619 & -3.148 & 3.3693 & 5 - 4 r3e9 + nh & 37041.139 & 2698.94526 & 1.619 & -3.108 & 3.3693 & 5 - 4 r2f10 + nh & 37041.176 & 2698.94261 & 1.618 & -3.069 & 3.3693 & 5 - 4 r1e11 + oh & 37050.462 & 2698.23581 & 1.031 & 3.682 & 4.413 & 2 - 1 p1e15.5 + oh & 37055.419 & 2697.87443 & 1.492 & 4.069 & 4.413 & 4 - 3 p2f9.5 + oh & 37059.961 & 2697.544 & 1.026 & 3.655 & 4.413 & 2 - 1 p1e16.5 + oh & 37063.156 & 2697.27596 & 1.492 & -4.069 & 4.413 & 4 - 3 p2e9.5 + k i & 37070.611 & 2696.765 & 3.3969 & 0.82 & & 2p@xmath40p@xmath43 - 2d@xmath44d@xmath45 + fe i & 37080.346 & 2696.057 & 5.0674 & 0.94 & & 3p@xmath33yd@xmath46 - 3d@xmath39p@xmath36 + oh & 37081.155 & 2696.00032 & 1.028 & 3.755 & 4.413 & 2 - 1 p2f16.5 + oh & 37117.883 & 2693.33812 & 1.487 & 4.024 & 4.413 & 4 - 3 p1e10.5 + the h@xmath0cl molecular line is not present in either arcturus or the solar photosphere due to the high photospheric temperatures , but is found in sunspot umbrae spectra . to fit the sunspot spectrum from @xcite , we used a similar methodology as @xcite to determine the effective temperature of the sunspot . we generated multiple atmospheric models with different temperatures and found an effective temperature of 3900 k allowed the best fit to the oh lines . we found an abundance of a(@xmath0cl)=5.31 @xmath30.12 best fit the spectral feature . an uncertainty of 0.10 dex was found from the model temperature by fitting the cl feature with models at 4000 k and 3800 k and the uncertainty on the fit was determined to be 0.06 dex . however , because we do not attempt to model the complex radiative transfer in sunspots , we chose to adopt a solar chlorine abundance of a(cl)=5.25 based on the meteoric cl abundance @xcite . the spectral range is not sufficiently large to derive atmospheric parameters for the target stars . the t@xmath47 , log g , [ fe / h ] , and @xmath48 values for each star were adopted from literature sources and can be found in table [ table::params ] along with references for each value . effective temperatures for these stars were mostly determined by photometry ( eg . @xcite ) and spectral energy distribution fitting ( e.g. @xcite ) . for stars with multiple temperatures cited in the literature , parameters were chosen based on the resulting oxygen abundance from the fit . the majority of the stars in the sample are metal rich and have oxygen abundances near [ o / fe]=0 . the adopted temperatures for each star were compared with temperatures derived using j - k@xmath49 colors from 2mass photometry @xcite and with a color - effective temperature relationship from @xcite . in our sample , 13 of the 15 giants are beyond the j - k@xmath49 @xmath50 0.9 limit for the color - temperature relation @xcite and the relation was extrapolated up to colors of 1.22 with most between 0.9 and 1.1 . the v magnitude was not used to minimize the effect of extinction from circumstellar dust . initial temperatures were compared to literature sources and are shown in figure [ fig::temp_colors ] ( left panel ) . the average difference between photometric temperatures and those from the literature is 51 k @xmath3 239 k. extinction due to circumstellar dust in the j and k@xmath49 bands were corrected using observed to model flux ratios from @xcite . three stars were not included in @xcite , @xmath25 and , @xmath26 cet , and @xmath27 vir . the average correction to the j - band magnitude was 0.01 mags @xmath3 0.10 mags and k@xmath49-band corrections had an average of 0.005 mags @xmath3 0.04 mags . the j - band standard deviation is dominated by the star hd 147923 with a correction of 0.28 mags . without this star , the average correction is @xmath51j= 0.01 mags @xmath3 0.06 mags . the temperatures derived from the corrected j - k@xmath49 colors are shown in the left panel of figure [ fig::temp_colors ] . the average difference between the temperatures adopted from the literature and the temperatures calculated from the j - k@xmath49 color is 76 k @xmath3 95 k. an additional test was performed by comparing temperatures derived by @xcite to those in the literature ( see table 3 ) for the 12 giants in our sample with measurable cl abundances , shown in the right panel of figure [ fig::temp_colors ] . some stars in table 3 have two citations ; one that represents a temperature from @xcite and another for the other atmospheric parameters . the temperature from the source associated with the atmospheric parameters is compared to the quoted @xcite temperature in those instances . the average difference between the @xcite temperature and those from other sources is -49 k @xmath3 143 k. excluding the outlier of 27 cnc , which has a difference of -374 k , brings the average down to -19 k @xmath3 105 k. the corrected j - k@xmath49 temperatures , temperatures from @xcite , and temperatures from other sources are consistent . l c c c c c c c c c c solar & 5780 & 4.40 & 0.00 & 0.75 & 8.43 & 7.83 & 8.69 & 7.51 & 5.13 & 6.34 + @xmath26 boo & 4290 & 1.60 & 0.52 & 1.67 & 8.34 & 7.69 & 8.60 & 7.32 & & 5.90 + sunspot umbra & 4000 & 4.40 & 0.00 & 0.75 & 8.43 & 7.83 & 8.69 & 7.51 & 5.13 & 6.34 + bd+680946 & 3600 & 5.00 & 0.00 & 1.00 & & & 8.74 & & 5.27 & + hd 6860 & 3842 & 0.90 & 0.20 & 2.20 & 8.33 & 7.96 & 8.62 & 7.09 & 4.90 & 6.11 + hd 18191 & 3340 & 0.30 & 0.25 & 2.45 & 8.33 & 7.99 & 8.47 & 7.36 & 4.82 & 6.43 + hd 18884 & 3796 & 0.91 & 0.45 & 2.00 & 8.33 & 7.63 & 8.60 & 6.97 & 4.86 & 6.25 + hd 30959 & 3450 & 0.80 & 0.15 & 2.95 & 8.38 & 8.09 & 8.56 & 7.41 & 4.88 & 6.13 + hd 35155 & 3650 & 0.80 & 0.72 & 2.30 & 8.30 & 7.64 & 8.45 & 7.04 & 4.57 & 5.91 + hd 44478 & 3640 & 1.00 & 0.00 & 2.45 & 8.43 & 8.03 & 8.83 & 7.43 & 5.07 & 6.37 + hd 49368 & 3700 & 1.00 & 0.45 & 2.50 & 8.42 & 7.99 & 8.56 & 7.26 & 4.90 & 6.24 + hd 71250 & 3574 & 1.00 & 0.00 & 2.35 & 8.53 & 7.73 & 8.77 & 7.31 & 4.88 & 6.04 + hd 112300 & 3700 & 0.80 & 0.17 & 2.00 & 8.70 & 8.07 & 8.96 & 7.23 & 5.27 & 6.44 + hd 119228 & 3684 & 1.00 & 0.00 & 1.80 & 8.69 & 7.94 & 8.89 & 7.42 & 5.05 & 6.32 + hd 120933 & 3625 & 0.98 & 0.09 & 2.00 & 8.47 & 7.71 & 8.65 & 7.08 & 4.88 & 6.00 + hd 138481 & 3970 & 1.60 & 0.20 & 2.40 & 8.90 & 8.18 & 9.09 & 7.32 & 5.50 & 6.29 + hd 147923 & 3800 & 0.80 & 0.19 & 2.20 & 8.44 & 8.19 & 8.64 & 7.25 & 5.11 & 6.25 + hd 216386 & 3800 & 1.00 & 0.05 & 2.10 & 8.61 & 8.17 & 8.92 & 7.26 & 5.38 & 6.21 + hd 224935 & 3620 & 1.20 & 0.05 & 2.75 & 8.89 & 8.17 & 8.99 & 7.47 & 5.07 & 6.34 + hd 10380 & 4110 & 1.93 & 0.17 & 2.10 & 8.47 & 7.83 & 8.67 & 7.22 & & 6.08 + hd 10824 & 3940 & 1.66 & 0.18 & 2.20 & 8.47 & 7.92 & 8.67 & 7.35 & & 6.18 + hd 15656 & 3990 & 1.65 & 0.21 & 2.20 & 8.51 & 7.70 & 8.70 & 7.30 & & 6.16 + hd 20468 & 4325 & 1.42 & 0.17 & 3.10 & 8.43 & 7.78 & 8.60 & 6.97 & & 6.07 + hd 20644 & 3980 & 1.56 & 0.31 & 2.60 & 8.36 & 7.88 & 8.49 & 7.36 & & 6.24 + hd 23413 & 4050 & 1.74 & 0.23 & 2.10 & 8.47 & 7.84 & 8.66 & 7.44 & & 6.23 + hd 29065 & 3990 & 1.77 & 0.35 & 2.30 & 8.08 & 8.02 & 8.44 & 7.44 & & 6.13 + hd 52960 & 4150 & 1.80 & 0.09 & 2.00 & 8.68 & 8.02 & 8.81 & 7.49 & & 6.28 + hd 58972 & 4000 & 1.82 & 0.37 & 2.20 & 8.27 & 7.87 & 8.49 & 7.17 & & 5.97 + hd 60522 & 3940 & 1.90 & 0.37 & 2.60 & 8.51 & 7.85 & 8.67 & 7.43 & & 6.25 + hd 62721 & 3940 & 1.67 & 0.27 & 1.90 & 8.47 & 7.61 & 8.65 & 7.33 & & 6.25 + hd 88230 & 3970 & 4.51 & 0.05 & 1.00 & & 8.59 & 7.85 & & & + hd 218792 & 4160 & 2.06 & 0.14 & 2.20 & 8.54 & 7.80 & 8.72 & 7.52 & & 6.21 + hd 223719 & 3960 & 1.63 & 0.24 & 2.40 & 8.37 & 7.93 & 8.59 & 7.30 & & 6.07 + hd 225212 & 4125 & 0.90 & 0.08 & 2.36 & 8.68 & 8.16 & 8.82 & 7.24 & & 6.42 + the aavso database was examined to determine whether our cool red giant stars were included . stars showing significant variability in the aavso database include rz ari , @xmath28 aqr , aw cvn , 83 uma , @xmath25 and , @xmath1 gem , and v613 mon . the least variable of this group is aw cvn with an amplitude of 0.028 mags in the v - band @xcite . other studies have shown v1261 ori with a v - band amplitude of 0.24 mags , an average v - band magnitude of 6.77 , a period of 624 days , and an observed minimum on jd 2454552 @xcite . two k - band measurements of k=2.03 mags on jd 2449653.5 @xcite and k=2.138 on jd 2451108.825 @xcite can be used to determine a v - k index and estimate the star s effective temperature range due to variability . the range of v - k for this star gives an effective temperature range 3527 k to 3594 k , a range less than 70 k @xcite . other work on m - giant pulsations shows that the amplitude of variability decreases at longer wavelengths , @xcite finds that the slope of the decrease in v can be represented by @xmath51v@xmath41.8 @xmath3 0.2@xmath51r and the slope of r is @xmath51r@xmath41.8 @xmath3 0.2@xmath51i for m - giants . extrapolating this trend would imply that the k - band amplitude decreases by a factor of 7.2 from the v - band to the k - band . for example rz ari , another moderately variable star in our sample , has a maximum v - band amplitude of @xmath40.25 and therefore would have an approximate k - band amplitude of 0.035 . this roughly corresponds to a temperature range of 3143 k to 3273 k , with an associated uncertainty of @xmath365 k @xcite . therefore , even in stars without measured k variability , their v - band variability can be used to characterize the uncertainty in their temperature . in the cases of our two most variable stars , rz ari and v613 mon , this temperature uncertainty is only on the order of @xmath450 k. c c c c c c c c fit & 0.13 & 0.08 & 0.07 & 0.11 & 0.05 & 0.06 + @xmath27 t@xmath47=@xmath375 k & 0.05 & 0.07 & 0.08 & 0.09 & 0.16 & 0.03 + @xmath27(log g ) = @xmath30.5 & 0.07 & 0.04 & 0.05 & 0.13 & 0.03 & 0.04 + @xmath27[fe / h ] = @xmath30.2 & 0.08 & 0.12 & 0.12 & 0.06 & 0.15 & 0.04 + @xmath27 @xmath48=@xmath30.5 ( km s@xmath52 ) & 0.02 & 0.02 & 0.06 & 0.06 & 0.03 & 0.07 + total uncertainty & 0.18 & 0.17 & 0.18 & 0.21 & 0.23 & 0.15 + due to the variety of sources used for atmospheric parameters and uncertainty in the @xmath27(v - k ) value , we adopt an uncertainty in temperature of @xmath375 k. the adopted uncertainty in log g is @xmath30.5 , in [ fe / h ] is @xmath30.2 and in microturbulence ( @xmath48 ) is @xmath30.5 km s@xmath52 . a conservative log g error was chosen because literature sources derived log g using a variety of methods . log g could be derived using the effective temperature , assumed mass , and luminosity of a star ( e.g. @xcite ) , analysis of atomic lines in the spectrum ( e.g. @xcite ) , or from models of stellar evolution ( e.g. @xcite ) , and not all sources have assigned uncertainties . the uncertainty on metallicity was set to reflect the uncertainty from multiple sources . for example , uncertainties derived from @xcite are on average 0.13 dex for ( fe / h ) ratios using @xmath26 tau as the reference star . the spectral analysis program moog @xcite 2014 version , was used in conjunction with marcs models @xcite to fit synthetic spectra to each star in table [ table::obslog ] . the synthetic spectra were convolved with a gaussian function to match both the depth and wings of the spectral features . the fit to the entire spectrum of rz ari can be found in figure [ fig::exspec2 ] . the upper panel of the figure shows an observed spectrum before telluric lines were removed and the middle panel shows only the telluric features contaminating the data . the fit to the h@xmath0cl feature for rz ari is shown in figure [ fig::rzari_clfeature ] . the abundances of c , n , o , si , cl , k , and ca were adjusted to obtain the best fit determined by eye to the spectrum and the resulting abundances are included in table [ table::params ] . the spectrum of the m - dwarf bd+68 946 resembles the sunspot umbra shown in figure [ fig::loggffits ] and we were unable to derive abundances for , , and without modeling the magnetic field and including the zeeman components for each line . atmospheric models were created for seven stars that span the full metallictity and temperature range of the sample to estimate the abundance uncertainties due to model parameters . the uncertainties for each element were averaged and the total errors can be seen in table [ table::uncertainties ] . the average errors for each parameter were assumed independent of one another and added in quadrature . the average uncertainty was adopted for each star . uncertainties on the fit were quantified for seven of the sample stars . by altering the abundances for each element to find fit the range of abundance that could reasonably fit each spectrum . the average uncertainties for each element can be found in table [ table::uncertainties ] . carbon showed the most uncertainty associated with the fit ; only one weak and blended ch line was present and reliable in each spectrum . other ch lines were potentially identified but could not be fit to features in the solar and arcturus spectra . abundant molecules in cool atmospheres include nh , ch , oh , cn and co. therefore , a change to the carbon abundance will affect the amount of co present in the star s atmosphere and will change the strength of observed oh lines . while the ch feature present in the spectra of these stars is weak , as seen in figure [ fig::exspec2 ] , it can estimate the carbon abundance . the ch , nh , and oh lines also depend on all the abundances of the cno trio . lowering the oxygen abundance increases the strength of ch features due to the reducing the number of co molecules and increasing the number of ch molecules . this resulted in 0.1 - 0.2 dex difference from the measured abundances and is added to the overall uncertainty . nitrogen abundances could be derived from nh molecular lines found in our spectra . we found when the nitrogen abundance changed the strength of the ch and oh line features were not strongly affected ; the change in c or o abundance associated with altering the n abundance was less than the error from fitting the synthetic spectra the oh and ch to molecular features . l c c c c c | c c c c c hd 6860 & 8.33 & 7.96 & 8.62 & 7.09 & 6.11 & 8.06 & 8.05 & 8.78 & 7.18 & 6.02 + hd 10380 & 8.47 & 7.83 & 8.67 & 7.22 & 6.08 & & & & 7.34 & 6.03 + hd 18191 & 8.33 & 7.99 & 8.47 & 7.36 & 6.43 & 8.10 & & 8.57 & & + hd 18884 & 8.33 & 7.63 & 8.60 & 6.97 & 6.25 & 8.64 & & 8.98 & & + hd 20468 & 8.43 & 7.78 & 8.60 & 6.97 & 6.07 & 8.16 & & 8.56 & & + hd 30959 & 8.38 & 8.09 & 8.56 & 7.41 & 6.13 & 8.49 & 8.26 & 8.75 & & + hd 35155 & 8.30 & 7.64 & 8.45 & 7.04 & 5.91 & 8.40 & 8.06 & 8.80 & & + hd 44478 & 8.43 & 8.03 & 8.83 & 7.43 & 6.37 & 8.32 & & 8.81 & & + hd 49368 & 8.42 & 7.99 & 8.56 & 7.26 & 6.24 & 8.43 & 8.50 & 8.80 & & + hd 112300 & 8.70 & 8.07 & 8.96 & 7.23 & 6.44 & 8.50 & & 8.84 & & + hd 147923 & 8.44 & 8.19 & 8.64 & 7.25 & 6.25 & 8.39 & 8.70 & 8.74 & & + hd 216386 & 8.61 & 8.17 & 8.92 & 7.26 & 6.21 & 8.56 & & 8.92 & & + the final [ @xmath0cl / fe ] abundances and their temperature dependence are shown in figure [ fig::results ] . the slope of [ @xmath0cl / fe ] versus t@xmath47 is flat , with no systematic dependence on temperature . additionally , the chosen h@xmath0cl feature is only measurable in stars cooler than 3900 k with the exception of the metal rich star hd 138481 . the dependence of the derived abundances for c , n , o , si , and ca on temperature is shown in figure [ fig::temp ] . the absence of any trend with temperature suggests our stellar effective temperatures are appropriate and that our atomic and molecular data are reliable . our c , n , o , si , and ca abundances , plotted in figure [ fig::otherelements ] , are consistent with the literature for stars near solar metallicity . the average abundances we derived for the coolest stars in our sample are consistent with those for the warmest stars . additionally , abundances for c , n , o , si , and ca measured from features in our spectra can be compared with previous determinations for stars in our sample from the literature . the abundances derived from the l - band spectra match other near infrared and infrared measurements , shown in table [ table::abundances_comparison ] . previous studies used equivalent width measurements of weak co , cn , and oh features between 1 - 4 @xmath1 m in infrared spectra @xcite . the absolute average difference between our carbon abundances and the literature values is 0.15 with a standard deviation of 0.09 . oxygen has an absolute average difference is 0.15 with a standard deviation of 0.12 . most stars therefore fall within our typical errors of 0.18 dex for both stars . only five of our sample stars have previously measured nitrogen abundances . two of our stars agree within one sigma while the three other stars disagree by @xmath40.5 dex . the dispersion in carbon may be due to difficulties fitting our single ch feature . three stars in our comparisons have oxygen abundances significantly larger than one standard deviation from the accepted literature values : hd 18884 , hd 35155 , and hd 49368 . additionally , both hd 35155 , and hd 49368 are 2 - 3 sigma larger than the our measured abundances . these are the three lowest metallicity stars in our sample , and the discrepancy may be due to inaccurate initial metallicities . follow up observations would be necessary to confirm the atmospheric parameters of these stars . additionally , while the oh features are strong , we note that even the strongest equivalent width is log(ew/@xmath28)@xmath24.77 our [ x / fe ] ratio are compared against [ fe / h ] for the program stars and f and g dwarfs in the thin disk @xcite in figure [ fig::otherelements ] . our derived stellar abundances are similar to thin disk f and g dwarfs , further demonstrating the reliability of our abundances and temperatures derived in this spectral range . again , the only significant discrepancy is with our oxygen abundances , which are generally @xmath40.1 dex lower than typical f and g dwarf values . additionally , si abundances at high metallicity are lower than the f and g dwarfs in @xcite . models of ca and si chemical evolution in the solar neighborhood ( adopted from fig . 10 in @xcite ) are shown in figure [ fig::otherelements ] . our abundances and the data from @xcite are consistent with the ca model but fall below the @xcite model for si . oxygen abundances derived from infrared molecular oh lines are sensitive to 3-d hydrodynamical effects in stars where convection is an important energy flow @xcite . in a t@xmath47=3660 k , log(g)=1.00 [ m / h]=0 star , @xcite found the abundance difference ( 3-d - 1-d ) for oh lines with excitation potentials between 1 - 2 ev is 0.06 to 0.08 dex at 1.6 @xmath1 m . this effect was noted to be less severe at increasing wavelength . any 3-d hydrodynamical effects in our oxygen abundances should be @xmath20.10 dex . the weak molecular features , lack of any systematic change with temperature , and comparisons to the literature all confirm our photospheric c , n , o , si , and ca abundances . our analysis demonstrates that the chlorine abundance can be measured in solar metallicity stars with temperatures below 3900 k. stars with temperatures above 3900 k are too warm to form the hcl molecule , with the one exception of hd 138481 , at an effective temperature of 3970 k but metal rich at [ fe / h]=+0.2 @xcite . the lowest metallicity star with detectable h@xmath0cl in our sample is hd 35155 at [ fe / h]=0.72 . the majority of stars with h@xmath0cl features in our sample are evolved stars of spectral types m , k , and s. late - type dwarfs are difficult to observe at 3.7 @xmath1 m due to their faint l - band apparent magnitude but the m dwarf bd+68 964 was bright enough to be included . this star has a measurable h@xmath0cl with [ o / fe ] and [ @xmath0cl / fe ] abundances that are consistent with the abundances measured in the evolved giants . the k8 dwarf hd 88230 is also included in our sample but h@xmath0cl is not measurable due this star s effective temperature of 3970 k. the evolutionary state and masses of the stars in our sample with chlorine abundance measurements can be determined from absolute magnitudes computed using hipparcos parallaxes @xcite and 2mass infrared photometry @xcite . the 2mass k@xmath49 magnitude was converted to the johnson k magnitude using methods of @xcite . a bolemetric correction for each star could then be determined from its j - k color and known atmospheric parameter @xcite . the input parameters and results for each star can be found in table [ table::bolcor ] . infrared photometry from iras @xcite , wise @xcite , and 2mass @xcite allowed the identification of any infrared excess due to circumstellar material around each star . stars with possible infrared excesses were then compared to @xcite ; only hd 147923 had j and k band magnitudes that were affected by more than 0.1 mags from the star s intrinsic magnitude @xcite . however , the distance to this star dominates the uncertainty on the luminosity rather than uncertainties in the j and k magnitudes . the stars position in the hertzsprung - russel diagram are compared to models for a star with z=0.017 and y=0.30 @xcite and the results are plotted in figure [ fig::stellarevol ] . the stars in our sample have masses between 1 @xmath53 and 3 @xmath53 and are consistent with the locus of stars on the red giant branch . all but one star is located less than 500 pc from the sun and the sample is therefore representative of stars in the solar neighborhood . l l l l l l bd+68 946 & - & 4.53 @xmath3 0.02 & 4.60 & 0.79 & 6.33 @xmath3 0.01 & 2.51 & -1.64 @xmath3 0.08 + @xmath25 and & 6860 & 60.5 @xmath3 3.0 & 1.85 & 0.89 & 5.69 @xmath3 0.11 & 2.71 & 3.09 @xmath3 0.09 + rz ari & 18191 & 107.8 @xmath3 3.5 & 0.87 & 1.10 & 5.96 @xmath3 0.07 & 2.85 & 3.14 @xmath3 0.08 + @xmath26 cet & 18884 & 76.4 @xmath3 3.0 & 1.82 & 1.10 & 6.17 @xmath3 0.09 & 2.69 & 3.29 @xmath3 0.09 + omicron ori & 30959 & 200@xmath54 28 & 0.66 & 1.15 & 7.10 @xmath3 0.30 & 2.85 & 3.60 @xmath3 0.15 + v1261 ori & 35155 & 288 @xmath3 70 & 2.14 & 1.20 & 5.10 @xmath3 0.53 & 2.85 & 2.80 @xmath3 0.23 + @xmath1 gem & 44478 & 71.0 @xmath3 3.6 & 1.86 & 0.95 & 6.05 @xmath3 0.11 & 2.69 & 3.24 @xmath3 0.09 + v613 mon & 49368 & 495 @xmath3 201 & 2.46 & 1.30 & 5.94 @xmath3 0.88 & 2.85 & 3.13 @xmath3 0.36 + 27 cnc & 71250 & 276 @xmath3 25 & 0.58 & 0.99 & 6.57 @xmath3 0.20 & 2.84 & 3.39 @xmath3 0.11 + @xmath27 vir & 112300 & 60.8 @xmath3 0.8 & -1.19 & 1.08 & 5.04 @xmath3 0.03 & 2.68 & 2.84 @xmath3 0.08 + 83 uma & 119228 & 160.5 @xmath3 5.7 & 0.34 & 1.03 & 5.63 @xmath3 0.08 & 2.68 & 3.08 @xmath3 0.09 + aw cvn & 120933 & 184 @xmath3 6.8 & 0.01 & 1.05 & 6.26 @xmath3 0.08 & 2.69 & 3.33 @xmath3 0.09 + 52 boo & 138481 & 257 @xmath3 17 & 1.21 & 0.85 & 5.78 @xmath3 0.14 & 2.54 & 3.19 @xmath3 0.10 + & 147923 & 1400 @xmath3 1100 & 3.46 & 1.25 & 7.21 @xmath3 1.70 & 2.84 & 3.64 @xmath3 0.70 + @xmath28 aqr & 216386 & 118.1 @xmath3 9.2 & 0.67 & 1.09 & 5.96 @xmath3 0.17 & 2.69 & 3.20 @xmath3 0.10 + 30 psc ( yy psc ) & 224935 & 132.5 @xmath3 10.4 & 0.40 & 1.02 & 5.94 @xmath3 0.17 & 2.85 & 3.13 @xmath3 0.11 + the @xmath0cl/@xmath5cl isotope ratio for the solar system has been measured to be 3.13 @xcite . a survey of hcl in the galaxy from molecular clouds , protoplanetary cores , and evolved stars ( such as irc+10216 ) revealed isotope ratios between 1@xmath55cl/@xmath5cl@xmath25 with most values between 1 and 3 @xcite . the range of values is attributed to different isotope production in type ia and ccsne @xcite . type ia supernova produce yield ratios of @xmath0cl/@xmath5cl @xmath4 3.5 to 5.5 depending on the model parameters , such as ignition , 2d versus 3d models , and model resolution @xcite . ccsne models produces chlorine yields with isotope ratios of 2.32 for z=0.004 , 18 m@xmath6 stars , 1.2 for z=0.02 , 25 m@xmath6 stars , and 1.75 for high energy supernova with z=0.02 and 25 m@xmath6 @xcite . our sample includes only one star , rz ari with a measurable h@xmath5cl feature at 3.7010 @xmath1 m , as shown in fig . [ fig::exspec2 ] . other stars , including the sunspot umbral spectra , were too warm to form an h@xmath5cl feature significantly above the noise level ; the h@xmath5cl feature is stronger at both lower temperature and higher abundance . the isotope ratio in rz ari was determined by comparing the equivalent widths of the two hcl features . the equivalent widths are ew(h@xmath0cl)=(81 @xmath3 6 ) m@xmath16 and ew(h@xmath5cl)=(36 @xmath3 6 ) m@xmath16 . uncertainties are found from the standard deviation of multiple measurements of the equivalent width at different continuum values . the final isotopic abundance ratio is @xmath0cl/@xmath5cl = 2.2 @xmath3 0.4 . this value was also confirmed by spectral synthesis ; this isotope ratio was used to fit the h@xmath5cl feature , shown in figure [ fig::exspec2 ] , with the best fit determined by eye . the isotope ratio for cl for different metallicities in the solar neighborhood has been predicted by @xcite : cl@xmath0/cl@xmath5 = 1.79 at [ fe / h]=0 , and 1.94 at [ fe / h ] = 0.5 . rz ari has an [ fe / h ] of 0.25 . our measured chlorine isotope value of 2.2 @xmath3 0.4 is near one standard deviation of these predicted values . this ratio is also consistent with the range of isotope measurements found in the interstellar medium @xcite and is less than two standard deviations from the solar isotopic ratio . further measurements in cool stars are necessary to explore the dispersion of the isotope ratio in the local solar neighborhood . figure [ fig::cl_vs_o ] compares the @xmath0cl abundance in our sample of stars with the total cl abundance measured in galactic planetary nebula and regions . studies of planetary nebula have found the radial galactic abundance gradients of cl and o to be statistically indistinguishable from one another @xcite and studies of cl in regions found o and cl had identical radial galactic abundance gradients of 0.043 dex kpc@xmath52 @xcite . this indicates both elements , cl and o , are produced in lockstep evolution @xcite . our sample of stars cover too narrow a range of galacto - centric distance to allow us to determine a gradient , but the average a(@xmath0cl ) of our sample is consistent with what is predicted for the solar neighborhood by @xcite and @xcite . the offset between the planetary nebula and stellar abundances shown in figure [ fig::cl_vs_o ] is estimated to be 0.16 dex . this offset is likely due to the difference between the total cl abundance reported in the atomic planetary nebula lines and @xmath0cl abundances measured from the h@xmath0cl molecular feature in our sample . for example the a(@xmath0cl ) abundance in rz ari is 4.82 and the full abundance is a(cl)=4.98 for an isotope ratio of @xmath0cl/@xmath5cl of 2.2 , an increase of 0.16 dex . additionally , the solar isotope ratio is 3.13 which gives a @xmath0cl abundance in the sun of 5.13 , a difference of 0.12 dex between the total abundance and @xmath0cl abundance . therefore while our results show scatter , the stellar abundances follow similar cl to o ratios to those seen in the nebular sources . future observations will use the methodology in this paper to determine cl isotope ratios in a larger sample of cool stars . we present the first direct measurements of chlorine in stellar sources , where cl can be directly compared to the iron abundance . figure [ fig::results ] shows the [ @xmath0cl / fe]cl / fe]=a(@xmath0cl)-a(cl)@xmath6+[fe / h ] ] ratio measured in our stellar sample versus a chemical evolution model from @xcite . the average chlorine abundance of our sample is [ @xmath0cl / fe]=0.10 @xmath3 0.14 . the chemical evolution model predicts that [ @xmath0cl / fe ] to be nearly constant with metallicity , declining by only 0.03 dex from [ fe / h]=0.6 to solar metallicity . our [ @xmath0cl / fe ] ratios are on average higher than the model by 0.16 @xmath3 0.15 dex . the lowest metallicity stars have an average [ @xmath0cl / fe ] ratio of 0.1 dex , about 0.2 dex higher than the solar metallicity but the sample contains too few metal - poor stars to conclude whether or not [ @xmath0cl / fe ] declines with decreasing metallicity . despite the small offset our results are still nearly consistent abundance of chlorine predicted for the solar neighborhood by chemical evolution models . chlorine abundances were also compared with the alpha elements silicon and calcium derived from our spectra . figure [ fig::cl_alpha ] shows our observed abundance ratios versus chemical evolution models ( reproduced from figure 10 ) from @xcite . our measured [ @xmath0cl / si ] and [ @xmath0cl / ca ] ratios are higher than predicted by @xcite by typically 0.4 dex . offsetting each chemical evolution model to match the observed abundance , as shown in fig . 10 , demonstrate that the slope of the fit is consistent with the measured abundances and that the discrepancies do not depend on metallicity in our sample . the bottom panel in figure [ fig::cl_alpha ] shows the [ ca / si ] ratio is also offset compared to the model value . the [ ca / si ] offset was found to be 0.20 dex . we find the [ @xmath0cl / ca ] ratio increases at higher [ fe / h ] . this is expected because [ @xmath0cl / fe ] is predicted to be constant with metallicity while [ ca / fe ] decreases as metallicity increases . the offset between models and chlorine abundance may be due to the additional production of cl by the @xmath29 process , which may affect yields of cl @xcite and is not included in the present model . these offsets between the chemical evolution models and the observed abundances put constraints on the strength of any additional process that produces cl . we are also comparing only @xmath0cl abundances to the cl chemical evolution models . including the @xmath5cl abundance would increase the overall abundance by @xmath40.1 to 0.2 dex depending on the isotope ratio . for example the a(@xmath0cl ) abundance in rz ari is 4.82 and the full abundance is a(cl)=4.98 for an isotope ratio of @xmath0cl/@xmath5cl of 2.2 . to determine if chlorine is produced using the s - process , two low temperature barium - enriched stars were included in our sample . hd 138481 is a class 0.5 ba star and hd 119228 is a class 0.5 ba star @xcite , represented as red diamonds in figure [ fig::results ] . the [ @xmath0cl / fe ] in those stars is similar to normal giants and suggests that additional cl production via the s - process is unlikely . we used l - band infrared spectroscopy to measure the abundance of chlorine in stars with effective temperatures below 3900 k , with the exception of the metal - rich hd 138481 which has a temperature of 3970 k. theoretical log gf values were used and compared to astrophysical log gf values derived from fitting synthetic spectra to arcturus , a quiet - sun photosphere spectrum , and a solar sunspot umbrae spectrum . of the sample , 16 stars showed measurable cl features in their spectra . the main results from the abundance analysis are summarized below . 1 . our sample consisted of m and k giants and dwarfs . the h@xmath0cl molecular feature is detected in 15 giants and one m - dwarf , all with effective temperatures below 3900 k except for hd 138481 which has a temperature of 3970 k. the masses of the evolved stars are determined from stellar evolution models and are found to be between 1 and 3 m@xmath6 2 . chlorine and oxygen abundances in stars are consistent with measurements made in planetary nebulae @xcite and regions @xcite . 3 . a @xmath0cl/@xmath5cl isotope ratio of 2.2@xmath30.4 is found in rz ari . this result is consistent with measurements in the interstellar medium @xcite and is near the solar value of 3.13 @xcite . it is also comparable to predicted isotope ratios of @xmath4 1.8 - 1.9 from supernova yields @xcite 4 . abundance measurements are determined for c , n , o , si , and ca are consistent with similar abundances in warmer stars in our sample ( @xmath4 4300 k ) . the abundances from our l - band spectra also matched abundances determined previously for these stars from the literature . our abundances also matched f and g dwarfs in the solar neighborhood @xcite . [ @xmath0cl / fe ] measurements are consistent with the slope of galactic chemical evolution models from @xcite but show on average a higher [ @xmath0cl / fe ] abundances than predicted by 0.16 dex . this offset is also seen in [ @xmath0cl / ca ] and [ @xmath0cl / si ] where chlorine is offset by @xmath40.35 dex compared to chemical evolution models @xcite . an additional processes producing chlorine may be necessary to explain the overabundance of cl compared to predictions . the @xmath29 process may affect cl production @xcite and should be considered . chlorine abundances in two ba - rich stars are similar to the chlorine abundance in other , non s - process rich stars in the sample . the similarity of these chlorine abundances suggests that additional a(@xmath0cl ) is not produced through the s - process we thank the anonymous referee for their thoughtful comments and suggestions on the manuscript . this paper is based on observations obtained at the kitt peak national observatory , a division of the national optical astronomy observatories ( noao ) . noao is operated by the association of universities for research in astronomy , inc . under cooperative agreement with the national science foundation . we are grateful to the kitt peak national observatory and particularly to colette salyk for her assistance at the start of the observing run and to anthony paat , krissy reetz , and doug williams during the run . this research has made use of the nasa astrophysics data system bibliographic services , the hitran database and the kurucz atomic line database operated by the center for astrophysics , and the vienna atomic line database operated at the institute for astronomy of the university of vienna . this research has made use of the simbad database , operated at cds , strasbourg , france . we thank eric ost for implementing the model atmosphere interpolation code . c. a. p. acknowledges the generosity of the kirkwood research fund at indiana university .

a tight binding model is introduced to describe the strong interaction limit of excitonic ordering . at stoichiometry , the model reduces in the strong coupling limit to a pseudo - spin model with approximate u(4 ) symmetry . excitonic order appears in the pseudo - spin model as in - plane pseudo - magnetism . the u(4 ) symmetry unifies all possible singlet and triplet order parameters describing such states . super - exchange , hunds - rule coupling , and other perturbations act as anisotropies splitting the u(4 ) manifold , ultimately stabilizing a paramagnetic triplet state . the tendency to ferromagnetism with doping ( observed experimentally in the hexaborides ) is explained as a spin - flop transition to a different orientation of the u(4 ) order parameter . the physical mechanism favoring such a reorientation is the enhanced coherence ( and hence lower kinetic energy ) of the doped electrons in a ferromagnetic background relative to the paramagnet . a discussion of the physical meaning of various excitonic states and their experimental consequences is also provided .

cond-mat0002259

the unexpected discovery of high-@xmath0 itinerant ferromagnetism in doped hexaborides@xcite has re - ignited interest in the problem of excitonic ordering near the semiconductor metal transition.@xcite excitonically ordered states are characterized by an off - diagonal order parameter describing pairing between conduction electrons and valence holes . early theoretical work by volkov _ et . al._@xcite anticipated the emergence of ferromagnetism on doping such an excitonic state . these authors considered the limit of nearly nested overlapping conduction and valence bands with _ weak _ repulsive electron - electron interactions . in this limit , the problem can be approximately cast into a form nearly identical to bcs theory , and studied using the techniques of mean - field theory . although this work ( and some subsequent recent studies@xcite ) suffers from the important physical mistake of neglecting the instability to phase separation , ferromagnetism remains nevertheless a generic feature in a corrected treatment.@xcite while the appearance of ferromagnetism in the weak - coupling limit is encouraging , it is far from a conclusive and complete theoretical explanation for the experiments . first , coulomb interactions in the hexaborides are not particularly weak , and most likely are comparable to the fermi energy and band overlap . second , the above explanation appears to hinge on the first - order nature of the excitonic to normal ( e - n ) transition in the bcs limit . while this feature , mathematically analogous to the first - order transition to the normal state due to pair - breaking by an external zeeman field in a superconductor,@xcite is present in the nested mean - field limit , there do not appear to be any general theoretical grounds mandating this behavior more generally . moreover , the universality of the experimental results , now observed in a large number of different compounds ( ca@xmath1la@xmath2b@xmath3 , bab@xmath3 , ca@xmath1ce@xmath2b@xmath3 , srb@xmath3, )@xcite , argues for the robustness of the phenomenon . to determine whether excitonic ferromagnetism is indeed more general than its weak - coupling theoretical basis , we consider here the completely opposite _ strong - coupling _ regime this is not expected to be directly applicable to the hexaborides , as these materials are most likely best described by an intermediate - coupling model . nevertheless , many useful insights are gained from this complementary limit . as usual , the principle assumption of the strong - coupling limit is the dominance of potential over kinetic energy . this is achieved concretely using a tight - binding model ( see eqs . [ tb1]-[nnn ] , in sec . [ sec : model ] ) , in which the conduction and valence bands of the conventional continuum theories are replaced by localized @xmath4 and @xmath5 orbitals , respectively . the analog of band gap in the continuum model is the level splitting @xmath6 . the order parameter characterizing excitonic ordering is then a matrix in spin space : @xmath7 where @xmath8 creates an electron with spin @xmath9 in the @xmath4 orbital , and @xmath10 annihilates an electron with spin @xmath11 in the @xmath5 orbital . excitonically ordered states thus have some partial occupation of the nominally excited @xmath4 states , as a result of coulombic repulsion . in general , @xmath12 is a proper order parameter ( i.e. one which characterizes a spontaneously broken symmetry ) if the @xmath4 and @xmath5 orbitals have different symmetries . in this paper , we consider a `` minimal model '' with this property , comprised of one @xmath4 and one @xmath5 orbital per unit cell see fig . [ fig : orbs ] . this mimics the situation in the hexaborides , for which the conduction and valence states also transform as different representations of the cubic point group@xcite . because of complications arising from orbital degeneracy , however , the appropriate representations for the hexaborides are three - dimensional rather than scalar . we defer the possible complications arising from these additional degrees of freedom to a future investigation . and @xmath13 orbitals , respectively.,width=192 ] as for the more familiar hubbard model ( see , e.g. ref . @xcite ) , the problem simplifies somewhat in the strong - coupling limit . considering first the undoped system ( half - filled = two electrons per unit cell ) , we obtain a novel quantum pseudo - spin model ( eqs . [ tj1]-[ising ] , sec . [ sec : undoped ] ) . within this model , the excitonic insulator ( ei ) appears as an intermediate state separating not a metal and a semiconductor but a _ mott insulator _ and a semiconductor ( or band insulator ) . in some respects , the behavior is argued to be quite similar to that of a quantum spin-@xmath14 xxz antiferromagnet in a magnetic field , with excitonic ordering analogous to xy antiferromagnetism . the `` spins '' of the model , however , can take on _ five _ distinct states per site : one singlet state with both electrons in the lower - energy @xmath5 orbital , and four different spin states with one @xmath4 and one @xmath5 electron . this is in contrast to the two states of a single spin-@xmath14 particle . in the strong coupling limit , this large hilbert space is `` unified '' by several approximate symmetries valid at different energy scales . at the largest energy scales this is an enormous su(4 ) group , corresponding to arbitrary complex rotations of the four components of @xmath12 . the approximate su(4 ) symmetry fully unifies all possible excitonic states , including singlet , triplet , and singlet - triplet coexistences . these are described by the general decomposition @xmath15 where @xmath16 , @xmath17 are the singlet and triplet order parameters , and @xmath18 and @xmath19 are the @xmath20 unit and pauli matrices in spin space , respectively . a system with approximate su(4 ) invariance contains the germ of ferromagnetism , since several possible excitonic states ( those with non - zero @xmath21 and/or @xmath22 ) give rise to net exchange fields , and hence a magnetic moment . su(4 ) symmetry implies that these states are low in energy . at intermediate energies the su(4 ) symmetry reduces to an su(2)@xmath23su(2 ) invariance , which reflects separate spin rotations of the @xmath4 and @xmath5 electrons . the latter is a symmetry of the conventional continuum models of eis , and transforms the order parameter in a `` chiral '' manner : @xmath24 , where @xmath25 and @xmath26 are su(2 ) matrices . finally , further weak interactions reduce this to a simple su(2)@xmath27 symmetry at the ( very ) lowest energies . these symmetry considerations underly the simple physical mechanism for ferromagnetism elucidated here@xcite . the dominant tendency imposed by coulomb interactions is to excitonic ordering . with approximate su(4 ) symmetry , however , the `` orientation '' ( form of @xmath12 ) of the order parameter is nearly free and fixed only by weak `` anisotropy '' terms . in the undoped material , these anisotropies favor a simple paramagnetic triplet state . doping introduces additional exchange energy contributions that modify the anisotropy , causing @xmath12 to `` flop '' into a different orientation with a ferromagnetic moment . in the present model , the excitonic order in the ferromagnet is of non - collinear triplet type , in which @xmath28 as shown in sec . [ sec : discussion ] , in addition to ferromagnetic magnetization , this state has additional spatially - varying local static moments and spin currents transverse to the axis of net magnetization . the transition to this state from the paramagnet is generally first order , and therefore coincides with a jump in the electronic density . since experiments are performed at fixed charge density ( dictated by the concentration of dopant ions ) , the intermediate `` forbidden '' range of dopings can be accommodated only by phase separation . with long - range coulomb interactions included , macroscopic phase separation is impossible , and charge domain formation is expected , as pointed out already in refs . @xcite . the _ detailed _ demonstration of this behavior with doping is non - trivial . as for many other strongly correlated systems , the problem of doping is much more difficult than that of the stoichiometric mott insulator . indeed , as the ei state lies intermediate between band and mott insulators , doping the ei is a sort of interpolation between doping a conventional band insulator and doping an antiferromagnetic insulator . the latter problem is of course at the crux of the physics of high - temperature superconductivity , so that perhaps the experimental and theoretical insights gained in the hexaborides will be helpful more generally . at any rate , doping the ei can be shown by very simple arguments to favor ferromagnetism in strong coupling . essentially , the physics of this behavior is similar to the `` nagaoka effect''@xcite in a doped antiferromagnet ferromagnetic alignment of the excitonic order parameters allows for more coherent propagation of the doped electrons , and hence a lowering of their kinetic energy . this mechanism is actually _ stronger _ in the ei than in the antiferromagnet , because of the global _ coherence _ of the excitonic condensate , and the near degeneracy ( due to approximate su(4 ) symmetry ) of ferromagnetic and paramagnetic states . ( half the bare splitting between @xmath4 and @xmath5 orbitals ) , and @xmath29 , the chemical potential . the shaded region is not analyzed in this paper . thick lines indicate the boundary between the undoped region below ( with two electrons per unit cell ) and the doped region above ( with more than two electrons per unit cell ) . the af@xmath30 and af@xmath31 phases are antiferromagnetic mott insulators with two and three electrons per unit cell , respectively . the bi state is the band insulator . intermediate between the bi and af@xmath30 phases are the excitonic insulator ( ei ) and an insulator with coexisting excitonic and nel order ( ei / af ) . the fm , fm@xmath32 , pfm , and pfm@xmath32 phases are all ferromagnetic metals ( see table [ tab : phases ] for the differences between these states ) , while pm indicates a paramagnetic metallic phase . all the metallic states above exhibit excitonic order.,width=312 ] to provide a concrete demonstration of these ideas , the strong coupling zero temperature phase diagram of the model is calculated in this paper using a `` free fermi gas '' approximation . this approximation captures the most important _ single quasiparticle _ physics of electronic propagation in an excitonically - ordered background , but neglects interactions between these quasiparticles . for simplicity , we also assume a _ fixed amplitude _ , @xmath33 , of the excitonic order parameter . the latter assumption is valid for weak doping , @xmath34 , in which the _ orientation _ of the ordering is of paramount importance . putting together the results of this calculation and the stoichiometric behavior , we arrive at the partial phase diagram in fig . [ partialfig ] . this is in agreement with the general expectations stated above . it should be stressed , however , that this analysis of doping is far from exhaustive . more detailed investigations of both the weak and strong coupling limits are currently underway@xcite the remainder of the paper is structured as follows . in sec . [ sec : model ] , we present a detailed exposition of the ( simplest ) tight - binding model capable of describing excitonically - ordered states , and consider the limit of infinite interaction strength . the bulk of the paper is contained in sec . [ sec : undoped ] , where the model is analyzed for large but finite interactions , focusing on the stoichiometric situation with two valence electrons per unit cell . for this electron density the model is insulating , but can sustain excitonic and other types of ordering . the properties of the model with doping are discussed in sec . [ sec : doping ] . we conclude in sec . [ sec : discussion ] with a clarifying discussion delineating the physical properties of various possible excitonic insulators , and the relation of the results of this paper to the hexaborides . we consider a minimal model capable of exhibiting excitonic order , which contains two orbitals per unit cell , so as to give rise to two bands in a non - interacting limit ( the actual situation in the hexaborides is more complex , with orbital degeneracy leading to multiple electron and hole pockets ) . a strong - coupling limit is obtained by first considering only _ local _ interactions within a unit cell : @xmath35 where @xmath36 are electron annihilation operators for the `` conduction '' and `` valence '' states , respectively , obeying @xmath37 . here and throughout the paper , we use latin indices @xmath38 to denote the lattice site , and greek indices @xmath39 to denote the spin state . labels will be suppressed and implicit wherever clarity allows . the parameters @xmath40,@xmath29,@xmath41,@xmath42 describe the `` band gap '' ( orbital energy difference ) , chemical potential , on - site `` hubbard '' repulsion , and nearest neighbor repulsion , respectively , within the unit cell . a crucial feature of @xmath43 is the absence of direct hopping between the @xmath4 and @xmath5 orbitals within the unit cell . for excitonic ordering to be well - defined , it is necessary at a minimum that the @xmath4 and @xmath5 states be distinguished by a discrete symmetry operation , e.g. parity . when this is the case , direct hopping between these orbitals is prohibited . it may be helpful to imagine an artificial situation in which the @xmath4 and @xmath5 orbitals represent @xmath44 and @xmath13 orbitals on a single site of a square lattice ( see fig . [ fig : orbs ] ) . in this situation , @xmath4 and @xmath5 orbitals are orthogonal both on the same site and on nearest neighbor sites . an overlap _ is _ possible , though for next - nearest neighbor pairs , i.e. on a diagonal . in general , an exchange interaction _ is _ allowed by symmetry , and takes the form @xmath45 where @xmath46 , @xmath47 . here and in the following , the pauli matrices @xmath48 act in the spin space . on physical grounds , a ferromagnetic exchange ( @xmath49 ) is most appropriate due to hund s rule effects for orthogonal orbitals . for pedagogical purposes , we may wish to consider instead the opposite antiferromagnetic sign for this exchange . from the discussion in sec . 1 , it is clear that an essential ingredient for excitonic ferromagnetism is the near - degeneracy of singlet and triplet states . to build this into the strong coupling model thus requires small @xmath50 . for the majority of the paper , therefore , we will neglect @xmath50 or treat it as a small perturbation . the analysis of the strong - coupling limit begins by first considering the on - site hamiltonian , @xmath51 , in the absence of electron hopping between adjacent unit cells . this may be thought of as the analog of the @xmath52 analysis of an ordinary hubbard model . in this case , the occupation of each orbital is a good quantum number , and the states can be straightforwardly enumerated . assuming @xmath53 , and at first also @xmath54 , one obtains the phase diagram shown in fig . [ fig : ultralocal ] . @xmath29 plane , neglecting exchange and all inter - cell hopping and interactions . regions with zero and four electrons per unit cell are not shown . in each phase , the lowest - energy states are pictured , with @xmath4 orbital above and @xmath5 orbital below.,width=264 ] for the present study , we are particularly interested in densities near two electrons per unit cell . note that the doping behavior ( i.e. on increasing @xmath29 ) in this regime depends crucially on the relative strength of @xmath55 and @xmath41 . in particular , for @xmath56 , the preferred charge @xmath57 state is one with both electrons in the lower orbital , corresponding to the band insulator . for @xmath58 , by contrast , the two - electron ground state has one electron in each orbital , and hence a net spin on each site . this is the ultra - strong coupling ( i.e. local ) version of a mott insulator . note that _ neither _ of these two states exhibits _ excitonic _ order . this can be seen by directly computing @xmath59 in either state . in fact , the operators @xmath60 and @xmath61 act to transform the two phases into one another , i.e. move an electron from the lower to upper orbital or vice versa . to investigate further , we must introduce hopping between adjacent cells . we will principally consider the simplest such term , @xmath62 where @xmath63 indicates that the sum is over nearest neighbor pairs of sites . different hopping integrals @xmath64 and @xmath65 between the two orbitals can also be easily included , but do not change the results significantly , so we will keep @xmath66 for simplicity ( see however , the discussion of particle - hole symmetry in sec . [ sec : discussion ] surrounding eqs . [ ph1]-[ph2 ] ) . in general , there are also hopping processes connecting @xmath4 and @xmath5 orbitals . due to the symmetry of the orbitals in fig . 1 , these occur only for next - nearest neighbors , @xmath67 \left ( a_i^\dagger b_j^{\vphantom\dagger } + { \rm h.c.}\right ) , \label{nnn}\ ] ] where the double angular brackets denote a sum over next nearest neighbors @xmath68 and @xmath69 . note that the hopping matrix elements in eq . [ nnn ] are real and vary in sign . the sign variations reflect the symmetry differences ( under rotations ) between the @xmath44 and @xmath70 orbitals . the reality of the coefficients is a matter of convention , which we fix by choosing the orbital wavefunctions to be real . we will assume , as appropriate in this example , that @xmath71 , so that @xmath72 is small , and can therefore be treated perturbatively . it is sometimes an important perturbation , because it reduces the symmetry of the hamiltonian . in particular , all of the terms in @xmath73 conserve the number of @xmath4 and @xmath5 particles separately . neglecting the @xmath4@xmath5 hopping , therefore , the model has @xmath74 continuous symmetries , corresponding to conservation of spin , and @xmath4 and @xmath5 charges . the perturbation @xmath72 reduces the continuous symmetries of the model down to @xmath75 corresponding to spin and total charge , which are required by the physics of the system . [ nnn ] actually still respects a number of discrete symmetries , such as @xmath76 simultaneously with a @xmath77 rotation . these symmetries , which in fact comprise the point - group operations of the square lattice , can be viewed as a residual discrete subgroup of the original @xmath78 present in the absence of @xmath72 . we will see in the next section that this gives rise to an ising symmetry under which the excitonic order parameters transform . in the central region of fig . 2 , e.g. for @xmath79 , all sites are doubly occupied in the strong coupling limit . nevertheless , for @xmath80 , the low energy states are highly degenerate . well to the left of the thick vertical line , each @xmath4 and @xmath5 orbital is singly occupied , so that there are effectively two spin-@xmath14 degrees of freedom in each unit cell . in the infinite coupling limit these are completely free , but they will interact due to virtual hopping processes when @xmath81 ( and @xmath72 ) is included . far to the right of the vertical line , the unique low energy state consists of a doubly occupied @xmath5 orbital in each unit cell , and hopping is unimportant . as the vertical line is approached from either side , virtual hopping processes can induce interactions involving all _ five _ low energy states . in this subsection , we develop an effective model for the interesting region near the vertical line . in this region , it is necessary and sufficient to truncate the hilbert space to just the five low - energy states in each unit cell ( although higher energy states must be kept in virtual processes ) . the physics is amusingly similar to a sort of generalized bosonic @xmath82@xmath83 model . on the left - hand side of the thick vertical line , each unit cell is occupied by two spins . at second order in @xmath81 , these interact via effective exchange interactions , @xmath84 where @xmath85 . this exchange constant may be obtained by computing perturbatively the energy difference between singlet and triplet states on a bond to second order in the hoppings , and neglecting the deviation from the vertical line ( i.e. setting @xmath86 ) in the denominators . the latter approximation is valid provided @xmath87 . well to the left of the vertical line ( in particular when @xmath88 ) , no doubly - occupied @xmath5 states are present , and eq . [ tj1 ] is a complete model . it describes two ferromagnetically bulk coupled heisenberg spin-@xmath14 antiferromagnets . on a hypercubic lattice ( square or cubic in two or three dimensions , respectively ) , one expects long - range antiferromagnetic order of spins on the same orbital sublattice , with @xmath4 and @xmath5 spins aligned parallel at each site . as the vertical line is approached , the energy cost of a doubly - occupied @xmath5 orbital is reduced towards zero , and they must be introduced into the lattice . unit cells with both electrons in the @xmath5 orbital act as `` holes '' , having no associated local moment . unlike the usual @xmath82-@xmath83 model holes , they are , however , bosonic and neutral ( relative to the magnetic state , they represent the removal of an @xmath4 electron and replacement with a @xmath5 electron ) . hole hopping occurs at second order in @xmath82 , : @xmath89 where @xmath90 $ ] is the hole `` chemical potential '' , @xmath91 , @xmath92 , and @xmath93 is the operator which interchanges the spin states at sites @xmath68 and @xmath69 . like in a conventional doped anti - ferromagnet , the presence of the @xmath94 operator in the hopping term leads to difficulties of hole motion in an antiferromagnetic spin background . naive successive hopping of a single hole in an antiferromagnetic state results in a generalization of the well - known `` string '' of misaligned spins in its wake . introducing the @xmath95 hopping term ( eq . [ nnn ] ) affects the system in several ways . there are renormalizations of the coupling constants in eq . [ tj2 ] and eq . [ tj1 ] , of order @xmath96 , @xmath97 . since , by assumption , @xmath71 , these are negligible . new exchange couplings are also generated between next - nearest - neighbor @xmath4 and @xmath5 spins , which were not previously present . because they are small , unfrustrating , and break no additional symmetries , these are also negligible . the most important effect is to introduce a term which violates @xmath98particle conservation : @xmath99 . \label{fugacity}\ ] ] here @xmath100 creates a triplet of spin one states of @xmath101 particles on the pair of sites @xmath102 , @xmath103 creates a singlet of @xmath101 particles on this pair , and the `` fugacity '' @xmath104 . note that although eq . [ fugacity ] violates conservation of the number of `` holes '' , it creates and annihilates them only in pairs . there thus remains a conserved ising charge or parity ( @xmath105 ) , signifying whether the number of holes is even or odd . this parity can be traced back to fact that the two orbitals on each site transform differently under spatial reflections . to understand the behavior of this model , we now introduce a useful reformulation . formally , the five possible states on each site can be viewed as different quantized values of a generalized pseudo - spin , and the above terms then take the form of nearest - neighbor interactions between these spins . in particular , we define five states per site via @xmath106 spin matrices @xmath107 , where @xmath108 . neglecting for the moment the hole non - conserving terms in eq . [ fugacity ] , @xmath109 , where @xmath110 and @xmath111 . the generalized exchange constants @xmath112 , @xmath113 , and zeeman field @xmath114 . this form of the hamiltonian exposes a strong similarity to the spin-@xmath14 xxz model in a zeeman field . in particular , the `` boson hopping '' @xmath115 is analogous to an antiferromagnetic in - plane exchange ( @xmath116 terms ) , spin - boson interaction @xmath117 to an antiferromagnetic ising exchange , and @xmath118 to a @xmath119-axis field . for @xmath120 and @xmath118 not too large , one expects the analog of canted xy antiferromagnetism , while for @xmath121 , one expects instead @xmath119-axis ising antiferromagnetism up to a threshold value of @xmath122 . for large fields , @xmath123 , one expects ultimately fully polarized states , which correspond to the mott and band insulators for @xmath124 and @xmath125 , respectively . surprisingly , @xmath126 displays an enormous su(4 ) invariance under @xmath127 , @xmath128 , where @xmath41 is an su(4 ) matrix . su(4 ) symmetry is expected to be a good approximation over a range of energies , because in the physical limit @xmath129 , @xmath130 . thus we will take the approach of first solving the su(4 ) invariant model , and considering successively the exchanges @xmath83 and @xmath131 , which reduce the symmetry of @xmath132 to su(2)@xmath23su(2 ) ( independent _ physical _ spin rotations of the @xmath4 and @xmath5 moments ) and su(2)@xmath23u(1 ) , respectively . lastly , we consider the effects of the hole - pair creation and annihilation terms in eq . [ fugacity ] , which can also be transcribed into the pseudo - spin language . one finds @xmath133 , where @xmath134 . \label{ising}\end{aligned}\ ] ] the coupling @xmath135 . while it is perhaps not completely transparent in this notation ( a better notation for this term will be introduced in next subsection [ ising2 ] ) , the effect of @xmath136 is to further break the su(2)@xmath23u(1 ) symmetry down to su(2)@xmath23z@xmath137 . the z@xmath137 invariance is the remnant of the physical parity symmetry discussed in the previous subsection . we expect that a simple weiss mean field theory ( mft ) gives qualitatively correct results for the stoichiometric phase diagram , as it does for the ordinary xxz+zeeman model.@xciteneglecting @xmath138 , the mft consists in replacing @xmath139 for each bond @xmath140 on the lattice , and similarly for the @xmath141 interaction . with this replacement , the hamiltonian decouples on different lattice sites , and the problem reduces to solving self - consistently the appropriate single - site problems . as an antiferromagnetic solution is expected , this amounts to equations for the ( @xmath142 component ) transverse staggered magnetization , defined by @xmath143 $ ] , and the uniform and staggered @xmath119-axis magnetizations , defined by @xmath144 . because of su(4 ) symmetry , all orientations of @xmath145 are degenerate , and it is sufficient to assume @xmath146 . in this subspace , the equations of mft become _ identical _ to those of the conventional spin-@xmath14 xxz antiferromagnet in a zeeman field . these equations were solved in ref . the resulting phase diagram is shown in fig . [ xxzfig ] . ( 4.0,3.0)(0,0 ) ( 0.4,0.1 ) ) stabilizes either the mott insulating ( mi ) or band insulating ( bi ) state , depending upon its sign . for `` in - plane '' anisotropy ( @xmath147 ) , the intervening phase is an excitonic insulator ( ei ) . in the opposite limit ( `` ising '' anisotropy ) it consists of a micro - phase - separated state with a checkerboard pattern of alternating band and mott insulating configurations at the lattice scale . in obtaining the pseudo - spin model from the strong - coupling limit of eqs . [ tb1]-[tb3 ] , one finds in - plane anisotropy , and the intermediate state is expected to be excitonically ordered . , title="fig:",width=288 ] ( 0.0,2.8)@xmath148 ( 2.8,1.7)@xmath149 ( 2.6,2.1)*ei * ( 1.0,1.7)*mps * ( 1.0,2.6)*mi @xmath150 * ( 1.0,0.5)*bi @xmath151 * since @xmath152 , we expect transverse pseudo - spin polarization , @xmath153 , provided @xmath154 . remarkably , the transverse components of the pseudo - spin operator are exactly the excitonic order parameters . in particular , straightforward algebra shows @xmath155 . thus for @xmath156 , mft predicts an excitonic insulator . we now turn to the evolution of the ground state in this regime on introducing the symmetry - breaking terms in @xmath138 . in their absence , the excitonic order parameter can `` point '' in any direction which is equivalent under the broken su(4 ) symmetry . within mft , this amounts to complete freedom to choose the four complex components of @xmath12 , subject to the constraint @xmath157 . in term of singlet and triplet components defined by eq . [ ddecomp ] , this constraint simply implies @xmath158 . the perturbations in @xmath138 can be viewed as `` anisotropies '' favoring sub - manifolds within this space . to clarify the nature of the anisotropy terms , it is helpful to work with the mean - field wavefunction , @xmath159 , where @xmath160 is the non - interacting band - insulating state , and @xmath161 is a local `` exciton creation operator '' . here it is now straightforward to evaluate the expectation value of @xmath138 in the mean - field ground state . up to a constant for fixed @xmath163 , on a hyper - cubic lattice one finds the bulk energy density @xmath164 where @xmath165 and @xmath166 . the above terms are essentially completely determined by the su(2)@xmath23su(2 ) and su(2)@xmath23u(1 ) symmetries . to proceed , we employ two identities derivable from eq . [ ddecomp ] : @xmath167 by assumption , @xmath168 , so that the first term in eq . [ anisotropy ] creates the dominant splitting of the su(4 ) ground - state degeneracy . the low - energy sub - manifold thus consists of the order parameters which minimize @xmath169 . [ id1 ] then implies @xmath170 ( note that the first term in eq . [ id1 ] is constant and equal to @xmath171 ) . the physical content of this condition is made clear by calculating the mean spin polarization on the @xmath4 site using the mean - field wavefunction in eq . [ creationop ] : @xmath172 thus the influence of the exchange coupling @xmath83 is to favor states with @xmath173 . this condition still allows a fairly large range of states , the simplest of which are pure singlet ( @xmath174 , @xmath175 ) and pure collinear triplet ( @xmath176 , @xmath177 , @xmath178 ) orderings . the additional effect of the hunds - rule ferromagnetic coupling @xmath131 is to introduce a small extra `` mass '' for the singlet order parameter , favoring a pure triplet state . the _ phase _ of the triplet order parameter is determined by the `` ising anisotropy '' terms in eq . [ ising ] . to see this , we rewrite @xmath179 and @xmath180 directly in terms of @xmath181 . one finds @xmath182 note that eq . [ ising2 ] explicitly breaks the u(1 ) symmetry of phase rotations of @xmath181 , down to the ising invariance @xmath183 . if @xmath136 is considered a weak perturbation , it can be treated by simply evaluating its expectation value in the mean - field ground state ( eq . [ creationop ] ) , giving @xmath184 , since @xmath68 and @xmath69 are next - nearest neighbors . using @xmath185 , one finds ( since @xmath186 ) that eq . [ ising2 ] favors an _ imaginary _ triplet order parameter @xmath187 . this is _ different _ from the weak - coupling treatment of ref . @xcite , in which a _ real _ triplet order parameter was found to be preferred . unlike in superconductivity , the phase of the excitonic order parameter has physical significance , as discussed by halperin and rice@xcite . in particular , it is straightforward to show that a _ real _ @xmath17 order parameter corresponds to a non - zero average spin density within the unit cell of the crystal , while for @xmath17 imaginary , the spin density is zero but there are instead non - zero spin _ currents_. the imaginary triplet state obtained here is therefore a sort of spin `` flux phase '' with non - zero spin currents . see sec . [ sec : discussion ] for a more in - depth discussion . apart from this difference , the strong - coupling results of this section are in very close agreement with the weak - coupling results of ref . indeed , not too much significance should be attached to the difference in phase of the order parameters , as indeed the models are in any case not completely identical . in fact , the detailed correspondence of results up to this point strongly argues for a continuous smooth interpolation ( `` adiabatic continuity '' ) of most physical properties of such systems as the overall interaction strength is increased from small to large values . finally , we comment on the modifications to the su(4)-invariant phase diagram in the presence of the symmetry - breaking terms in eqs . [ tj1 ] , [ fugacity ] . as argued above , these favor an imaginary triplet state when @xmath188 . inside the mott insulator , these terms stabilize an antiferromagnetically ordered magnetic state . on approaching the mott insulator boundary , therefore , we expect the emergence of magnetic ordering . this implies the existence of at least one additional phase boundary separating the triplet ei ( which has no non - zero spin density ) from a magnetically ordered ei with non - zero average spin density , somewhere inside the region in which the ei phase occurs in the su(4)-invariant model . in this section , we consider the behavior as a low density of electrons is added to the system . in the strong - coupling limit , this reduces to an effective @xmath82-@xmath83like model , in which the hilbert space is restricted to states in which all sites ( unit cells ) are either doubly ( corresponding to the excitonic pseudo - spins modeled above ) or triply occupied , the latter containing one @xmath4 and two @xmath5 electrons . the system is then governed by an effective hamiltonian @xmath189 , where @xmath190 projects onto this restricted hilbert space . as many years of work on high-@xmath0 superconductivity has taught us , the problem of doping a correlated ( mott ) insulator , particularly with spin ( and here pseudospin ) ordering , is extremely complex and difficult . here , we will adopt the absolute simplest approach extending the above mft to the low electron density limit . we assume , as suggested by the weak - coupling analysis , that the essential ingredient for excitonic ferromagnetism is the approximate enhanced ( in this case su(4 ) ) symmetry of the effective hamiltonian . in considering the doped state , then , it is crucial to determine in what way the added electrons affect the splitting of the degenerate su(4 ) ground - state manifold . in the strong - coupling limit , the majority of the energy of an added electron is kinetic , since @xmath191 , etc .. just as in the simpler but much studied @xmath82@xmath83 model for the cuprates , coherent motion of an added electron , however , is greatly hindered by ( pseudo)-spin ordering of the insulating background . moreover , coherent motion is possible to a varying degree depending upon the precise nature of the background . we first consider this effect for a single added electron using the variational method . a natural variational ansatz is @xmath192 where _ both _ the doped electron s wavefunction @xmath193 and the excitonic order parameter @xmath12 ( implicit in @xmath194 ) are considered as variational parameters . for fixed @xmath195 , the energy depends only upon @xmath83 , @xmath131 , and @xmath82 . in particular , one finds @xmath196 where @xmath197 is the lattice spacing , @xmath198 and the matrix @xmath199 . physically , we identify the first term in eq . [ evar ] as the bulk energy density , reduced by the presence of a single doped electron ( occupying the volume fraction @xmath200 ) . in the second term , the quantity @xmath201 is then readily interpreted as the energy of the added electron . [ hopping ] is then a hopping hamiltonian for this electron . in a polarized excitonic background , this hopping is in general non - diagonal in spin . in terms of singlet and triplet components , @xmath202 where @xmath203 , the mean spin polarization on the @xmath4 site , is given by eq . [ meansa ] . minimizing eq . [ hopping ] in the space of normalized wavefunctions @xmath193 gives the tight - binding schrdinger equation , @xmath204 where the angular brackets indicate a sum over the nearest neighbors @xmath69 of site @xmath68 . the single - particle eigenstates of this equation are plane waves with spins polarized parallel and antiparallel to @xmath203 , with eigenvalues @xmath205 \sum_{i=1}^d \left[\cos k_i a_0 \right ] , \label{vardisp}\ ] ] where @xmath197 is the lattice spacing . the location of the minimum - energy electronic excitations depends crucially on the magnitude of @xmath203 , and hence @xmath118 . when @xmath206 , electrons with spin parallel and antiparallel to @xmath203 have minimum energy at different points in momentum space . such large values of @xmath118 correspond to strongly overlapping bands , close to the boundary between the mott and excitonic insulators . for simplicity , we will specialize to the case when @xmath207 , which occurs for @xmath208 . in this case , the minimal energy single - particle energy excitations for both spin orientations have momentum @xmath209 . furthermore , the optimal spin orientation is parallel to @xmath203 . such an electron takes advantage of the `` zeeman '' energy due to the exchange field ( proportional to @xmath203 ) generated by the `` core '' spins ( i.e. the spins of the two electrons per unit cell present in the insulator ) . in the undoped system , however , @xmath173 , due to the anisotropy in eq . [ anisotropy ] . we therefore expect that the optimal order parameter in the doped system is determined by a competition between these two terms . with some algebra , it is straightforward to verify that , due to the hunds - rule term @xmath131 , the complex pure triplet state ( i.e. with @xmath210 but @xmath211 ) is always more energetically favorable than a singlet - triplet coexistence ( with @xmath212).@xcite without loss of generality , it is thus convenient to choose a spin quantization axis , letting @xmath213 one then finds @xmath214 . in any such state , @xmath215 , so that the core spins also contribute to the ferromagnetic moment . it remains to determine the optimal angle @xmath216 . to proceed , we need to extend eq . [ evar ] to a small but non - zero _ density _ of doped electrons . at low densities , it seems natural to neglect interactions between doped electrons , and use the simplest possible _ free fermi gas _ estimate for the electronic dopant energy . in particular , we approximate the energy of the system as the sum of two contributions : a `` bulk '' contribution from the undoped unit cells containing two electrons and a spatially uniform order parameter @xmath12 , and a `` dopant '' contribution , approximated by the energy of a free fermi gas of electrons with dispersion given by eq . [ vardisp ] . for concreteness , the detailed formulae are presented in the following for three spatial dimensions ( @xmath217 ) . at low densities , only single - particle states near @xmath218 are occupied , so it is convenient to expand around this point , @xmath219 , yielding the dispersion @xmath220 \left [ 3 - { { q^2 a_0 ^ 2 } \over 2 } \right ] - \tilde\mu.\ ] ] here we have re - instated a ( shifted ) chemical potential @xmath221 to control the density of doped electrons . it is both convenient and physically helpful to work at fixed chemical potential rather than fixed charge density , as this allows naturally for the possibility of phase separation . as is perhaps not surprising based on the results of weak - coupling analysis@xcite , we will see that phase separation does indeed occur in a physically interesting parameter range of the model ( at least within this approximation ) . because we are interested in the energy density only insofar as to determine the angle @xmath216 , we neglect in the following all terms independent of @xmath216 . inserting eq . [ nonunitary ] into eq . [ anisotropy ] gives the bulk energy @xmath222\sin^2 2\theta + { \rm const.}\ ] ] ( in three dimensions ) . this must be added to the ground state energy of the free fermi gas of doped electrons . simple but tedious algebraic calculations lead to the final expression for the total energy density of the system : @xmath223 , \label{ef}\ ] ] where @xmath224 the function @xmath225 is straightforwardly related to the energy density of the three - dimensional free electron gas in a zeeman field . in general it depends not only on @xmath226 and @xmath227 , but also on @xmath228 . for simplicity , we will assume @xmath229 , which holds near to the excitonic insulator band insulator boundary , and is satisfied more generally in the interesting region of the phase diagram ( where @xmath228 is @xmath230 , since @xmath231 in the strong coupling limit @xmath232 see fig . [ dlfig ] ) . in this case , @xmath225 becomes independent of @xmath228 . its functional form is @xmath233 where @xmath234 is the heavyside step function . [ ef][cale ] give the energy density of the system as a function of chemical potential @xmath221 ( through @xmath227 ) and order parameter angle @xmath216 ( through @xmath226 ) . the optimal excitonic angle @xmath216 is determined by minimizing @xmath235 at fixed @xmath221 . if @xmath221 and @xmath216 are known , the density of doped electrons @xmath236 and itinerant magnetization density @xmath237 are then given by the free - fermion results : @xmath238 note that the system is doped ( i.e. @xmath239 ) whenever @xmath240 . one should keep in mind also that the full magnetization density @xmath241 includes a contribution @xmath242 from the core spins . [ ef]-[mag ] completely determine the state of the system at zero - temperature as a function of @xmath243 and @xmath228 . the mathematical problem of minimizing @xmath244 is algebraically quite tedious , and significant care must be taken to avoid spurious local minima and saddle points . the results of a careful study are shown in fig . [ dlfig ] . all the phases shown are excitonically ordered , but differ in doping @xmath236 , excitonic angle @xmath216 , and magnetization @xmath245 . the properties of each are summarized in table [ tab : phases ] . in the strong - coupling limit , we expect ( see sec . [ sec : doping ] ) @xmath246 , in which case there is a direct _ first - order _ transition from a paramagnetic excitonic insulator ( ei ) to a fully - polarized ferromagnetic metal ( fmfp@xmath32 ) . ( 4.0,3.0)(0,0 ) ( 0.3,0.5 ) is the non - dimensionalized chemical potential , and the ordinate @xmath247 indicates the strength of excitonic ordering relative to super - exchange interactions ( see eqs . [ ef]-[gammadef ] for the precise definitions ) . the properties of the various phases shown are listed in table [ tab : phases ] . for larger @xmath243 ( not shown ) , both excitonic ordering and ferromagnetism disappear , owing to reduction of the amplitude of @xmath181.,title="fig:",width=288 ] ( 0.8,2.5)@xmath247 ( 2.8,0.2)@xmath248 the phase boundaries in fig . [ dlfig ] variously indicate first ( discontinuous ) and second(discontinuous ) order transitions . all the vertical phase boundaries denote continuous transitions , while most of the transitions on curved phase boundaries are discontinuous . the exceptions are the ppfm@xmath32fpfm@xmath32 boundary ( which is everywhere second order ) and the lower - portion of the fpfm fpfm@xmath32 transition line , which is continuous below the tricritical point indicated in the figure . .phases of the doped ei in the free - fermi gas approximation . the five columns list the abbreviation , presence or absence of doping , magnetic order , degree of polarization , and excitonic angle ( @xmath216 ) , respectively , for the six phases . [ cols="^,^,^,^,^",options="header " , ] which portion of this phase diagram is most physically significant ? in the strong - coupling limit , @xmath231 , and it therefore seems reasonable to suppose @xmath246 , so that the system undergoes a simple and direct first order transition from the undoped and paramagnetic ei to the fully - rotated half - metallic ferromagnet , fpfm*. coincident with this transition is a jump in the electronic charge density @xmath236 , from zero in the insulator to a non - zero value in the metal . the model introduced in sec . [ sec : model ] contains many possible excitonically ordered states in various regions of its phase diagram . in the undoped case , we have argued that a simple paramagnetic collinear triplet ordering is most likely , while a state with @xmath249 obtains for electron densities slightly greater than two per unit cell . nevertheless , if , as supposed , su(4 ) symmetry is a good approximation , then many other possible states must necessarily be nearly as low in energy . in the hope that the truth may ultimately be decided by experimental measurements , it seems useful to delineate the physical characteristics of each of these phases . with the exception of the non - collinearly ordered states , the analysis of the next few paragraphs is identical ( though in somewhat different notation ) to that of halperin and rice@xcite . first , let us consider the existence of a time - averaged magnetic moment . in the tight - binding formulation , the electron field operator is expanded in terms of wannier orbitals , @xmath250 , \label{decomp}\ ] ] where @xmath251 is the wannier function for the @xmath101 orbital , and we neglect the other ( unoccupied ) states . consider next the spin density operator . we will assume for simplicity ( though this is not essential ) that each wannier function has support only within one unit cell . [ decomp ] then leads to a representation for the spin density operator @xmath252 , @xmath253 to proceed , we choose both wannier functions to be _ real_. then _ for the undoped case _ , the spin density can be rewritten in terms of @xmath254 and @xmath17 : @xmath255 recall further eq . [ meansa ] and its analog for @xmath256 : @xmath257 . \label{sab}\ ] ] there are thus non - zero static local moments whenever @xmath258 , @xmath259 , or @xmath260 are non - zero . in the simplest such states , @xmath261 axis . the _ net _ moment ( integrated over the unit cell ) is , however , zero , unless @xmath262 , in which case the real and imaginary parts of @xmath17 are both non - zero and not parallel . in addition to the net ferromagnetic polarization along @xmath263 , such states have a non - collinear static spin density in the unit cell . the net moment along these other directions remains zero . to see why such states sustain a net polarization , consider the particular case given in eq . [ nonunitary ] , with excitonic angle @xmath216 . one can then use eq . [ ddecomp ] to rewrite the order parameter matrix as @xmath264 . \label{rewrite}\ ] ] inspection of the mean - field wavefunction , eq . [ creationop ] and eq . [ rewrite ] immediately shows that the amplitude for up and down spins are unequal , so long as @xmath216 is not a multiple of @xmath265 . some confusion may arise in the reader with regard to time - reversal symmetry . it appears surprising to have @xmath17 and @xmath266 , the latter containing a cross - product , both contributing to @xmath254 . in fact , both terms transform like a spin under time - reversal . this is simplest to see in the path - integral representation of the quantum system , in which the fermion operators are replaced by time - dependent grassman fields @xmath267 , @xmath268 , and similarly for @xmath269 . the grassman fields then transform under time - reversal according to @xmath270 note the important minus sign in the above transformation , which is possible because @xmath271 , @xmath272 ( @xmath273 , @xmath274 ) are independent fields ( not related by complex conjugation ) in the path integral . this reflects the anti - unitary nature of time - reversal symmetry . at any rate , [ deldef]-[ddecomp ] then imply that @xmath275 under time - reversal . the combination of complex conjugation and the minus sign for @xmath17 imply that both terms in eq . [ sab]are odd under time - reversal . indeed , the necessary and sufficient conditions for broken time - reversal symmetry is @xmath276 and/or @xmath277 . a perhaps surprising consequence of eqs . [ spineqns ] is that apparently if @xmath278 but @xmath279 , there is no net magnetization . in fact , this result applies only to the particular undoped model consider here , and is a consequence of a special variety of particle - hole symmetry ( which we denote ph ) . to make this explicit , define a hole creation operator @xmath280 . then the electron number operator can be rewritten as @xmath281 in the undoped system , the mean number of electrons per unit cell is two , so that @xmath282 . thus precisely at this density , and _ only _ at this density , we may entertain the possibility of symmetry under the transformation ph : @xmath283 in the new variables , the excitonic order parameter becomes @xmath284 . thus @xmath285 ( here the superscript @xmath286 indicates the matrix transpose ) under ph . also useful is the operator @xmath287 ( proportional to @xmath288 in the @xmath289 subspace ) , which is invariant under ph . thus @xmath126 ( see eq . [ pseudospin ] ) is ph - invariant . similarly , it is straightforward to show that under ph , the two spin operators are exchanged : @xmath290 thus @xmath138 is also ph - invariant , as is @xmath136 , as can be easily shown . thus the undoped hamiltonian is invariant under ph . considering the order parameters , we find that @xmath291 thus the combination @xmath292 is _ odd _ under ph , and hence can not give rise to a total moment , since @xmath293 is ph - invariant . @xmath294 , however , is ph - invariant , and can hence couple directly to a ferromagnetic moment . it should be stressed that ph _ is not a microscopically exact symmetry _ , even in the stoichiometric situation . it occurred in the above analysis only because of the arbitrary choice of equal hopping between @xmath4 and @xmath5 orbitals , @xmath295 , in eq . [ tb3 ] . in general , one expects @xmath296 , which leads to different anti - ferromagnetic exchange constants between @xmath4 and @xmath5 spins in @xmath138 , eq . different exchange constants destroy the invariance of the hamiltonian under the interchange of @xmath4 and @xmath5 spins , eq . [ spinic ] , which is the effect of ph . it is straightforward to show that , when this asymmetry is included in the microscopic hamiltonian , states with @xmath297 are also ferromagnetic . in addition , even in the model with @xmath298 , _ doping _ breaks the ph symmetry , and gives rise to a ferromagnetic moment in the @xmath297 state . considerations similar to those above eq . [ spineqns ] apply to the electronic charge density ( @xmath299 ) , current density ( @xmath300 ) , and spin current density ( @xmath301 ) operators . one finds @xmath302 , \nonumber \\ j^{\mu\nu}({\bf r } ) & = & \!\!\!\!\!{1 \over 2 m } { \rm im}\ , \delta_t^\mu \left [ \phi_a({\bf r } ) \partial_\nu \phi_b ( { \bf r})\ ! -\ ! \phi_b({\bf r } ) \partial_\nu\phi_a ( { \bf r } ) \right ] . \label{currentsetc}\end{aligned}\ ] ] in the final equation above , @xmath301 is the current density for spin polarized along the @xmath29 axis propagating in the @xmath303 direction . for completeness , the mean - field expressions for the number of @xmath4 and @xmath5 particles are @xmath304 from eqs . [ currentsetc ] , we can read off the physical interpretation of the various other types of ordering . if @xmath305 , there is a spontaneous _ spin current _ in the unit cell . this is necessarily the case for any state with @xmath262 , which , as discussed above , also exhibits non - collinear static moments . the simpler state with @xmath306 has _ only _ the spin currents , and is the magnetic analog of a `` flux phase '' in modern terminology . similarly , if the singlet order parameter has an imaginary part @xmath307 , there are non - zero charge currents within the unit cell . this is exactly a flux phase . finally , a real singlet order parameter , @xmath308 , gives rise to a charge - density @xmath309 that breaks the point group symmetry of the crystal , since @xmath310 is not a scalar . another importance characteristic of the phases with triplet ordering is a finite ( transverse ) uniform spin susceptibility . this is a very general consequence of broken spin - rotational invariance . in the simplest collinear triplet states , @xmath311 , where @xmath312 is a real vector . the elementary excitations of the symmetry - broken state can then be classified _ only _ by their spin along the triplet axis , @xmath313 . the transverse components of @xmath314 , however , do not commute with @xmath17 . an applied zeeman field along one of these axes therefore immediately acts to mix together the former ground and excited states . it is fairly straightforward to demonstrate by this mechanism a constant transverse spin susceptibility for the collinearly - order triplet states . for non - collinearly ordered triplets , we conjecture that _ all _ components of the uniform susceptibility are finite . this distinction is most likely primarily academic , as experimentally available samples would presumably break up into domains with random orientations of @xmath17 , thus effectively isotropizing the bulk susceptibility . very crude estimates for the magnitude of @xmath315 can be obtain in both the strong and weak coupling limits of excitonically ordered states . in strong coupling , the susceptibility can be computed by naive perturbation theory in the mean - field ground state . in the optimal case ( @xmath188 ) , in which the excitonic ordering is maximal , one finds @xmath316 , where @xmath317 is the characteristic stiffness for excitonic ordering ( see sec . [ sec : undoped ] ) , and @xmath318 is the bohr magneton . in weak coupling , the susceptibility is approximately equal to the free - electron value , @xmath319 , where @xmath320 is the density of states at the fermi energy . a puzzling aspect of the experimental data on the hexaborides is the absence of a substantial gap in optical conductivity measurements in the undoped materials . in general , the excitonically ordered insulators discussed here _ are _ expected to exhibit hard optical gaps ( i.e. complete absence of weight in @xmath321 at small @xmath322 ) at low frequencies and zero temperature , so this is an important point which such a theory must contend with . several possible physical situations can , however , resolve this apparent discrepancy . in the weak and intermediate coupling limits , it is possible to sustain a metallic state simultaneously with excitonic order . this requires imperfectly nested fermi surfaces a detailed investigation of this possibility is underway . even in the strong coupling limit , it is also possible that a gap exists but is anomalously small . indeed , in the present model , the optical gap can be estimated by considering the energy cost required to transfer an electron from one unit cell to its neighbor . at the optimal conditions for excitonic ordering , one has @xmath323 and a straightforward calculation of energies from eq . [ tb1 ] gives the optical gap @xmath324 . thus one expects at least @xmath325 . note that in the strong - coupling limit , there is no universal relation between @xmath326 and the excitonic order parameter . the models and discussions in this paper demonstrate the feasibility of a strong - coupling approach to excitonic ordering . a direct application of the results to the hexaborides is , however , not appropriate , due to the simplified nature of the hamiltonian discussed here . it is possible to generalize the tight - binding model discussed here to a `` two - band '' ( p and d orbital ) hamiltonian which more accurately models the physics of these materials . this model contains the significant new ingredient of orbital degeneracy , and hence considerable additional richness . such orbital degeneracy is the tight - binding analog of the valley degeneracy encountered in band - theoretic treatments . the _ methods _ of this paper , however , remain applicable in this case as well . a thorough treatment of this problem presents an attractive and challenging theoretical opportunity . it is reasonable to ask at this point whether there are any experimental consequences of the excitonic scenario which are relatively model - independent , and can therefore be firmly stated in advance of more accurate results ? for this , we look to the discussion of the previous subsection , focusing particularly on the properties in the undoped material . the excitonic scenario postulates symmetry breaking even without doping , which distinguishes it from , e.g. low - density ferromagnetism ala wigner . all calculations so far appear to favor triplet ordering , which implies first a _ constant ( temperature independent at low @xmath286 ) susceptibility _ in the insulator . second , triplet ordering necessarily gives rise to _ either _ static spin moments or static spin currents ( or both ) within the unit cell . because the latter are presumably difficult to observe , this is a less strong condition . third , because the triplet state breaks spin - rotational invariance , it implies the existence of two low - energy collective `` magnon '' modes ( presumably dispersing as @xmath327 at low energies ) , which could be observable via inelastic neutron or raman scattering . fourth , an excitonic explanation for ferromagnetism upon doping requires that the `` pseudo - spin flop '' phenomena occur , and hence ( in this sense ) approximate su(4 ) symmetry . this approximate symmetry implies the existence of additional collective modes with small excitation gaps . it is natural to ask , given the above emphasis on the undoped state , whether the excitonic ferromagnet is itself truly a distinct phase of matter separate from the more familiar ( theoretically ) wigner ferromagnet ? the answer depends upon the extent to and manner in which the dopant _ ions _ influence the behavior of the electrons . in the models investigated to date , the dopants influence the material only insofar as to donate extra charge carriers , providing no perturbation to the _ potential _ felt by the electrons except to slightly increase the neutralizing positive background charge . in this treatment , the lattice point group symmetries are strictly maintained , and the excitonic ferromagnet is indeed a distinct state of matter : it exhibits more broken ( point group ) symmetries than the wigner ferromagnet . in reality , the dopant ions most likely distribute randomly throughout the crystal , and thereby perturb the potential experienced by the electrons . this random potential explicitly breaks the lattice invariances , and washes out this sharp distinction between the excitonic and wigner ferromagnets . whether this effect is of practical importance is unclear . the large _ increase _ of conductivity upon doping suggests that the electrons are not strongly scattered by the dopant ions . in any case , the physics at low electron density is quite subtle . in particular , the environment around a nearly isolated dopant atom retains a large fraction of the symmetries of the pure lattice ( e.g. a lanthanum dopant replacing strontium preserves cubic point group symmetries around the lanthanum ion ) . because electrons interact with only one impurity at a time at low densities , the symmetry of the local environment is expected to improve the distinction between wigner and excitonic ferromagnets . clearly , further predictions are possible within more specific models . several authors@xcite have recently pointed out the likelihood of phase separation at low electron densities . this occurs naturally in the pseudo - spin flop picture , but because it has already been discussed , we will not dwell on it here . probably most importantly , any excitonically - ordered state by definition breaks the point - group symmetry of the lattice . this symmetry - breaking is directly observable , but unfortunately depends in detail on the way it occurs . in particular , many of the triplet states that appear to be favored have less obvious order parameters , so that more work needs to be done to ascertain the appropriate experimental probes . further modeling using the strong coupling approach promises to help resolve these and other issues . thanks to c. m. varma for stimulating interest in this problem , and to z. fisk for providing copies of experimental data . this research was supported by the nsf career program under grant nsf - dmr-9985255 .

mono- and multilayer fese thin films grown on srtio and bitio substrates exhibit a greatly enhanced superconductivity over that found in bulk fese . a number of proposals have been advanced for the mechanism of this enhancement . one possibility is the introduction of a cross - interface electron - phonon (- ) interaction between the fese electrons and oxygen phonons in the substrates that is peaked in the forward scattering ( small direction due to the two - dimensional nature of the interface system . motivated by this , we explore the consequences of such an interaction on the superconducting state and electronic structure of a two - dimensional system using migdal - eliashberg theory . this interaction produces not only deviations from the expectations of conventional phonon - mediated pairing but also replica structures in the spectral function and density of states , as probed by angle - resolved photoemission spectroscopy , scanning tunneling microscopy / spectroscopy , and quasi - particle interference imaging . we also discuss the applicability of migdal - eliashberg theory for a situation where the- interaction is peaked at small momentum transfer and in the fese / sto system . = 4 _ keywords _ : superconductivity , forward scattering , eliashberg theory , fese films

1602.00656

bulk iron selenide ( fese ) is an unconventional superconductor with a relatively modest transition temperature @xmath4 k at ambient pressures @xcite ; however , its @xmath5 can be increased by upwards of a factor of ten when a monolayer of fese is deposited on srtio@xmath6 ( sto ) @xcite or batio@xmath6 ( bto ) substrates @xcite . this discovery has attracted considerable scientific interest @xcite as it provides not only a new route to high - temperature ( high-@xmath5 ) superconductivity but also a new lens through which unconventional superconductivity can be studied . several proposals have been advanced for the microscopic origin of this enhancement and the most widely discussed scenarios are broadly divided into two categories . the first category involves charge transfer between the substrate and the film , which dopes the monolayer with excess electrons . this may result in modifications of the electronic structure , which in turn enhances an unconventional pairing mechanism @xcite or suppresses competing phases @xcite . in addition to this , the charge transfer shifts the holelike bands centered at @xmath7 below the fermi level @xcite , creating an electronic structure similar to the intercalated fese systems @xcite . this electronic structure challenges the fermi - surface - nesting driven , purely electronic pairing mechanism common to the iron - based superconductors @xcite . the second category of proposals for fese / sto encompasses interface - related effects , where a more direct role is played by the substrate . the main proposals here include the suppression of a competing phase via strain or modifications of the electron - phonon ( @xmath1-@xmath2 ) interaction in the fese layer @xcite , or the introduction of one across the interface @xcite . in this paper we focus on the last scenario , which is motivated by several key experimental observations . the first is that the intercalated fese systems @xcite and fese / sto thin films post - treated with k @xcite or na @xcite adatoms all have maximum @xmath5 values @xmath8 k. this is despite the fact that all of these systems have a similar electronic structure to fese monolayers on sto . this indicates that any common unconventional pairing mechanism in these systems is insufficient to account for the observed high-@xmath9 k in monolayer fese / sto and fese / bto and an additional contribution to pairing is likely needed @xcite . the second ( and more important ) observation is that of replica bands in the electronic structure of superconducting fese / sto @xcite . these are interpreted as shake - off states arise from a coupling between the fese 3@xmath10 electrons and an optical o bond stretching phonon branch in the oxide substrate . the crucial observation here is the size and shape of these replica bands being complete copies of the main electron band shows that the @xmath1-@xmath2 coupling constant must be strongly peaked in the forward scattering direction ( small @xmath11 ) . because of this momentum structure , the cross - interface coupling is strongly intraband in nature and can contribute to pairing in most channels , even those commonly associated with repulsive pairing mechanisms relevant for the iron - based superconductors @xcite . these observations make the cross - interface coupling a likely candidate for the additional pairing needed to produce the observed @xmath5 in fese / sto @xcite . as this is a special issue on _ superconductivity in the two - dimensional ( 2d ) limit _ it is worth stressing that the forward focus of the cross - interface @xmath1-@xmath2 interaction can be understood as a consequence of dimensionality of the fese thin films and the anisotropic dielectric properties of the interface . a more complete discussion of this aspect can be found in ref . @xcite . here , we summarize the main points . the fese films sit a distance @xmath12 above the tio terminated sto substrate @xcite . the motion of the oxygen atoms in the direction perpendicular to the fese film induce a local dipole moment with an effective charge @xmath13 at the surface of the substrate . as a result , the optical o modes create an electric dipole potential on the fese electrons . in the continuum limit , this perturbation will dynamically scatter the fese electrons from states with in - plane momentum @xmath14 to @xmath15 with a matrix element @xmath16 here , @xmath11 is the in - plane momentum transfer , @xmath17 sets the range of the interaction in momentum space , and @xmath18 ( @xmath19 ) is the dielectric constant parallel ( perpendicular ) to the interface . in fese / sto ( and also fese / bto ) one expects @xmath20 @xcite , since the fese film can provide additional metallic screening parallel to the interface . in this limit , @xmath21 is small , resulting in an interaction that is peaked at @xmath22 . previously , some of the present authors showed that a strong forward - focused interaction not only reproduces the replica band structure but also acts as a remarkably effective pairing mediator with a superconducting @xmath23 in the weak coupling limit @xcite . as such , the cross - interface @xmath1-@xmath2 coupling implied by the replica bands can provide a substantial contribution to @xmath5 and this can work in conjunction with an unconventional pairing mechanism . this outlook is supported by recent projector quantum monte carlo results @xcite and multiband bcs results involving the incipient hole bands at the @xmath24-point @xcite . it is therefore important to understand and confirm the exact role played by the cross - interface coupling . motivated by this , we build upon our previous work @xcite and present several aspects of a 2d system of electrons coupled to an optical phonon branch by forward focused interaction . our aim is to draw a more complete picture of the consequences of such interaction and to provide further spectroscopic signatures that can be used to confirm its presence . we note that while our focus is on the fese / sto system , similar forward scattering may occur in other low - dimensional correlated systems @xcite . we hope that our results will provide additional means to confirm or rule out this possibility . the paper is organized as follows . section [ sec : formalism ] provides the details of the model and migdal - eliashberg formalism used throughout the paper . section [ sec : migdal ] discusses the validity of this approach in fese / sto , which has been questioned recently @xcite . our major results are presented in section [ sec : results ] , beginning in section [ sec : akw ] , where we review our prior results for the formation of the replica bands and present new details . in section [ sec : gap ] we then examine the properties of the superconducting state and the anisotropy of the superconducting order parameter . finally , in sections [ sec : dos ] and [ sec : qpi ] we present results for electronic density of states and quasiparticle interference patterns , respectively . finally , section [ sec : conclusions ] provides additional discussion and concluding remarks . we consider a simplified single - band model for fese electrons coupled to an optical phonon branch via a strongly momentum - dependent interaction . the system is described by the hamiltonian @xmath25 where @xmath26 ( @xmath27 ) creates ( annihilates ) an electron with wavevector @xmath14 and spin @xmath28 , @xmath29 ( @xmath30 ) creates ( annihilates ) a phonon with wavevector @xmath11 ; @xmath31 is the electronic band dispersion measured relative to the chemical potential @xmath32 ; @xmath33 is the phonon dispersion ( @xmath34 ) ; and @xmath35 is the momentum dependent @xmath1-@xmath2 interaction . the electronic band dispersion has a simple form @xmath36 - \mu$ ] , where @xmath37 is the in - plane lattice constant . as we are primarily motivated by the case of fese / sto , we set @xmath38 mev and @xmath39 mev unless otherwise stated . this parameter set produces a single electron - like fermi surface with @xmath40 and a fermi velocity @xmath41 ev@xmath42 along the @xmath43 line , which is similar to the electron pocket at the @xmath44 point in fese monolayers on sto @xcite . we approximate the relevant phonon branch with a dispersionless @xmath45 mev optical oxygen phonon mode . this is supported by several _ ab initio _ calculations that indicate that the highest lying phonon branch in bulk sto @xcite and fese / sto interfaces @xcite is relatively flat in the vicinity of @xmath22 . the @xmath1-@xmath2 matrix element is taken to depend on the momentum transfer only , as derived microscopically in previous works @xcite . specifically , we parameterize the coupling as @xmath46 , where @xmath47 is adjusted to fix the total dimensionless coupling strength of the interaction ( see section [ sec : lambda ] ) and @xmath21 sets the range of the interaction in momentum space . previously , we showed that @xmath21 must be small in order to reproduce the spectral properties of the replica band @xcite ( see also fig . [ fig : akwlmd ] ) . this implies that the @xmath1-@xmath2 coupling is strongly peaked in the forward scattering direction ( small @xmath48 ) . we study the model hamiltonian using migdal - eliashberg ( me ) theory . ( the validity of this approach for the fese / sto system is discussed further in section [ sec : migdal ] . ) with the nambu notation the electronic self - energy in the superconducting state and on the matsubara frequency axis can be partitioned as @xmath49\hat{\tau}_0 + \chi({{\bf k}},i\omega_n)\hat{\tau}_3 \nonumber\\ & + \phi({{\bf k}},i\omega_n)\hat{\tau}_1,\end{aligned}\ ] ] where @xmath50 are the pauli matrices , @xmath51 is the quasiparticle weight , @xmath52 renormalizes the band dispersion , and @xmath53 is the anomalous self - energy ( which is zero in the normal state ) . in me theory , the self - energy is obtained by summing the single - loop diagrams shown in fig . [ fig : diagrams](a ) . this leads to a self - consistency equation @xmath54 here , @xmath55 and @xmath56 are the dressed electron and bare phonon green s functions , respectively ; @xmath57 and @xmath58 are fermionic and bosonic matsubara frequencies , respectively ; and @xmath59 ( @xmath60 ) is the inverse temperature . the self - energy on the real frequency axis is obtained from the analytic continuation of the imaginary axis solutions with @xmath61 . we carry out this procedure using the efficient two - step iterative procedure given in ref . @xcite , where the self - energy is obtained from self - consistently solving @xmath62.\end{aligned}\ ] ] in this case , the @xmath1-@xmath2 coupling constant depends on the momenutum transfer @xmath11 only . the momentum summations can therefore be evaluated efficiently using fast fourier transforms . the discrepancy of the average value of mass enhancement @xmath63 and the conventional definition for the dimensionless coupling @xmath64 obtained from our numerical calculations at @xmath65 k. these results are obtained on a @xmath66 site lattice and with gaussian broadening of 15 mev . the dashed line indicates the line for @xmath67 as would be expected for a momentum - independent coupling . ] it is common practice to define a dimensionless measure of the interaction strength @xmath63 when studying electron - boson coupled systems . physically , this value characterizes the average value of the quasiparticle mass renormalization on the fermi surface due to the interaction . it is defined as @xmath68 , where @xmath69.\ ] ] an alternative measure of this coupling , denoted @xmath64 , is commonly employed in the literature . it is given by a double fermi - surface - average of the @xmath1-@xmath2 coupling constant latexmath:[\[\label{eq : lambdawrong } \lambda \equiv \frac{2}{n_\mathrm{f}\omega n^2 } \sum_{{{\bf k}},{{\bf k}}^\prime } in the context of this work it is important to stress that these two definitions are _ not _ equivalent ; only when the momentum dependence of @xmath35 is weak and @xmath71 does @xmath72 @xcite . this is illustrated for our forward scattering interaction in fig . [ fig : lambda ] , which plots the physical @xmath63 vs. the common definition @xmath64 for various values of @xmath21 . one sees clearly that @xmath63 differs significantly from @xmath64 and the size of the discrepancy grows as the interaction becomes increasingly peaked at @xmath22 . this indicates that a strongly forward - focused @xmath1-@xmath2 interaction is less effective at dressing electrons , resulting in dressed carriers with lighter effective masses . this is one factor contributing to the large @xmath5 values obtained from this interaction . for the remainder of this work we use @xmath63 to characterize the strength of the @xmath1-@xmath2 coupling . in me theory the self - energy is computed by summing the diagrams shown in fig . [ fig : diagrams](a ) , while the crossing diagrams like the one shown in fig . [ fig : diagrams](b ) [ involving vertex corrections like the one shown in fig . [ fig : diagrams](c ) ] are neglected . justification for this approximation is typically provided by migdal s theorem , which states that the vertex corrections in a metallic system with a wide bandwidth are proportional to the product of the dimensionless coupling @xmath63 times the adiabatic ratio @xmath73 @xcite . here , @xmath74 parameterizes the typical energy of the relevant phonons and @xmath75 is the fermi energy of the electron . in a metallic system the lowest order vertex correction is proportional to @xmath76 , where @xmath77 and @xmath44 are the electron and ion masses , respectively . thus the self - energy corrections due to the crossing diagrams are much smaller than the rainbow diagrams and can be neglected at all orders . the validity of this approach has been questioned @xcite in fese / sto , where @xmath78 mev and @xmath79 mev . it turns out , however , that the strong forward - scattering peak in the interaction removes the adiabatic ratio from the prefactor of the vertex correction . as a result , migdal s theorem has the potential to break down in _ any _ system dominated by small @xmath11 scattering , regardless of the value of @xmath80 . to demonstrate this , we consider the lowest order vertex correction @xmath81 [ fig . [ fig : diagrams](c ) ] in the limit of perfect forward scattering , where @xmath82@xcite . in this limit , no momentum is transferred to the electron and we can set the external momenta to be equal @xmath83 . the vertex correction is then given by @xmath84 \left[i(\omega_{n}-\omega_m)-\xi_{{\bf k}}\right]}.\end{aligned}\ ] ] since we are interested in the behavior of electrons near the fermi level , we set @xmath85 ( @xmath86 ) . the summation over the internal matsubara frequencies can now be performed exactly , yielding @xmath87 . \end{aligned}\ ] ] here , we have reintroduced the dimensionless coupling @xmath88 @xcite . for inelastic scattering @xmath89 , the vertex is bounded by @xmath90 since @xmath91 in the relevant regime @xmath92 , we conclude that the vertex correction is bounded by the dimensionless coupling strength @xmath63 . note that this expression does not include the ratio @xmath93 as one would naively expect from migdal s theorem . for elastic scattering @xmath94 , the first order correction diverges at low temperatures . ( here we are refering to the vertex function for elastic scattering from phonons , which is different from the self - energy effects created by elastic forward scattering scattering from off - plane impurities that has been discussed in the context of the cuprates @xcite . ) this is particularly troublesome , as it leads to a @xmath95 contribution to the electronic self - energy when one includes the first crossing diagram ( fig . [ fig : diagrams]b ) . ( see also the appendix . ) fortunately , this divergence also occurs at higher orders in the self - energy expansion and in the limit of perfect forward scattering we sum the elastic vertex corrections in the ladder approximation ( fig . [ fig : diagrams]d ) . we obtain @xmath96 ^ 2 \gamma_{{{\bf k}}_\mathrm{f } } ( \omega_n - \omega_m ) \\ = 1 - g_0 ^ 2 t \sum_m \frac{2 \omega}{\omega^2 + \omega_m^2 } \frac{1}{(\omega_n - \omega_m)^2 } \gamma_{{{\bf k}}_\mathrm{f } } ( \omega_n - \omega_m).\end{aligned}\ ] ] by performing the matsubara summation we observe that the dominant term in the low - temperature expansion originates from the residue of the double pole at @xmath97 . we find @xmath98 it only remains to fix @xmath99 , and so the full elastic vertex function is given by @xmath100 this final result shows that the divergence has disappeared in the summation of the diagrams , and the elastic vertex function is bounded between @xmath101 . we conclude that for strong forward scattering , we need to make a distinction between elastic and inelastic scattering . for inelastic scattering , the vertex corrections are bounded by the dimensionless coupling @xmath102 , which is already the case for the first order correction . for elastic scattering , we need to sum up to infinite order in perturbation theory . the resulting elastic vertex function is bounded by @xmath103 . furthermore , at low temperatures , the vertex in eq . ( 7 ) tends towards zero for @xmath104 near the fermi level . we therefore conclude that the elastic contribution to the vertex does not play a strong role in modifying the self - energy due to the interaction with the optical phonon . ( a ) a false colour image of @xmath105 at @xmath106 k for coupling to a 100 mev optical mode with @xmath107 and @xmath108 . ( b ) a waterfall stack of the data shown in ( a ) . ] there are two important lessons to be drawn from this analysis . first , and most important , the adiabatic ratio @xmath93 does not appear as a prefactor to @xmath109 . therefore , the standard measure for validity of migdal s theorem does not apply to the case of strong forward scattering , regardless of the location of the band minimum with respect to the phonon energy . this , however , leads to the second observation . namely , the vertex correction is bounded by a term proportional to the dimensionless coupling @xmath63 . therefore , the contributions to self - energy due to these corrections will be of order @xmath110 and higher . in our previous work we showed that @xmath111 is sufficient to account for the observed replica bands in the arpes spectra . as such , the parameter regime relevant to the fese / sto system is in the weak - coupling limit , where the vertex corrections are expected to be small and can be treated perturbatively . in this context , it is worth noting that this perturbative limit was examined extensively in refs . @xcite and @xcite , where it was found that the vertex corrections have a complicated effect on the superconducting @xmath5 , which depends on the momentum structure of the interaction . in the perturbative regime , @xmath5 was found to be suppressed when @xmath112 is dominated by large @xmath11 processes while it was enhanced when @xmath112 is dominated by small @xmath11 . the latter case is the one considered here . due to these considerations , we proceed by summing the diagrams traditionally considered by migdal - eliashberg theory and neglect for the time being the vertex corrections . in this case we are required to sum the rainbow diagrams in order to obtain the superconducting state ; however , we note that the majority of our spectral features are determined by the first order rainbow diagram , which supports the notion that the @xmath1-@xmath2 spectral features can be captured in the perturbative regime . we begin by presenting details of the spectral properties of our model , which illustrates the formation of the replica bands . this section will also serve to briefly review some of our results from ref [ fig : akw ] shows @xmath105 for @xmath113 and @xmath114 . results are shown for @xmath106 k @xmath115 k , and the superconducting gap is apparent in the main band crossing @xmath80 . two replica bands are also present , with each offset by approximately @xmath74 and @xmath116 , respectively , from the main band . here , the forward - focused @xmath1-@xmath2 interaction produces replicas that are complete copies of the main band ; each replica terminates at @xmath117 , and has the characteristic back bending associated with the bogliubov quasiparticle dispersion . the results shown in fig . [ fig : akw ] are consistent with arpes data of lee _ et al . _ this points to the @xmath22 nature of the coupling in fese / sto @xcite . if the @xmath1-@xmath2 interaction is broadened ( increasing @xmath21 ) then the replica band is significantly smeared in momentum space , as shown in the lower row of fig . [ fig : akwlmd ] . for @xmath118 the replica nature of the shake - off band is maintained . for @xmath119 , however , the shake - off band is significantly broadened and no longer follows the same curvature of the main band . this confirms that @xmath21 must be small in order to reproduce the experimentally observed clear and complete copies of the main band . the total strength of the interaction can also be extracted from the data by examining the relative spectral weight of the two bands . we define the ratio @xmath120 as the ratio of the maximum intensity in the replica band relative to the main band . the top row of fig . [ fig : akwlmd ] shows the spectral function at @xmath121 for several values of the dimensionless coupling , which allows us to extract this ratio from our data as a function of @xmath63 . the results are summarized in the insets of that figure . a conservative estimate of @xmath122 was obtained from the experimental data in ref . we can compare this to our results , where we find extract @xmath123 for @xmath124 and @xmath125 for @xmath107 . these values confirm that the cross - interface coupling in fese / sto is in the perturbative regime . the interface effects leading to the strong forward scattering peak in the @xmath1-@xmath2 interaction do not require a shallow electron band as realized in fese / sto . it is therefore interesting to examine how this interaction will manifest itself in a system where the main band crosses the phonon energy . in fig . [ fig : bandmin ] we explored such a scenario by adjusting the chemical potential of our model such that the band minimum moves from above the phonon energy ( left side ) to below the phonon energy ( right side ) . results are shown for @xmath124 ( top row ) , @xmath107 ( middle row ) , and @xmath119 ( bottom row ) . in all cases , the replica bands remain located below the main band , however , as the main band begins to cross the phonon energy scale , we see the formation of the traditional kink features commonly seen in the high-@xmath5 cuprates @xcite . interestingly , the kink structure of the main band is also copied completely in the replica band when @xmath21 is sufficiently small . we now exam the superconducting state mediated by the forward - focused @xmath1-@xmath2 interaction . our aim here is not to advocate for a purely phonon mediated pairing interaction in fese / sto at this time . rather , we wish to examine the effects of an @xmath1-@xmath2 interaction with a strong forward scattering peak in order to understand the properties of the superconducting state formed by such an interaction . we will discuss how our results fit into the larger context of the fese / sto discussion in section [ sec : conclusions ] . the superconducting gap function at the @xmath126 matsubara frequency and on the fermi surface @xmath127 is plotted in fig . [ fig : gap](a ) as a function of temperature for several values of @xmath21 . only the maximal value on the fermi surface is plotted here , since the gap function has @xmath128-wave symmetry and is quite isotropic for this set of parameters . here , results are shown for a series of increasing @xmath21 values , while the total coupling is fixed at @xmath108 . for the broadest interaction ( @xmath119 ) we obtain a relatively modest @xmath129 k , which is larger than the @xmath130 k one would estimate from the weak coupling bcs result . as the interaction becomes more focused around @xmath22 , @xmath5 increases rapidly , with @xmath131 k for @xmath124 ( see inset ) . this result demonstrates the high-@xmath5 s that can be produced by a strong forward scattering @xmath1-@xmath2 interaction and are consistent with our prior results @xcite , as well as those obtained from simplified bcs treatments @xcite . we obtain @xmath132 ratios of @xmath133 , @xmath134 , @xmath135 , and @xmath136 for @xmath119 , @xmath137 , @xmath138 , and @xmath139 , respectively , indicating that the deviation from the weak coupling bcs value increases as the interaction becomes increasingly peaked . ( a ) the temperature dependence of the superconducting gap at @xmath117 as a function of temperature for increasingly forward - focused interactions . different symbols indicate different values of @xmath21 used , as indicated by the same legend of ( b ) . ( b ) the gap value at low temperature ( @xmath140 k ) vs. the total coupling strength . the dash - dotted line shows zero temperature gap value expected from the perfect forward scattering . the dashed line mark the value of the @xmath63 used in ( a ) . ] the magnitude of the superconducting gap at @xmath141 k as a function of @xmath63 is plotted in fig . [ fig : gap](b ) for several values of @xmath21 . for a perfect forward scattering ( @xmath142 ) in the weak coupling limit , the analytic expression of the zero temperature gap on the fermi surface is @xmath143 while @xmath144 @xcite . for a strong forward scattering @xmath124 we see a linear dependence on @xmath63 at weak coupling , in agreement with the result @xmath23 for the same limit . as the value of @xmath21 is increased , however , there is a crossover to a more exponential - like dependence on @xmath63 , which reflects the change from a @xmath145 behavior in the strong forward scattering limit to a @xmath146 behavior for a more uniform interaction . ( a ) the momentum dependence of the superconducting gap function at the first matsubara frequency . the results are plotted along the high symmetry directions @xmath147 @xmath148 ( left ) and @xmath147 @xmath149 ( right ) . ( b ) the spectral function obtained for @xmath124 at @xmath150 ( solid black ) and @xmath151 k ( dashed red ) . results are shown for @xmath152 , @xmath153 , and @xmath154 when read from the bottom to the top of panel ( b ) . ] another interesting aspect is the momentum structure of @xmath155 , which is shown in fig . [ fig : deltak ] . for small @xmath21 the superconducting gap has maximum values on the fermi surface but it is rapidly suppressed for @xmath14 away from @xmath117 . this is in contrast to the case of a momentum independent interaction , where @xmath156 is independent of momentum . as @xmath21 increases ( with @xmath63 fixed ) , the total magnitude of the gap decreases and broadens in momentum space ; however , the maximum remains on the fermi surface . this behavior can be easily understood in the limit of perfect forward scattering , where the vertex reduces to @xmath157 with @xmath158 @xcite . in the weak coupling limit ( @xmath159 , @xmath160 , @xmath161 ) , the gap equation then reduces to @xmath162 this modified gap equation shows that @xmath156 depends most sensitively on the electronic structure at @xmath14 in the limit of strong forward scattering . hence the value of the gap at @xmath163 has decoupled from the values at the other momenta . as discussed in our previous work @xcite , this momentum decoupling " @xcite is the ultimate cause of the large @xmath5 values produced by this type of interaction . it is also the origin of the momentum structure shown in fig . [ fig : deltak ] , as one can immediately see that @xmath164 will be suppressed as @xmath31 deviates from the fermi surface . this is confirmed by the numerical calculations with finite width @xmath21 of the interaction shown in fig . [ fig : deltak](a ) , which also show that this effect becomes less pronounced as the width of the interaction is made more uniform . the momentum structure of @xmath164 implies that only the dispersion of states near @xmath80 will be affected by the transition to the superconducting state . this is illustrated in fig . [ fig : deltak](b ) , where we show the spectral function @xmath165 at @xmath152 , @xmath153 , and @xmath154 in the normal ( @xmath65 k , dashed red ) and superconducting ( @xmath150 k , solid black ) states . the peaks of the main and replica bands at @xmath117 shift with the opening of the superconducting gap . in contrast , the two spectral functions away from @xmath117 are hardly affected in the superconducting state , since the gap function decays to zero very fast away from the fermi surface . thus , the lack of shift at the band minimum is a characteristic of a superconducting state mediated by forward scattering bosons . the absence of this shift at the band minimum may be discernible from the case of a uniform gap if the band is shallow relative to the size of the gap [ @xmath166 . this condition would be met in fese / sto , if the observed @xmath167 mev gap @xcite is momentum independent throughout the whole first brillioun zone . ( solid line ) and local dos @xmath168 ( dashed line ) for chemical potential ( a ) @xmath169 mev and ( b ) @xmath170 mev . the insets show the corresponding fermi surfaces and the definition of the angle around the fermi surface . ] at @xmath140 k along the path in the @xmath171 ( solid line ) and @xmath172 directions ( dashed line ) for chemical potentials ( a ) @xmath169 mev and ( b ) @xmath170 mev . the insets show the ratio of the maximal gap in the @xmath171 and @xmath172 directions . the colour of lines indicates different values of @xmath21 as indicated the insets . in all cases @xmath173 . ] another consequence of the momentum decoupling in the gap equation is that a significant anisotropy in @xmath174 can occur , which is determined by the anisotropy of the fermi velocity or local dos in @xmath14-space @xcite . to illustrate this point , we calculated the gap function for our model with @xmath175 mev such that @xmath80 lies in close proximity to the van hove singularity in the band structure . in this case , the local dos at fermi level , defined as @xmath176 , is extremely anisotropic , as shown in fig . [ fig : vfndos](a ) . ( @xmath177 defines the angle around the fermi surface measured from the positive @xmath178-axis and @xmath179 is the momentum in the direction tangential to the fermi surface . ) for comparison , @xmath168 and @xmath180 for @xmath170 mev are shown in fig . [ fig : vfndos](b ) , where each has a much weaker variation around the fermi surface . the corresponding gap functions @xmath181 at @xmath182 k along the @xmath171 ( solid lines ) and @xmath172 ( dashed lines ) directions are plotted in fig . [ fig : gap0kpath ] . results are shown for several values of @xmath21 , as indicated by the colour of the lines . the insets of fig . [ fig : gap0kpath ] show the ratio of the gap maximum along the @xmath171 ( @xmath183 ) and @xmath172 ( @xmath184 ) directions , which defines a measure of the gap anisotropy . in the @xmath169 mev case , we see a non - trivial evolution of the gap anisotropy as a function of @xmath21 . for large @xmath21 ( red ) , i.e. closer to uniform scattering , @xmath164 is somewhat anisotropic with the maximum value along ( @xmath185 ) being @xmath186 times the maximum along the diagonal . these values are inversely proportional to the fermi velocity , with @xmath187 , which can be inferred from the gap equation in the perfect forward scattering limit . as the value of @xmath21 initially decreases , the gap anisotropy becomes more pronounced until a maximum is reached for @xmath188 . this reflects the momentum decoupling effect seen in ref . further decreases in @xmath21 , however , result in a more uniform gap . this can be traced to an increase in the gap magnitude along the zone diagonal relative to the gap along @xmath171 . we speculate that the decreasing anisotropy in this region is related to an increase in the phase stiffness of the condensate when @xmath21 is very small and larger total gap values are obtained . these results are in contrast to those obtained when @xmath39 mev and the anisotropy in the band structure is much less pronounced . in this case , the resulting gap function is relatively isotropic with at most a @xmath189 variation occurring when @xmath107 . in both cases the gap has an @xmath128-wave symmetry , as can be expected for an attractive interaction . figs . [ fig : vfndos ] and [ fig : gap0kpath ] demonstrate that a strong forward scattering can produce an anisotropic gap if the variations in the fermi velocity are significant . in such a case the gap magnitude is inversely related to the fermi velocity . early arpes measurements on fese / sto found that the superconducting gap was fairly isotropic @xcite ; however , a more recent arpes study has observed a @xmath190 variation in @xmath156 around the electron pockets @xcite . this is larger than we infer from our single - band model , but our calculation neglects the nearly degenerate electron pockets , their relative orbital character , and any hybridization effects between them , and the anisotropy from an unconventional pairing mechanism that may be working in conjunction with the forward scattering . it would be interesting to carry out our calculation for a more realistic band structure that captures these aspects . the last topic that we wish to discuss in this section is the relevance for forward scattering to pairing channels that are traditionally considered in the context of purely electronic , repulsive interactions . for any momentum dependent interaction , one can define a projection of the coupling in the @xmath191-pairing channel @xcite @xmath192 where @xmath193 for an @xmath128-wave gap , @xmath194 for an @xmath195-wave gap , @xmath196/{2}$ ] for a @xmath197 gap , and so forth . if @xmath198 then the coupling is attractive in that channel . for a single - band system with strong forward scattering one can convince oneself that @xmath199 is positive for many @xmath191 since @xmath200 and @xmath201 have a better chance to have the same sign if @xmath14 is close to @xmath202 when @xmath203 is large ( this is somewhat sensitive to shape and size of the fermi surface ) while @xmath204 is the largest . as a result , strong forward scattering on its own will realize an @xmath128-wave gap but has non - zero contributions in other pairing channels . therefore , if a dominant pairing interaction arising from repulsive interactions is present , then the forward scattering @xmath1-@xmath2 interaction can contribute to the resulting unconventional superconducting state in most instances . these considerations also show that if the @xmath128-wave channel is blocked by a repulsive interaction in a singleband system , then the forward scattering will produce a gap symmetry in the channel with the next leading value of @xmath205 . this was pointed out in ref . @xcite , where it was discussed in the context of forward scattering in the cuprates using a simplified eliashberg treatment . we demonstrate this here for our model by introducing the morel - anderson coulomb pseudopotential @xmath206 , which suppresses the @xmath128-wave pairing tendency . [ fig : gapfwvdwv ] shows the resulting gap function for two different choices in parameters and @xmath207 . in this case the @xmath128-wave channel is effectively blocked by @xmath208 and a @xmath10-wave or @xmath209-wave gap is realized , depending on our parameter choice . we stress that these solutions were not imposed during our self - consistent calculations . rather , they were found naturally during the iterative procedure . the reader should also note that the @xmath5 is not strongly suppressed by the coulomb pseudopotential @xcite . left : @xmath209-wave gap @xmath210 for @xmath211 . right : @xmath10-wave gap @xmath210 for @xmath212 . @xmath213 , @xmath214 for @xmath211 , and @xmath215 for @xmath212 are used in the calculation . the green line indicates the fermi surface . ] in a boson - mediated superconductor , the electron - boson interaction produces fine - structure in the single - particle density of states ( dos ) @xmath216 . the energy at which these structures appear is determined by boson energy shifted by the maximum value of the superconducting gap @xmath217 @xcite . in addition , these structures can appear either as shoulders or as local minima in @xmath216 , depending on the gap symmetry and momentum dependence of the interaction @xcite . as such , the renormalized electronic structure carries valuable information regarding the pairing mediator . motivated by this , we calculated @xmath218 for our model in the case of a strongly forward focused interaction . the result is shown in fig . [ fig : dos ] , where the model parameters are @xmath219 mev , @xmath124 , and @xmath220 . these results were obtained on an @xmath221 @xmath222-grid and at a temperature of @xmath150 k , well below @xmath5 . the dos shows clear coherence peaks associated with the fully gapped @xmath128-wave state , which are broadened by a @xmath223 mev contribution that was added to the imaginary part of @xmath224 . the dos also has several additional structures arising from the replica bands . the insets of fig . [ fig : dos ] zoom in on the relevant energy ranges . in this case , the structures are different from those expected for a conventional @xmath1-@xmath2 mediated superconductor with an @xmath128-wave gap . in the conventional case , the boson renormalizations form shoulders in @xmath216 , where the dos is enhanced ( suppressed ) for energies below ( above ) @xmath225 , resulting in a characteristic minima in @xmath226 @xcite . in the forward scattering case , however , we see step - like features located at approximately @xmath74 above and below @xmath80 , which correspond to the onset of the spectral weight of the replica bands . this results in a local maxima in @xmath226 at the energy @xmath74 , occurring for both for the unoccupied and occupied dos . this difference provides another structure in the electronic structure that might be used for identifying replica bands in real materials . the density of states @xmath216 for @xmath124 , @xmath220 , and @xmath219 mev , calculated at @xmath150 k. a small @xmath227 mev contribution was also added to the imaginary part of the single - particle self - energy in order to broaden the spectra at this @xmath14-resolution . the two insets on the left zoom in on the spectra in the regions where the boson - induced features are found and the inset on the right shows @xmath228.,scaledwidth=45.0% ] fourier transform scanning tunneling spectroscopy , also commonly known as quasi - particle interference ( qpi ) , has become a popular method for studying the electronic structure and gap symmetry of unconventional superconductors @xcite . this technique exploits the fact that the incoming and outgoing wavefunction of an electron scattered from an impurity will interfere with each other to form ripples in the charge density . the power spectrum of the resulting density modulations measured at a bias voltage @xmath229 will have peaks in its intensity at wave vectors @xmath230 , which are related to spanning vectors across the constant energy contours of the band @xmath231 . by tracking the dispersion of the @xmath230 peaks , one can infer information about the underlying electronic band structure . one of the advantages of this technique over arpes is that it can access unoccupied states @xcite . this technique can therefore probe both the occupied and unoccupied replica bands produced by a strong forward scattering mode . motivated by this , we calculated the qpi spectrum for our model . our qpi calculations are carried out using the standard @xmath232-matrix formalism for a single - band superconductor , as outlined in a number of references @xcite . in the presence of an impurity , the fourier transform ( ft ) of the local electron density ( ft - ldos ) is partitioned as @xmath233 , where @xmath234 is the ft - ldos for the homogeneous system and @xmath235 is the ft of the induced density modulations due to the impurity . the latter part is given by @xmath236 , where @xmath237.\end{aligned}\ ] ] here , @xmath238 is the green s function in the presence of the impurity ; @xmath239 is the @xmath232-matrix ; and @xmath240 is the bare green s function in the absence of the impurity . for our purposes in this section , @xmath240 includes the self - energy due to the @xmath1-@xmath2 interaction . the assumption here is that the impurities and @xmath1-@xmath2 interaction do not significantly influence one another . for simplicity , we treat the impurity as a weak non - magnetic point - like potential scatter located at the origin . in this limit the @xmath232-matrix is momentum independent and is given by @xmath241^{-1},\ ] ] where @xmath242 and @xmath243 characterizes the strength of the impurity potential . for our calculations we took @xmath244 mev ; however , our results are not too sensitive to this value . our qpi results are summarized in figs . [ fig : qpi ] and [ fig : qpi_dispersion ] . the spectral function for our bare " electronic structure , including the effects of the @xmath1-@xmath2 interaction , is shown in fig . [ fig : qpi](a ) for reference . the relevant model parameters are given in the figure caption . constant energy cuts of the spectral function are shown figs . [ fig : qpi](b1)(b4 ) for energies located inside the occupied replica band [ @xmath245 mev , panel ( b1 ) ] , inside the occupied states of the main band [ @xmath246 mev , panel ( b2 ) ] , inside the unoccupied states of the main band [ @xmath247 mev , panel ( b3 ) ] , and inside the unoccupied replica band [ @xmath248 mev , panel ( b4 ) ] . in fig . [ fig : qpi](b1)(b3 ) one can see a clear ring - like feature corresponding to the single replica or main band that is intersected at these energies ; however , in fig . [ fig : qpi](b4 ) , a double ring structure is observed where the inner ring corresponds to the replica band and the outer ring corresponds to the main band . this electronic structure is reflected in the corresponding qpi spectra , shown in fig . [ fig : qpi](c1)(c4 ) , respectively . the maps in figs . [ fig : qpi](c1)(c3 ) have a clear ring - like structure with a radius @xmath249 , where @xmath250 is a vector spanning the constant energy contours . these features resemble what one would expect for a simple single band system @xcite . at @xmath251 mev , however , the qpi pattern develops a double ring structure , as one would might expect from the topology of the main and replica bands shown in fig . [ fig : qpi](b4 ) . based on the geometry of the features we can assign the outer qpi peak to intraband scattering within the main band and inner qpi peak to interband scattering between the unoccupied states of the replica and main bands . we also see a very weak feature due to intraband scattering within the replica band , however , it is very difficult to resolve in this image . the intensity of these features also follows the hierarchy of spectral weight associated with the type of band . the relative intensity of these features are also shown in fig . [ fig : qpi_dispersion ] . these results establish that the replica bands can be observed in the qpi spectra and therefore ft - sts could in principle study these features , particularly on the unoccupied side of @xmath80 . in this case , the qpi peaks due to the replica bands will appear similar to regular bands at a given bias voltage albeit with a reduced intensity ; however , the dispersion and spectral weight of these features will be quite different . to illustrate this , we plot the qpi intensity along the @xmath252 line as a function of energy in order to track the dispersion of the qpi peaks . [ fig : qpi_dispersion](a1 ) presents this information as a false colour image while fig . [ fig : qpi_dispersion](a2 ) presents it as a contour plot in order to highlight the features with weaker intensity . if a qpi peak is due to intraband scattering in the main band , then it will disperse towards @xmath22 as the bias voltage is tuned towards the top or bottom of the band @xcite . this can be clearly seen in fig . [ fig : qpi_dispersion](a1 ) from the most intense feature . in contrast , the dispersion of a peak due to inter or intraband scattering to a replica band will be different due to the extinction of the replica band s spectral weight at @xmath117 . this is apparent in fig . [ fig : qpi_dispersion](a1 ) and fig . [ fig : qpi_dispersion](a2 ) , where the intensities of the interband scattering feature at positive energies and the intraband scattering feature at negative energies sudden disappears at the energy where the replica band disappears . this is another unique spectral fingerprint of a replica band . while there have been several qpi studies conducted on the fese / sto system @xcite , to the best of our knowledge no indications of the replica bands have been observed . one study by huang _ et al . _ ( ref . @xcite ) found evidence for an electron - like and located 75 mev above @xmath80 ; however , the measurements of the tunneling decay length indicated that this band was centered at the @xmath24-point rather than the @xmath44-point . this rules out the possibility that this feature is due to the replica expected for the electron pockets at the @xmath44-point . however , we also note that nearly all of the qpi studies completed to date have focused on energy windows of less than 80 mev of the fermi level @xcite . this is outside the region where the replica band are expected , based on the phonon energy scale inferred by arpes @xcite . it would therefore be interesting to extend these studies to higher bias voltages in order to determine if the features shown in fig . [ fig : qpi_dispersion ] can be observed . this will be a challenging task , however , as our calculations have shown that the replica qpi peaks are much weaker in intensity than the main qpi peaks . we have examined several aspects of a system of electrons coupled to an optical phonon mode via a strongly momentum dependent interaction , which is peaked for small momentum transfers @xmath11 . this interaction is remarkably effective at mediating pairing , resulting in a superconducting state with a high-@xmath5 for very modest values of the total dimensionless coupling @xmath63 and a rich phenomenology . in the absence of a coulomb pseudopotential the superconducting state has an @xmath128-wave symmetry . but the gap function can be anisotropic when there are significant variations in the system s fermi velocity and the interaction is sufficiently peaked at @xmath22 . in addition , other gap symmetries including @xmath10- and @xmath209- wave symmetries can be realized when the @xmath128-wave channel is blocked by the coulomb interaction . this result highlights the property that a forward scattering interaction can be attractive in other pairing channels that are traditionally mediated by repulsive interactions . the forward - focused interaction leaves several distinct replica bands in the electronic structure , in agreement with those observed in two recent arpes studies @xcite here , we have further demonstrated that these features can also be found in the single - particle dos and qpi patterns probed by stm / sts . more importantly , the manifestation of these features in the dos is qualitatively different than the usual dos modulations induced by the conventional attractive @xmath1-@xmath2 interaction mediated by phonons . similarly , the intensity and dispersion of the qpi peaks due to scattering from the replica bands have a distinctive energy dependence . these facts allow the case of forward scattering to be differentiated from the case of a more conventional ( uniform ) @xmath1-@xmath2 interaction . these predictions can help future experiments to confirm or rule out the presence of such an interaction in fese / sto , fese / bto , or other novel interface systems . we close with some comments on the situation of fese films on oxide substrates . the forward scattering mechanism is very effective at producing high values of @xmath5 . moreover , the richness of the pairing symmetries we observed here further confirms that this interaction can also help mediate unconventional pairing channels . combined , these facts mean that the cross interface coupling with a strong @xmath22 peak can be an effective means to engineer superconductivity at higher transition temperatures in low - dimensional interfaces . as discussed in the introduction , and in ref . @xcite , the intercalated fese systems have band structures remarkably similar to fese / sto and a @xmath253 k. similarly , fese thin films post treated with k and na adatoms result in comparable values of @xmath5 . these factors suggest that an unconventional pairing mechanism is most likely present ; however , the maximal @xmath5 that such a pairing mechanism produces appears to be pinned below @xmath247 k. as argued in ref . @xcite , the cross interface coupling can provide the additional increase in superconductivity needed to reach the observed @xmath254 k. there is also the strong possibility that the incipient holelike band at the @xmath24 point contributes to the traditional @xmath195 pairing scenario due to the large energy scale of the spin fluctuations @xcite . in this case , the small @xmath11 scattering introduced by the phonons would be primarily intraband in nature and thus benificial to pairing . multichannel scenarios such as these may also shed light on the recent observation of two superconducting domes @xcite in k doped - fese / sto , a result that highlights the richness of the superconducting state in this system and the possibility of two pairing mechanisms . as we concluded in ref . @xcite , one obvious experiment that would potentially help sort out this situation is an @xmath255o isotope measurement for @xmath5 . if a joint pairing mechanism is realized then the @xmath256 isotope coefficient should be reduced by a significant amount @xcite . it should be noted , however , that vertex corrections to the @xmath1-@xmath2 interaction can also modify the expected isotope coefficient @xcite . we thank m. berciu , p. j. hirschfeld , a. linscheid , and d. j. scalapino for useful conversations . l. r. acknowledges funding from rubicon ( dutch science foundation ) . a portion of this research was conducted at the center for nanophase materials sciences , which is a doe office of science user facility . this manuscript has been authored by ut - battelle , llc under contract no . de - ac05 - 00or22725 with the u.s . department of energy . the united states government retains and the publisher , by accepting the article for publication , acknowledges that the united states government retains a non - exclusive , paid - up , irrevocable , world- wide license to publish or reproduce the published form of this manuscript , or allow others to do so , for united states government purposes . the department of energy will provide public access to these results of federally sponsored research in accordance with the doe public access plan ( http://energy.gov / downloads / doepublic- access - plan ) . cpu time was provided in part by resources supported by the university of tennessee and oak ridge national laboratory joint institute for computational sciences ( http://www.jics.utk.edu ) . in the perfect forward scattering limit , the contribution to the self - energy at the fermi level from the first crossing diagram ( fig . [ fig : diagrams]c ) is @xmath257 performing the matsubara sum yields @xmath258 . \label{crossingpfs}\end{aligned}\ ] ] for temperatures @xmath259 , the hyperbolic cosecant of @xmath260 approaches zero and the crossing diagram can be approximated as @xmath261 the above is completely contributed by the elastic term @xmath262 in the sum of internal matsubara frequencies @xmath263 . this shows that the diverging elastic vertex correction produces a self - energy contribution at the fermi level that diverges as @xmath264 . this can be understood from the structure of the crossing diagram in fig . [ fig : diagrams]b . whenever one of the phonons has a zero frequency , the total diagram is multiplied by the greens function @xmath265 . this contribution is usually suppressed by a factor @xmath232 ; however , here the vertex correction diverges as @xmath266 and we obtain a non - vanishing @xmath267 contribution . 10 f.- c. hsu _ et al . _ , proceedings of the national academy of sciences * 105 * , 14262 ( 2008 ) . q. l. wang _ et al . _ , * 29 * , 037402 ( 2012 ) . d. f. liu _ et al . _ , nature commun . * 3 * , 931 ( 2012 ) . k. liu , z .- y . lu , and t. xiang , phys . b * 85 * , 235123 ( 2012 ) . f. zheng , z. wang , w. kang , and p. zhang , scientific reports * 3 * , 2213 ( 2013 ) . _ et al_. , nature mater . * 12 * , 605 ( 2013 ) . _ , nature mater . * 12 * , 634640 ( 2013 ) . w. h. zhang _ et al . _ , b * 89 * , 060506(r ) ( 2014 ) . et al . _ , nature commun . * 5 * , 5044 ( 2014 ) . et al . _ , lett . * 112 * , 107001 ( 2014 ) . j. j. lee _ _ , nature * 515 * , 245 ( 2014 ) . b. li , z. w. xing , g. q. huang , and d. y. xing , journal of applied physics * 115 * , 193907 ( 2014 ) . _ , nature commun . * 5 * , 5047 ( 2014 ) . lee , chinese physics b * 11 * , 117405 ( 2015 ) . l. rademaker , y. wang , t. berlijn , and s. johnston , arxiv:1507.03967 ( 2015 ) . _ et al_. , nature mater . * 14 * , 285 ( 2015 ) . y. miyata , k. nakayama , k. sugawara , t. sato , and t. takahashi , nat . * 14 * 775 ( 2015 ) . s. coh , m. l. cohen , s. g. louie , new j. phys . * 17 * , 073027 ( 2015 ) . et al . _ , lett . * 115 * , 017002 ( 2015 ) . q. fan _ et al_. , nature phys . * 11 * , 946 ( 2015 ) . song , et al . , arxiv:1511.02007 ( 2015 ) . j. j. seo , _ et al . _ , arxiv:1511.07950 ( 2015 ) . j. shiogai , y. ito , t. mitsuhashi , t. nojima , and a. tsukazaki , arxiv:1510.00175 ( 2015 ) . et al_. , arxiv:1512.01949 ( 2015 ) . et al_. , arxiv:1512.01948 ( 2015 ) . y. zhang , j. j. lee , r. g. moore , w. li , m. yi , m. hashimoto , d. h. lu , t. p. devereaux , d .- h . lee , and z .- x . shen , _ superconducting gap anisotropy in monolayer fese thin film_. arxiv:1512.06322 ( 2015 ) . li , f. wang , h. yao , d .- h . lee , arxiv:1512.06179 ( 2015 ) . e. dagotto , rev . phys . * 85 * , 849 ( 2013 ) . j. g. guo _ et al . _ , b * 82 * , 180520 ( 2010 ) . et al . _ , arxiv:1505.06361 ( 2015 ) . i. mazin , d.j . singh , m.d . johannes , and m .- h . du , phys . * 101 * , 057003 ( 2008 ) . p. j. hirschfeld , m. m. korshunov , i. i. mazin , rep . . phys . * 74 * , 124508 ( 2011 ) . m. burrard - lucas _ et al . _ , nature mater . * 12 * , 15 ( 2013 ) . g. varelogiannis , a. perali , e. cappelluti , and l. pietronero , phys . b * 54 * , r6877 ( 1996 ) . m. l. kuli and o. v. dolgov , phys . stat ( b ) * 242 * , 151 ( 2005 ) and references therein . m. l. kuli and r. zeyher , phys . b * 49 * , 4395(r ) ( 1994 ) . a. aperis , p. kotetes , g. varelogiannis , and p. m. oppeneer , phys . b * 83 * , 092505 ( 2011 ) . n. bulut and d. j. scalapino , phys . b * 54 * , 14971 ( 1996 ) . x. chen , a. linscheid , s. maiti , and p. j. hirschfeld , phys . b * 92 * , 224514 ( 2015 ) ; l. p. gorkov , arxiv:1510.03327 ( 2015 ) . n. choudhury , e. j. walter , a. i kolesnikov , and c .- k . loong , phys . b * 77 * , 134111 ( 2008 ) . b. li , z. w. xing , g. q. huang , and d. y. xing , journal of applied physics * 115 * , 193907 ( 2014 ) . y. wang , a. linscheid , t. berlijn , and s. johnston ( to be published ) . f. marsiglio , m. schossmann , and j. p. carbotte , phys . b * 37 * , 4965 ( 1988 ) . g. d. mahan , _ many - particle physics_. 3rd ed . ( 2000 ) . a. b. migdal , sov . jetp * 34 * , 996 ( 1958 ) . l. zhu , p. j. hirschfeld , and d. j. scalapino , phys . b * 70 * , 214503 ( 2004 ) . s. h. hong _ et al_. , phys . rev . lett . * 113 * , 057001 ( 2014 ) . c. grimaldi , l. pietronero , and s. strssler , phys . b * 52 * , 10530 ( 1995 ) . l. pietronero , s. strssler , and c. grimaldi , phys . b * 52 * , 10516 ( 1995 ) . for recent reviews , see t. cuk _ et al_. , phys . ( b ) * 242 * , 11 ( 2005 ) ; j.p . carbotte , t. timusk , and j. hwang , rep . 74 , 066501 ( 2011 ) . o. v. danylenko , o. v. dolgov , m. l. kuli , and v. oudovenko , eur . j. b * 9 * , 201 ( 1999 ) . s. johnston and t. p. devereaux , phys . b * 81 * , 214512 ( 2010 ) . g. santi , t. jarlborg , m. peter , and m. weger , physica c * 259 * , 253 ( 1996 ) . w. mcmillan and j. rowell , in _ superconductivity _ , edited by r. d. parks ( dekker , new york , 1969 ) vol . 1 . g. l. de castro , c. berthod , a. piriou , e. giannini , and . fischer , phys . * 101 * , 267004 ( 2008 ) . j. e. hoffman , _ et al_. , science * 297 * , 1148 ( 2002 ) . et al . _ , science * 327 * , 181 ( 2010 ) s. chi _ et al_. , phys . lett . * 109 * , 078002 ( 2012 ) . m. p. allan , _ et al . _ , science * 336 * , 563 ( 2012 ) . et al . _ , . lett . * 108 * , 127001 ( 2012 ) . p. j. hirschfeld , d. altenfeld , i. eremin , i. i. mazin , phys . b * 92 * , 184513 ( 2015 ) . q wang and d .- h . lee , phys . b * 67 * , 020511 ( 2003 ) . s. grothe , s. johnston , s. chi , p. dosanjh , s. a. burke , and y. pennec , phys . lett . * 111 * , 246804 ( 2013 ) . l. capriotti , d. j. scalapino , and r. d. sedgewick , phys . b * 68 * , 014508 ( 2003 ) . y. bang , new j. phys . * 16 * , 023029 ( 2014 ) ; f. wang , f. yang , m. gao , z .- y . lu , t. xiang , and d .- h . lee , europhys . lett . * 93 * , 57003 ( 2011 ) ; t. s. nunner , j. schmalian , and k. h. bennemann , phys . b * 59 * , 8859 ( 1999 ) . y. bang , phys . b * 78 * , 075116 ( 2008 ) .

the-point statistics of singularity strength variables for multiplicative branching processes is calculated from an analytic expression of the corresponding multivariate generating function . the key ingredient is a branching generating function which can be understood as a natural generalisation of the multifractal mass exponents . various random multiplicative cascade processes pertaining to fully developed turbulence are discussed .

chao-dyn9804028

many complex processes occurring in nature are believed to be organized in a selfsimilar way : turbulence , large - scale structure formation in the universe , high - energetic multiparticle dynamics and diffusion limited aggregation are but a few examples showing scaling over a wide range . a convenient tool to characterize the spatial configurations of these processes is the multifractal formalism @xcite@xcite which analyses the one - point statistics of singularity strengths under the assumption of multiple local scale invariance . typically , scale - invariance is a consequence of some hierarchical organisation of the underlying process . since this process occurs in space , its characterisation in terms of multifractals can be expected also to yield information about spatial correlation functions . multivariate correlations , on the other hand , have long been known to contain ( at least in principle ) complete information of the underlying random process . moments and especially cumulants are capable , in their multivariate form , of providing this information by means of well - tested statistical procedures . as in the case of multifractals , cumulants can detect and characterize scale invariance , but unlike the former do not depend on its presence . the existence of these two sets of tools naturally leads to the question how they are interrelated : given a selfsimilar process , does the multifractal characterisation completely determine the spatial correlation functions and vice versa , or does one provide more information than the other ? some attempts have been made to elucidate the relationship between multifractals and correlation functions : two - point statistics of multifractal measures have been discussed in refs . @xcite and applied to turbulence in @xcite . it is to be expected , though , that only the full @xmath0-point statistics of singularity strengths would provide the equivalent information contained in the spatial correlation functions of all orders . it is known that different selfsimilar processes can lead to identical multifractal exponents @xcite . this can be understood as a hint that the multifractal characterisation is indeed incomplete and that the spatial correlation functions may contain more information about the underlying process . we shall show that this is indeed the case for several examples of multiplicative branching processes . the ultimate goal for any analysis , of course , would be a characterisation of the multivariate statistics gained from a physical process to a degree of completeness that would permit precise reconstruction of the `` branching rule '' that governs each step of the selfsimilar process . as far as theory is concerned , this holy grail would be achieved with the specification of a multivariate generating function whose @xmath0-fold derivatives would precisely reproduce the corresponding experimentally measured @xmath0-variate cumulants . clearly , finding this real - life generating function remains beyond present - day capabilities , for both theoretical and experimental reasons . it is possible , though , to make progress towards the ideal by inventing simple branching models whose generating functions can be calculated . an approach to calculate the @xmath0-point spatial correlations to arbitrary order within some of these models was presented in refs . @xcite : there , the multivariate generating function of the spatial correlations was constructed iteratively from a backward evolution equation , leading to a recursive derivation of spatial correlations . an important further step has been the recent discovery of a large class of analytic solutions for generating functions of selfsimilar multiplicative branching processes @xcite . in this paper , we expand on the latter discovery and attempt to cast further light on said relationship between multifractals and correlations . after a brief explanation of the mechanics of multiplicative branching processes in sec . [ sec - multip ] , we review in sec . [ sptcor ] the formalism of multivariate correlations and derive the abovementioned analytic solutions for the branching generating functions . while singularity strength variables are already treated in sec . [ sptcor ] , the relationship between spatial correlations and multifractals is explored more fully in sec . [ mfrct ] , where we also show how the vaunted splitting function ( `` branching rule '' ) can be reconstructed via a two - dimensional laplace transform . the question of reconstructing this splitting function from data is also briefly discussed . four examples of branching processes , treated in sec . [ xmpl ] , drive home the message that multifractals contain less information than the spatial correlations . we discuss our results in more general terms in sec . [ concll ] . random multiplicative branching processes , sometimes also called weight - curdling models or simply cascade models , can be constructed for any number of branches per splitting . for simplicity , we concentrate on binary processes ; generalizations are straightforward . the branching rule governing a binary cascade is described as follows : a starting energy density @xmath1 , uniformly distributed on the unit interval , is split up into two daughter densities @xmath2 and @xmath3 living on adjacent subintervals of length @xmath4 . the random weights @xmath5 and @xmath6 , often also called multipliers or splitting parameters , are drawn from a joint probability density ( or `` splitting function '' ) @xmath7 . in the next generation , each of the two daughter energy densities is itself split up by the same branching rule into two granddaughters distributed uniformly over adjacent subintervals of length @xmath8 . generally , energy densities of the @xmath9-th generation , @xmath10 , are characterized by the binary index @xmath11 , with each @xmath12 taking on possible values 0 or 1 . successive repetition of this prescription for each @xmath13 of the @xmath9-th generation yields , at the next branching , @xmath14 energy densities @xmath15 and @xmath16 , with the multipliers drawn from a splitting function @xmath17 , which usually ( although not necessarily ) is identical with @xmath7 . the process is completed after @xmath18 cascade steps . when the splitting function @xmath19 does not depend on the generation @xmath9 or branch location @xmath20 , the above prescription gives rise to a selfsimilar multiplicative branching process , wholly characterized by the choice of splitting function @xmath21 . we list four examples for later use . the choice @xmath22 \ ; \delta ( q_0 + q_1 -2)\ ] ] leads to the binomial multiplicative process , also known as the @xmath21-model ( often the equivalent parameter @xmath23 is used ) @xcite . the delta function @xmath24 ensures that this process conserves energy at every branching . the splitting function for the non - energy conserving counterpart of the @xmath21-model , the @xmath25-model @xcite , is given by @xmath26 \quad .\ ] ] for the energy dissipation process in fully developed turbulence , the parametrisation @xmath27 with @xmath28 has been directly deduced from experiments @xcite . the corresponding splitting function not conserving energy , is given by @xmath29 these four splitting functions are shown in fig . 1 . for branching processes , the success in extracting useful information or presenting a clear picture depends strongly on the choice of variable . in refs . @xcite , forward and backward evolution equations for a multivariate generating function were employed to calculate the correlations directly in terms of the energy densities , @xmath30 . in the context of multifractals @xcite , on the other hand , another set of variables , the so - called singularity strengths @xmath31 , are in use . the definition of these strengths , @xmath32 , means that they are related to the energy density variables @xmath33 by @xmath34 so that correlations @xmath35 between the singularity strengths @xmath31 are intimately related to corresponding correlations @xmath36 between the logarithms of the energy densities @xmath33 . we have shown previously @xcite that correlations between the @xmath37 lead to a particularly transparent structure for multiplicative branching processes . starting with correlations of the logarithms of energy densities , we show below how this transparent structure arises before translating these results into the language of multifractals . we consider the spatial correlations @xmath38 of order @xmath0 between the variables @xmath39 . the bracket @xmath40 indicates the averaging over all possible configurations . moments @xmath41 are conveniently calculated from the multivariate generating function @xmath42 = \left\langle \exp\biggl ( \sum_{k_1 , \ldots , \atop k_{j}=0}^1 \lambda_{k_1 \cdots k_j}^{(j ) } \ln\epsilon_{k_1 \cdots k_j}^{(j ) } \biggr ) \right\rangle\ ] ] by taking appropriate derivatives with respect to the conjugate variables @xmath43 : @xmath44 } { \partial \lambda_{\bkappa_1}^{(j ) } \cdots \partial \lambda_{\bkappa_n}^{(j ) } } \right| _ { \lambda = 0 } \quad .\ ] ] the corresponding multivariate cumulant generating function @xmath45 = \ln z [ \lambda^{(j ) } ] \ ] ] yields the cumulant correlation densities in the same way via @xmath46 \over \partial\lambda_{\bkappa_1}^{(j ) } \cdots \partial\lambda_{\bkappa_n}^{(j ) } } \right|_{\lambda^{(j ) } = 0 } = \left\langle \ln\epsilon_{\bkappa_1}^{(j ) } \cdots \ln\epsilon_{\bkappa_n}^{(j ) } \right\rangle_{c } \quad ; \ ] ] here , we have introduced the index @xmath47 ( for `` cumulant '' ) on the last bracket to distinguish ( [ 22fuenf ] ) from the expression ( [ 22eins ] ) for the ordinary correlation densities . the lowest four orders of cumulant correlation densities are @xmath48 with brackets @xmath49 indicating summation over permutations of indices . the derivation of the cumulant generating function depends crucially on a rearrangement of terms in the sum @xmath50 entering the generating function in ( [ 22zwei ] ) . the trick lies in recognising that the cascade prescription of section [ sec - multip ] implies that the energy density in the @xmath18-th generation is the product of all the splitting parameters in its ancestry , @xmath51 so that its logarithm is additive in the multipliers logarithms , @xmath52 inserting this into the sum in the exponential of the generating function in ( [ 22zwei ] ) and defining @xmath53 we find , after rearrangement and judicious regrouping of terms , that @xmath54 this can now be utilized to rewrite the configuration average @xmath55 of any function @xmath56 in terms of the splitting parameters as follows . we start with the definition @xmath57 where @xmath58 is the joint probability of finding a given logarithmic energy density @xmath59 in bin ( @xmath60 ) , a given @xmath61 in the next bin , etc . since each energy density @xmath33 is completely determined by its @xmath18 ancestral splitting parameters , the multivariate probability density @xmath62 can be constructed from the splitting function @xmath19 by using ( [ 221drei ] ) : @xmath63 \nonumber \\ & & \qquad \times \biggl [ \prod _ { { k_1 , \ldots , \atop k_{j}=0 } } ^1 \delta \bigl ( \ln\epsilon_{k_1 \cdots k_j}^{(j ) } - \sum_{j=1}^j \ln q_{k_1 \cdots k_j}^{(j ) } \bigr ) \biggr ] \quad .\end{aligned}\ ] ] inserting ( [ hceb ] ) , ( [ 221eins ] ) and ( [ 221vier ] ) into eq . ( [ 22vier ] ) , we find that the cumulant generating function becomes @xmath64 & = & \ln\biggl [ \int \bigl ( \prod _ { { k_1 , \ldots , \atop k_{j}=0 } } ^1 d(\ln \epsilon_{k_1\cdots k_j}^{(j ) } ) \bigr ) \ ; p(\ln\epsilon_{0\cdots 0}^{(j ) } , \ldots , \ln\epsilon_{1\cdots 1}^{(j ) } ) \biggr . \nonumber \\ & & \biggl . \qquad \times \exp \biggl ( \sum_{k_1,\ldots , \atop k_j = 0}^1 \lambda_{k_1\cdots k_j}^{(j ) } \ln \epsilon_{k_1\cdots k_j}^{(j ) } \biggr ) \biggr ] \nonumber \\ & = & \sum_{j=1}^j \sum_{k_1,\ldots , \atop k_{j-1 } = 0}^1 q [ \lambda_{k_1\cdots k_{j-1 } 0}^{(j ) } , \lambda_{k_1\cdots k_{j-1 } 1}^{(j ) } ] \quad,\end{aligned}\ ] ] where @xmath65 & = & \ln \bigl [ \int dq_{0}\ , dq_{1}\ , p(q_{0 } , q_{1 } ) \exp \left ( \lambda_{0 } \ln q_{0 } + \lambda_{1 } \ln q_{1 } \right ) \bigr ] \nonumber \\ & = & \ln \bigl\langle \exp \left ( \lambda_{0 } \ln q_{0 } + \lambda_{1 } \ln q_{1 } \right ) \bigr\rangle \end{aligned}\ ] ] eqs . ( [ 221fuenf])([221sechs ] ) represent the long - sought analytic expression for multiplicative cascades : they show that the cumulant generating function of the entire cascade can be written as the sum of all branching generating functions @xmath66 , one for every branching . in most cases , the two - fold integrals ( [ 221sechs ] ) can be solved analytically or parametrically to yield a complete analytic solution for @xmath67 and thereby for all cumulants . the scope and limitations of the solution ( [ 221fuenf ] ) are as follows . clearly , the generating function in terms of @xmath37 is applicable to any functional form of the splitting function or b.g.f . it does not depend on the number of branches either : trivariate or even higher - variate splitting functions can be implemented . due to the additive nature of the b.g.f.s , the splitting functions can differ from generation to generation and even from branch to branch . the only ( and important ) precondition for the applicability of eq . ( [ 221fuenf ] ) is that the splitting variables of every branching must be independent of those of the other branchings and generations . the cumulant generating function can also be constructed from a forward evolution equation . in ref . @xcite , the forward evolution equation for the generating function was derived in terms of energy densities @xmath68 . in terms of the @xmath69 , the cumulant generating function after @xmath9 cascade steps , @xmath70 $ ] , can be expressed in terms of @xmath71 $ ] as follows . at each of the @xmath72 independent branchings , eq . ( [ 221drei ] ) relates the daughter energy densities to their parent by @xmath73 in analogy to ( [ hceb ] ) , @xmath74\end{aligned}\ ] ] and assigning , as in ( [ 221sieben ] ) , @xmath75 we arrive at @xmath76 & = & k^{(j-1 ) } [ \lambda^{(j-1 ) } ] \nonumber \\ & & + \sum_{k_1,\ldots , \atop k_{j-1 } = 0}^1 \ln \bigl [ \int dq_{k_1\cdots k_{j-1 } 0}^{(j)}\ , dq_{k_1\cdots k_{j-1 } 1}^{(j)}\ ; p(q_{k_1\cdots k_{j-1 } 0}^{(j ) } , q_{k_1\cdots k_{j-1 } 1}^{(j ) } ) \bigr . \nonumber \\ & & \qquad\qquad\qquad \bigl . \times \exp \left ( \lambda_{k_1\cdots k_{j-1 } 0}^{(j ) } \ln q_{k_1\cdots k_{j-1 } 0}^{(j ) } + \lambda_{k_1\cdots k_{j-1 } 1}^{(j ) } \ln q_{k_1\cdots k_{j-1 } 1}^{(j ) } \right ) \bigr ] \nonumber \\ & = & k^{(j-1 ) } [ \lambda^{(j-1 ) } ] + \sum_{k_1,\ldots , \atop k_{j-1 } = 0}^1 q [ \lambda_{k_1\cdots k_{j-1 } 0}^{(j ) } , \lambda_{k_1\cdots k_{j-1 } 1}^{(j ) } ] \quad .\end{aligned}\ ] ] this forward evolution equation can be iterated from @xmath77 down to @xmath78 , where we arrive at @xmath79 = \ln ( z^{(0 ) } [ \lambda^{(0 ) } ] ) = 0 $ ] ; the outcome of this iteration is identical to the solution ( [ 221fuenf ] ) . similarly , a backward evolution equation for the cumulant generating functions can also be used to achieve identical results ; we leave this as an exercise to the reader . having found an analytic expression of the multivariate generating function @xmath80 $ ] , it is now straightforward to calculate the cumulant correlation densities ( [ 22sechs])([22neun ] ) between the @xmath39-variables via eq . ( [ 22fuenf ] ) . in order to give compact expressions , we define an ultrametric distance between bins . assume the two bins are together for the first @xmath9 steps of the binary cascade before splitting at the @xmath81-th generation , @xmath82 and @xmath83 , where @xmath84 for all @xmath85 and @xmath86 . then the ultrametric distance between them is @xmath87 for @xmath0 bins , the generalised ultrametric distance is @xmath88 in the following , we assume that the same functional form for the b.g.f . is used at all branchings . we also assume the splitting function @xmath19 to be symmetric under exchange of the splitting parameters , @xmath89 ; the four splitting functions given in eqs . ( [ 21eins])([21vier ] ) fall into this category . in consequence , the branching generating function @xmath90 = q [ \lambda_1 , \lambda_0 ] $ ] is also symmetric , so that also @xmath91 } { \partial\lambda_0^{n_1 } \partial\lambda_1^{n_2 } } \right|_{\lambda=0 } = \left . \frac { \partial^{n_1+n_2 } q[\lambda_0,\lambda_1 ] } { \partial\lambda_0^{n_2 } \partial\lambda_1^{n_1 } } \right|_{\lambda=0 } \quad .\ ] ] this relation results in compact expressions for the cumulant correlation densities @xmath92 . inserting ( [ 221fuenf ] ) into eq . ( [ 22fuenf ] ) and taking into account ( [ 221sieben ] ) , the cumulant correlation density of first order is found to be @xmath93 } { \partial\lambda_0 } \right|_{\lambda=0 } \right ) \quad .\ ] ] for second order we get , with ( [ 223eins ] ) , @xmath94 } { \partial\lambda_0 ^ 2 } \right|_{\lambda=0 } \right ) + ( 1 - \delta_{d_2,0 } ) \left ( \left . \frac { \partial^2 q[\lambda_0,\lambda_1 ] } { \partial\lambda_0 \partial\lambda_1 } \right|_{\lambda=0 } \right ) \quad , \ ] ] while with the generalisation ( [ 223zwei ] ) , the third order cumulant density becomes @xmath95 } { \partial\lambda_0 ^ 3 } \right|_{\lambda=0 } \right ) + ( 1 - \delta_{d_3,0 } ) \left ( \left . \frac { \partial^3 q[\lambda_0,\lambda_1 ] } { \partial\lambda_0 ^ 2 \partial\lambda_1 } \right|_{\lambda=0 } \right ) \quad .\ ] ] in the case of fourth order , two cases have to be distinguished , depending on whether the first splitting of the four bins @xmath96 goes into three and one , say ( @xmath97 ) with @xmath98 and @xmath99 , or into two and two , say ( @xmath100 ) with @xmath101 , @xmath102 and @xmath103 : @xmath104 } { \partial\lambda_0 ^ 4 } \right|_{\lambda=0 } \right ) + ( 1 - \delta_{d_4,0 } ) \left ( \left . \frac { \partial^4 q[\lambda_0,\lambda_1 ] } { \partial\lambda_0^{n_1 } \partial\lambda_1^{n_2 } } \right|_{\lambda=0 } \right ) \quad .\ ] ] even under the stated assumptions , the expressions ( [ 223vier])([223sieben ] ) are still very general for selfsimilar binary multiplicative cascade models . the one and only input is the splitting function @xmath19 determining the branching generating function @xmath105 $ ] . the cumulant correlation densities for the splitting functions ( [ 21eins])([21vier ] ) will be discussed in more detail in sec . [ xmpl ] . it is clear from eqs . ( [ 223vier])([223sieben ] ) that the derivatives of the branching generating function completely fix the spatial cumulant correlation densities . using the definition ( [ 221sechs ] ) these derivatives can be expressed in terms of generic branching moments : @xmath106 } { \partial\lambda_0^{n_1 } \partial\lambda_1^{n_2 } } \right|_{\lambda=0 } = \left\langle ( \ln q_0)^{n_1 } ( \ln q_1)^{n_2 } \right\rangle_c \quad .\ ] ] in the lowest three orders , they read explicitly : @xmath107 we have defined the cumulant correlation densities @xmath108 in terms of the @xmath39-variables . with the help of ( [ 21fuenf ] ) , they can be transformed into cumulant correlation densities @xmath109 between the singularity strength variables @xmath31 . we find for the corresponding generating function @xmath110 & = & \ln \left\langle \exp\biggl ( \sum_{k_1 , \ldots , \atop k_j = 0}^1 \lambda_{k_1 \cdots k_j}^{(j ) } \alpha_{k_1 \cdots k_j}^{(j ) } \biggr ) \right\rangle \nonumber \\ & = & \ln \left\langle \exp\biggl ( \sum_{k_1 , \ldots , \atop k_j = 0}^1 \lambda_{k_1 \cdots k_j}^{(j ) } \left ( 1 - \frac{1}{j \ln 2 } \ln\epsilon_{k_1 \cdots k_j}^{(j ) } \right ) \biggr ) \right\rangle \nonumber \\ & = & \sum_{k_1 , \ldots , \atop k_j = 0}^1 \lambda_{k_1 \cdots k_j}^{(j ) } + k \left [ \frac{(-1)}{j \ln 2 } \lambda^{(j ) } \right ] \quad .\end{aligned}\ ] ] taking derivatives , the relations between the cumulant densities are found to be , for first order , @xmath111 and , for @xmath112 , @xmath113 in other words , with the exception of first order , the @xmath114 are directly proportional to @xmath92 . one way to approach multifractals is to count how often specific values @xmath25 of the singularity strengths occur ; this leads to the @xmath115-spectrum @xcite . the latter is related by a legendre transformation to a sequence of mass exponents @xmath116 which are defined by the moment scaling behavior of bin energies @xmath117 : @xmath118 in order to avoid confusion we have written @xmath119 instead of the more familiar notation @xmath120 . since the bin energies are related to the singularity strengths by @xmath121 , these moments are to be understood as scale - dependent measures for the one - point statistics of the singularity strengths . we first derive the expression for the exponents @xmath116 for the special case of a symmetric splitting function @xmath89 of a binary multiplicative cascade process which conserves energy . ( note that for energy non - conserving splitting functions , as for example those given in ( [ 21zwei ] ) and ( [ 21vier ] ) , the multifractal exponents can not be defined unambiguously ; see ref . then we can use ( [ 221zwei ] ) and the statistical independence of splitting parameters to deduce @xmath122 insertion into eq . ( [ 31eins ] ) yields : @xmath123 for a @xmath5/@xmath6-asymmetric , energy - conserving splitting function we have to replace the sum appearing in ( [ 31eins ] ) by @xmath124^j \quad , \ ] ] which gives rise to @xmath125 the close relationship between the branching generating function @xmath105 $ ] and the multifractal exponents @xmath116 is revealed by simple inspection of ( [ 221sechs ] ) and ( [ 31fuenf ] ) . we find that @xmath126 \right ) + \exp\left ( q[\lambda_0=0,\lambda_1=\nu ] \right ) \bigr ] \quad .\ ] ] this relationship simplifies further once we consider an energy - conserving , @xmath127 symmetric splitting function . through ( [ 31drei ] ) we arrive at @xmath128 \quad .\ ] ] the two relations ( [ 32eins ] ) and ( [ 32zwei ] ) express @xmath116 in terms of @xmath105 $ ] . it appears that @xmath129 is more limited than @xmath66 in that the latter is defined over the full @xmath130@xmath131 plane while @xmath119 is restricted to the axes @xmath132 or @xmath133 . let us consider the point in more detail . in particular , we will demonstrate that different energy - conserving splitting functions may lead to identical multifractal exponents while their branching generating functions do differ . given an energy - conserving , @xmath134-symmetric splitting function @xmath135 , we construct a related splitting function by breaking the @xmath134-symmetry : @xmath136 the splitting function @xmath137 is almost identical to @xmath19 , their only difference being that the smaller splitting parameter @xmath5 is now always drawn for the branching into the left subinterval whereas the the larger splitting parameter @xmath138 goes into the right subinterval . the one - point statistic is not affected by this modification of the splitting function , since it does not care whether the larger splitting parameter goes to the left or right subinterval . hence the multifractal exponents are not changed either . with ( [ 31fuenf ] ) we get @xmath139 we are hence forced to conclude that a multifractal analysis does not suffice completely to characterize the dynamics , i.e. the splitting function , of the underlying multiplicative cascade process . this statement holds even more strongly once we also give up energy conservation in the splitting function , with the consequence that properly defined backward moments show deviations from perfect multifractal scaling . in order to attain the full information , there is no way to get around spatial correlations in general , and the cumulant correlation densities @xmath92 in particular , to extract the branching moments @xmath140 and hence the branching generating function @xmath141 = \sum_{n_1,n_2=0}^{\infty } \frac{1}{n_1!n_2 ! } \left\langle ( \ln q_0 ) ^{n_1 } ( \ln q_1 ) ^{n_2 } \right\rangle_c \lambda_0^{n_1 } \lambda_1^{n_2 } \quad .\ ] ] using eq . ( [ 221sechs ] ) the branching generating function can then be uniquely inverted into the splitting function via a two - dimensional inverse laplace transformation , @xmath142 - ( \lambda_0 + \lambda_1 + 2 ) \ln 2 } \quad .\ ] ] as a consequence we view @xmath105 $ ] as the natural and complete generalisation of the multifractal mass exponents @xmath119 for random multiplicative binary cascade processes . the @xmath21-model splitting function given in eq . ( [ 21eins ] ) , when inserted into eq . ( [ 221sechs ] ) , determines the branching generating function @xmath105 $ ] : @xmath143 & = & \ln \left ( \frac{1}{2 } \left\ { \exp\left [ \lambda_0 \ln ( 1+\beta ) + \lambda_1 \ln ( 1-\beta ) \right ] \right . \right . \nonumber \\ & & \left . \left . \qquad + \exp\left [ \lambda_0 \ln ( 1-\beta ) + \lambda_1 \ln ( 1+\beta ) \right ] \right\ } \rule{0mm}{1em } \right ) \quad .\end{aligned}\ ] ] we introduce the transformation @xmath144 which leads to @xmath145 = \frac { \ln ( 1-\beta^2 ) } { 2 } \lambda_+ + \ln \left ( \cosh \left ( \frac{1}{2 } \lambda_- \ln\left ( \frac{1+\beta}{1-\beta } \right ) \right ) \right ) \quad .\ ] ] fig . 2(a ) illustrates this branching generating function . now we calculate the derivatives of the branching generating function with respect to the conjugate variables @xmath130 and @xmath131 , which are needed for the cumulant densities @xmath92 . for the first derivative we find : @xmath146 } { \partial \lambda_0 } \right|_{\lambda=0 } & = & \left . \frac { \partial q[\lambda_0,\lambda_1 ] } { \partial \lambda_1 } \right|_{\lambda=0 } \nonumber \\ & = & \frac { \ln ( 1-\beta^2 ) } { 2 } + \frac{1}{2 } \ln\left ( \frac{1+\beta}{1-\beta } \right ) \left . \tanh \left ( \frac{1}{2 } \lambda_- \ln\left ( \frac{1+\beta}{1-\beta } \right ) \right ) \right|_{\lambda=0 } \nonumber \\ & = & \frac{1}{2 } \ln(1 - \beta^2 ) \quad .\end{aligned}\ ] ] for an arbitrary derivative of @xmath105 $ ] of order @xmath147+@xmath148 , we get the following relationship : @xmath149 } { \partial \lambda_0^{n_1 } \partial \lambda_1^{n_2 } } \right|_{\lambda=0 } & = & \left . \frac { \partial^{n_1+n_2 } q[\lambda_0,\lambda_1 ] } { \partial \lambda_-^{n_1+n_2 } } \right|_{\lambda=0 } \left ( \frac { \partial \lambda_- } { \partial \lambda_0 } \right)^{n_1 } \left ( \frac { \partial \lambda_- } { \partial \lambda_1 } \right)^{n_2 } \nonumber \\ & = & ( -1)^{n_2 } \left . \frac { \partial^{n_1+n_2 } q[\lambda_0,\lambda_1 ] } { \partial \lambda_0^{n_1+n_2 } } \right|_{\lambda=0 } \quad .\end{aligned}\ ] ] except for an alternating sign , the two - point branching moments @xmath150 are identical to the one - point branching moments @xmath151 ; this only holds for the @xmath21-model and not for the other three models associated with the splitting functions ( [ 21zwei])([21vier ] ) . as a consequence of ( [ 41fuenf ] ) , we only need to calculate derivatives of @xmath105 $ ] with respect to @xmath130 . we use the intermediate step of ( [ 41vier ] ) and write for @xmath152 : @xmath153 } { \partial \lambda_0^{n } } \right|_{\lambda=0 } = \left . \frac{1}{2 } \ln\left ( \frac{1+\beta}{1-\beta } \right ) \ ; \left ( \frac { \partial^{n-1 } } { \partial \lambda_0^{n-1 } } \left\ { \tanh \left [ \frac{1}{2 } \lambda_0 \ln\left ( \frac{1+\beta}{1-\beta } \right ) \right ] \right\ } \right ) \right|_{\lambda_0=0 } \quad .\ ] ] since @xmath154 is an odd function in @xmath155 , i.e.@xmath156 with the bernoulli numbers @xmath157 , all odd derivatives of the branching generating function @xmath105 $ ] with respect to @xmath130 vanish : @xmath158 } { \partial \lambda_0^{n } } \right|_{\lambda=0 } = 0 \qquad\qquad ( n=3,5,7,\ldots ) \quad .\ ] ] the same conclusion could have also been obtained directly from ( [ 41fuenf ] ) : for odd @xmath159 we have @xmath160 / \partial \lambda_0^{n}|_{\lambda=0 } = - \partial^{n } q[\lambda_0,\lambda_1 ] / \partial \lambda_1^{n}|_{\lambda=0}$ ] , but from @xmath5/@xmath6-symmetry considerations of the splitting function we expect @xmath161 ; hence , @xmath160 / \partial \lambda_0^{n}|_{\lambda=0 } = \langle ( \ln q_0)^{n } \rangle_c = 0 $ ] . an immediate consequence of this result is that , for example , the spatial cumulant densities of third order vanish , @xmath162 ; see eq . ( [ 223sechs ] ) . for even derivatives of @xmath105 $ ] with respect to @xmath130 we find from ( [ 41sechs ] ) and ( [ 41sieben ] ) that @xmath163 } { \partial \lambda_0^{n } } \right|_{\lambda=0 } = \frac{(2^n-1)b_n}{n } \left ( \ln \left ( \frac{1+\beta}{1-\beta } \right ) \right)^n \qquad ( n=2,4,6,\ldots ) \quad .\ ] ] the results ( [ 41vier ] ) , ( [ 41fuenf ] ) , ( [ 41acht ] ) and ( [ 41neun ] ) can now be inserted into eqs . ( [ 223vier])([223sieben ] ) to determine the cumulant correlation densities @xmath164 . 3 illustrates the results for the two - point statistics : @xmath165 of second order , and @xmath166 and @xmath167 of fourth order ; note again that the third order @xmath168 , vanishes . 4(a ) and 4(b ) show the second - order and fourth order cumulants as a function of the ultrametric distances respectively . since the @xmath21-model splitting function is energy - conserving and @xmath134-symmetric we can use relation ( [ 32zwei ] ) to extract the multifractal exponents @xmath119 from the branching generating function @xmath105 $ ] . this leads to the well - known result @xcite @xmath169 \quad .\ ] ] it is illustrated in fig . 5 . via @xmath170 } { \partial \lambda_0^n } \right|_{\lambda=0 } \nonumber \\ & = & - \delta_{n,1 } + \frac{1}{\ln 2 } \left\langle ( \ln q_0 ) ^n \right\rangle_c\end{aligned}\ ] ] the derivatives of @xmath119 with respect to @xmath171 are linked to the derivatives ( [ 41vier ] ) , ( [ 41acht ] ) and ( [ 41neun ] ) of the @xmath21-model branching generating function ( [ 41drei ] ) . the symmetric @xmath25-model is similar to the @xmath21-model except that it does not conserve energy in a cascade splitting . from the splitting function ( [ 21zwei ] ) , its b.g.f . is found to be @xmath172 & = & { \textstyle{1\over 2 } } ( \lambda_0+\lambda_1 ) \ln(1-\beta^2 ) \nonumber \\ & & + \ln \left\ { \cosh\left [ { 1\over 2 } \lambda_0 \ln\left ( 1 + \beta \over 1 - \beta \right ) \right ] \right\ } \nonumber \\ & & + \ln \left\ { \cosh\left [ { 1\over 2 } \lambda_1 \ln\left ( 1 + \beta \over 1 - \beta \right ) \right ] \right\ } \quad , \end{aligned}\ ] ] clearly different from the b.g.f . ( [ 41drei ] ) for the @xmath21-model . 2(b ) depicts the b.g.f . ( [ 42eins ] ) . note that for @xmath173 or @xmath174 the two expressions ( [ 41drei ] ) and ( [ 42eins ] ) become identical ; consequently , the one - point branching moments of the @xmath25-model are identical to those of the @xmath21-model given in ( [ 41vier ] ) , ( [ 41acht ] ) and ( [ 41neun ] ) : @xmath175 \over \partial \lambda_0^n } \right|_{\lambda = 0 } \right)_{\alpha{\rm -model } } = \left ( \left . { \partial^n q[\lambda_0 , \lambda_1 ] \over \partial \lambda_0^n } \right|_{\lambda = 0 } \right)_{p{\rm -model } } \,.\ ] ] this is the reason why , in a multifractal approach , the two models look the same asymptotically . to see differences between them , one must consider two - point branching moments : for the @xmath25-model they all vanish , @xmath176 \over \partial \lambda_0^{n_1 } \partial \lambda_1^{n - n_1 } } \right|_{\lambda = 0 } = \left\langle ( \ln q_0)^{n_1 } ( \ln q_1)^{n - n_1 } \right\rangle_c = 0 \quad , \ ] ] where @xmath177 . we hence see that the two - point branching moments are sensitive to the violation of energy conservation in the splitting function . as a consequence of ( [ 42zwei ] ) and ( [ 42drei ] ) , the cumulant densities @xmath92 of the @xmath25- and @xmath21-model now look slightly different . fig . 4(a ) compares @xmath178 of second order as a function of the ultrametric distance ( [ 223eins ] ) . the @xmath21-model is able to describe the multifractal aspects of the intermittent fluctuations occurring in the energy dissipation field in fully developed turbulence . however , due to its simplicity , the experimental multiplier distributions can not be reproduced . in ref . @xcite , a modification of the @xmath21-model was proposed which accounts for the correct multiplier distributions ; we call this modification the srst cascade model . insertion of its splitting function ( [ 21drei ] ) into eq . ( [ 221sechs ] ) yields an analytic expression for its b.g.f . , @xmath179 = \ln\left [ \int_0 ^ 1 { \rm d}z z^{\lambda_0+\beta-1 } ( 1-z)^{\lambda_1+\beta-1 } \right ] + \ln\left [ \frac{\gamma(2\beta)}{\gamma(\beta)^2 } \right ] + ( \lambda_0 + \lambda_1 ) \ln 2 \quad .\ ] ] since @xmath180 and @xmath181 , so that @xmath182 and @xmath183 , the integral appearing in the first term of the right hand side can be identified with the beta function @xmath184 . this leads to @xmath185 = ( \lambda_0 + \lambda_1 ) \ln 2 + \ln\left ( \frac{\gamma(2\beta)}{\gamma(\lambda_0+\lambda_1 + 2\beta ) } \frac{\gamma(\lambda_0+\beta)}{\gamma(\beta ) } \frac{\gamma(\lambda_1+\beta)}{\gamma(\beta ) } \right ) \quad .\ ] ] this result is illustrated in fig . 2(c ) . from ( [ 43zwei ] ) the spatial cumulant densities @xmath92 can be calculated in the straightforward manner presented in sec . [ cumcord ] . in the lowest even orders , the results are very similar to those obtained for the @xmath21-model ; the odd orders , however , are not equal to zero anymore . making use of the relationship ( [ 32zwei ] ) the result ( [ 43zwei ] ) translates into the following multifractal exponents : @xmath186 note that this expression , which is illustrated in fig . 5 , is only defined for @xmath187 ; for @xmath188 the integral in ( [ 43eins ] ) diverges . this is equivalent to the statement that the negative moments with @xmath189 of the splitting functions ( [ 21drei ] ) and ( [ 21vier ] ) do not exist . for the multifractal spectrum to exist over the full @xmath171 range , however , moments of all orders , both positive and negative , must be finite . the absence of finite negative moments hence implies that it is not possible to construct full @xmath119 and @xmath115 curves for this specific splitting function . it is thus an example of a well - defined selfsimilar cascade process which can not be described fully by the multifractal formalism . experimentally , the intermittent structures in the three - dimensional energy dissipation field of fully developed turbulence are observed on a one - dimensional cut . although energy is conserved in three dimensions , this is probably not the case in one dimension . for the multiplicative branching models , this has the consequence that the splitting function @xmath19 can not be expected to conserve energy . in this spirit , the expression ( [ 21vier ] ) represents an untested extrapolation of the srst multiplier distribution ( [ 21drei ] ) , which has been deduced from one - dimensional data under the assumption of energy conservation @xcite . insertion of the splitting function ( [ 21vier ] ) into eq . ( [ 221sechs ] ) yields the corresponding branching generating function : @xmath190 = ( \lambda_0 + \lambda_1 ) \ln 2 + \ln\left ( \frac{\gamma(2\beta)}{\gamma(\lambda_0 + 2\beta ) } \frac{\gamma(\lambda_0+\beta)}{\gamma(\beta ) } \right ) + \ln\left ( \frac{\gamma(2\beta)}{\gamma(\lambda_1 + 2\beta ) } \frac{\gamma(\lambda_1+\beta)}{\gamma(\beta ) } \right ) \quad ; \ ] ] the explicit derivation is analogous to the one given in sec . [ srstec ] . the illustration of ( [ 44eins ] ) is given in fig . 2(d ) ; it differs from the branching generating function ( [ 43zwei ] ) of the srst cascade model with energy conservation . as in the @xmath21/@xmath25-model comparison , the one - point derivatives @xmath191 \over \partial \lambda_0^n } \right|_{\lambda = 0 } \right)_{\rm srst \atop ( no ec ) } = \left ( \left . { \partial^n q[\lambda_0 , \lambda_1 ] \over \partial \lambda_0^n } \right|_{\lambda = 0 } \right)_{\rm srst \atop ( ec)}\ ] ] of the srst cascade model with ( ec ) and without ( no ec ) energy conservation are identical , while for the two - point derivatives we find ( @xmath177 ) @xmath192 \over \partial \lambda_0^{n_1 } \partial \lambda_1^{n - n_1 } } \right|_{\lambda = 0 } \right)_{\rm srst \atop ( no ec ) } = 0 \neq \left ( \left . { \partial^n q[\lambda_0 , \lambda_1 ] \over \partial \lambda_0^{n_1 } \partial \lambda_1^{n - n_1 } } \right|_{\lambda = 0 } \right)_{\rm srst \atop ( ec ) } \quad .\ ] ] consequently , the cumulant densities @xmath92 of the srst - cascade model with and without energy conservation are different . with a clever change of variables from energy densities @xmath193 to the singularity strengths @xmath194 or @xmath195 , we have derived an analytic expression for the multivariate generating function of binary multiplicative cascade models . the latter completely describes the @xmath0-point statistics , i.e. the spatial ( cumulant ) correlation densities of arbitrary order @xmath0 . the key input has been a bivariate branching generating function , which is related to the underlying splitting function of the binary multiplicative cascade process via a two - dimensional laplace transform . this branching generating function can be understood as a natural and , for selfsimilar binary cascade processes , complete generalisation of the multifractal mass exponents . while its properties completely fix the spatial correlation densities , the multifractal mass exponents do not . various cascade models , relevant to fully developed turbulence , have been discussed to underpin this point . we have shown that , given that the experimentally measurable cumulants in @xmath196 are @xmath0-fold derivatives of the branching generating function , the latter can in principle be reconstructed from the former . with the help of eqs . ( [ 32neun ] ) and ( [ 32zehn ] ) , the b.g.f . can then be inverted into the splitting function via a two - dimensional inverse laplace transform . in this way , the violation of energy conservation along one - dimensional cuts through the three - dimensional energy dissipation field can be inspected . before this new approach is applied directly to ( turbulence ) data , however , a number of complications will have to be dealt with . in order to infer the branching generating function from the cumulant densities @xmath92 ( to all orders , in principle ) , a very effective representation of the latter has to be found ; here , as in refs . @xcite , a wavelet transformation might be useful to compress the information contained in the cumulant densities . moreover , there is the problem of non - homogeneity : as a consequence of the hierarchical nature of the cascade evolution , the theoretical correlation functions are not invariant with respect to spatial translations , in contradiction to experimental measurements . it remains to be seen whether and in what way a scheme to restore homogeneity , as for example the one used in @xcite , influences or destroys the capability of inferring the branching generating function . furthermore , the observed statistical dependence of multipliers @xcite has to be taken into account . recent simulations @xcite indicate that this issue is closely linked to the restoration of homogeneity . this is in agreement with the conclusions reached by nelkin and stolovitzky @xcite by a different route , who argue that the experimentally proven dependence of multiplier distributions on the position of the subinterval implies that the multipliers are not statistically independent . since any scheme to restore homogeneity will necessarily average out subinterval positions in some way , it will likely influence the multipliers statistical dependence also . this remains to be explored in detail @xcite . finally , the binary structure of the selfsimilar cascade processes discussed in this paper may not be appropriate : assuming that the physical processes themselves are , indeed , selfsimilar , the best selfsimilar basis for a scaling analysis ( such as a specific wavelet ) should be selected by the data itself . once these points are clarified , new information can hopefully be gleaned from the analysis of `` fully developed turbulence data '' . besides fully - developed turbulence , we envisage many and diverse applications of our analytic solution in other branches of physics . the case of qcd branching processes immediately comes to mind . for the latter , the @xmath25- and @xmath21-models have already been used in this context as simulation toy models @xcite . implications in this and , for example , random multiplicative process calculations in large - scale structure formation in the universe @xcite remain to be explored .

in this paper , we make use of the relations between the braid and mapping class groups of a compact , connected , non - orientable surface without boundary and those of its orientable double covering to study embeddings of these groups and their ( virtual ) cohomological dimensions . we first generalise results of to show that the mapping class group of relative to a-point subset embeds in the mapping class group of relative to a-point subset . we then compute the cohomological dimension of the braid groups of all compact , connected aspherical surfaces without boundary , which generalises results of . finally , if the genus of is greater than or equal to , we give upper bounds for the virtual cohomological dimension of . [ multiblock footnote omitted ] [ multiblock footnote omitted ] [ multiblock footnote omitted ]

1610.03288

let @xmath1 be a compact , connected surface , and let @xmath7 be a finite ( possibly empty ) subset of @xmath1 of cardinality @xmath8 . let @xmath9 denote the group of homeomorphisms of @xmath1 for the operation of composition that leave @xmath10 invariant . if @xmath1 is orientable , let @xmath11 denote the set of elements of @xmath9 that preserve orientation . note that @xmath11 is a subgroup of @xmath9 of index two . we define the _ mapping class group @xmath12 of @xmath1 relative to @xmath10 _ by : @xmath13 if @xmath1 is orientable ( resp . non orientable ) , the group @xmath12 is thus the set of isotopy classes of @xmath11 ( resp . @xmath9 ) , the isotopies being relative to the set @xmath10 . if @xmath14 ( resp . @xmath9 ) then we let @xmath15 $ ] denote its mapping class in @xmath12 . up to isomorphism , @xmath12 only depends on the cardinality @xmath3 of the subset @xmath10 , and we shall often denote this group by @xmath16 . if @xmath10 is empty , then we omit it from the notation , and shall just write @xmath17 , @xmath18 , @xmath19 etc . the mapping class group may also be defined in other categories ( pl category , smooth category ) , the groups obtained being isomorphic @xcite . the mapping class group has been widely studied from different points of view see @xcite for example . if @xmath20 , there is a surjective homomorphism @xmath21 defined by @xmath22 for all @xmath23 and @xmath24 , where @xmath25 is the symmetric group on @xmath3 letters . this homomorphism induces a surjective homomorphism from @xmath12 to @xmath25 that we also denote by @xmath26 , and its kernel is called the _ pure mapping class group of @xmath1 relative to @xmath10 _ , denoted by @xmath27 . so we have a short exact sequence : @xmath28 and if @xmath23 then @xmath15\in pmcg(s;x)$ ] if and only if @xmath29 fixes @xmath10 pointwise . mapping class groups are closely related to surface braid groups . if @xmath20 , the _ @xmath3^th^ ordered configuration space _ @xmath30 of @xmath1 is the set of all ordered @xmath3-tuples of distinct points of @xmath1 that we may consider as a subspace of the @xmath3-fold product of @xmath1 with itself . the group @xmath31 acts freely on @xmath30 by permuting coordinates , and the associated quotient is the _ @xmath3^th^ unordered configuration space _ of @xmath1 , denoted by @xmath32 . the fundamental group @xmath33 ( resp . @xmath34 ) , denoted by @xmath35 ( resp . @xmath36 ) is the _ braid group _ . _ pure braid group _ ) of @xmath1 on @xmath3 strings @xcite , and @xmath35 and @xmath36 are related by a short exact sequence similar to that of ( [ eq : sigmamcg ] ) . it is well known that @xmath37 ( resp . @xmath38 ) is isomorphic to _ artin s braid group _ _ artin s pure braid group _ ) on @xmath3 strings . if @xmath39 is a @xmath40-fold covering map , where @xmath41 , then for all @xmath42 , there is a continuous map : @xmath43 defined by @xmath44 for all @xmath45 , and the induced homomorphism @xmath46 on the level of fundamental groups is injective @xcite . taking @xmath47 to be the standard @xmath6-fold ( universal ) covering , where @xmath48 is the @xmath6-sphere and @xmath49 is the real projective plane , this result was then applied in @xcite to classify the isomorphism classes of the finite subgroups of @xmath50 , and to show that @xmath50 and @xmath51 are linear groups for all @xmath20 . if @xmath52 ( resp . @xmath53 ) , let @xmath54 ( resp . @xmath55 ) be a compact , connected orientable ( resp . non - orientable ) surface of genus @xmath56 without boundary . in the non - orientable case , @xmath56 is the number of projective planes in a connected sum decomposition . if @xmath53 and @xmath8 , the orientable double covering @xmath57 induces a homomorphism @xmath58 , where the @xmath5-point subset of marked points in @xmath1 is equal to the inverse image by @xmath59 of a @xmath3-point subset of @xmath0 . one of the main aims of this paper is to generalise the injectivity result of @xcite to this homomorphism . if @xmath60 and @xmath61 , it was shown in @xcite that there exists an injective homomorphism @xmath62 , and that via @xmath63 , @xmath64 may be identified with the subgroup of @xmath65 that consists of isotopy classes of symmetric homeomorphisms . in section [ sec : embeddings ] , we show that a similar result holds for all @xmath8 , namely that the homomorphism @xmath66 induced by @xmath59 is injective , and that via @xmath66 , @xmath67 may be identified with the subgroup of @xmath68 that consists of isotopy classes of symmetric homeomorphisms . [ th : braidmcg ] let @xmath69 , let @xmath55 be a compact , connected , non - orientable surface of genus @xmath56 without boundary , and let @xmath1 be its orientable double covering . then the homomorphism @xmath70 induced by the covering @xmath71 is injective . further , if @xmath61 then we have a commutative diagram of the following form : @xmath72{d}{\varphi_{k\ # } } & mcg(n;k)\arrow{r}{\tau_{k}}\arrow[hook]{d}{\phi_k } & mcg(n)\arrow[hook]{d}{\phi } \arrow{r } & 1\\ 1\arrow{r } & b_{2k}(s)\arrow{r } & mcg(s;2k ) \arrow{r}{{\widetilde{\tau}}_{k } } & mcg(s ) \arrow{r } & 1 , \end{tikzcd}\ ] ] where @xmath73 and @xmath74 are the homomorphisms induced by forgetting the markings on the sets of marked points . note that in contrast with @xcite , theorem [ th : braidmcg ] holds for all @xmath53 , not just for @xmath61 . as we recall in remark [ rem : notinjective ] , the result of @xcite does not hold if @xmath75 . the proof of this exceptional case @xmath75 in theorem [ th : braidmcg ] will turn out to be the most difficult , in part due to the non injectivity of @xmath76 . in section [ sec : vcd ] , we compute the cohomological dimension of the braid groups of all compact surfaces without boundary different from @xmath48 and @xmath49 , and we give an upper bound for the virtual cohomological dimension of the mapping class group @xmath77 for all @xmath78 and @xmath42 . recall that the _ virtual cohomological dimension _ @xmath79 of a group @xmath80 is a generalisation of the cohomological dimension @xmath81 of @xmath80 , and is defined to be the cohomological dimension of any torsion - free finite index subgroup of @xmath80 @xcite . in particular , if @xmath80 is a braid or mapping class group of finite ( virtual ) cohomological dimension then the corresponding pure braid or mapping class group has the same ( virtual ) cohomological dimension . in ( * theorem 5 ) , it was shown that if @xmath82 ( resp . @xmath83 ) then @xmath84 ( resp . @xmath85 ) . these results are generalised in the following theorem , the proof being a little more straightforward due to the fact that the braid groups of @xmath86 and @xmath87 are torsion free if @xmath53 . [ th : cdbraid ] let @xmath88 . then : @xmath89 we then turn our attention to the mapping class groups . if @xmath90 , j. harer proved that ( * ? ? ? * theorem 4.1 ) : @xmath91 using the embedding of @xmath92 in @xmath65 given in @xcite , g. hope and u. tillmann showed that if @xmath61 then @xmath93 ( * ? ? ? * corollary 2.2 ) . using ( * ? ? ? * theorem 5 ) , it was shown in ( * ? ? ? * corollary 6 ) that if @xmath82 ( resp . @xmath83 ) then @xmath94 ( resp . @xmath95 ) . in the case of @xmath48 , we thus recover the results of harer . in the second part of section [ sec : vcd ] , we generalise some of these results to non - orientable surfaces with marked points . [ prop : vcdmap ] let @xmath96 . the mapping class groups @xmath97 and @xmath98 have the same ( finite ) virtual cohomological dimension . further : a. @xmath99 . b. if @xmath61 then @xmath100 . this paper is organised as follows . in section [ sec : odc ] , we recall some definitions and results about orientation - true mappings and the orientable double covering of non - orientable surfaces , we describe how we lift an element of @xmath101 to one of @xmath18 in a continuous manner , and we show how this correspondence induces a homomorphism from the mapping class group of a non - orientable surface to that of its orientable double covering ( proposition [ prop : hopetillmann ] ) . in section [ sec : embeddings ] , we prove theorem [ th : braidmcg ] using long exact sequences of fibrations involving the groups that appear in equation ( [ eq : mcgtop ] ) and the corresponding braid groups @xcite . in most cases , using @xcite , we obtain commutative diagrams of short exact sequences , and the conclusion is obtained in a straightforward manner . the situation is however much more complicated in the case where @xmath0 is the klein bottle and @xmath1 is the @xmath6-torus @xmath102 , due partly to the fact that the exact sequences that appear in the associated commutative diagrams are no longer short exact . this case requires a detailed analysis , notably in the case @xmath103 . in section [ sec : vcd ] , we prove theorem [ th : cdbraid ] and proposition [ prop : vcdmap ] . finally , in an appendix , we provide presentations of @xmath104 and @xmath105 that we use in the proof of theorem [ th : braidmcg ] . the first - named author was partially supported by fapesp - fundao de amparo a pesquisa do estado de so paulo , projeto temtico topologia algbrica , geomtrica 2012/24454 - 8 . he and the second - named author would like to thank the ` rseau franco - brsilien en mathmatiques ' for financial support for their respective visits to the laboratoire de mathmatiques nicolas oresme umr cnrs 6139 , universit de caen normandie , from the 2^nd^ to the 15^th^ of september 2016 , and to the instituto de matemtica e estatstica , universidade de so paulo , from the 9^th^ of july to the 1^st^ of august 2016 . the third - named author was supported by the consejo nacional de ciencia y tecnologa to visit the laboratoire de mathmatiques nicolas oresme umr cnrs 6139 , universit de caen normandie , from the 1^st^ of february 2014 to the 31^st^ of january 2015 . during that period , he was also partially supported by the laboratorio internacional solomon lefschetz ( laisla ) . let @xmath106 be manifolds with base points @xmath107 , respectively . following @xcite , a pointed map @xmath108 is called _ orientation true _ if the induced homomorphism @xmath109 sends orientation - preserving ( resp . orientation - reversing ) elements to orientation - preserving ( resp . orientation - reversing ) elements . in other words , the map @xmath29 is orientation - true if for all @xmath110 , either @xmath111 and @xmath112 are both orientation preserving or they are are both orientation reversing . in the case of a branched covering @xmath113 it follows by ( * proposition 1.4 ) that @xmath29 is orientation - true . in what follows , we will be interested in the case of maps between surfaces . we will start with the case of non - orientable surfaces . [ lem : l2 ] every homeomorphism @xmath114 of a non - orientable surface is orientation true . consequently , the subgroup of all orientation - preserving loops is invariant with respect to @xmath115 . lemma [ lem : l2 ] is an obvious consequence of ( * ? ? ? * proposition 1.4 ) , but we shall give a more direct proof . let @xmath116 be a base point and consider @xmath117 , which we represent by a loop @xmath118 . we fix a local orientation @xmath119 at @xmath120 , and we consider the orientation on @xmath0 induced by @xmath29 , denoted by @xmath121 . moving the orientation @xmath119 along the path @xmath122 by a finite sequence of small open sets @xmath123 , the images @xmath124 may be used to transport the local orientation @xmath121 along the path @xmath125 . consider the maps @xmath126 and the tangent bundle @xmath127 of @xmath0 . pulling back by these maps , we obtain bundles over @xmath128 that are homeomorphic , and that are the trivial bundle if the loop is orientation - preserving or the twisted bundle if the loop is orientation - reversing . let @xmath53 , and let @xmath129 be as defined in the introduction . the unique orientable double covering of @xmath130 may be constructed as follows ( see @xcite for example ) . let the orientable surface @xmath131 be embedded in @xmath132 in such a way that it is invariant under the reflections with respect to the @xmath133- , @xmath134- and @xmath135-planes . consider the involution @xmath136 defined by @xmath137 . the orbit space @xmath138 is homeomorphic to the surface @xmath130 , and the associated quotient map @xmath57 is a double covering . to simplify the notation , from now on , we will drop the subscripts @xmath56 and @xmath139 from @xmath0 and @xmath1 respectively unless there is risk of confusion , so @xmath1 will be the orientable double covering of @xmath0 . as indicated previously , the map @xmath59 gives rise to a map on the level of configuration spaces that induces an injective homomorphism @xmath140 of the corresponding braid groups , and this allows us to study the braid groups of a non - orientable surface in terms of those of its orientable double covering @xcite . the following result is an immediate consequence of lemma [ lem : l2 ] and is basically contained in ( * ? ? ? * key - lemma 2.1 ) . [ lem : th1 ] let @xmath114 be a homeomorphism of a non - orientable surface , and let @xmath71 be the orientable double covering . then @xmath29 admits a lift , and the number of lifts is exactly two . using lemma [ lem : l2 ] , we know that @xmath141 , where we identify @xmath142 with a subgroup of @xmath143 . by basic properties of covering spaces , the map @xmath29 lifts to a map @xmath144 , and since @xmath145 is a double covering , there are two lifts . the lifts of @xmath29 are in one - to - one correspondence with the group @xmath146 of deck transformations . there is thus a natural way to choose a lift in a continuous manner simply by choosing @xmath147 to be the lift of degree @xmath148 ( the other lift is of degree @xmath149 since @xmath150 is of degree @xmath149 ) . let @xmath151 denote this choice of lift . we may use lemma [ lem : th1 ] to compare mapping class groups of orientable and non - orientable surfaces . [ prop : hopetillmann ] let @xmath0 be a non - orientable surface , and let @xmath1 be its orientable double covering . then there is a homomorphism @xmath152 such that the following diagram commutes : @xmath153^{\rho}\ar[d]&{\text{top}}^{+}(s)\ar[d]\\mcg(n)\ar[r]^{\phi}&mcg(s ) . } \end{gathered}\ ] ] further , if the genus of @xmath0 is greater than or equal to @xmath154 then @xmath63 is injective . if @xmath155 are isotopic homeomorphisms then an isotopy between them lifts to an isotopy between the orientation - preserving homeomorphisms @xmath156 and @xmath157 of @xmath1 . this proves the first part of the statement . for the second part , suppose that the genus of @xmath0 is greater than or equal to @xmath154 . let @xmath158 denote the extended mapping class group of @xmath1 consisting of the isotopy classes of all homeomorphisms of @xmath1 , and let @xmath159 be the subgroup of @xmath158 defined by : @xmath160\in mcg^{\pm}(s ) \,\right\rvert\,\text{$f\in { \text{top}}(s)$ , and there exists $ f'\simeq f$ such that $ f'\circ j = j\circ f'$}\right\},\ ] ] where @xmath161 means that @xmath162 is isotopic to @xmath29 . by @xcite , @xmath163\right\rangle$ ] . let @xmath164\right\rangle$ ] denote the quotient map . let @xmath165 . then @xmath166)=[\rho(f)]$ ] by diagram ( [ eq : commrho ] ) . by lemma [ lem : th1 ] and the comment that follows it , @xmath29 admits exactly two lifts , @xmath156 and @xmath167 , the first ( resp . second ) preserving ( resp . reversing ) orientation . now @xmath168 is an orientation - reversing lift of @xmath29 , from which we conclude that @xmath169 . hence @xmath170\in c\!\left\langle j\right\rangle\cap mcg(s)$ ] , and so @xmath171 . on the other hand , @xmath172 is an isomorphism using the proof of ( * ? ? ? * key - lemma 2.1 ) . it then follows that @xmath173 , in particular @xmath63 is injective . in the case of surfaces with marked points , there is another continuous way to choose a lift . for example , given a finite subset @xmath174 of @xmath0 , denote its preimage under @xmath59 by : @xmath175 where @xmath176 for all @xmath177 . pick a subset of @xmath178 that contains exactly one element of @xmath179 , denoted by @xmath180 , for all @xmath177 . if @xmath181 then the restriction of @xmath29 to @xmath10 is a given permutation of @xmath10 , and if @xmath182 is such that @xmath183 , we may define @xmath184 to be the unique lift of @xmath29 such that @xmath185 . the two choices for @xmath186 correspond to the two possible lifts of @xmath29 . [ rem : notinjective ] suppose that the genus of @xmath0 is @xmath6 , in which case @xmath0 is the klein bottle @xmath187 . if @xmath102 is the torus , the homomorphism @xmath188 is not injective , and it may be described as follows . set @xmath189 . the group @xmath190 is isomorphic to @xmath191 @xcite , and its elements are given by the mapping classes of the following automorphisms : @xmath192 let @xmath193 be the orientable double covering of @xmath187 and let @xmath194 \right.\right\rangle$ ] , so that @xmath195 and @xmath196 . given a map @xmath197 such that the induced homomorphism on the fundamental group is given by @xmath198 , @xmath199 , if @xmath200 is odd , then @xmath29 lifts to a map from @xmath102 to @xmath102 , and there are exactly two lifts . the matrices of the induced homomorphisms of these lifts on @xmath201 are @xmath202 and @xmath203 . observe that the determinant of one of these two matrices is positive . identifying @xmath204 with @xmath205 , we conclude that @xmath63 sends the first ( resp . second ) two automorphisms of equation ( [ eq : eltsmcgk ] ) to the matrix @xmath206 ( resp . to @xmath207 ) . in particular , the second part of proposition [ prop : hopetillmann ] does not hold in this case . as in all of this paper , the surfaces @xmath0 and @xmath1 under consideration are compact and without boundary , and @xmath208 is the double covering defined in section [ sec : odc ] . if @xmath10 is a finite @xmath3-point subset of @xmath0 , @xmath209 and @xmath181 , then we define a map @xmath210 by @xmath211 , where we consider @xmath29 to be an element of @xmath101 and @xmath212 is as defined in section [ sec : odc ] . by proposition [ prop : hopetillmann ] , the map @xmath213 induces a homomorphism @xmath214 defined by @xmath215)=[\rho_{k}(f)]$ ] . on the other hand , the map @xmath216 ( resp . @xmath217 ) defined by @xmath218 for all @xmath165 ( resp . @xmath219 for all @xmath220 ) is a locally - trivial fibration @xcite whose fibre is @xmath221 ( resp . @xmath222 ) . the long exact sequence in homotopy of these fibrations gives rise to the following exact sequences : @xmath223 where @xmath224 are induced by suppressing the markings on @xmath10 and @xmath178 respectively @xcite . in theorem [ th : braidmcg ] , we will see that the injectivity of @xmath63 given by proposition [ prop : hopetillmann ] carries over to that of the homomorphism @xmath225 between mapping class groups of marked surfaces and holds for _ all _ @xmath53 , in contrast to the non injectivity in the case @xmath75 described in remark [ rem : notinjective ] . let @xmath1 be a compact , connected surface without boundary , let @xmath226 , and let @xmath227 be a subset of cardinality @xmath228 . if @xmath1 is orientable , let : @xmath229\in pmcg(s;l)\right . \ } \ ; \text{and}\\ { \text{top}}^{+}_{{\widetilde{f}}}(s;2l ) & = \left\ { { \widetilde{f}}\in { \text{top}}^{+}(s ) \,\left\lvert \text{${\widetilde{f}}(\pi^{-1}(x_{j}))=\pi^{-1}(x_{j})$ for all $ j=1,\ldots , l$ } \right . \right\},\end{aligned}\ ] ] and if @xmath1 is non orientable , let @xmath230\in pmcg(s;l)\right . \}.\end{aligned}\ ] ] observe that if @xmath1 is orientable ( resp . non orientable ) , @xmath231 ( resp . @xmath232 ) . before giving the proof of theorem [ th : braidmcg ] , we state and prove the following two results . [ prop : pi1topt ] let @xmath42 , and consider the homomorphism @xmath233 induced by the map @xmath234 defined above . then @xmath235 , and @xmath236 . first , by ( * ? * theorem 2 , p. 63 ) ) , @xmath237 . secondly , taking @xmath238 to be equipped with @xmath239 as its basepoint , a representative loop of an element @xmath240 is a path in @xmath238 from @xmath239 to itself . it follows from the definition of @xmath234 that @xmath241 . it remains to show that @xmath242 . since @xmath243 , the homomorphism @xmath244 coincides with the homomorphism of ( * ? ? ? * theorem 1 ) . using the exact sequence of that theorem and ( * corollary 1.3 ) , we see that @xmath245 , where @xmath246 and @xmath247 are the generators of @xmath248 defined in ( * ? ? ? * theorem 5 ) . but by ( * ? ? ? * proposition 4.2 ) , these two elements generate the centre of @xmath249 as required . the first part of the following lemma generalises results of @xcite . [ lem : hamstrom ] a. [ it : hamstroma ] let @xmath1 be a compact , connected orientable ( resp . non - orientable ) surface without boundary for which @xmath250 , ( resp . @xmath251 ) is trivial , and let @xmath252 . then @xmath253 , ( resp . @xmath254 ) is trivial . in particular , @xmath255 ( resp . @xmath256 ) is trivial for all @xmath252 . b. [ it : hamstromb ] @xmath257 is trivial for all @xmath252 . a. assume first that @xmath1 is orientable . we prove the result by induction on @xmath228 . if @xmath258 then the result holds by the hypothesis , using the fact that @xmath231 . suppose by induction that the result holds for some @xmath252 . the map @xmath259 given by evaluating an element of @xmath260 at the point @xmath261 is a fibration with fibre @xmath262 . taking the long exact sequence in homotopy of this fibration and using the fact that @xmath263 is trivial , it follows that the homomorphism @xmath264 induced by the inclusion of the fibre in the total space is injective , and since @xmath253 is trivial by the induction hypothesis , @xmath265 is too . this proves the first part of the statement if @xmath1 is orientable . if @xmath1 is non orientable , it suffices to replace @xmath260 by @xmath266 in the proof of the orientable case . the second part of the statement is a consequence of the first part and ( * ? ? ? * corollary , p. 65 ) and ( * ? ? ? * theorem 4.1 ) . b. let @xmath267 . the map @xmath268 that to a homeomorphism @xmath269 associates the set @xmath270 is a fibration whose fibre is @xmath271 . if @xmath272 then @xmath273 is just @xmath238 . taking the long exact sequence in homotopy , and using the fact that @xmath274 is a @xmath275 , we obtain the following exact sequence : @xmath276 where the homomorphism @xmath277 is induced by the map @xmath278 that forgets the marking on the last two points . if @xmath257 is trivial for some @xmath252 then clearly @xmath279 is also trivial . so applying induction on @xmath228 , it suffices to prove the result for @xmath258 . it follows from proposition [ prop : pi1topt ] that the map @xmath280 sends @xmath281 isomorphically onto the centre of @xmath105 . it follows from exactness of ( [ eq : ftilde ] ) that @xmath282 is trivial , and this completes the proof of the lemma . if @xmath1 is a surface that satisfies the hypotheses of lemma [ lem : hamstrom]([it : hamstroma ] ) then @xmath1 is different from the @xmath6-sphere ( * ? ? ? * , p. 303 ) and the real projective plane ( * ? ? ? * theorem 3.1 ) , so @xmath1 is an eilenberg mac lane space of type @xmath275 . consider the double covering @xmath283 , fix a @xmath3-point subset @xmath10 of @xmath0 , and let @xmath209 . by the comments preceding the statement of the theorem , we obtain the following commutative diagram of fibrations : @xmath284\ar[d]^{\rho_{k } } & { \text{top}}(n ) \ar[r]^{\psi_{k}}\ar[d]^{\rho } & d_{k}(n ) \ar[d]^{\varphi_{k}}\\ { \text{top}}^+(s;{\widetilde{x } } ) \ar[r ] & { \text{top}}^{+}(s ) \ar[r]^{{\widetilde{\psi}}_{2k } } & d_{2k}(s ) , } \end{gathered}\ ] ] where @xmath212 is as defined in section [ sec : odc ] , and @xmath285 is given by equation ( [ eq : liftconfig ] ) . the left - hand square clearly commutes because @xmath213 is the restriction of @xmath212 to @xmath221 . we claim that the right - hand square also commutes . on the one hand , @xmath286 , using the fact that @xmath156 is a lift of @xmath29 , so @xmath287 . conversely , if @xmath288 , then there exists @xmath289 such that @xmath290 . if @xmath291 is a lift of @xmath292 then @xmath293 , which proves the claim . let @xmath294 denote the homomorphism induced by @xmath212 on the level of fundamental groups . we now take the long exact sequence in homotopy of the commutative diagram ( [ eq : topdiag ] ) . the form of the resulting commutative diagram depends on the genus @xmath56 of @xmath0 , hence we consider the following three cases . a. suppose that @xmath61 . then by @xcite , @xmath295 and @xmath296 are trivial , and we obtain the commutative diagram of ( [ eq : mcgdiag ] ) . since @xmath297 ( resp . @xmath63 ) is injective by @xcite ( resp . by proposition [ prop : hopetillmann ] ) , the injectivity of @xmath225 follows from the 5-lemma . b. suppose that @xmath298 . using @xcite , we obtain the following commutative diagram of short exact sequences : @xmath299 where in both cases , @xmath300 is identified with the subgroup generated by the full - twist braid of the corresponding braid group . since @xmath297 is injective by @xcite , so is @xmath301 , and a routine diagram - chasing argument shows that @xmath225 is injective . c. finally suppose that @xmath75 , so that @xmath0 is the klein bottle @xmath187 and @xmath1 is the torus @xmath102 . let @xmath42 . we claim that @xmath302 . to prove this , let @xmath303 . let @xmath304 and @xmath305 denote the usual homomorphisms onto the symmetric groups of @xmath10 and @xmath178 respectively as described in equation ( [ eq : sigmamcg ] ) . from the geometric construction of @xmath213 , if @xmath306 and @xmath307)(x_{i})=x_{j}$ ] then @xmath308)(y_{i } ) \in \pi^{-1}(x_{j})$ ] . if @xmath15\in { \ensuremath{\operatorname{\text{ker}}(\phi_{k})}}$ ] , where @xmath309 , then @xmath310=\phi_{k}([f])=[\operatorname{\text{id}}_{s}]$ ] . in particular , @xmath308)= \operatorname{\text{id}}_{s_{2k}}$ ] , so for all @xmath177 , @xmath308)(y_{i } ) = y_{i}$ ] , and it follows that @xmath307)(x_{i})=x_{i}$ ] , whence @xmath15\in pmcg(k;k)$ ] as claimed . it thus suffices to prove that the restriction @xmath311 of @xmath225 to @xmath312 is injective . now assume that @xmath313 . let @xmath314 ( resp . @xmath315 ) be the map given by forgetting the last coordinate ( resp . the marking on the last point ) . let @xmath316 , let @xmath317 . let @xmath318 be the map given by forgetting the last two coordinates , and let @xmath319 be the map defined in the proof of lemma [ lem : hamstrom]([it : hamstromb ] ) . then we have the following commutative diagram whose rows are fibrations : @xmath320 & { \text{top}}^{+}(t ) \ar[r]^{{\widetilde{\delta}}_{2(k-1 ) } } & d_{2(k-1)}^{(2)}(t)\\ { \text{top}}^{+}_{{\widetilde{f}}}(t;2k ) \ar[r ] \ar[u]_{{\widetilde{q}}_{2k } } & { \text{top}}^{+}(t ) \ar@{=}[u ] \ar[r]^{{\widetilde{\delta}}_{2k } } & d_{2k}^{(2)}(t ) \ar[u]_{{\widetilde{p}}_{2k}}\\ { \text{top}}_{f}(k;k ) \ar[r ] \ar[u]_{\rho_{k}\left\lvert_{{\text{top}}_{f}(k;k)}\right . } \ar[d]^{q_{k } } & { \text{top}}(k ) \ar[u]_{\rho } \ar@{=}[d ] \ar[r]^{\delta_{k } } & f_{k}(k ) \ar[u]_{\widehat{\varphi}_{k } } \ar[d]^{p_{k}}\\ { \text{top}}_{f}(k;k-1 ) \ar[r ] \ar[d]^{\rho_{k-1}\left\lvert_{{\text{top}}_{f}(k;k-1)}\right . } & { \text{top}}(k ) \ar[d]^{\rho } \ar[r]^{\delta_{k-1 } } & f_{k-1}(k)\ar[d]^{\widehat{\varphi}_{k-1 } } \\ { \text{top}}^{+}_{{\widetilde{f}}}(t;2(k-1 ) ) \ar[r ] & { \text{top}}^{+}(t ) \ar[r]^{{\widetilde{\delta}}_{2(k-1 ) } } & d_{2(k-1)}^{(2)}(t ) . } \end{gathered}\ ] ] the map @xmath321 ( resp . @xmath322 ) is defined by : @xmath323 for all @xmath324 ( resp . for all @xmath325 ) , and the map @xmath326 is defined by @xmath327 for all @xmath328 . note also that the diagram remains commutative if we identify the corresponding terms of the first and last rows . taking the long exact sequence in homotopy of the diagram ( [ eq : bigcommdiag ] ) and applying lemma [ lem : hamstrom ] , we obtain the following commutative diagram whose rows are exact : @xmath329 where @xmath330 and @xmath331 are the boundary homomorphisms associated with the corresponding fibrations of ( [ eq : bigcommdiag ] ) , @xmath332 , @xmath333 , @xmath334 and @xmath73 are the restrictions of these homomorphisms to the given groups , @xmath335 and @xmath336 are the maps induced by @xmath337 and @xmath338 respectively on the level of @xmath339 , and @xmath340 , @xmath341 , @xmath342 and @xmath343 are the homomorphisms induced by @xmath344 and @xmath345 on the level of @xmath346 . the homomorphism @xmath63 is given by proposition [ prop : hopetillmann ] , and @xmath347 is the restriction of @xmath297 to @xmath348 , so is injective by @xcite . we remark that @xmath349 ( resp . @xmath350 ) , and may be identified with the centre of @xmath351 ( resp . of @xmath249 ) , which in fact lies in @xmath348 ( resp . in @xmath248 ) . once more , the diagram ( [ eq : bigcommdiagexact ] ) remains commutative if we identify the corresponding groups of the first and last rows . in particular , the second and third columns yield respectively : @xmath352 we recall at this point that our aim is to show that the homomorphism @xmath311 is injective . note that by diagram ( [ eq : bigcommdiagexact ] ) , @xmath353 , and it follows from the exactness of ( [ eq : bigcommdiagexact ] ) and the injectivity of @xmath347 that @xmath301 is also injective . we claim that the restriction @xmath354 is an isomorphism . indeed : a. it is well defined . indeed , if @xmath355 then by the commutative diagram ( [ eq : bigcommdiagexact ] ) , @xmath356 , so @xmath357 . b. suppose that @xmath358 . by exactness of the second row of the diagram ( [ eq : bigcommdiagexact ] ) , there exists @xmath359 such that @xmath360 , and by commutativity of the same diagram , @xmath361 . but @xmath340 is injective , so @xmath362 , whence @xmath363 . thus @xmath364 is injective too . c. let @xmath365 . using commutativity and exactness of the second row of the diagram ( [ eq : bigcommdiagexact ] ) , we have @xmath366 , so there exists @xmath367 such that @xmath368 , and thus @xmath369 . hence there exists @xmath359 such that @xmath370 . it follows that @xmath371 , and that @xmath372 by exactness of the second row of ( [ eq : bigcommdiagexact ] ) , which proves that @xmath364 is surjective . + in a similar manner , one may show that the restriction @xmath373 is also an isomorphism , and if @xmath374 then @xmath375 by equation ( [ eq : thirdcolumn ] ) , and so @xmath376 . we thus obtain the following commutative diagram of short exact sequences : @xmath377&{\ensuremath{\operatorname{\text{ker}}((q_{k})_{\#})}}\ar[r ] \ar[d]^{\phi_{k}\left\lvert_{{\ensuremath{\operatorname{\text{ker}}((q_{k})_{\#})}}}\right . } & pmcg(k;k ) \ar[r]^(0.45){(q_{k})_{\ # } } \ar[d]^{\phi_{k}\left\lvert_{pmcg(k;k)}\right . } & pmcg(k;k-1 ) \ar[r ] \ar[d]^{\phi_{k-1}\left\lvert_{pmcg(k;k-1)}\right . } & 1\\ 1\ar[r]&{\ensuremath{\operatorname{\text{ker}}(({\widetilde{q}}_{2k})_{\#})}}\ar[r ] & \pi_{0}{\text{top}}^{+}_{{\widetilde{f}}}(t;2k ) \ar[r]_(0.45){({\widetilde{q}}_{2k})_{\ # } } & \pi_{0}{\text{top}}^{+}_{{\widetilde{f}}}(t;2(k-1 ) ) \ar[r ] & 1 , } \end{gathered}\ ] ] the second arrow in each row being inclusion . we claim that @xmath378 is injective . this being the case , if @xmath379 is injective then @xmath311 is also injective by the commutativity and exactness of the rows of ( [ eq : mcgdiag4 ] ) . by induction on @xmath3 , to complete the proof in the case @xmath75 , it will thus suffice to prove that the homomorphism @xmath380 is injective , which we shall do shortly ( note that @xmath381 , so we may remove the restriction symbol from @xmath382 ) . we first prove the claim . let @xmath383 be such that @xmath384 . from the isomorphism ( [ eq : iso2 ] ) , there exists a unique @xmath385 such that @xmath386 . by ( [ eq : lastcolumn ] ) , we have @xmath387 , so @xmath388 . on the other hand , @xmath389 by commutativity of the diagram ( [ eq : bigcommdiagexact ] ) , hence @xmath390 . so @xmath391 by the isomorphism ( [ eq : iso1 ] ) . but @xmath347 is the restriction of @xmath297 to @xmath348 , so is injective @xcite . thus @xmath363 , @xmath392 , and therefore @xmath378 is injective as claimed . + it thus remains to show that the homomorphism @xmath380 is injective . suppose that @xmath393 . taking @xmath103 , the second and third rows of diagram ( [ eq : bigcommdiagexact ] ) become : @xmath394 & z(p_1(k ) ) \ar[r]^(0.55){(\delta_{1})_{\ # } } \ar[d]^{\phi_{1 } } & p_1(k)\ar[r]^(.38){\partial_{1 } } \ar[d]^{\varphi_{1\ # } } & mcg(k;1)\ar[r]^(.54){\tau_{1}}\ar[d]^{\phi_{1 } } & { \underbrace{mcg(k)}_{\mathbb{z}_{2}\oplus \mathbb{z}_{2 } } } \ar[d]^(.6){\phi}\ar[r]&1\\ 1\ar[r ] & z(p_2(t ) ) \ar[r]^(0.55){({\widetilde{\delta}}_{2})_{\ # } } & b_{2}(t)\ar[r]^(.38){{\widetilde{\partial}}_{2 } } & mcg(t;2)\ar[r]^(.52){{\widetilde{\tau}}_{2 } } & { \underbrace{mcg(t)}_{\operatorname{\text{sl}}(2,\mathbb{z } ) } } \ar[r ] & 1 , } \end{gathered}\ ] ] since @xmath395 , we see that @xmath396 , and so by remark [ rem : notinjective ] , @xmath397 is either equal to the mapping class of the identity of @xmath190 , in which case we say that @xmath398 is of _ type one _ , or is equal to the mapping class of the automorphism @xmath399 in which case we say that @xmath398 is of _ type two_. recall that @xmath400 and @xmath401 , and every element of @xmath402 may be written ( uniquely ) in the form @xmath403 , where @xmath404 . by the appendix , @xmath405 and @xmath406 , where the given generators of @xmath105 are subject to the relations of theorem [ presfull ] . the group @xmath105 contains @xmath407 as an index two subgroup , where @xmath408 is the free group on @xmath409 . the homomorphism @xmath410 is given by @xmath411 and @xmath412 . using the relations of theorem [ presfull ] , one may check that @xmath413 . + first assume that @xmath414 is of type one . then @xmath415 , and by exactness of the first row of ( [ eq : mcgdiag2 ] ) , there exists @xmath416 such that @xmath417 , and so @xmath418 , where @xmath404 are unique . if @xmath419 then @xmath420 by exactness of the first row of ( [ eq : mcgdiag2 ] ) , and the conclusion clearly holds . so suppose that either @xmath421 is odd or @xmath422 . then @xmath423 using the fact that @xmath424 . since @xmath425 , by observing the induced permutation of @xmath426 , we conclude that @xmath421 must be even , so @xmath427 , and that @xmath428 by exactness of the second row of ( [ eq : mcgdiag2 ] ) . using the decomposition ( [ eq : decompp2 t ] ) , it follows that @xmath429 , which yields a contradiction . so if @xmath414 is of type one then @xmath430 . now suppose that @xmath414 is of type two . consider the basepoint - preserving homeomorphism @xmath122 of @xmath187 illustrated in figure [ fig : homeoh ] . observe that @xmath431)=\tau_{1}(w)$ ] , so by exactness of the first row of ( [ eq : mcgdiag2 ] ) , there exists @xmath432 such that @xmath433\,\partial_{1}(w')$ ] . let @xmath404 be such that @xmath418 . + ( 0,0 ) ( 1,0 ) ; ( 1,0 ) ( 2,0 ) ; ( 0,2 ) ( 1,2 ) ; ( 1,2 ) ( 2,2 ) ; ( 0,0 ) ( 0,1 ) ; ( 0,1 ) ( 0,2 ) ; ( 2,2 ) ( 2,1 ) ; ( 2,1 ) ( 2,0 ) ; ( 2,1 ) node ( a ) [ right ] @xmath111 ( 1,0 ) node ( b ) [ below ] @xmath434 ( 1,2 ) node ( c ) [ above ] @xmath434 ( 0,1 ) node ( d ) [ left ] @xmath111 ; + ( 0,0 ) ( 1,0 ) ; ( 1,0 ) ( 2,0 ) ; ( 0,2 ) ( 1,2 ) ; ( 1,2 ) ( 2,2 ) ; ( 0,0 ) ( 0,1 ) ; ( 0,1 ) ( 0,2 ) ; ( 2,2 ) ( 2,1 ) ; ( 2,1 ) ( 2,0 ) ; ( 0,0 ) ( 2,2 ) ; ( 0,0 ) ( 1,1 ) ; ( 1,1 ) ( 2,2 ) ; ( 2,1 ) node ( a ) [ right ] @xmath111 ( 1,0 ) node ( b ) [ below ] @xmath434 ( 1,2 ) node ( c ) [ above ] @xmath434 ( 0,1 ) node ( d ) [ left ] @xmath435 ( 1.25,1.4 ) node ( e ) [ left ] @xmath436 ; + now @xmath122 lifts to an orientation - preserving homeomorphism @xmath437 that is illustrated in figure [ fig : lifthomeoh ] , where @xmath438 is the inverse image of the basepoint of @xmath187 under @xmath59 . it follows that @xmath439 , and hence @xmath440=[\rho_{1}(h)]=\phi_{1}([h])$ ] . + ( 0,0 ) ( 3,0 ) ; ( 3,0 ) ( 4,0 ) ; ( 0,2 ) ( 3,2 ) ; ( 3,2 ) ( 4,2 ) ; ( 0,0 ) ( 0,1 ) ; ( 0,1 ) ( 0,2 ) ; ( 4,0 ) ( 4,1 ) ; ( 4,1 ) ( 4,2 ) ; ( 2,0 ) ( 2,1 ) ; ( 2,1 ) ( 2,2 ) ; ( 0,0 ) circle [ radius=0.06 ] ; ( 2,0 ) circle [ radius=0.06 ] ; ( 4,1 ) node ( a ) [ right ] @xmath441 ( 3,0 ) node ( b ) [ above ] @xmath40 ( 3,2 ) node ( c ) [ above ] @xmath40 ( 0,1 ) node ( d ) [ left ] @xmath441 ( 2,1 ) node ( g ) [ left ] @xmath442 ( 0.4,-0.4 ) node ( e ) [ left ] @xmath443 ( 2.4,-0.4 ) node ( f ) [ left ] @xmath444 ; + ( 0,0 ) ( 3,0 ) ; ( 3,0 ) ( 4,0 ) ; ( 0,2 ) ( 3,2 ) ; ( 3,2 ) ( 4,2 ) ; ( 0,0 ) ( 0,1 ) ; ( 0,1 ) ( 0,2 ) ; ( 4,0 ) ( 4,1 ) ; ( 4,1 ) ( 4,2 ) ; ( 0,0 ) ( 1,1 ) ; ( 1,1 ) ( 2,2 ) ; ( 2,2 ) ( 3,1 ) ; ( 3,1 ) ( 4,0 ) ; ( 4,1 ) node ( a ) [ right ] @xmath441 ( 2,0 ) node ( b ) [ below ] @xmath40 ( 2,2 ) node ( c ) [ above ] @xmath40 ( -0.2,1 ) node ( d ) [ left ] @xmath445 ( 2,1 ) node ( e ) @xmath446 ; + on the other hand , relative to the chosen generators of @xmath104 , the generator @xmath292 is represented by the loop @xmath441 based at @xmath443 , and @xmath447 is represented by two loops , namely @xmath441 and a parallel copy @xmath442 of @xmath441 based at @xmath444 . the loop @xmath442 represents the element @xmath448 of @xmath104 . using the definition of @xmath449 , we must find a homotopy @xmath450 of @xmath102 starting at the identity , _ i.e. _ @xmath451 , for which the evaluation of the homotopy at the points @xmath452 and @xmath453 yields the braid @xmath448 . let @xmath450 be the homotopy obtained by pushing @xmath444 along the loop @xmath442 and that satisfies @xmath454 . this homotopy clearly has the desired trace ( or evaluation ) , and so @xmath455 $ ] . since @xmath456 , we have : @xmath457\,\partial_{1}(w'))= [ { \widetilde{h } } ] \ , { \widetilde{\partial}}_{2}\circ \varphi_{1\#}(w')= { \widetilde{\partial}}_{2}(x^{-1}a \varphi_{1\#}(w')),\ ] ] by commutativity of the diagram ( [ eq : mcgdiag2 ] ) , where @xmath426 is given by equation ( [ eq : phiwprime ] ) . once more , by considering the induced permutation of @xmath458 , @xmath421 must be even , and so @xmath459 , where @xmath460 is in the normal form of equation ( [ eq : decompp2 t ] ) . since @xmath461 , we thus obtain a contradiction . we conclude that @xmath462 is injective , and this completes the proof of the theorem . in this section , we study the ( virtual ) cohomological dimension of surface braid groups and mapping class groups with marked points , and we prove theorem [ th : cdbraid ] and proposition [ prop : vcdmap ] . let @xmath88 , and let @xmath54 or @xmath87 . since @xmath36 is of ( finite ) index @xmath463 in @xmath35 , and @xmath464 is a finite - dimensional cw - complex and an eilenberg - mac lane space of type @xmath275 @xcite , the cohomological dimensions of @xmath36 and @xmath35 are finite and equal , so it suffices to determine @xmath465 . let us prove by induction on @xmath3 that @xmath466 and that @xmath467 . the result is true if @xmath103 since then @xmath468 , @xmath469 and @xmath470 . so suppose that the induction hypothesis holds for some @xmath42 . the fadell - neuwirth fibration @xmath471 given by forgetting the last coordinate gives rise to the following short exact sequence of braid groups : @xmath472 where @xmath473 , @xmath474 being an element of @xmath464 , and @xmath475 is defined geometrically by forgetting the last string . since @xmath476 has the homotopy type of a bouquet of circles , @xmath477 is trivial for all @xmath478 and for any choice of local coefficients @xmath479 , and @xmath480 , hence @xmath481 . by * chapter viii ) , it follows that @xmath482 . to conclude the proof of the theorem , it remains to show that there exist local coefficients @xmath479 such that @xmath483 . we will show that this is the case for @xmath484 . by the induction hypothesis , we have @xmath485 . consider the serre spectral sequence with integral coefficients associated to the fibration @xmath486 . then we have that : @xmath487 since @xmath466 and @xmath488 from above , this spectral sequence has two horizontal lines whose possible non - vanishing terms occur for @xmath489 and @xmath490 . we claim that the group @xmath491 is non trivial . to see this , first note that @xmath492 is isomorphic to the free abelian group of rank @xmath493 , where @xmath494 if @xmath54 and @xmath495 if @xmath496 , so @xmath497 , and hence @xmath498 , where we identify @xmath499 with ( the dual of ) @xmath500 , the abelianisation of @xmath0 . the action of @xmath36 on @xmath0 by conjugation induces an action of @xmath36 on @xmath500 . let @xmath450 be the subgroup of @xmath500 generated by the elements of the form @xmath501 , where @xmath502 and @xmath503 and @xmath504 represents the action of @xmath111 on @xmath292 . then we obtain a short exact sequence @xmath505 of abelian groups , and the long exact sequence in cohomology applied to @xmath36 yields : @xmath506 the last term is zero since @xmath466 , and so the map between the remaining two terms is surjective . so to prove that @xmath491 is non trivial , it suffices to show that @xmath507 is non trivial . to do so , we first determine @xmath508 . suppose first that @xmath54 , and consider the presentation of @xmath36 on @xmath500 given in ( * ? ? ? * corollary 8) . with the notation of that paper , a basis @xmath509 of @xmath0 is given by : @xmath510 and a set of coset representatives in @xmath511 of a generating set @xmath512 of @xmath36 is given by @xmath513 . using ( * ? ? ? * table 1 and theorem 7 or corollary 8) , one sees that the commutators of the elements of @xmath512 with those of @xmath509 project to the trivial element of @xmath500 , with the exception of @xmath514 $ ] and @xmath515 $ ] that project to ( the coset of ) @xmath516 for all @xmath517 ( we take the opportunity here to correct a couple of small misprints in these results : relation ( 19 ) of theorem 7 should read as in relation ( @xmath518 ) of corollary 8 ; and in table 1 , in each of the three rows , the first occurrence of @xmath519 should read @xmath520 ) . note that we obtain a similar result for @xmath521 , but using the surface relation ( * ? ? ? * equation ( 1 ) ) in @xmath511 and taking @xmath522 yields no new information . it follows that @xmath450 is the subgroup of @xmath500 generated by the @xmath516 , where @xmath523 , and that @xmath508 is the free abelian group generated by the coset representatives of the @xmath524 and @xmath525 , where @xmath526 . in particular , @xmath527 . since the induced action of @xmath36 on @xmath508 is trivial , using the induction hypothesis , we conclude that : @xmath528 it then follows from ( [ eq : lescohom ] ) that @xmath529 . since @xmath530 for all @xmath531 and @xmath532 , we have @xmath533 , thus @xmath534 is non trivial , and hence @xmath535 . this proves the result in the orientable case . now let us turn to the non - orientable case . the idea of the proof is the same as in the orientable case , but the computations for @xmath500 and @xmath450 are a little different . we use the presentation of @xmath36 given in @xcite . with the notation of that paper , a basis @xmath509 of @xmath0 is given by @xmath536 , and a set of coset representatives in @xmath511 of a generating set @xmath512 of @xmath36 is given by @xmath537 . using ( * ? ? ? * theorem 3 ) , one sees that the commutators of the elements of @xmath512 with those of @xmath509 project to the trivial element of @xmath500 , with the exception of @xmath538 $ ] that projects to ( the coset of ) @xmath539 for all @xmath517 and @xmath540 . we obtain a similar result for @xmath521 , and using the surface relation ( * ? ? ? * relation ( c ) , theorem 3 ) in @xmath511 and taking @xmath522 implies that the element @xmath541 also belongs to @xmath450 using additive notation . it follows that @xmath450 is the subgroup of @xmath500 generated by the @xmath542 , where @xmath523 , and the element @xmath541 . we conclude that @xmath543 . the argument then goes through as in the orientable case . the exact sequences given by ( [ eq : mcgdiag ] ) and ( [ eq : bigcommdiagexact ] ) also allow us to obtain information about the virtual cohomological dimension of @xmath77 if @xmath78 and @xmath42 . a. since @xmath312 is a subgroup of finite index of @xmath190 , it suffices to show that @xmath544 . from the exact sequence given by the third row of ( [ eq : bigcommdiagexact ] ) : @xmath545 since @xmath190 is finite , @xmath312 has the same vcd as the quotient @xmath546 , where @xmath547 . we now prove the result by induction on @xmath3 . first , the result is true for @xmath103 since @xmath548 . now suppose that the result holds for some @xmath42 . since the projection @xmath549 maps the centre @xmath550 of @xmath551 isomorphically onto the centre @xmath552 of @xmath348 , we have the following short exact sequence of groups : @xmath553 now @xmath554 by induction , hence @xmath555 , because @xmath556 . using the short exact sequence @xmath557 and the fact that @xmath558 by theorem [ th : cdbraid ] , we see that @xmath559 , so @xmath560 , and the result follows . b. let @xmath61 . consider the short exact sequence : @xmath561 given by equation ( [ eq : mcgdiag ] ) . then by ( * ? ? ? * corollary 2.2 ) and theorem [ th : cdbraid ] , we have : @xmath562 as required . in this appendix , we provide presentations of @xmath563 and @xmath564 that are adapted to our situation . from ( * ? ? ? * section 4 ) , we have : if necessary , the generator @xmath575 may be suppressed from the list of generators of @xmath563 . using theorem [ thm : presp2 t ] , we may obtain the following useful relations in @xmath563 : @xmath576 setting @xmath577 , @xmath578 , @xmath579 and @xmath580 , we obtain a new presentation of @xmath563 whose generators are @xmath575 , @xmath581 , @xmath582 , @xmath583 and @xmath584 , that are subject to the following relations : if we let @xmath589 @xmath590 , @xmath591 and @xmath592 , then it is not hard to show that @xmath104 has a presentation whose generators are @xmath292 , @xmath593 , @xmath447 and @xmath594 that are subject to the following relations : it follows from this presentation that @xmath563 is isomorphic to @xmath598 , where @xmath599 is a basis of the free group @xmath600 , and @xmath601 is a basis of the free abelian group @xmath602 . we then obtain the following presentation of @xmath564 . the generator @xmath26 given in the statement of theorem [ presfull ] is the artin generator @xmath614 . the proof of the theorem is standard , and makes use of the above presentation of @xmath104 , the short exact sequence : @xmath615 standard results on presentations of group extensions ( * ? ? ? * proposition 1 , p. 139 ) , and computations of relations between the given generators by means of geometric arguments . fadell , e. , husseini , s. , _ the nielsen number on surfaces _ , topological methods in nonlinear functional analysis ( toronto , ont . , 1982 ) , _ contemp . _ , amer . math . soc . , providence , ri , * 21 * ( 1983 ) , 5998 . farb , b. , _ some problems on mapping class groups and moduli space _ , in problems on mapping class groups and related topics , proc . sympos . pure math . * 74 * , 1155 , amer . soc . , providence , ri , 2006 . gonalves , d. l. , guaschi , j. , _ on the structure of surface pure braid groups _ , j. pure appl . algebra * 182 * ( 2003 ) , 3364 ( due to a printer s error , this article was republished in its entirety with the reference * 186 * ( 2004 ) 187218 ) . gonalves , d. l. , guaschi , j. , _ inclusion of configuration spaces in cartesian products , and the virtual cohomological dimension of the braid groups of @xmath48 and @xmath49 _ , pacific j. math . , to appear .

in energy harvesting ( eh ) relay networks , the _ coherent _ communication requires accurate estimation / tracking of the _ instantaneous _ channel state information ( csi ) which consumes extra power . as a remedy , we propose two _ noncoherent _ eh relaying protocols based on the amplify - and - forward ( af ) relaying , namely , power splitting noncoherent af ( ps - ncaf ) and time switching noncoherent af ( ts - ncaf ) , which do not require any instantaneous csi . we develop a _ noncoherent _ framework of simultaneous wireless information and power transfer ( swipt ) , embracing ps - ncaf and ts - ncaf in a _ unified _ form . for arbitrary-ary noncoherent frequency - shift keying ( fsk ) and differential phase - shift keying ( dpsk ) , we derive maximum - likelihood detectors ( mlds ) for ps - ncaf and ts - ncaf in a _ unified _ form , which involves integral evaluations yet serves as the optimum performance benchmark . to avoid expensive integral computations , we propose a _ closed - form _ detector using the gauss - legendre approximation , which achieves almost identical performance as the mld but at substantially lower complexity . these eh - based noncoherent detectors achieve full diversity in rayleigh fading . numerical results demonstrate that our proposed ps - ncaf and ts - ncaf may outperform the conventional grid - powered relay system under the same total power constraint . various insights which are useful for the design of practical swipt relaying systems are obtained . interestingly , ps - ncaf outperforms ts - ncaf in the single - relay case , whereas ts - ncaf outperforms ps - ncaf in the multi - relay case . energy harvesting , maximum - likelihood , noncoherent , simultaneous wireless information and power transfer ( swipt ) .

1505.07194

conventional battery powered wireless communications systems require periodic recharging or replacement of the batteries , which incurs a high operation burden @xcite and can be cumbersome or even impossible ( e.g. , for biomedical devices implanted in the human body @xcite ) . recently , energy harvesting ( eh ) from the ambient radio - frequency ( rf ) signals has been developed as one of the attractive alternatives to prolong the lifetime of energy - constrained nodes in wireless networks @xcite . the dual use of rf signals for eh and information delivery ( i d ) has led to the novel architecture of simultaneous wireless information and power transfer ( swipt ) @xcite , which allows wireless nodes to scavenge energy as well as extract information simultaneously from the rf signals , thus constituting an appealing solution for energy - constrained applications such as wireless relay networks . practical swipt receiver architectures make use of two different circuits performing eh and i d individually @xcite . the receiver may either switch between the eh and i d circuits in a time - division fashion , a scheme known as time switching @xcite , or split the received rf signals into two streams fed to the eh and i d circuits at the same time , a scheme known as power splitting @xcite . typically , eh and i d circuits operate with rather different receiver power sensitivities , i.e. , @xmath1 dbm for information receivers and @xmath2 dbm for energy receivers @xcite . since the eh circuit is designed to maximize the eh efficiency while the i d circuit typically aims for maximum information rate , a fundamental _ rate - energy tradeoff _ exists for swipt systems @xcite . various resource allocation and beamforming schemes were designed to achieve different tradeoffs between rate and energy in swipt systems . in particular , the optimum power allocation maximizing the information rate subject to eh constraint was studied for broadband systems with perfect instantaneous channel state information ( csi ) @xcite . the optimum resource allocation with service differentiation between low- and high - priority data was studied for eh networks in @xcite . in @xcite , a joint optimization and game - theoretic framework was developed to optimize the packet delivery policy and cooperation strategy for delay tolerant networks with eh . moreover , beamforming schemes which maximize the information rate subject to eh constraint or maximize the harvested energy subject to rate constraint were studied under imperfect csi in @xcite and @xcite , respectively . recently , swipt has also been applied in wireless relay systems which allows energy - constrained relay nodes to harvest energy from the source rf signals and consequently brings in substantial benefits in wireless networks @xcite . various theoretical analysis and practical design have been conducted for eh relay systems in the literature . specifically , for dual - hop amplify - and - forward ( af ) networks , a greedy switching protocol between data relaying and eh was proposed in @xcite , and the joint relay selection and power allocation scheme was developed in @xcite . moreover , the outage probability and ergodic capacity for eh af relaying were derived for delay - limited and delay - tolerant applications , respectively , in @xcite . besides af relay systems , the eh has also been considered for decode - and - forward ( df ) networks . in particular , the outage probability of eh df relaying was studied , taking into account the spatial randomness of the source - destination pairs , for various network topologies in @xcite . the aforementioned studies on eh relay systems have either implicitly or explicitly assumed the instantaneous csi availability to allow for _ coherent _ information decoding . however , the coherent swipt requires the source to periodically send training symbols , which incurs an increased signaling overhead and processing burden . moreover , the relays may need to relay the training symbols and/or estimate the source - relay channels @xcite , which results in additional energy consumptions and poses serious issues especially for energy - constrained relay nodes . as a remedy , the _ noncoherent _ swipt eliminating the need for the instantaneous csi was first studied for the eh df relay systems @xcite , where two eh relaying protocols , namely , the power splitting noncoherent df ( ps - ncdf ) and time switching noncoherent df ( ts - ncdf ) were proposed , and the corresponding maximum - likelihood detectors ( mlds ) facilitating noncoherent swipt were obtained . however , these protocols and the detectors are applicable only for the df relay systems . it is still unknown about how the noncoherent af relaying can benefit from eh , and how the noncoherent af relaying performs as compared to the noncoherent df relaying in eh relay systems . in addition , for eh af relay systems , the noncoherent mld minimizing the symbol - error rate ( ser ) in noncoherent swipt is still unknown . realizing that none of the previous works have tackled the design challenges of noncoherent swipt in eh af relay systems , we aim to fill the gap . in this paper , we study the _ noncoherent _ eh relay systems consisting of multiple af relays , which can harvest energy from the ambient source signals and utilize the harvested energy to assist the communication . the main contributions of this paper are summarized as follows : * for arbitrary @xmath0-ary noncoherent frequency - shift keying ( fsk ) or differential phase - shift keying ( dpsk ) signalings , we propose a unified _ noncoherent _ swipt framework embracing two eh relaying protocols , namely , the power splitting noncoherent af ( ps - ncaf ) and time switching noncoherent af ( ts - ncaf ) . * following the proposed swipt framework , we develop noncoherent mlds for ps - ncaf and ts - ncaf in a _ unified _ form , which involves integral evaluations yet serves as the optimum performance benchmark for noncoherent swipt . to avoid integral computations , we also develop a _ closed - form _ gauss - legendre approximation based detector , which achieves almost identical ser to the mld at substantially lower complexity . it is demonstrated that the proposed eh - based noncoherent detectors achieve full diversity in rayleigh fading . in terms of the error performance , the proposed ps - ncaf and ts - ncaf may outperform conventional grid - powered noncoherent relay systems under the same total power constraint . * the choice of the time switching or power splitting parameters represents a tradeoff between eh and information transmission , and there exist unique optimum values of these parameters which minimize the ser . * the number of eh relay nodes is a key factor on the performance of the noncoherent eh relay systems : ps - ncaf outperforms ts - ncaf in the single - relay case , whereas ts - ncaf outperforms ps - ncaf in the multi - relay case . * the optimum relay position is closer to the source than to the destination , regardless of the eh relaying protocols ( this is in contrast to the conventional self - powered relay systems where the optimum relay position is closer to the destination for af and is closer to the source for df ) . furthermore , as the path - loss exponent increases , the optimum relay position shifts slightly towards the destination , but is still closer to the source . * the @xmath0-fsk signaling with @xmath3 is a more suitable solution for eh relay systems due to its higher energy efficiency compared to @xmath0-dpsk , regardless of the eh relaying protocols . * comparison of the af- and df - based eh relaying protocols demonstrates that the ser performance of af and df relaying is almost the same in noncoherent eh relay systems , irrespective of the swipt architecture ( e.g. , power splitting or time switching ) . this is in contrast to the conventional self - powered relay systems where either af or df may outperform each other depending on the relay locations . the remainder of the paper is organized as follows . section [ sec : sysmodel ] describes the system model and develops noncoherent eh relaying protocols . section [ sec : unifiedsysmodel ] obtains a unified noncoherent swipt framework for eh af relay system . section [ sec : mld ] derives the noncoherent detectors achieving swipt . section [ sec : numerical ] gains useful insights into the noncoherent swipt through simulations and section [ sec : con ] concludes the paper . _ notation _ : we use @xmath4 , @xmath5 , @xmath6 , @xmath7 , @xmath8 , @xmath9 , and @xmath10 to denote the conjugate , transpose , conjugate transpose , expectation , real part , 2-norm , and natural logarithm , respectively . also , @xmath11 , @xmath12 , and @xmath13 denote an all - zero column vector , an @xmath14 identity matrix , and a column vector with 1 at its @xmath15-th entry and 0 elsewhere , respectively . moreover , @xmath16 denotes that @xmath17 is defined by @xmath18 . finally , @xmath19 means that @xmath20 is a circularly symmetric complex gaussian ( cscg ) random vector with mean @xmath21 and covariance @xmath22 . consider an eh relay network where the source terminal @xmath23 communicates with the destination terminal @xmath24 through the help of a set of potential relay candidates , where the selected relays can harvest energy from the source rf signals and forward the source information . suppose that a total of @xmath25 eh relays , @xmath26 , @xmath27 , are predetermined . the source may communicate directly with the destination through the source - destination link @xmath28 or indirectly via the two - hop relay channels @xmath29 ( first hop ) and @xmath30 ( second hop ) . the relays , which do not have fixed power supplies , can only harvest energy from the rf signals radiated by the source and utilize that harvested energy to assist the source - destination communication through the noncoherent af relaying , thus enabling swipt in the eh relay networks . we consider a composite fading model comprising the large - scale path loss @xmath31 as well as the small - scale fading @xmath32 , @xmath33 . the path loss component @xmath31 is a distance - dependent constant and will be specified later . the small - scale fading coefficients are modeled as @xmath34 , which corresponds to the rayleigh fading scenario . we assume that the instantaneous csis , @xmath32 , @xmath33 , are unavailable to any terminals in the network , and thus , _ noncoherent _ signalings such as @xmath0-fsk and @xmath0-dpsk are considered . note that the quadrature amplitude modulation ( qam ) is inconsistent with the noncoherent swipt framework considered here as it typically requires the instantaneous csi at the receiver for coherent information delivery . . [ fig : sysmodel ] in the ps - ncaf protocol , each eh relay node splits its received rf signal into two streams , which are fed to the eh and i d circuits at the same time . suppose that the total communication block time @xmath35 ( sec ) is divided into @xmath36 sub - blocks for ps - ncaf , each of length @xmath37 is allocated to one of the transmitters ( including the source and relays ) , as illustrated in fig . [ fig : ps_ncaf ] . in the first sub - block , the source broadcasts the rf signal with power @xmath38 ( watts ) . at each eh relay node , the @xmath39 portion of the received rf signal is used for eh , and the harvested energy is used for relaying the source signal in one of the following @xmath25 sub - blocks ; thus , the average harvested power @xmath40 available for data relaying at @xmath26 is @xmath41 @xmath42 , where @xmath43 is the power splitting factor ( psf ) and @xmath44 is the eh efficiency @xcite . the remaining @xmath45 portion of the received rf signal is fed to the i d circuit at each relay node , where the signal is amplified using the harvested power @xmath40 in and forwarded to the destination in one of the following @xmath25 sub - blocks . for @xmath0-dpsk transmission , each source message @xmath46 is differentially encoded into two consecutive information - bearing symbols @xmath47 and @xmath48 according to @xmath49 , where @xmath50 is the initial reference signal . since the detection of each source message @xmath51 is based on two consecutive received symbols at the destination , it is convenient to represent the signals as @xmath52 vectors . let @xmath53^t$ ] denote the source transmitted signal . then , the received signal , @xmath54^t$ ] , at the relay node @xmath26 , @xmath42 , which is the intended for i d , can be expressed as @xmath55 where @xmath56 ( sec ) is the symbol time and the power - scaling factor @xmath45 is due to the power splitting at @xmath26 . the additive white gaussian noise ( awgn ) at @xmath26 ( when @xmath23 serves as the transmitter ) is made up of two components : the awgn due to the receive antenna ( which is introduced before the power splitter ) , modeled as @xmath57 , and the awgn due to the i d circuit ( which is introduced after the power splitter ) , modeled as @xmath58 @xcite . at the destination , no power splitting is needed as the received signal is only used for i d . thus , the received sinal at the destination can be written as @xmath59 where @xmath60 and @xmath61 are the awgns due to the receive antenna and i d circuit , respectively . each relay amplifies its received signal @xmath62 with an amplifying gain @xmath63 and forwards that signal to the destination . thus , the received signal at the destination is given by @xmath64 where @xmath65 and @xmath66 are the awgns due to the receive antenna and i d circuit , respectively . the amplifying gain @xmath63 ensures that the average transmission power for data relaying at @xmath26 is fixed to @xmath40 in @xcite . for ps - ncaf with @xmath0-dpsk , the amplifying gain @xmath63 is given by @xmath67 for @xmath0-fsk transmission , the message @xmath46 is transmitted over one of the @xmath0 orthogonal carriers . the baseband equivalent received signal intended for i d at the receiving terminal @xmath68 which is transmitted from @xmath69 is denoted as an @xmath70 vector @xmath71^t$ ] , @xmath72 . then , the signal model for ps - ncaf employing noncoherent @xmath0-fsk can be represented as @xmath73 where @xmath74 and @xmath75 , @xmath72 , are the awgns due to the receive antenna and the i d circuit at @xmath68 , respectively . for ps - ncaf with @xmath0-fsk , the amplifying gain @xmath63 , which ensures that the average transmission power for data relaying at @xmath26 is fixed to @xmath40 in , is given by @xcite @xmath76}}.\label{eq : ps_af_fsk_gain}\end{aligned}\ ] ] in the ts - ncaf protocol , the total communication block time @xmath35 ( sec ) is divided into two consecutive phases : the eh phase of length @xmath77 and the i d phase of length @xmath78 , as illustrated in fig . [ fig : ts_ncaf ] , where @xmath79 is the time switching coefficient ( tsc ) . in the eh phase , the source sends rf energy flow to the relays with power @xmath38 . each eh relay can harvest energy from the received rf signal and utilize that harvested energy for data relaying in the subsequent i d phase . the i d phase is further divided into @xmath36 sub - blocks of length @xmath80 each , where the first sub - block is assigned to the source for new information transmission and each of the remaining @xmath25 sub - blocks is assigned to one of the relays for data relaying . then , the average harvested power @xmath40 for data relaying at @xmath26 is given by@xmath81 @xmath42 , where @xmath82 is the energy conversion efficiency @xcite . in the ts - ncaf protocol , the received signal during the i d phase at each node is solely used for information processing , i.e. , no power splitting is involved . thus , the baseband equivalent signal model for ts - ncaf with @xmath0-dpsk is represented in ( @xmath52)-vector form as follows : @xmath83 @xmath42 , where @xmath84 and @xmath85 , @xmath33 , are the awgns due to the receive antenna and the i d circuit , respectively . for ts - ncaf with @xmath0-dpsk , the amplifying gain @xmath63 , which ensures that the average transmission power at @xmath26 is fixed to @xmath40 in , is given by @xcite @xmath86 the baseband equivalent signal model for ts - ncaf employing noncoherent @xmath0-fsk can be represented as [ eq : by0d_ts ] @xmath87 for @xmath42 , where @xmath74 and @xmath75 , @xmath33 , are the awgns due to the receive antenna and i d circuit at @xmath68 , respectively . for ts - ncaf with @xmath0-fsk , the amplifying gain @xmath63 , which ensures that the average transmission power at @xmath26 is fixed to @xmath40 in , is given by @xcite @xmath88}}.\label{eq : ts_af_fsk_gain}\end{aligned}\ ] ] in the last section , we developed _ two _ eh relaying protocols , namely , ps - ncaf and ts - ncaf , which can be easily applied in conjunction with _ two _ widely adopted noncoherent signalings such as noncoherent @xmath0-fsk and @xmath0-dpsk , thus resulting in a total of _ four _ different noncoherent swipt schemes for eh relay systems . from section [ sec : sysmodel ] , we see that each scheme has a different system model with a set of different parameters . in particular , the amplifying gain expressions @xmath63 corresponding to the four possible schemes in , , , and are all different . these distinct system models and inconsistent parameters make it rather cumbersome for further design and analysis . to resolve this problem , in this section , we will develop a unified swipt framework embracing both ps - ncaf and ts - ncaf , which enables unified further development . the key to the unifying process is to unify the definitions of all different system parameters . first of all , we introduce the _ effective noise variance _ for each receiving terminal as follows : [ eq : noise_variances ] @xmath89 furthermore , let us introduce the information rate , @xmath90 ( bps ) , which is the total transmitted information bits normalized by the total communication block time @xmath35 ( sec ) , where @xmath91 is the total number of transmitted information - bearing symbols . for ps - ncaf , the @xmath91 new symbols are sent over the first sub - block of length @xmath37 . for ts - ncaf , due to the time switching effect , the @xmath91 new symbols are transmitted over the first sub - block of length @xmath80 . by definition of @xmath92 , we have @xmath93 where @xmath56 depends on @xmath35 through @xmath91 as follows : @xmath94 for the purpose of performance comparison of ps - ncaf and ts - ncaf , one must ensure that the information rate @xmath92 is the same for both schemes , which is accomplished by choosing the common parameters @xmath35 ( total communication block time ) , @xmath91 ( total number of symbols ) , and @xmath0 ( modulation alphabet size ) for both schemes . this implies that ps - ncaf and ts - ncaf must have different symbol duration @xmath56 , in order to keep the same information rate @xmath92 . with the above definitions , the average signal - to - noise ratios ( snrs ) of the direct source - destination link , the source - relay link , and relay - destination link can be expressed , respectively , as and @xmath95 are defined as the actual link snrs associated with the direct link and the first - hop link , respectively ; but @xmath96 is defined as the average snr of the second - hop when the received signal from the first - hop is of unit energy , i.e. , @xmath97 . ] [ eq : snrs ] @xmath98r } , & \text{ps - ncaf } , \\ \frac{(1-\alpha)p_0\mathcal{l}_{0r}\log_2m}{(k+1)(\sigma_{0r,1}^2+\sigma_{0r,2}^2)r } , & \text{ts - ncaf } , \end{array } \right . \\ \gamma_{rd } & { \stackrel{\delta}{=}}\left \ { \begin{array}{ll } \frac{\rho \eta p_0\mathcal{l}_{0r}\mathcal{l}_{rd}\log_2m}{\big[(1-\rho)p_0\mathcal{l}_{0r}\log_2m+(k+1)\big((1-\rho)\sigma_{0r,1}^2+\sigma_{0r,2}^2\big)r\xi\big](\sigma_{rd,1}^2+\sigma_{rd,2}^2 ) } , & \text{ps - ncaf } , \\ \frac{(k+1)\alpha \eta p_0\mathcal{l}_{0r}\mathcal{l}_{rd}\log_2m}{\big[(1-\alpha)p_0\mathcal{l}_{0r}\log_2m+(k+1)(\sigma_{0r,1}^2+\sigma_{0r,2}^2)r\xi\big](\sigma_{rd,1}^2+\sigma_{rd,2}^2 ) } , & \text{ts - ncaf } , \end{array } \right.\label{eq : gamma_rd}\end{aligned}\ ] ] for @xmath99 , where @xmath100 for @xmath0-dpsk and @xmath101 for @xmath0-fsk . following the parameter definitions in , the signal models for ps - ncaf and ts - ncaf employing @xmath0-dpsk can be expressed in a unified form as follows : [ eq : dpsk_unified ] @xmath102 where @xmath103 for @xmath33 . note that the unified model in can represent either ps - ncaf or ts - ncaf by appropriately choosing the parameters according to . following the definitions in , the signal models for ps - ncaf and ts - ncaf employing noncoherent @xmath0-fsk can be unified as follows : [ eq : fsk_unified ] @xmath104 where @xmath105 for @xmath106 and @xmath42 . the main benefit of the unified noncoherent swipt frameworks in and lies in the fact that it enables unified design and analysis for the ps - ncaf and ts - ncaf protocols . for example , following these proposed unified noncoherent swipt frameworks , the noncoherent detectors for both ps - ncaf and ts - ncaf can be obtained in a unified form , which is treated in more detail in the next section . in this section , the main objective is to develop the noncoherent detection schemes for eh af relay systems . to proceed , we first tackle the mathematical challenges involved in the development of the noncoherent detectors . then , the ( exact ) mlds for ps - ncaf and ts - ncaf are derived in a unified form , which characterizes the optimum performance benchmark for noncoherent swipt in eh af networks . finally , low - complexity noncoherent detectors are derived , which achieves almost identical performance to the mlds at dramatically lower complexity . the unified signal models in and involve complicated transformations of the cscg random variables / vectors . to facilitate the development of the ml detection schemes , it is useful to study the probability density functions ( pdfs ) of those gaussian transformations . [ lemma : pdf ] consider random variables @xmath107 , @xmath52 random vectors @xmath108 , and @xmath70 random vectors @xmath109 , @xmath110 , all of which are mutually independent . let @xmath111 where @xmath112 is any integer number between 1 and @xmath0 , @xmath113^t$ ] , and @xmath114 is any complex number . the pdfs of @xmath115 , @xmath116 , and @xmath117 , @xmath118 , are given by @xmath119\frac{\omega_2\sigma_1 ^ 2}{\sigma_2 ^ 2 } , \frac{|x_2-cx_1|^2}{(1+|c|^2)\sigma_2 ^ 2},\frac{|x_1+c^*x_2|^2}{(1+|c|^2)\sigma_2 ^ 2},1\bigg),\label{eq : pdf_x0}\\ f_{\bm{y}_0}(\bm{y } ) & = \frac{1}{(\pi\sigma_2 ^ 2)^m}i\bigg(\frac{\omega_2\sigma_1 ^ 2}{\sigma_2 ^ 2},\big(1+\frac{\omega_1}{\sigma_1 ^ 2}\big)\frac{\omega_2\sigma_1 ^ 2}{\sigma_2 ^ 2 } , \frac{\|\bm{y}\|^2-|y_p|^2}{\sigma_2 ^ 2},\frac{|y_p|^2}{\sigma_2 ^ 2},m-1\bigg),\label{eq : pdf_y0}\end{aligned}\ ] ] where @xmath120^t$ ] , @xmath121^t$ ] , and @xmath122 is defined as @xmath123 for any @xmath124 . see appendix [ app : pdf ] . the pdf analysis in lemma [ lemma : pdf ] is very useful for finding the mlds for eh af relay systems , which will be addressed in the next subsection . furthermore , the generic analytical results obtained in lemma [ lemma : pdf ] may be useful for other different applications involving the same gaussian transformations as considered here . following the unified noncoherent swipt frameworks in and and the generic pdf analysis in lemma [ lemma : pdf ] , in this subsection , the mlds for ps - ncaf and ts - ncaf are obtained in a unified form . [ thm : exact_mld ] the mlds for both ps - ncaf and ts - ncaf employing @xmath0-dpsk can be expressed in a unified form as @xmath125 similarly , the mlds for both ps - ncaf and ts - ncaf employing @xmath0-fsk are given in a unified form as @xmath126 see appendix [ app : exact_mld ] . the mlds in and involve the computations of the integral @xmath122 in , which results in a high computational complexity . nevertheless , the mlds are very useful as they characterize the optimum performance benchmark for noncoherent swipt in eh af networks . for example , one may use the mlds as a theoretical performance upper bound to test the performance of any suboptimum detectors , if developed in the literature . in this subsection , to avoid integral evaluations and make the mlds suitable for practical implementation , we adopt the gauss - legendre ( gl ) quadrature to approximate the integral , resulting in the so - called gauss - legendre detectors ( glds ) in closed - form . [ th : approx_mld ] the glds for both ps - ncaf and ts - ncaf employing @xmath0-dpsk are given in the same form as , except that the multivariate function @xmath127 is replaced by @xmath128 , which is given by @xmath129 the parameters @xmath130 and @xmath131 are , respectively , the weights and nodes of the gl quadrature of order 5 , where @xmath132 , @xmath133 , @xmath134 , @xmath135 , @xmath136 , @xmath137 , and @xmath138 is defined as follows : @xmath139 similarly , the glds for ps - ncaf and ts - ncaf employing @xmath0-fsk are given in , where @xmath127 is replaced by @xmath128 . see appendix [ app : approx_mld ] . the mlds in theorem [ th : approx_mld ] require the computation of @xmath0 integrals for detecting one information - bearing symbol , whereas the proposed glds avoid the integral computations and involve no special functions . thus , the glds can be easily implemented in practice . furthermore , we will demonstrate in the next section that the glds achieve almost the same error performance as the mlds . thus , the glds constitute practical solutions for eh noncoherent relaying systems . in this section , we evaluate the performance of the proposed noncoherent swipt schemes from various aspects through monte carlo simulations . a @xmath25-relay line network model is considered , where the relays are distributed over the straight line between the source and destination . given the source - destination distance @xmath140 , the source - relay distance @xmath141 and the relay - destination distance @xmath142 may vary arbitrarily while guaranteeing @xmath143 , for @xmath42 . the eh efficiency is set to @xmath144 as in @xcite . for ease of simulations , the antenna noise and i d circuit noise are assumed to have equal variances , i.e. , @xmath145 , @xmath33 , which ensures that each receiving node has a total noise variance of @xmath146 . since the transmit power @xmath38 applied to the source is the only energy supply to the whole network , the error performance of the whole system is parameterized by @xmath147 . unless otherwise stated , the communication distance is set to @xmath148 ( m ) and the bounded path - loss model @xcite is adopted , i.e. , @xmath149 , where the path loss exponent is @xmath150 as in @xcite . we will evaluate the ser performance of the mlds and glds for the proposed ps / ts - ncaf framework . in addition , the ps / ts - ncdf framework and the corresponding mlds developed in @xcite will be considered as a benchmark for the purpose of comparison . ) or tsc ( @xmath151 ) for ps - ncaf and ts - ncaf employing bdpsk and bfsk at information rate @xmath152 bps and @xmath153 db in a single - relay network ( @xmath154 ) with @xmath155 . ] in this subsection , we compare the error performance of ps - ncaf and ts - ncaf by investigating the impacts of the psf @xmath39 , the tsc @xmath151 , and the number of relays @xmath25 . the comparisons will be carried out for binary dpsk ( bdpsk ) and binary fsk ( bfsk ) . the ser performance of the proposed eh relaying protocols is evaluated by varying the psf @xmath39 for ps - ncaf and the tsc @xmath151 for ts - ncaf while fixing the snr as @xmath156 db for binary ( @xmath157 ) noncoherent signalings at the information rate @xmath152 bps . [ fig : ps_ts_vary_coeff_35db_m2_k1 ] illustrates the sers versus the psf @xmath39 or tsc @xmath151 for the _ single - relay _ ( @xmath154 ) case where the relay is located at @xmath158 . we see that there exist ( unique ) optimum values of the psf ( i.e. , @xmath159 ) and the tsc ( i.e. , @xmath160 ) which minimize the sers of ps - ncaf and ts - ncaf , respectively . this is because the choices of the psf or tsc result in some tradeoff between eh and i d in eh relay systems . specifically , for larger @xmath39 ( or @xmath151 ) , more energy can be harvested and utilized for data relaying , whereas less signal energy ( or time ) can be used for i d . for smaller @xmath39 ( or @xmath151 ) , more signal energy ( or time ) is available for i d ; but less energy can be harvested and used for data relaying . due to this tradeoff , the optimum values of @xmath39 and @xmath151 always lie between 0 and 1 , which balances the operation of eh and i d such that the overall ser is minimized . allocated to each terminal versus eh relay system with @xmath38 supplied to the source only for binary noncoherent signalings with @xmath152 bps , where @xmath161 , @xmath162 , @xmath159 , and @xmath163 . ] for the single - relay network as illustrated in fig . [ fig : ps_ts_vary_coeff_35db_m2_k1 ] , the minimum ser of ps - ncaf ( achieved around @xmath164 ) is lower than the minimum ser of ts - ncaf ( achieved around @xmath160 ) . however , this relationship is reversed in the _ multi_-relay case . specifically , for the 2-relay case ( @xmath165 ) employing binary noncoherent signalings where @xmath166 and @xmath167 bps , fig . [ fig : ps_ts_vary_coeff_30db_m2_k2 ] shows that the minimum ser of ts - ncaf achieved around @xmath168 is slightly smaller than the minimum ser of ps - ncaf achieved around @xmath169 , for both bdpsk and bfsk . in addition , for the 3-relay ( @xmath161 ) case where @xmath162 and @xmath152 bps , as shown in fig . [ fig : ps_ts_vary_coeff_35db_m2_k3 ] , we can clearly see that the minimum ser of ts - ncaf achieved around @xmath163 is much smaller than the minimum ser of ps - ncaf achieved around @xmath159 . these comparisons indicate that the number of relays @xmath25 is a key factor which dictates to the superior eh protocol . specifically , ps - ncaf outperforms ts - ncaf in the single - relay case ( @xmath154 ) , while ts - ncaf outperforms ps - ncaf in the multi - relay case ( @xmath170 ) . the reason for this is explained as follows . for ps - ncaf , since the eh time and the i d time are equal ( both equal to @xmath37 ) , the harvested power in for data relaying at each relay is independent of @xmath25 , which makes the second - hop snr @xmath96 independent of @xmath25 as well , as illustrated in . in contrast , for ts - ncaf , due to the time switching operation , the eh time , @xmath77 , is generally unequal to the i d time , @xmath80 , at each relay . thus , for fixed harvested energy , the harvested power for data relaying at each relay becomes proportional to @xmath36 , and consequently , the second - hop snr @xmath96 is a monotonically increasing function of @xmath25 , as illustrated in . hence , as @xmath25 increases , the second - hop link for ts - ncaf becomes increasingly more reliable as compared to that for ps - ncaf , which results in extra performance improvement over ps - ncaf for @xmath170 . for the multi - relay case , we are also interesting in comparing ts - ncaf and ps - ncaf at different snrs , such that the performance gain of ts - ncaf over ps - ncaf can be quantified at any target sers . fig . [ fig : grid_power_af_k3_total_p0 ] shows the sers versus the snr for the 3-relay network employing the optimum eh parameters , i.e. , @xmath171 for ps - ncaf and @xmath163 for ts - ncaf according to fig . [ fig : ps_ts_vary_coeff_35db_m2_k3 ] , where @xmath162 and @xmath152 bps . for the target ser of @xmath172 , we see that ts - ncaf outperforms ps - ncaf by about 1.5 db , for both bdpsk and bfsk . this significant gain is very attractive as the energy efficiency is of major concern in eh systems . furthermore , the conventional grid - powered relay system is evaluated as a benchmark . for a fair comparison in terms of the total power consumption , the powers allocated to the source and all relays are equal to @xmath173 , ensuring the same total power of @xmath38 as for the eh relay system . we observe that both eh protocols , namely ps - ncaf and ts - ncaf , outperform the grid - powered af relaying protocol , for the same noncoherent modulation . this is because given the same total power , the eh ability allows the relays to obtain extra powers from the ambient rf signals , thus boosting the overall performance over the grid - powered relay system with no eh . for conventional self - powered relay systems where the eh capability is disabled @xcite , the impact of the relay position on the error performance is well understood . specifically , the optimum relay position minimizing the ser in _ coherent _ relay systems is at the midpoint between the source and destination , for both af and df @xcite . for _ noncoherent _ relay systems , however , the optimum relay positions differ for af and df . for df relaying , the optimum relay position is closer to the source than to the destination @xcite , whereas the optimum relay position for af relaying is closer to the destination than to the source @xcite . [ fig : ps_ts_vary_d ] illustrates the ser versus the relative relay position @xmath174 for noncoherent eh relay systems . for both af - based ps / ts - ncaf proposed in this paper and the df - based ps / ts - ncdf in @xcite , the optimum relay positions are @xmath175 when the path loss exponent is @xmath176 , as shown in fig . [ fig : ps_ts_vary_d_26db_2p7 ] , and @xmath177 when the path loss exponent is @xmath150 , as shown in fig . [ fig : ps_ts_vary_d_35db_4 ] . that is , unlike the conventional noncoherent relay networks where the optimum relay positions depend on the relaying protocols ( af or df ) , the optimum relay positions for noncoherent eh relay systems are invariant with respect to af and df , and are always closer to the source than to the destination . this major difference is due to the fact that eh relays are solely powered by the source , and thus , the dominant performance limiting factor in eh relay networks is the harvested energy at the relays , which is invariant with respect to af and df . moreover , the overall performance is affected by the path loss , which is a distance - dependent constant . since both the eh operation and the path loss effect are irrelevant to the relaying protocols , the optimum relay positions are the same with respect to af and df in eh relay systems . we note that the eh operation and the path loss have different impacts on the relay position . on one hand , the relays must be sufficiently close to the source in order to harvest enough energy . on the other hand , the relays must not be infinitely close to the source ( e.g. , located at the source ) because otherwise the relay - to - destination path loss would become very significant , which degrades the overall performance . considering both the eh and path - loss effects , the optimum relay position for eh relay systems is relatively closer to the source than to the destination . as illustrated in fig . [ fig : ps_ts_vary_d ] , the optimum relay position , which is always closer to the source than to the destination , shifts slightly towards the destination as the path loss exponent @xmath178 increases from 2.7 to 4 . this is because a larger path loss exponent corresponds to a larger relay - to - destination path loss , which can degrade the overall error performance . as such , the relay should step towards the destination a little bit to diminish the effect of the increased path loss . nevertheless , the prerequisite of eh operation can not be obviated , and thus , the relays must always be deployed closer to the source than to the destination , with some slight movements towards the destination as the path loss exponent increases . so far , we have only considered noncoherent binary signalings with @xmath157 , and a 3 db performance gain of bdpsk over bfsk can be observed in fig . [ fig : grid_power_af_k3_total_p0 ] . in this subsection , we will evaluate the performance of higher order noncoherent signalings for @xmath179 . specifically , for a single - relay network with @xmath180 , the ser of @xmath181 at @xmath182 db and @xmath183 bps is illustrated in fig . [ fig : ps_ts_vary_coeff_38db_m4_k1 ] , where 4-dpsk slightly outperforms 4-fsk , irrespective of the eh relaying protocols ( e.g. , ps / ts - ncaf or ps / ts - ncdf ) . moreover , the ser for @xmath184 at @xmath185 db and @xmath186 bps is illustrated in fig . [ fig : ps_ts_vary_coeff_40db_m8_k1 ] , which shows that 8-fsk substantially outperforms 8-dpsk , irrespective of the eh relaying protocols . observing the trend from @xmath157 to 4 and 8 , we conclude that @xmath0-fsk is more efficient in terms of energy consumption than @xmath0-dpsk , which results in significant performance gain for @xmath3 . in particular , the performance gain of 8-fsk over 8-dpsk is about 4 db in the target ser of @xmath187 , as demonstrated in @xcite . ) case with @xmath155 , @xmath159 , and @xmath152 bps and 2 ) two - relay ( @xmath165 ) case with @xmath188 , @xmath169 , and @xmath167 bps . ] ) case with @xmath155 , @xmath160 , and @xmath152 bps and 2 ) two - relay ( @xmath165 ) case with @xmath188 , @xmath168 , and @xmath167 bps . ] since both af and df relaying protocols are useful candidates for achieving swipt in eh relay systems , it is meaningful to compare the performance of the af - based ps / ts - ncaf proposed in this paper with that of the df - based ps / ts - ncdf developed in @xcite . the ser performance of the eh relaying protocols generally depend on the relay positions , the eh parameters ( such as psf and tsc ) , and snrs . thus , performance comparisons can be carried out from various angles . in particular , fig . [ fig : ps_ts_vary_d ] illustrates the comparison of ps / ts - ncaf and ps / ts - ncdf by varying the relay positions for fixed eh parameters and fixed snr . in addition , fig . [ fig : ps_ts_higer_constellations ] illustrates the sers for a variety of eh parameter settings ( psf or tsc ) , where the relay position and snr are fixed . the remaining possible comparison is therefore to fix the relay position and eh parameters , while varying the snrs . to this end , we compare ps - ncaf with ps - ncdf , and ts - ncaf with ts - ncdf at different snrs in figs . [ fig : ps_ncaf_validation ] and [ fig : ts_ncaf_validation ] , respectively , where two typical network geometries are considered : 1 ) single - relay ( @xmath154 ) case with @xmath155 , @xmath160 , @xmath159 , and @xmath152 bps and 2 ) two - relay ( @xmath165 ) case with @xmath188 , @xmath168 , @xmath169 , and @xmath167 bps . note that the psf @xmath39 and tsc @xmath151 employed in the comparisons of figs . [ fig : ps_ncaf_validation ] and [ fig : ts_ncaf_validation ] are optimum values determined in figs . [ fig : ps_ts_vary_coeff_35db_m2_k1 ] and [ fig : ps_ts_vary_coeff_30db_m2_k2 ] . for all the comparisons carried out from various angles in figs . [ fig : ps_ts_vary_d][fig : ts_ncaf_validation ] , we observe that the ser curves associated with ps - ncaf and ps - ncdf almost overlap , so do the ser curves associated with ts - ncaf and ts - ncdf . these observations reveal that the noncoherent af and df relaying protocols yield almost the same error performance in eh relay networks , regardless of the swipt architectures ( e.g. , power splitting and time switching ) , the modulation types ( e.g. , dpsk and fsk ) , and system parameters ( e.g. , information rate @xmath92 , number of relays @xmath25 , relay positions , and snrs ) . this is in contrast to the conventional self - powered relay networks , where either af or df may outperform each other depending on the relay positions . in particular , for self - powered relay networks , df might outperform af when the relay is much closer to the source than to the destination , because the noise amplification due to af substantially degrades the performance . on the other hand , af might outperform df when the relay is much closer to the destination , due to the high relay decoding error probabilities in df . however , both effects of noise amplification ( associated with af ) and error propagation ( associated with df ) are significantly weakened and become less important in eh relay systems , because the dominant performance limiting factor is the harvested energy at the relay , which is invariant with respect to the af and df relaying protocols . hence , for eh relay systems , there is almost no difference between the noncoherent af and df relaying in terms of the error performance . in figs . [ fig : ps_ncaf_validation ] and [ fig : ts_ncaf_validation ] , we also observe that the proposed noncoherent detectors achieve full diversity orders of 2 and 3 for the single- and two - relay networks , respectively , with the direct link . furthermore , we would like to note that the ser performance of the gld is in excellent agreement with the ser performance of the mld , as demonstrated in all figs . [ fig : ps_ts_vary_coeff_35db_m2_k1][fig : ts_ncaf_validation ] . moreover , the gld is given in closed - form , involving no integrals or special functions at all . this suggests that the closed - form gld is an appealing practical solution for noncoherent swipt in eh af relay systems . finally , this also indicates that @xmath189 of is a very accurate approximation for the integral @xmath122 in . we now evaluate the performance of the proposed detectors under a practical indoor communication scenario operating at 900 mhz , where the typical communication distance of @xmath190 ( m ) is considered . note that according to the current state - of - the - art of rf energy harvesting / transfer , the maximum line - of - sight ( los ) operating distance for some rf energy harvesting chips is 1214 meters @xcite . in this paper , we consider rayleigh fading with blocked los due to a double plasterboard wall ; thus , the communication distance of 10 meters is a very reasonable and practical choice . according to empirical measurements carried out at 900 mhz , , we set the path loss exponent as @xmath191 , which corresponds to the communication inside the typical office building ( * ? ? ? * table 2.2 ) . the partition loss is set to 3.4 db assuming that the transmit and receive terminals are separated by the double plasterboard wall ( * ? ? ? * table 2.1 ) . also , we assume that the communication occurs on the same floor ; thus , the floor loss can be ignored . according to the indoor path loss model featuring at 900 mhz ( * eq . ( 2.38 ) ) , we set the path loss as @xmath192 in this subsection . [ fig : indoor_3_relays ] shows the ser performance for an eh relay system with @xmath161 relays , all located at @xmath193 , @xmath194 . note that we choose this relay position because the optimum relay position is much closer to the source than to the destination , as indicated in fig . [ fig : ps_ts_vary_d ] . also , the eh parameters are chosen as @xmath163 and @xmath159 , which are optimum values for @xmath161 relays according to fig . [ fig : ps_ts_vary_coeff_35db_m2_k3 ] . our simulation in fig . [ fig : indoor_3_relays ] demonstrates that for binary modulations with @xmath152 bps , the diversity order of 4 is achieved for @xmath161 eh relays ( with the direct link ) , which confirms the full diversity performance . , where @xmath190 ( m ) , @xmath195 for @xmath194 , @xmath163 , @xmath159 , and @xmath152 bps . in this paper , we have developed a noncoherent swipt framework for eh af relay systems , which embraces both ps - ncaf and ts - ncaf in a unified form and supports arbitrary @xmath0-ary noncoherent signalings including @xmath0-fsk and @xmath0-dpsk . the main advantage of the proposed swipt scheme is that it eliminates the need for the instantaneous csi , which alleviates the system overhead and reduces the energy consumption . following this framework , we developed noncoherent mlds for ps - ncaf and ts - ncaf in a _ unified _ form , which involved integral evaluations yet served as the optimum performance benchmark for noncoherent swipt . to avoid integral computations , we also proposed _ closed - form _ glds , which achieved almost identical ser performance to the mlds at substantially lower computational complexity . it was demonstrated that these noncoherent detectors achieve full spatial diversity in rayleigh fading . moreover , in terms of the error performance , the proposed swipt relaying schemes may outperform conventional grid - powered relay systems under the same total power constraint . finally , numerical results led to useful design insights into the noncoherent swipt in various aspects , including the effects of the time switching or power splitting parameters , relaying protocols , the number of relays , relay positions , and the modulation alphabet size , which are summarized as follows : * the choice of the power slitting factor or time switching coefficient represents some tradeoff between energy harvesting and information delivery . unique optimal values of the psf or tsc minimizing the ser exist between 0 and 1 . * the number of eh relays is a key factor on the performance of noncoherent swipt : ps - ncaf outperforms ts - ncaf in the single - relay case , whereas ts - ncaf outperforms ps - ncaf in the multi - relay case . * the optimum relay position is closer to the source than to the destination , with some slight movements towards the destination as the path loss exponent increases . * the noncoherent @xmath0-fsk with @xmath3 is more energy - efficient and thus might be more suitable than @xmath0-dpsk for eh relay systems , regardless of the eh relaying protocols . * the noncoherent af and df relaying protocols yield almost the same error performance in eh relay networks , regardless of the swipt architectures , the modulation types , and system parameters . an interesting extension of this work is to analyze the ser of the mld or gld , and using the obtained ser expression to _ analytically _ determine the optimum time switching coefficient or power splitting factor . in addition , more sophisticated and intelligent coordination of the source and relays transmissions ( such as relay selection ) may also be considered in order to further boost the performance . in this proof , we will first study the distribution of @xmath115 , followed by a brief derivation of the pdf of @xmath117 . it is not hard to show that the conditional distribution of @xmath115 given @xmath196 is @xmath197 , where @xmath198.\ ] ] we denote by @xmath199 the determinant of @xmath200 , and it can be shown that @xmath201z+\sigma_2 ^ 2\big),\label{eq : det}\\ \bm{\sigma}_0^{-1}&=\frac{1}{\det(\mathrm{\sigma_0})}\left[\begin{array}{cc } ( \omega_1|c|^2+\sigma_1 ^ 2)z+\sigma_2 ^ 2 , & -\omega_1c^*z \\ -\omega_1cz , & ( \omega_1+\sigma_1 ^ 2)z+\sigma_2 ^ 2 \end{array } \right].\label{eq : inverse_matrix}\end{aligned}\ ] ] it follows that the conditional pdf of @xmath115 given @xmath196 is @xmath202 . substituting in the expressions of and , we can show that @xmath203\left[\begin{array}{cc } ( \omega_1|c|^2+\sigma_1 ^ 2)z+\sigma_2 ^ 2 , & -\omega_1c^*z \\ -\omega_1cz , & ( \omega_1+\sigma_1 ^ 2)z+\sigma_2 ^ 2 \end{array } \right]\left[\begin{array}{c } x_1 \\ x_2 \end{array } \right]\label{eq : app_matrix_prod1}\\ & = \frac{[(\omega_1|c|^2+\sigma_1 ^ 2)z+\sigma_2 ^ 2]|x_1|^2+[(\omega_1+\sigma_1 ^ 2)z+\sigma_2 ^ 2]|x_2|^2 - 2\omega_1z\re(cx_1x_2^*)}{(\sigma_1 ^ 2z+\sigma_2 ^ 2 ) \big([\omega_1(1+|c|^2)+\sigma_1 ^ 2]z+\sigma_2 ^ 2\big)}\label{eq : app_matrix_prod2}\\ & = \frac{a_1}{\sigma_1 ^ 2z+\sigma_2 ^ 2}+\frac{a_2}{[\omega_1(1+|c|^2)+\sigma_1 ^ 2]z+\sigma_2 ^ 2},\label{eq : app_matrix_prod3}\end{aligned}\ ] ] where follows by the partial fraction expansion . it can be shown that @xmath204z+\sigma_2 ^ 2\big)\bm{x}^h\bm{\sigma}_0^{-1}\bm{x}\big|_{z=-\frac{\sigma_2 ^ 2}{\omega_1(1+|c|^2)+\sigma_1 ^ 2 } } = \frac{|x_1+c^*x_2|^2}{1+|c|^2}.\label{eq : app_a2}\end{aligned}\ ] ] the unconditional pdf of @xmath205 can be computed as follows : @xmath206z+\sigma_2 ^ 2}\big)}}{(\sigma_1 ^ 2z+\sigma_2 ^ 2 ) \big([\omega_1(1+|c|^2)+\sigma_1 ^ 2]z+\sigma_2 ^ 2\big ) } { \mathrm{d } } z.\label{eq : app_uncon_pdf_dpsk2 } \end{aligned}\ ] ] applying the change of variable @xmath207 in and taking some algebraic manipulations , we can show that @xmath208}{\sigma_2 ^ 2}t}\big)\bigg ] } { \left(1+\frac{\omega_2\sigma_1 ^ 2}{\sigma_2 ^ 2}t\right)\left(1+\frac{\omega_2[\omega_1(1+|c|^2)+\sigma_1 ^ 2]}{\sigma_2 ^ 2}t\right ) } { \mathrm{d } } z,\label{eq : app_uncon_pdf_dpsk3 } \end{aligned}\ ] ] where @xmath209 finally , following the definition of @xmath122 in , we can simplify into . the derivation of the pdf of @xmath117 follows similar lines as the derivation of @xmath116 , which is summarized in two main steps . in the first step , the conditional pdf of @xmath117 given @xmath196 is obtained as @xmath210}\prod_{\substack{i=1\\ i\neq p}}^m\frac { { \mathrm{e } } ^{-\frac{|y_i|^2}{\sigma_1 ^ 2z+\sigma_2 ^ 2}}}{\pi(\sigma_1 ^ 2z+\sigma_2 ^ 2 ) } \nonumber\\ & = \frac { { \mathrm{e } } ^{-\big(\frac{\|\bm{y}\|^2-|y_p|^2}{\sigma_1 ^ 2z+\sigma_2 ^ 2}+\frac{|y_p|^2}{(\omega_1+\sigma_1 ^ 2)z+\sigma_2 ^ 2}\big)}}{\pi^m ( \sigma_1 ^ 2z+\sigma_2 ^ 2)^{m-1 } [ ( \omega_1+\sigma_1 ^ 2)z+\sigma_2 ^ 2]}.\label{eq : app_cond_pdf_y0_1}\end{aligned}\ ] ] in the second step , the unconditional pdf of @xmath117 is obtained as @xmath211}{\big(1+\frac{\omega_2\sigma_1 ^ 2}{\sigma_2 ^ 2}z\big)^{m-1}\big[1+\frac{\omega_2(\omega_1+\sigma_1 ^ 2)}{\sigma_2 ^ 2}z\big ] } { \mathrm{d } } z , \label{eq : app_uncon_pdf_fsk}\end{aligned}\ ] ] which simplifies to following the definition of @xmath122 in . in this proof , we will start with the derivation of the mld for @xmath0-dpsk . then the mld for @xmath0-fsk will be derived in a similar procedure . recall the unified signal model in for both ps - ncaf and ts - ncaf employing @xmath0-dpsk . it is not hard to see that @xmath212 , where @xmath213.\end{aligned}\ ] ] thus , the conditional pdf of @xmath214 given @xmath51 can be expressed as @xmath215 using some algebraic manipulations , the signal model for @xmath216 in can be rewritten in the desired form as @xmath115 in @xmath217 where @xmath218^t\\ \tilde{x}_1&{\stackrel{\delta}{=}}\sigma_{0r}\sqrt{\gamma_{0r}}h_{0r}s(l-1)\sim\mathcal{cn}(0,\sigma_{0r}^2\gamma_{0r})\\ \tilde{x}_2&{\stackrel{\delta}{=}}\sigma_{rd}\sqrt{\gamma_{rd}}h_{rd}\sim\mathcal{cn}(0,\sigma_{rd}^2\gamma_{rd})\\ \tilde{\bm{x}}_1&{\stackrel{\delta}{=}}\sigma_{0r}\bm{n}_{0r}\sim\mathcal{cn}(\bm{0},\sigma_{0r}^2\bm{i}_2)\\ \tilde{\bm{x}}_2&{\stackrel{\delta}{=}}\sigma_{rd}\bm{n}_{rd}\sim\mathcal{cn}(\bm{0},\sigma_{rd}^2\bm{i}_2).\end{aligned}\ ] ] applying lemma [ lemma : pdf ] , the conditional pdf of @xmath216 given @xmath51 can be written as @xmath219 following the definition of the mld , we have @xmath220 where @xmath221 is the likelihood function when @xmath51 is transmitted , and follows by the independence among all different signal branches . substituting and into yields the mld in for @xmath0-dpsk . the mld for @xmath0-fsk can be derived with similar steps as in the derivation of the mld for @xmath0-dpsk . specifically , we first obtain the pdf of @xmath222 in , conditioned on the transmitted message @xmath51 , as @xmath223 then , @xmath224 in can be rewritten in the desired form as @xmath117 as @xmath225 , where @xmath226 , @xmath227 , @xmath228 , and @xmath229 . then , applying lemma [ lemma : pdf ] , the conditional pdf of @xmath224 given @xmath51 can be expressed as @xmath230 finally , following the definition of mld , we can show that the mld for @xmath0-fsk is given by . the main difficulty involved in the mld is the integral computation of @xmath122 in . thus , we will develop a tight closed - form approximation for this integral . to that end , we first transform the original integral into a desired form which is easy to approximate . specifically , we carry out the following algebraic operations : @xmath231 where follows by the change of variable @xmath232 and @xmath138 is defined in . the integrand @xmath138 is a well - behaved function : first , it is a bounded positive function with @xmath233 ; second , it is a smooth function with continuous derivatives with respect to @xmath234 of any order . furthermore , the integral limits are finite . these elegant properties enable very accurate approximation of the integral using the gl quadrature with a finite order . indeed , the choice of the approximation order must always compromise between the accuracy and computational cost @xcite . the higher order , the more accurate the gl quadrature is . on the other hand , however , higher order requires the computation of more terms which involves higher computational cost . considering both the accuracy and computational cost , we choose the fifth order gl quadrature for our purpose @xcite , which results in @xmath235 , where @xmath128 is given in closed - form in . finally , using @xmath128 in place of @xmath127 in the mlds yields the corresponding glds . demonstrate that the performance of glds are almost the same as that of the mlds , indicating that the fifth order gl approximation adopted here is highly accurate . ] x. lu , p. wang , d. niyato , d. i. kim , and z. han , `` wireless networks with rf energy harvesting : a contemporary survey , '' _ _ to appear in _ ieee commun . surveys and tutorials_. [ online ] . available : http://arxiv.org/abs/1406.6470 a. a. nasir , x. zhou , s. durrani , and r. a. kennedy , `` wireless - powered relays in cooperative communications : time - switching relaying protocols and throughput analysis , '' _ _ submitted for publication in _ ieee trans . [ online ] . available : http://arxiv.org/abs/1310.7648 d. niyato , p. wang , h. tan , w. saad , and d. i. kim , `` cooperation in delay tolerant networks with wireless energy transfer : performance analysis and optimization , '' _ _ to appear in _ ieee trans . _ a. a. nasir , x. zhou , s. durrani , and r. a. kennedy , `` relaying protocols for wireless energy harvesting and information processing , '' _ ieee trans . wireless commun . _ , vol . 12 , pp 36223636 , july 2013 . z. ding , i. krikidis , b. sharif , and h. v. poor , `` wireless information and power transfer in cooperative networks with spatially random relays , '' _ ieee trans . wireless commun . _ , vol . 13 , pp . 44404453 , aug . 2014 . p. liu and i .- kim , `` optimum / sub - optimum detectors for multi - branch dual - hop amplify - and - forward cooperative diversity networks with limited csi , '' _ ieee trans . wireless commun . _ , vol . 9 , pp . 7885 , jan . p. liu , s. gazor , i .- kim , and d. i. kim , `` energy harvesting noncoherent cooperative communications , '' _ _ submitted for publication in _ ieee trans . wireless commun . _ [ online ] . available : http://arxiv.org/abs/1505.07188 m. r. souryal and b. r. vojcic , `` performance of amplify - and - forward and decode - and - forward relaying in rayleigh fading with turbo codes , '' in _ proc . speech , signal process . ( icassp ) _ , vol . 4 , may 2006 , pp .

we discuss the time - dependent formulation of perturbation theory in the context of the interacting zeroth - order hamiltonians that appear in multi - reference situations . as an example , we present a time - dependent formulation and implementation of second - order-electron valence perturbation theory . the resulting t - nevpt2method yields the fully uncontracted-electron valence perturbation wavefunction and energy , but has a lower computational scaling than the usual contracted variants , and also avoids the construction of high - order density matrices and the diagonalization of metrics . we present results of t - nevpt2for the water , nitrogen , carbon , and chromium molecules , and outline directions for the future .

1512.09213

a standard strategy in the quantum chemistry of strongly correlated systems is to first compute an approximate zeroth - order wavefunction within an active space of frontier molecular orbitals . @xcite the description of the remaining correlation outside of the active space is then the purview of multi - reference dynamic correlation theories . @xcite in principle , it is possible to directly extend single - reference dynamic correlation theories to the multi - reference setting . however , several complications arise . first , the dimension of the space in which the perturbed first - order wavefunction resides is much larger in the multi - reference case : @xmath1 , where @xmath2 is the dimension of the active hilbert space , and @xmath3 is the number of external orbitals . thus explicitly representing a general first - order wavefunction is costly . second , the starting zeroth - order wavefunction is no longer an eigenstate of a one - electron hamiltonian , but rather an interacting hamiltonian . this means that functions of the zeroth - order hamiltonian , such as the resolvent operator in perturbation theory , are not known in explicit computational form . third , the standard wick s theorem , which reduces expectation values of fermionic operator strings to products of single - particle density matrices , does not apply.@xcite this leads to the need to compute expensive high - order density matrices . it is common to circumvent these complications by introducing additional approximations which are not used in the single - reference setting . some of these are internally contracted wavefunctions,@xcite alternative zeroth - order hamiltonians,@xcite and approximations to high - order density matrices.@xcite while enormously useful , these approximations can lead to new problems of their own , such as intruder states associated with the choice of zeroth - order hamiltonian , or metric instabilities associated with internal contraction.@xcite in this work , we show that using a time - dependent formulation ameliorates the complexity of the multi - reference dynamic correlation framework , without the need to introduce additional approximations . as an example , we derive and implement the time - dependent @xmath0-electron valence second - order perturbation theory ( t - nevpt2 ) . the resulting theory , although equivalent to the fully _ uncontracted _ @xmath0-electron valence perturbation theory,@xcite exhibits a lower computational scaling than the common contracted approximations for large active spaces , and avoids the need to diagonalize metric matrices or compute four - particle density matrices . since the problem is time - independent , introducing time - dependence may also be viewed as a numerical trick equivalent to the well - known laplace transformation of perturbation theory.@xcite however , because the resolvent operator is not known explicitly for an interacting hamiltonian , while time - evolution with the same hamiltonian is a relatively simple operation , the advantages of the laplace transform are greater in the multi - reference as compared to single - reference setting . in section [ sec : theory ] we review time - independent and time - dependent many - body perturbation theory , taking note of the relevant considerations for the multi - reference setting . in section [ sec : implementation ] we describe the efficient computational implementation of the time - dependent second - order @xmath0-electron valence perturbation theory . in sections [ sec : compdetails ] and [ sec : results ] we describe the computational details and investigate the performance of t - nevpt2for a number of multi - reference problems with significant dynamic correlation : ( i ) the bond dissociation in and ; ( ii ) the ground and excited states in ; and ( iii ) the potential energy curve of the chromium dimer . finally , in section [ sec : conclusions ] we present our conclusions and future outlook for the formulation . we begin with a brief overview of time - independent perturbation theory for multi - reference problems . we define the electronic hamiltonian in second - quantized form as @xmath4 where @xmath5 and @xmath6 are the usual one- and antisymmetrized two - electron integrals : @xmath7 indices @xmath8 run over the entire spin - orbital basis @xmath9 . to formulate multi - reference dynamic correlation theories , we partition the spin - orbitals into three sets : ( i ) _ core _ ( doubly - occupied ) with indices @xmath10 ; ( ii ) _ active _ with indices @xmath11 ; and ( iii ) _ external _ ( unoccupied ) with indices @xmath12 ( [ fig : mo_diagram ] ) . .,scaledwidth=25.0% ] the hamiltonian can be divided into two parts : @xmath13 where @xmath14 includes terms involving the core or external labels and @xmath15 contains only the active - space contributions : @xmath16 where an additional term is included to describe interaction between the core and active electrons . we now assume that we have determined a starting complete active - space ( cas ) wavefunction @xmath17 . to construct a multi - reference perturbation theory , we consider the wavefunction @xmath17 to be an eigenfunction of a zeroth - order hamiltonian @xmath18 . while the choice of @xmath18 is flexible , it is clear that it must be an interacting hamiltonian . a convenient choice , which does not lead to intruder - state problems , and the one we will assume in later development , is the dyall hamiltonian@xcite @xmath19 , defined as : @xmath20 where @xmath21 is the one - particle density matrix of @xmath17 . without loss of generality , we can work in the diagonal basis of the core and external generalized fock operators , @xmath22 , @xmath23 , and henceforth we will assume that we do so . we can now consider a direct expansion with respect to the perturbation @xmath24 . for example , writing the ground - state energy of @xmath25 as @xmath26 , we define the @xmath0th - order perturbation to the energy as : @xmath27 starting from our choice of @xmath28 as the reference hamiltonian , the time - independent perturbation series corresponds to the fully uncontracted @xmath0-electron valence perturbation theory ( nevpt).@xcite denoting @xmath29 as the part of @xmath30 that contributes explicitly to the first - order wavefunction @xmath31 ( @xmath32 , @xmath33 ) , the second - order energy can be written as : @xmath34 we can now observe the complications in the uncontracted multi - reference perturbation theory described above , which arise in evaluating the expression . the first - order wavefunction involves the resolvent @xmath35 , however , unlike in the single - reference setting , the inversion appears to formally require inverting in the many - particle space . in addition , @xmath31 lives in a space of determinants with @xmath36 or @xmath37 electrons in the core and active orbitals and @xmath38 or @xmath39 electrons in the external orbitals , respectively . except for very small basis sets , the resulting number of determinants is even larger than in the @xmath40-electron complete active space used to determine @xmath17 . based on these concerns , the above uncontracted multi - reference perturbation theory formulation is almost never used . instead , most implementations use an _ internal contraction _ approximation , where @xmath31 is expanded in a set of contracted many - particle basis functions as @xmath41 where @xmath42 are operators which generate excitations . different choices of @xmath42 may be made , giving rise to , for example , the strongly - contracted ( sc- ) and partially - contracted ( pc- ) variants of @xmath0-electron valence perturbation theory.@xcite although the numerical error due to internal contraction is generally small,@xcite a significant drawback is that the basis states @xmath43 are no longer orthogonal . this introduces a many - particle metric into the theory , which can either be diagonalized at large cost ( e.g. @xmath44 , where @xmath45 is the number of active orbitals ) or which leads to numerical instabilities . further , expectation values such as the energy in eq . involve the excitation operators @xmath42 . these then require the evaluation of long strings of fermion operators and the computation of high - order density matrices . our thesis here is that , when considering larger active spaces , it can be more natural and computationally efficient to work with the original uncontracted formulation , which is a direct extension of the single - reference formalism , rather than with the more commonly implemented and approximate , contracted formulations . the key is to organize the uncontracted algorithm in an appropriate way . such an organization is conveniently provided by the time - dependent formulation of the perturbation theory . we now consider multi - reference perturbation expansion from a time - dependent perspective . the general time - dependent perturbation theory@xcite writes the ground - state wavefunction as : @xmath46\ket{\psi_0}}{\langle \psi_0 | \mathcal{t } \exp \left[-i \int_{-\infty}^0 \mathrm{d}t \ ( \hat{h}^{(0 ) } + \lambda \hat{v}(t))\right]\ket{\psi_0 } } \ , \end{aligned}\ ] ] where the perturbation is switched on adiabatically at @xmath47 , i.e. @xmath48 , and the operator @xmath49 ensures time - ordering . since the perturbation @xmath30 does not depend on time ( aside from the exponentially vanishing adiabatic factor ) , the time - dependent perturbation expansion must be identical to the time - independent expansion discussed in sec . ii a. for example , the @xmath0th - order wavefunction @xmath50 is identical to the @xmath0th - order wavefunction evaluated from the rayleigh - schrdinger perturbation theory . similarly , the @xmath0th - order energy given by @xmath51 where @xmath52 and @xmath53 denotes linked contributions , is identical to [ eq : e_n ] . restricting our attention to second - order perturbation theory , [ eq : e_tdpt ] can be written more simply as : @xmath54 where only the @xmath55 contributions are included , and we have made the standard deformation of the integral from the real to imaginary @xmath56 axis with the substitution @xmath57 ( wick rotation ) . eq . is no other than the well - known laplace transform expression of the second - order energy , and is transparently equivalent to the time - independent result in [ eq : e_tipt2 ] . however , the presence of the reference hamiltonian in the exponent rather than in a denominator is a major advantage in the multi - reference setting . in particular , since the zeroth - order hamiltonian for a multi - reference problem must take the form @xmath58 where @xmath59 and @xmath15 commute , then the exponent factorizes as @xmath60 and the time - evolution can be carried out completely separately in the active and core / external spaces . ( note that this does not require using the dyall hamiltonian specifically ) . because of this decoupling we do not need to consider the large @xmath61 dimension of the first - order interacting space , and this removes the barrier to working with the uncontracted formulation of multi - reference perturbation theory . further , time - evolution with @xmath15 is an operation of similar complexity to determining @xmath17 itself , and thus does not lead to an increase in the computational scaling of the method . while it is possible , in principle , to use the structure of @xmath18 without referring to time - evolution , the time - dependent language provides a natural organization for efficient algorithms . let us now analyze the time - dependent formulation specializing to the choice of @xmath62 . we first analyze the structure of contributions to @xmath63 in [ eq : e_tdpt2 ] . it is convenient to divide @xmath55 into 8 different terms : @xmath64 } + \hat{v}^{[+1 ] } + \hat{v}^{[-1 ] } + \hat{v}^{[+2 ] } + \hat{v}^{[-2 ] } \\ \notag & + \hat{v}^{[+1 ' ] } + \hat{v}^{[-1 ' ] } + \hat{v}^{[0 ' ] } \ , \end{aligned}\ ] ] where the superscripts @xmath65 $ ] , @xmath66 $ ] , @xmath67 $ ] , @xmath68 are the conventional nevpt2 notation,@xcite and do not denote the perturbation order . explicit equations for each term of [ eq : v_prime ] are given in the appendix . operators in [ eq : v_prime ] give rise to 8 contributions to the second - order energy @xmath63 : @xmath69 } + e^{[+1 ] } + e^{[-1 ] } + e^{[+2 ] } + e^{[-2 ] } \\ \notag & + e^{[+1 ' ] } + e^{[-1 ' ] } + e^{[0 ' ] } \ .\end{aligned}\ ] ] as an example , we consider the @xmath70}$ ] energy contribution arising from double excitations from active to external space ( @xmath71 ) : @xmath72 } & = -\int_{0}^{\infty } \braket { \psi_0 | \hat{v}^{[-2]\dag}_h(\tau ) \hat{v}^{[-2]}_h(0 ) |\psi_0 } \mathrm{d}\tau \notag \\ & = -\frac{1}{16 } \int_{0}^{\infty } \sum_{\substack{wxyz\\abcd}}{{v}_{{zw}}^{{cd } } } { { v}_{{ab}}^{{xy } } } \braket { \psi_0 | { a^\dagger_{z}}(\tau){a^\dagger_{w}}(\tau){a^{\ } _ { d}}(\tau){a^{\ } _ { c}}(\tau ) { a^\dagger_{a}}{a^\dagger_{b}}{a^{\ } _ { y}}{a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \notag \\ & = -\frac{1}{8 } \int_{0}^{\infty } \sum_{\substack{wxyz\\ab}}{{v}_{{zw}}^{{ab } } } { { v}_{{ab}}^{{xy } } } e^{-({\ensuremath{\varepsilon_{a } } } + { \ensuremath{\varepsilon_{b } } } ) \tau } \braket { \psi_0 | { a^\dagger_{z}}(\tau){a^\dagger_{w}}(\tau){a^{\ } _ { y}}{a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \notag \\ & = -\frac{1}{8 } \int_{0}^{\infty } \sum_{wxyz } { \epsilon_{{zw}}^{{xy}}}(\tau ) { g_{{xy}}^{{zw}}}(\tau ) \ , \mathrm{d}\tau \ , \label{eq : e_-2_deriv}\end{aligned}\ ] ] where we defined an intermediate @xmath73 and used the property of the time - dependence of the dyall hamiltonian in the external space : @xmath74 . in [ eq : e_-2_deriv ] , @xmath75 is a two - particle one - time green s function of the dyall hamiltonian in the active space . in general , the time - ordered @xmath76-particle @xmath0-time green s function is defined in imaginary time as : @xmath77 the highest rank active - space green s function that appears in [ eq : e_tdpt2 ] arises from the @xmath55 contributions that involve 3 active - space labels , namely @xmath78}$ ] and @xmath79}$ ] . contractions of these operators yield three - particle one - time green s functions @xmath80 and @xmath81 , respectively , as shown in the appendix . the one- , two- , and three - particle , one - time , active - space green s functions are the central objects to compute in time - dependent second - order @xmath0-electron valence perturbation theory ( t - nevpt2 ) . the reduction of the energy to these compact quantities is an important strength of the time - dependent view . physically , this simplicity arises because electron correlation effects are implicitly described by the time - dependence of the green s functions . we stress that although the time - dependent second - order energy is expressed entirely in terms of reduced green s functions , reminiscent of the internal contraction approximation , there is no approximation being made [ eq : e_tdpt2 ] yields the fully uncontracted theory . thus , there is no inversion of a metric tensor , which is a bottleneck for large active spaces and basis sets in internally - contracted theories , and the most complicated object to appear is the three - particle green s function , in contrast to the four - particle density matrix in time - independent internally - contracted nevpt2 . additional costs arise from the need to compute a time - evolution and time - integration in [ eq : e_tdpt2 ] , however , as we will demonstrate in sec . iii , these tasks are not computationally difficult and can be carried out very efficiently . we now briefly discuss the implementation of t - nevpt2for complete active - space self - consistent field ( casscf ) reference wavefunctions.@xcite the t - nevpt2algorithm is summarized below and can be easily implemented using any available full configuration interaction ( fci ) or casscf computer program . the individual steps are : 1 . compute initial active - space wavefunctions at @xmath82 ( e.g. , @xmath83 , @xmath84 , etc ) . there are seven types of active - space states to be computed , namely @xmath85 , @xmath86 , @xmath87 , @xmath88 , @xmath89 , @xmath90 , and @xmath91 ( see appendix for details ) . 2 . loop over time steps starting with @xmath92 . compute reduced green s functions by evaluating the overlap matrix elements @xmath93 , where @xmath94 and @xmath95 are the active - space states ( @xmath85 , @xmath87 , @xmath68 ) at @xmath96 and @xmath82 , respectively . for example , for @xmath97 , the corresponding two - particle green s function is @xmath98 . 3 . compute correlation energy contributions at @xmath99 as the product of one- and two - electron integrals and the active - space green s functions : @xmath100 . for example , for @xmath97 , the integral prefactor @xmath101 and the corresponding energy contribution @xmath70}(\tau')= \sum_{wxyz } { \epsilon_{{zw}}^{{xy}}}(\tau ' ) \braket{\psi_{wz}(\tau')|\psi_{yx}}$ ] . 4 . if the magnitude of the correlation energy at @xmath99 is less than the energy convergence threshold ( @xmath102 ) , proceed to step 5 . if @xmath103 , propagate the active - space states @xmath95 according to the time - dependent schrdinger equation : @xmath104 where @xmath105 and @xmath106 is a step in imaginary time . return to step 2 . 5 . compute correlation energy by time - integration : @xmath107 [ eq : integrate ] is evaluated by fitting the computed values @xmath108 to an exponential function @xmath109 , followed by the analytic integration of the obtained result . depending on the desired accuracy , fitting typically requires a linear combination of 6 to 12 exponentials . the basic t - nevpt2algorithm outlined above has @xmath110 computational cost , where @xmath2 is the dimension of the active - space hilbert space , @xmath111 is the number of time steps , and @xmath45 is the number of active orbitals . the efficiency of the algorithm is greatly improved if we avoid the explicit computation and storage of the @xmath90 and @xmath91 states by contracting these wavefunctions with the corresponding two - electron integral tensors . for example , the three - particle contribution to @xmath112}$ ] can be evaluated by computing a vector @xmath113 propagating @xmath114 , and evaluating the integral : @xmath115 } \leftarrow - \frac{1}{4 } \int_{0}^{\infty } \sum_{a } e^{-{\ensuremath{\varepsilon_{a } } } \tau } \braket{v_a(\tau)|v_a } \mathrm{d}\tau \ .\end{aligned}\ ] ] avoiding the explicit time - propagation of the @xmath90 and @xmath91 states lowers the computational scaling of the t - nevpt2algorithm to @xmath116 + @xmath117 , where @xmath3 is the number of external orbitals . this is significantly less than the cost of computing the four - particle density matrix in internally - contracted time - independent theories such as sc- and pc - nevpt2 ( @xmath118 ) , when using large active spaces . for further efficiency improvements , it is desirable to reduce the number of time steps @xmath111 while maintaining the accuracy of time - propagation and time - integration . in our implementation , the time - propagation in [ eq : propagate ] is performed using the embedded runge - kutta ( erk ) ( 4,5 ) algorithm,@xcite which automatically determines the time step @xmath106 to use in a fourth - order runge - kutta propagation based on the error estimate of the fifth - order propagator approximation , as well as the specified convergence threshold @xmath119 . we begin time - propagation by setting @xmath106 to a small value ( @xmath120 ) . at each iteration , the new time step is determined as @xmath121 , where @xmath122 is the time step estimated using the erk method . the resulting algorithm typically requires 10 - 15 time - propagation steps to achieve 0.1 @xmath123accuracy in the correlation energy @xmath63 . [ fig : algorithm_errors ] demonstrates the errors in @xmath63 for the dissociation of the molecule , relative to the exact uncontracted nevpt2 correlation energy computed , for different values of @xmath119 . finally , we stress that each step of the t - nevpt2algorithm can be efficiently parallelized . in particular , computation , memory storage and time - propagation of the active - space states @xmath95 can be performed in an `` embarrassingly '' parallel fashion , with zero communication between processors . in our parallel implementation , each processor is assigned to compute a subset of states @xmath124 ( @xmath125 ) , where @xmath126 is the total number of processors ( step 1 of the algorithm above ) . at every time step @xmath99 , the @xmath127th subset @xmath128 is propagated on the @xmath127th processor ( step 4 ) . the correlation energy contributions from each subset @xmath128 can be computed by evaluating the green s function matrix elements @xmath129 , where the states @xmath95 at @xmath82 are either recomputed or accessed from memory ( steps 2 and 3 ) . we do not exploit parallelism for the energy evaluation in our current implementation , since this is a relatively inexpensive step of the algorithm . t - nevpt2was implemented as a standalone python code interfaced with the pyscf program.@xcite for details about the t - nevpt2implementation , see sec . [ sec : implementation ] . our t - nevpt2code reproduces the uncontracted nevpt2 results using the mps - pt2 algorithm reported by sharma and chan.@xcite all multi - reference methods employed the complete active - space self - consistent field ( casscf ) wavefunctions@xcite as the reference , with active spaces of @xmath0 electrons in @xmath76 orbitals denoted as ( @xmath0e , @xmath76o ) . computations using the complete active - space second - order perturbation theory ( caspt2),@xcite partially - contracted @xmath0-electron valence second - order perturbation theory ( pc - nevpt2),@xcite as well as multi - reference configuration interaction theory with single and double excitations ( mrci)@xcite were performed using the molpro package.@xcite the mrci energies were supplied with the davidson correction ; the resulting method is denoted as mrci+q.@xcite in all computations , the cc - pv@xmath130z ( @xmath130 = d , t , q , 5 ) basis sets@xcite were used , unless noted otherwise . for the chromium dimer , the t - nevpt2correlation energies were converged to @xmath119 = @xmath131 @xmath132 , while for all other systems the tighter @xmath119 = @xmath133 @xmath132parameter was used . the total energies were extrapolated to the complete basis set limit ( cbs ) using the following equations:@xcite @xmath134 where @xmath135 and @xmath136 are the casscf reference and the corresponding correlation energies , respectively , computed using the cc - pv@xmath130z basis sets . to obtain @xmath137 , the @xmath138 ( @xmath130 = q , 5 ) correlation energies were used . we begin by investigating the accuracy of t - nevpt2for the description of the bond dissociation of and the symmetric bond stretching of . [ tab : h2o , tab : n2 ] present the total energies of and computed using t - nevpt2 , mrci+q , caspt2 , as well as strongly- and partially - contracted nevpt2 . we use mrci+q as the benchmark and plot the error in energy ( @xmath139 $ ] ) along the ground - state potential energy curves ( pecs ) in [ fig : h2o , fig : n2 ] . we first compare performance of t - nevpt2with respect to caspt2 . for , pecs of both methods exhibit similar mean absolute deviations ( @xmath140 ) relative to mrci+q ( [ fig : h2o ] ) , while the t - nevpt2curve is more parallel , as demonstrated by the smaller non - parallelity error ( @xmath141= @xmath142 = 4.5 @xmath123 ) , compared to caspt2 ( @xmath141= 7.3 @xmath123 ) . in the case of ( [ fig : n2 ] ) , this situation is reversed , with caspt2 giving the smaller @xmath140errors than t - nevpt2(@xmath140= 11.8 and 34.0 @xmath123 , respectively ) , while the @xmath141errors for both methods are similar ( 9.7 and 10.4 @xmath123 ) . c1.5 cm c2.8 cm c2.8 cm c2.8 cm c2.8 cm c2.8 cm & & & & & + 0.8 & @xmath1430.32231 & @xmath1430.30424 & @xmath1430.30584 & @xmath1430.30683 & @xmath1430.30685 + 0.9 & @xmath1430.38528 & @xmath1430.36722 & @xmath1430.36799 & @xmath1430.36920 & @xmath1430.36924 + 1.0 & @xmath1430.38902 & @xmath1430.37094 & @xmath1430.37158 & @xmath1430.37305 & @xmath1430.37309 + 1.1 & @xmath1430.36348 & @xmath1430.34540 & @xmath1430.34634 & @xmath1430.34807 & @xmath1430.34812 + 1.2 & @xmath1430.32479 & @xmath1430.30684 & @xmath1430.30813 & @xmath1430.31018 & @xmath1430.31022 + 1.3 & @xmath1430.28175 & @xmath1430.26418 & @xmath1430.26536 & @xmath1430.26774 & @xmath1430.26780 + 1.4 & @xmath1430.23886 & @xmath1430.22233 & @xmath1430.21924 & @xmath1430.22212 & @xmath1430.22217 + 1.5 & @xmath1430.19900 & @xmath1430.18321 & @xmath1430.17842 & @xmath1430.18146 & @xmath1430.18151 + 1.6 & @xmath1430.16342 & @xmath1430.14836 & @xmath1430.14311 & @xmath1430.14610 & @xmath1430.14615 + 1.7 & @xmath1430.13263 & @xmath1430.11834 & @xmath1430.11277 & @xmath1430.11568 & @xmath1430.11573 + 1.8 & @xmath1430.10676 & @xmath1430.09326 & @xmath1430.08752 & @xmath1430.09029 & @xmath1430.09034 + 1.9 & @xmath1430.08570 & @xmath1430.07292 & @xmath1430.06715 & @xmath1430.06976 & @xmath1430.06981 + 2.0 & @xmath1430.06910 & @xmath1430.05693 & @xmath1430.05128 & @xmath1430.05371 & @xmath1430.05376 + 2.1 & @xmath1430.05645 & @xmath1430.04473 & @xmath1430.03931 & @xmath1430.04158 & @xmath1430.04163 + 2.2 & @xmath1430.04709 & @xmath1430.03570 & @xmath1430.03056 & @xmath1430.03271 & @xmath1430.03275 + 2.3 & @xmath1430.04035 & @xmath1430.02919 & @xmath1430.02435 & @xmath1430.02637 & @xmath1430.02641 + 2.4 & @xmath1430.03558 & @xmath1430.02456 & @xmath1430.02001 & @xmath1430.02193 & @xmath1430.02196 + 2.5 & @xmath1430.03224 & @xmath1430.02131 & @xmath1430.01701 & @xmath1430.01883 & @xmath1430.01887 + 2.6 & @xmath1430.02990 & @xmath1430.01903 & @xmath1430.01493 & @xmath1430.01668 & @xmath1430.01672 + 2.7 & @xmath1430.02826 & @xmath1430.01742 & @xmath1430.01348 & @xmath1430.01518 & @xmath1430.01522 + 2.8 & @xmath1430.02710 & @xmath1430.01628 & @xmath1430.01247 & @xmath1430.01413 & @xmath1430.01416 + c1.5 cm c2.8 cm c2.8 cm c2.8 cm c2.8 cm c2.8 cm & & & & & + 0.9 & @xmath1430.28956 & @xmath1430.27199 & @xmath1430.25600 & @xmath1430.25900 & @xmath1430.25936 + 1.0 & @xmath1430.43066 & @xmath1430.41272 & @xmath1430.39751 & @xmath1430.40049 & @xmath1430.40084 + 1.1 & @xmath1430.46373 & @xmath1430.44560 & @xmath1430.43088 & @xmath1430.43386 & @xmath1430.43421 + 1.2 & @xmath1430.44276 & @xmath1430.42467 & @xmath1430.41012 & @xmath1430.41314 & @xmath1430.41349 + 1.3 & @xmath1430.39769 & @xmath1430.37989 & @xmath1430.36513 & @xmath1430.36824 & @xmath1430.36860 + 1.4 & @xmath1430.34495 & @xmath1430.32775 & @xmath1430.31235 & @xmath1430.31559 & @xmath1430.31597 + 1.5 & @xmath1430.29323 & @xmath1430.27699 & @xmath1430.26046 & @xmath1430.26388 & @xmath1430.26427 + 1.6 & @xmath1430.24689 & @xmath1430.23201 & @xmath1430.21382 & @xmath1430.21743 & @xmath1430.21783 + 1.7 & @xmath1430.20792 & @xmath1430.19480 & @xmath1430.17441 & @xmath1430.17817 & @xmath1430.17859 + 1.8 & @xmath1430.17703 & @xmath1430.16575 & @xmath1430.14268 & @xmath1430.14649 & @xmath1430.14695 + 1.9 & @xmath1430.15385 & @xmath1430.14369 & @xmath1430.11270 & @xmath1430.11691 & @xmath1430.11746 + 2.0 & @xmath1430.13759 & @xmath1430.12873 & @xmath1430.09558 & @xmath1430.09968 & @xmath1430.10020 + 2.1 & @xmath1430.12673 & @xmath1430.11832 & @xmath1430.08447 & @xmath1430.08825 & @xmath1430.08871 + 2.2 & @xmath1430.11970 & @xmath1430.11121 & @xmath1430.07734 & @xmath1430.08080 & @xmath1430.08118 + 2.3 & @xmath1430.11521 & @xmath1430.10637 & @xmath1430.07285 & @xmath1430.07602 & @xmath1430.07633 + 2.4 & @xmath1430.11234 & @xmath1430.10307 & @xmath1430.07004 & @xmath1430.07296 & @xmath1430.07323 + 2.5 & @xmath1430.11048 & @xmath1430.10078 & @xmath1430.06827 & @xmath1430.07100 & @xmath1430.07122 + 2.6 & @xmath1430.10924 & @xmath1430.09918 & @xmath1430.06715 & @xmath1430.06972 & @xmath1430.06991 + 2.7 & @xmath1430.10839 & @xmath1430.09804 & @xmath1430.06643 & @xmath1430.06887 & @xmath1430.06903 + 2.8 & @xmath1430.10779 & @xmath1430.09721 & @xmath1430.06596 & @xmath1430.06829 & @xmath1430.06843 + 2.9 & @xmath1430.10736 & @xmath1430.09661 & @xmath1430.06565 & @xmath1430.06789 & @xmath1430.06801 + ) as a function of the o h bond length for the symmetric bond dissociation of water , relative to multi - reference configuration interaction ( mrci+q ) . all methods employed the ( 6e , 9o ) casscf reference wavefunction and the basis set . the h o h angle was fixed at 104.5@xmath144 . the inset shows deviation of the strongly and partially - contracted nevpt2 energies from the uncontracted t - nevpt2results.,scaledwidth=45.0% ] we now compare the performance of t - nevpt2with sc- and pc - nevpt2 . since the t - nevpt2results are uncontracted , the difference of the t - nevpt2and nevpt2 energies allows us to observe the errors of the internal contraction approximation . these errors are plotted in the inset of [ fig : h2o , fig : n2 ] . the effect of strong contraction in sc - nevpt2 amounts to errors of 1 to 5 @xmath123 in correlation energy for the systems studied , with significant deviations from parallelity ( 2 to 3 @xmath123 ) relative to t - nevpt2 . the errors of internal contraction in pc - nevpt2 are much smaller : 0.04 and 0.42 @xmath123for and respectively , on average . nevertheless , in the case of , the pc - nevpt2 energies exhibit a noticeable non - parallelity error ( @xmath145 0.7 @xmath123 ) , relative to t - nevpt2 , with the largest deviation of @xmath145 0.9 @xmath123at 1.88 , where both methods show steep increase of their errors with respect to mrci+q ( [ fig : n2_zoomed ] ) . e , @xmath123 ) of the three singlet states of , relative to dmrg . for caspt2 , sc - nevpt2 , and t - nevpt2 , the ( 8e , 8o ) casscf reference wavefunction was used , with all three states averaged with equal weights . the dmrg results employing the ( 12e , 28o ) active space were taken from ref . . all computations employed the cc - pvdz basis set.,scaledwidth=45.0% ] in this section , we analyze the performance of t - nevpt2for predicting the properties of excited states in the carbon dimer ( ) . specifically , we consider the three singlet electronic states of : the ground @xmath146state and the two low - lying excited states ( @xmath147and @xmath148 ) . for each state , reference wavefunctions were obtained from the state - averaged casscf computation using the ( 8e , 8o ) active space , with all three states averaged with equal weights . we employ the recent density matrix renormalization group ( dmrg ) results by wouters _ et al._@xcite as the benchmark . [ fig : c2_pec ] compares pecs of the three states computed using dmrg and t - nevpt2 , while the errors of caspt2 , sc - nevpt2 , and t - nevpt2 , relative to dmrg , are shown in [ fig : c2_diff ] . out of the three methods considered , sc - nevpt2 exhibits the smallest non - parallelity errors ( @xmath141 ) , while @xmath141of t - nevpt2is intermediate between that of sc - nevpt2 and caspt2 . both t - nevpt2and sc - nevpt2 show a more consistent performance across pecs of different electronic states than caspt2 , which can be observed by comparing the spread of the error curves in [ fig : c2_diff ] for each method . as a result , the t - nevpt2and sc - nevpt2 vertical excitation energies are in significantly better agreement with dmrg , compared to caspt2 . at r(c c ) = 2.4 @xmath149 ( near equilibrium ) , the errors in the ( @xmath150 ; @xmath151 ) vertical excitation energies are ( 0.03 ; 0.03 ) , ( 0.06 ; 0.05 ) , and ( 0.17 ; 0.11 ) ev for t - nevpt2 , sc - nevpt2 , and caspt2 , respectively . in addition , t - nevpt2and sc - nevpt2 show much smaller non - parallelity errors than caspt2 for the description of avoided crossing of the @xmath146and @xmath148states ( 2.9 3.4 @xmath149 ) . finally , we turn our attention to the chromium dimer , whose ground - state pec is notoriously difficult to describe well theoretically . the molecule has been the subject of many computational studies.@xcite it has been shown that , for a proper description of the dissociation curve , a combination of high levels of theory with very large basis sets is required . with respect to the choice of multi - reference methodology , accurate results have recently been obtained using the multi - reference - averaged quadratic coupled cluster method,@xcite auxiliary - field quantum monte carlo,@xcite and the combination of the internally - contracted perturbation theories ( caspt2 and nevpt2 ) with dmrg.@xcite the latter perturbation theory studies have demonstrated that multi - reference wavefunctions based on the valence ( 12e , 12o ) active space , which contains the @xmath152 and @xmath153 orbitals of chromium atoms , generally do not provide a sufficiently good zeroth - order approximation for perturbative treatments,@xcite and including extra @xmath154 shells is necessary to achieve quantitative agreement with the experiment.@xcite however , an open question is the effect of contraction in the perturbation theory treatment of the curve . here , we perform the first uncontracted perturbation theory study of the pec by employing our t - nevpt2algorithm . as we currently can only use casscf reference wavefunctions , we will limit our analysis to the ( 12e , 12o ) active space , and investigate the effect of the strong contraction approximation by comparing the results of t - nevpt2with sc - nevpt2 in the same active space . [ fig : cr2_pec ] shows the ground - state pecs for the dissociation curve computed using the cc - pv@xmath130z basis sets ( @xmath130 = t , q , 5 ) and extrapolated to the complete basis set limit ( cbs ) . using the small ( 12e , 12o ) active space , both sc - nevpt2 and t - nevpt2significantly overbind ( by @xmath145 0.9 ev ) around the equilibrium region ( 1.5 2.1 ) , relative to experiment.@xcite also , as previously seen , the computed pecs strongly depend on the basis set , giving rise to the large correction of the cbs extrapolation . when using the cc - pvtz basis set , both perturbative methods produce an incorrect shape of the pecs , with a barrier in the range of 2.0 2.3 . however , in the case of t - nevpt2 , the magnitude of the barrier is significantly lowered compared to sc - nevpt2 . increasing the basis set results in the disappearance of the barrier and the appearance of the characteristic `` shoulder '' on the pecs ( 2.1 2.8 ) . in this region , the cbs - extrapolated pecs for both methods exhibit a good agreement with experiment , with errors of less than 0.1 ev . [ fig : cr2_diff ] compares the sc - nevpt2 and t - nevpt2binding energies relative to experiment . the effect of the strong contraction approximation amounts to @xmath145 0.15 ev in dissociation energy in the equilibrium region and decreases monotonically towards the dissociation limit . interestingly , the recent results of guo _ et al._@xcite using sc - nevpt2 on top of a dmrg reference wavefunction in a ( 12e , 22o ) active space ( including the @xmath154 shells and relativistic effects ) yields a slight underbinding relative to experiment , by about 0.05 ev . assuming a smaller effect of contraction in the larger active space employed by guo _ , the underbinding may be corrected by using the uncontracted formulation . finally , as seen from [ fig : cr2_diff ] , in the shoulder region the uncontracted t - nevpt2pec is slightly more parallel to experiment than that of sc - nevpt2 . in this work we argued for considering a time - dependent implementations of multi - reference perturbation theory . the essential reason is that it is complicated to represent the resolvent operator for a complicated zeroth - order hamiltonian , while it is simple to represent the corresponding time - evolution of the same operator . as an example , we provided the formulation and implementation of second - order perturbation theory using the dyall hamiltonian . the corresponding theory is equivalent to fully uncontracted @xmath0-electron valence perturbation theory , but reduces the scaling relative to contracted variants of the theory , particularly with respect to the number of active orbitals ( @xmath155 versus @xmath156 scaling ) and further avoids the need to diagonalize large metric matrices . using this formulation we examined the effect of contraction in multi - reference perturbation theory in a variety of model problems , including water , nitrogen dimer , carbon dimer , and the chromium dimer , in large and realistic basis sets . several extensions of the current work can be imagined . an immediate extension is to combine the time - dependent formulation of the @xmath0-electron valence perturbation theory with dmrg reference wavefunctions , thus providing an uncontracted analogue of the recently reported combination of strongly - contracted @xmath0-electron valence perturbation theory with the dmrg reference.@xcite such an uncontracted theory will also be closely related to the recently reported mps - pt2 theory , although it would use the time - dependent dmrg algorithm , rather than relying on mps compression of the hylleraas functional as in mps - pt2.@xcite there , the efficiency and accuracy of the dmrg compression during imaginary - time evolution remains to be studied . alternatively , the generalization of higher - order diagrammatic perturbation theories and resummations using interacting zeroth - order hamiltonians is a natural extension of this work . in this context , comparison with the recent renormalized green s function techniques based on impurity formulations , which incorporate different kinds of contributions into the diagrammatic expansion , would appear interesting and fruitful.@xcite this work was supported by the us department of energy , office of science through award de - sc0008624 ( primary support for a.y.s . ) , and award de - sc0010530 ( additional support for g.k .- l.c . ) . a.y.s . would like to thank dr . qiming sun , dr . enrico ronca and sheng guo for insightful discussions . components of the perturbation operator @xmath55 ( [ eq : v_prime ] ) are defined as : @xmath158 } & = \frac{1}{4 } \sum_{ijab } { { v}_{{ab}}^{{ij } } } { a^\dagger_{a } } { a^\dagger_{b } } { a^{\ } _ { j } } { a^{\ } _ { i } } \ , \\ \hat{v}^{[+1 ] } & = \frac{1}{2 } \sum_{ijxa } { { v}_{{ax}}^{{ij } } } { a^\dagger_{a } } { a^\dagger_{x } } { a^{\ } _ { j } } { a^{\ } _ { i } } \ , \\ \hat{v}^{[-1 ] } & = \frac{1}{2 } \sum_{ixab } { { v}_{{ab}}^{{ix } } } { a^\dagger_{a } } { a^\dagger_{b } } { a^{\ } _ { x } } { a^{\ } _ { i } } \ , \\ \hat{v}^{[+2 ] } & = \frac{1}{4 } \sum_{ijxy } { { v}_{{xy}}^{{ij } } } { a^\dagger_{x } } { a^\dagger_{y } } { a^{\ } _ { j } } { a^{\ } _ { i } } \ , \\ \hat{v}^{[-2 ] } & = \frac{1}{4 } \sum_{xyab } { { v}_{{ab}}^{{xy } } } { a^\dagger_{a } } { a^\dagger_{b } } { a^{\ } _ { y } } { a^{\ } _ { x } } \ , \\ \hat{v}^{[+1 ' ] } & = \sum_{ix } { h_{{x}}^{{i } } } { a^\dagger_{x } } { a^{\ } _ { i } } + \frac{1}{2 } \sum_{ijx } { { v}_{{xj}}^{{ij } } } { a^\dagger_{x } } { a^\dagger_{j } } { a^{\ } _ { j } } { a^{\ } _ { i } } \notag \\ & + \frac{1}{2 } \sum_{ixyz } { { v}_{{yz}}^{{ix } } } { a^\dagger_{y } } { a^\dagger_{z } } { a^{\ } _ { x } } { a^{\ } _ { i } } \ , \\ \hat{v}^{[-1 ' ] } & = \sum_{xa } { h_{{a}}^{{x } } } { a^\dagger_{a } } { a^{\ } _ { x } } + \sum_{ixa } { { v}_{{ai}}^{{xi } } } { a^\dagger_{a } } { a^\dagger_{i } } { a^{\ } _ { i } } { a^{\ } _ { x } } \notag \\ & + \frac{1}{2 } \sum_{xyaz } { { v}_{{az}}^{{xy } } } { a^\dagger_{a } } { a^\dagger_{z } } { a^{\ } _ { y } } { a^{\ } _ { x } } \ , \\ \hat{v}^{[0 ' ] } & = \sum_{ia } { h_{{a}}^{{i } } } { a^\dagger_{a } } { a^{\ } _ { i } } + \frac{1}{2 } \sum_{ija } { { v}_{{aj}}^{{ij } } } { a^\dagger_{a } } { a^\dagger_{j } } { a^{\ } _ { j } } { a^{\ } _ { i } } \notag \\ & + \sum_{ixya } { { v}_{{ay}}^{{ix } } } { a^\dagger_{a } } { a^\dagger_{y } } { a^{\ } _ { x } } { a^{\ } _ { i } } \ .\end{aligned}\ ] ] the t - nevpt2correlation energy contributions ( [ eq : e_contributions ] ) can be written as : @xmath159 } & = \frac{1}{4 } \sum_{ijab } \frac{{{v}_{{ij}}^{{ab } } } { { v}_{{ab}}^{{ij}}}}{{\ensuremath{\varepsilon_{i } } } + { \ensuremath{\varepsilon_{j } } } - { \ensuremath{\varepsilon_{a } } } - { \ensuremath{\varepsilon_{b } } } } \ , \\ \label{eq : e_+1 } e^{[+1 ] } & = -\frac{1}{2 } \sum_{xy } \int_{0}^{\infty } \sum_{ija } { { v}_{{ij}}^{{ay } } } { { v}_{{ax}}^{{ij } } } e^{({\ensuremath{\varepsilon_{i } } } + { \ensuremath{\varepsilon_{j } } } - { \ensuremath{\varepsilon_{a } } } ) \tau } \braket{\psi_0| { a^{\ } _ { y } } ( \tau ) { a^\dagger_{x } } |\psi_0 } \mathrm{d}\tau \ , \\ \label{eq : e_-1 } e^{[-1 ] } & = -\frac{1}{2 } \sum_{xy } \int_{0}^{\infty } \sum_{iab } { { v}_{{iy}}^{{ab } } } { { v}_{{ab}}^{{ix } } } e^{({\ensuremath{\varepsilon_{i } } } - { \ensuremath{\varepsilon_{a } } } - { \ensuremath{\varepsilon_{b } } } ) \tau } \braket{\psi_0| { a^\dagger_{y } } ( \tau ) { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \ , \\ \label{eq : e_+2 } e^{[+2 ] } & = -\frac{1}{8 } \sum_{wxyz } \int_{0}^{\infty } \sum_{ij } { { v}_{{ij}}^{{zw } } } { { v}_{{xy}}^{{ij } } } e^{({\ensuremath{\varepsilon_{i } } } + { \ensuremath{\varepsilon_{j } } } ) \tau } \braket{\psi_0| { a^{\ } _ { z } } ( \tau ) { a^{\ } _ { w } } ( \tau ) { a^\dagger_{y } } { a^\dagger_{x } } |\psi_0 } \mathrm{d}\tau \ , \\ \label{eq : e_-2 } e^{[-2 ] } & = -\frac{1}{8 } \sum_{wxyz } \int_{0}^{\infty } \sum_{ab } { { v}_{{zw}}^{{ab } } } { { v}_{{ab}}^{{xy } } } e^{-({\ensuremath{\varepsilon_{a } } } + { \ensuremath{\varepsilon_{b } } } ) \tau } \braket{\psi_0| { a^\dagger_{z } } ( \tau ) { a^\dagger_{w } } ( \tau ) { a^{\ } _ { y } } { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \ , \\ \label{eq : e_+1p } e^{[+1 ' ] } & = -\sum_{xy } \int_{0}^{\infty } \sum_{i } { \tilde{h}_{{i}}^{{y } } } { \tilde{h}_{{x}}^{{i } } } e^{{\ensuremath{\varepsilon_{i } } } \tau } \braket{\psi_0| { a^{\ } _ { y } } ( \tau ) { a^\dagger_{x } } |\psi_0 } \mathrm{d}\tau \notag \\ & - \sum_{wxyz } \int_{0}^{\infty } \sum_{i } { \tilde{h}_{{i}}^{{w } } } { { v}_{{zy}}^{{ix } } } e^{{\ensuremath{\varepsilon_{i } } } \tau } \braket{\psi_0| { a^{\ } _ { w } } ( \tau ) { a^\dagger_{z } } { a^\dagger_{y } } { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \notag \\ & - \frac{1}{4 } \sum_{\substack{uvw\\xyz } } \int_{0}^{\infty } \sum_{i } { { v}_{{iu}}^{{wv } } } { { v}_{{zy}}^{{ix } } } e^{{\ensuremath{\varepsilon_{i } } } \tau } \braket{\psi_0| { a^\dagger_{u } } ( \tau ) { a^{\ } _ { v } } ( \tau ) { a^{\ } _ { w } } ( \tau ) { a^\dagger_{z } } { a^\dagger_{y } } { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \ , \\ \label{eq : e_-1p } e^{[-1 ' ] } & = -\sum_{xy } \int_{0}^{\infty } \sum_{a } { \tilde{h}_{{y}}^{{a } } } { \tilde{h}_{{a}}^{{x } } } e^{-{\ensuremath{\varepsilon_{a } } } \tau } \braket{\psi_0| { a^\dagger_{y } } ( \tau ) { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \notag \\ & - \sum_{wxyz } \int_{0}^{\infty } \sum_{a } { \tilde{h}_{{w}}^{{a } } } { { v}_{{az}}^{{xy } } } e^{-{\ensuremath{\varepsilon_{a } } } \tau } \braket{\psi_0| { a^\dagger_{w } } ( \tau ) { a^\dagger_{z } } { a^{\ } _ { y } } { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \notag \\ & - \frac{1}{4 } \sum_{\substack{uvw\\xyz } } \int_{0}^{\infty } \sum_{a } { { v}_{{uv}}^{{aw } } } { { v}_{{az}}^{{xy } } } e^{-{\ensuremath{\varepsilon_{a } } } \tau } \braket{\psi_0| { a^\dagger_{u } } ( \tau ) { a^\dagger_{v } } ( \tau ) { a^{\ } _ { w } } ( \tau ) { a^\dagger_{z } } { a^{\ } _ { y } } { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \ , \\ \label{eq : e_0p } e^{[0 ' ] } & = \sum_{ia } \frac{{\tilde{h}_{{i}}^{{a } } } { \tilde{h}_{{a}}^{{i}}}}{{\ensuremath{\varepsilon_{i } } } - { \ensuremath{\varepsilon_{a } } } } + 2 \sum_{ixya } \frac{{\tilde{h}_{{a}}^{{i } } } { { v}_{{ix}}^{{ay } } } { \gamma_{{y}}^{{x } } } } { { \ensuremath{\varepsilon_{i } } } - { \ensuremath{\varepsilon_{a } } } } \notag \\ & - \sum_{wxyz } \int_{0}^{\infty } \sum_{ia } { { v}_{{iz}}^{{aw } } } { { v}_{{ay}}^{{ix } } } e^{({\ensuremath{\varepsilon_{i } } } - { \ensuremath{\varepsilon_{a}}})\tau } \braket{\psi_0| { a^\dagger_{z } } ( \tau ) { a^{\ } _ { w } } ( \tau ) { a^\dagger_{y } } { a^{\ } _ { x } } |\psi_0 } \mathrm{d}\tau \ .\end{aligned}\ ] ] in [ eq : e_0,eq : e_+1,eq : e_-1,eq : e_+2,eq : e_-2,eq : e_+1p , eq : e_-1p , eq : e_0p ] , the @xmath160 matrix elements are defined as : @xmath161 h .- j . werner , p. j. knowles , g. knizia , f. r. manby , m. schtz , p. celani , t. korona , r. lindh , a. mitrushenkov , g. rauhut , k. r. shamasundar , t. b. adler , r. d. amos , a. bernhardsson , a. berning , d. l. cooper , m. j. o. deegan , a. j. dobbyn , f. eckert , e. goll , c. hampel , a. hesselmann , g. hetzer , t. hrenar , g. jansen , c. kppl , y. liu , a. w. lloyd , r. a. mata , a. j. may , s. j. mcnicholas , w. meyer , m. e. mura , a. nicklass , d. p. oneill , p. palmieri , k. pflger , r. pitzer , m. reiher , t. shiozaki , h. stoll , a. j. stone , r. tarroni , t. thorsteinsson , m. wang , and a. wolf , molpro , version 2010.1 , a package of ab initio programs ( 2010 ) , see http://www.molpro.net ( accessed december 30 , 2015 ) .

we investigate the structure of the two permutation classes defined by the sets of forbidden patterns and . by considering how the hasse graphs of permutations in these classes can be built from a sequence of rooted source graphs , we determine their algebraic generating functions . our approach is similar to that of `` adding a slice '' , used previously to enumerate various classes of polyominoes and other combinatorial structures . to solve the relevant functional equations , we make extensive use of the kernel method

1407.0570

we consider a permutation to be simply an arrangement of the numbers @xmath2 for some positive @xmath3 . a permutation @xmath4 is said to be _ contained _ in another permutation @xmath5 if @xmath5 has a subsequence whose terms have the same relative ordering as those of @xmath4 . for example , @xmath6 is contained in @xmath7 because the subsequence @xmath8 is ordered in the same way as @xmath6 . if @xmath4 is not contained in @xmath5 then we say that @xmath5 _ avoids _ @xmath4 . for example , @xmath7 avoids @xmath9 . in the context of containment and avoidance , a permutation is often called a _ pattern_. the containment relation is a partial order on the set of all permutations , and a set of permutations closed downwards ( a down - set ) in this partial order is called a _ permutation class_. it is natural to define a permutation class by the minimal set of permutations that it avoids . this minimal forbidden set of patterns is known as the _ basis _ of the class . the class with basis @xmath10 is denoted @xmath11 . given a permutation class @xmath12 , we denote by @xmath13 the set of permutations in @xmath12 of length @xmath3 . the ( univariate ) _ generating function _ of @xmath12 is then @xmath14 , where @xmath15 is the length of @xmath5 . the _ growth rate _ of @xmath12 is defined by the limit @xmath16{|\ccc_n| } , $ ] if it exists . it is widely believed ( see the first conjecture in @xcite ) that all permutation classes have a growth rate . in the study of permutation classes , there has been significant interest in deriving the generating functions for classes with a few small basis elements ( see @xcite for an up - to - date list of results ) . this has led to the enrichment of the theory of permutation classes due to the requisite development of a variety of enumeration techniques . we add to this work by proving the following two theorems : * hasse graphs * corresponding to each permutation @xmath5 , we define an ordered plane graph @xmath17 , which we call its _ hasse graph_. if @xmath18 is the poset on the points @xmath19 in which @xmath20 if both @xmath21 and @xmath22 , then @xmath17 is the graph corresponding to the hasse diagram of @xmath18 . see the figures throughout this paper for illustrations showing the hasse graphs of permutations . in practice , we tend not to distinguish between a permutation and its hasse graph . the minimal elements of the poset @xmath18 are known as the _ left - to - right minima _ of the permutation @xmath5 . similarly , maximal elements of @xmath18 are called _ right - to - left maxima _ of @xmath5 . hasse graphs of permutations were previously considered by bousquet - mlou & butler @xcite , who determined the algebraic generating function of the family of _ forest - like _ permutations whose hasse graphs are acyclic . more recently , they have been used by the present author @xcite to establish a new lower bound for the growth rate of @xmath23 . given a permutation @xmath5 , we partition the vertices of @xmath17 by spanning it with a sequence of graphs , which we call the _ source graphs _ of @xmath5 . there is one source graph for each left - to - right minimum of @xmath5 . suppose @xmath24 are the vertices of @xmath17 corresponding to the left - to - right minima of @xmath5 , listed from left to right . then the @xmath25th source graph @xmath26 is the graph induced by @xmath27 and those vertices of @xmath17 lying above and to the right of @xmath27 that are not in @xmath28 . we refer to @xmath27 as the _ root _ of source graph @xmath26 . see figure [ figfperm ] for an illustration . the structure of the source graphs of permutations in a specific permutation class is constrained by the need to avoid the patterns in the basis of the class . if the source graphs for some class are acyclic , we refer to them as _ source trees_. the _ bottom subgraph _ of a hasse graph is the graph induced by its lowest vertex ( the least entry in the permutation ) and all the vertices lying above and to its right . observe that the bottom subgraph may contain vertices from more than one source graph . for example , the bottom subgraph of the hasse graph in figure [ figfperm ] contains vertices from three source graphs . bottom subgraphs of permutations in a specific permutation class satisfy the same structural restrictions as do the source graphs . we refer to an acyclic bottom subgraph as a _ bottom subtree_. we build the hasse graph of a permutation by starting with a source graph and then repeatedly adding another source graph to the lower right . the technique is similar to that of `` adding a slice '' , which has been used to enumerate constrained compositions and other classes of polyominoes , a topic of interest in statistical mechanics ( see , for example , bousquet - mlou s review paper @xcite , the books of van rensburg @xcite and guttmann @xcite , and flajolet & sedgewick ( * ? ? ? * examples iii.22 and v.20 ) ) . when a source graph is added , its vertices are interleaved horizontally with the non - root vertices of the bottom subgraph of the graph built from the previous source graphs . typically , the positioning of the vertices of the new source graph is constrained by the need to avoid forbidden patterns . in order to derive the univariate generating functions we require , we make use of multivariate functions involving additional `` catalytic '' variables that record certain parameters of the bottom subgraph of the permutations . these additional variables enable us to establish recurrence relations which we can then solve using the kernel method . typically , when employing a multivariate generating function , we treat it simply as a function of the relevant catalytic variable , writing , for example , @xmath29 rather than @xmath30 . occasionally , we also make use of a variant of the symbolic structural notation presented in flajolet & sedgewick @xcite to establish functional equations . in particular , @xmath31 is the atomic class consisting of a single vertex , and we use @xmath32 to represent a possibly empty sequence of elements of @xmath33 and @xmath34 to represent a non - empty sequence of elements of @xmath33 . the two classes we enumerate are quite distinct structurally . a source graph in @xmath35 consists of a root together with a @xmath36avoider formed from the non - root vertices . however , the presence of a @xmath36 forces any subsequent source graph to be simply a fan . in contrast , @xmath37 has plane source graphs and a much more uniform structure . we enumerate @xmath35 in section [ sectf ] . in doing so , the kernel method is used six times to solve the relevant functional equations . the class @xmath37 is enumerated in section [ secte ] . this requires an unusual simultaneous double application of the kernel method . let us use @xmath38 to denote @xmath35 . the structure of class @xmath38 depends critically on the presence or absence of occurrences of the pattern @xmath36 . in light of this , to enumerate this class , we partition it into three sets @xmath33 , @xmath39 and @xmath12 as follows : @xmath40 . every permutation in @xmath39 contains at least one occurrence of a @xmath36 , but avoids @xmath42 and @xmath43 . every permutation in @xmath12 contains a @xmath42 or a @xmath43 . we refer to a permutation in @xmath33 as an @xmath33permutation , and similarly for @xmath39 and @xmath12 . @xmath45 \draw [ red , thin ] ( 3,11)--(5,16 ) ; \draw [ red , thin ] ( 5,16)--(4,4)--(7,14 ) ; \draw [ red , thin ] ( 12,13)--(8,8)--(13,12)--(9,6)--(12,13 ) ; \draw [ red , thin ] ( 6,1)--(7,14 ) ; \draw [ red , thin ] ( 8,8)--(6,1)--(9,6 ) ; \draw [ red , thin ] ( 10,3)--(12,13)--(11,2)--(13,12)--(10,3)--(14,10)--(11,2)--(15,9)--(10,3)--(16,7)--(11,2)--(17,5)--(10,3 ) ; \draw [ blue , very thick ] ( 2,17)--(1,15)--(5,16 ) ; \draw [ blue , very thick ] ( 7,14)--(3,11 ) ; \draw [ blue , very thick ] ( 12,13)--(3,11)--(13,12 ) ; \draw [ blue , very thick ] ( 16,7)--(9,6)--(4,4)--(8,8)--(14,10)--(9,6)--(15,9)--(8,8 ) ; \draw [ blue , very thick ] ( 4,4)--(17,5 ) ; \draw [ blue , very thick ] ( 10,3)--(6,1)--(11,2 ) ; \plotpermnobox{}{15,17,11 , 4,16 , 1,14 , 8 , 6 , 3 , 2,13,12,10 , 9 , 7 , 5 } \draw [ thin ] ( 1,15 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 3,11 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 4,4 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 6,1 ) circle [ radius=0.4 ] ; \end{tikzpicture}\ ] ] the addition of a source graph to a @xmath12permutation can only yield another @xmath12permutation ( since it ca nt cause the removal a @xmath42 or @xmath43 pattern ) . similarly , the addition of a source graph to a @xmath39permutation ca nt result in an @xmath33permutation . hence , we can enumerate @xmath33 without first considering @xmath39 or @xmath12 , and can enumerate @xmath39 before considering @xmath12 . before investigating the structure of permutations in @xmath33 , @xmath39 and @xmath12 , let us briefly examine what a typical source graph in @xmath38 looks like . firstly , the avoidance of @xmath46 means that the non - root vertices of any source graph form a @xmath36avoider . secondly , the avoidance of @xmath47 presents no additional restriction on the structure of a source graph , because the presence of a @xmath47 would imply the presence of a @xmath36 in the non - root vertices . thus a source graph in @xmath38 consists of a root together with a @xmath36avoider formed from the non - root vertices . we begin by looking at @xmath40 . as is very well known , this class is enumerated by the catalan numbers . however , we need to keep track of the structure of the bottom graph . so we must determine the appropriate bivariate generating function . let @xmath48 denote the set of source graphs in set @xmath33 . now , each member of @xmath48 is simply a _ fan _ , a root vertex connected to a ( possibly empty ) sequence of pendant edges . bottom subgraphs are also fans . thus source graphs and bottom subgraphs of @xmath33 are acyclic . when enumerating @xmath33 , we use the variable @xmath49 to mark the _ number of leaves _ ( non - root vertices ) in the bottom subtree . the generating function for @xmath48 is thus given by @xmath50 we now consider the process of building an @xmath33permutation from a sequence of source trees . when a source tree is added to an @xmath33permutation , the root vertex of the source tree may be inserted to the left of zero or more of the leaves of the bottom subtree . see figure [ figabuild ] for an illustration . note that , in this and other similar figures , the original bottom subgraph is displayed to the upper left , with the new source graph to the lower right . the action of adding a source tree is thus seen to be reflected by the linear operator @xmath51 whose effect on @xmath52 is given by @xmath53 \;=\ ; a_\sss(u)\+(1+u+\ldots+u^k ) \;=\ ; a_\sss(u)\+\frac{1-u^{k+1}}{1-u}.\ ] ] hence , the bivariate generating function @xmath54 for @xmath33 is defined by the following recursive functional equation : @xmath55 @xmath45 \draw [ very thick ] ( 2,11)--(1,6)--(3,10 ) ; \draw [ very thick ] ( 5,9)--(1,6)--(6,8 ) ; \draw [ very thick ] ( 7,7)--(1,6 ) ; \draw [ blue , very thick ] ( 8,5)--(4,1)--(9,4 ) ; \draw [ blue , very thick ] ( 10,3)--(4,1)--(11,2 ) ; \draw [ red , thin ] ( 5,9)--(4,1)--(6,8 ) ; \draw [ red , thin ] ( 7,7)--(4,1 ) ; \plotpermnobox{}{6,11,10 , 1 , 9 , 8 , 7 , 5 , 4 , 3 , 2 } \draw [ thin ] ( 1,6 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 4,1 ) circle [ radius=0.4 ] ; \end{tikzpicture}\ ] ] this equation can be solved using the _ kernel method_. to start , we express @xmath54 in terms of @xmath56 , by expanding and rearranging to give @xmath57 equivalently , we have the equation @xmath58 now , if we set @xmath49 to be a root of the multiplier of @xmath54 , we obtain a linear equation for @xmath56 . this is known as `` cancelling the kernel '' ( the multiplier of @xmath54 being the kernel ) . the appropriate root to use can be identified from the combinatorial requirement that the series expansion of @xmath56 contains no negative exponents and has only non - negative coefficients . in this case , the correct root is @xmath59 , which yields the univariate generating function for @xmath33 , @xmath60 this is the generating function for the catalan numbers as expected . finally , by substituting for @xmath56 back into we get the following explicit algebraic expression for @xmath54 : @xmath61 on this occasion , we have explained every step of the derivation . on subsequent occasions , we present fewer details of the algebraic manipulations . we now consider set @xmath39 . recall that sets @xmath39 and @xmath12 consist of those permutations in class @xmath38 that contain at least one occurrence of a @xmath36 . we need to keep track of the position of the leftmost occurrence of a @xmath62 in such a pattern . given a permutation in @xmath39 or @xmath12 , let us call the vertex corresponding to the leftmost @xmath62 in a @xmath36 the _ spike_. in the figures , the spike is marked with a star . we now make a key observation.[obskey ] when adding a source graph to a permutation containing a @xmath36 , no vertex of the source graph may be positioned to the right of the spike , or else a @xmath47 would be created . hence , the spike in any permutation in classes @xmath39 or @xmath12 occurs in its bottom subgraph . when enumerating sets @xmath39 and @xmath12 , we use the variable @xmath49 to mark _ the number of vertices to the left of the spike _ in its bottom subgraph . @xmath45 \draw [ blue , very thick ] ( 2,15)--(1,1)--(3,14 ) ; \draw [ blue , very thick ] ( 4,13)--(1,1)--(5,10 ) ; \draw [ blue , very thick ] ( 6,9)--(1,1)--(7,7 ) ; \draw [ blue , very thick ] ( 8,4)--(1,1 ) ; \draw [ blue , very thick ] ( 14,3)--(1,1)--(15,2 ) ; \draw [ blue , very thick ] ( 9,12)--(5,10)--(10,11)--(6,9)--(9,12)--(7,7)--(10,11)--(8,4)--(9,12 ) ; \draw [ blue , very thick ] ( 7,7)--(11,8)--(8,4 ) ; \draw [ blue , very thick ] ( 12,6)--(8,4)--(13,5 ) ; \plotpermnobox{}{1,15,14,13,10,9,7,4,12,11 , 8 , 6 , 5 , 3 , 2 } \draw [ thin ] ( 1,1 ) circle [ radius=0.4 ] ; \node[above ] at ( 9,11.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture}\ ] ] let @xmath63 be the set of source graphs in set @xmath39 . since @xmath39permutations contain a @xmath36 but avoid @xmath42 and @xmath43 , the non - root vertices of a permutation in @xmath63 consist of two descending sequences , the second sequence beginning ( with the spike ) above the last vertex in the first sequence . see figure [ figbsource ] for an illustration . if we consider the non - root vertices in order from top to bottom , then it can be seen that @xmath63 is defined by the structural equation @xmath64 the first term on the right corresponds to the root and the remaining terms deal with the non - root vertices in order from top to bottom , vertices to the left of the spike being marked with @xmath49 . the third term corresponds to the spike and the fifth represents the lowest point to the left of the spike ( the rightmost @xmath65 of a @xmath36 ) . hence , the generating function for @xmath63 is @xmath66 we now study the process of building a @xmath39permutation from a sequence of source graphs . there are two cases . a permutation in @xmath39 may result either from the addition of a source graph to an @xmath33permutation , or else from adding a source graph to another @xmath39permutation . we address these two cases in turn . @xmath45 \draw [ very thick ] ( 2,17)--(1,12)--(4,16 ) ; \draw [ very thick ] ( 5,15)--(1,12)--(6,14 ) ; \draw [ very thick ] ( 7,13)--(1,12 ) ; \draw [ blue , very thick ] ( 8,11)--(3,2)--(9,9 ) ; \draw [ blue , very thick ] ( 10,8)--(3,2)--(11,5 ) ; \draw [ blue , very thick ] ( 12,4)--(3,2 ) ; \draw [ blue , very thick ] ( 16,3)--(3,2 ) ; \draw [ blue , very thick ] ( 9,9)--(13,10)--(10,8 ) ; \draw [ blue , very thick ] ( 11,5)--(13,10)--(12,4)--(14,7)--(11,5)--(15,6)--(12,4 ) ; \draw [ red , thin ] ( 5,15)--(3,2)--(4,16 ) ; \draw [ red , thin ] ( 7,13)--(3,2)--(6,14 ) ; \plotpermnobox{}{12,17 , 2,16,15,14,13,11 , 9 , 8 , 5 , 4,10 , 7 , 6 , 3 } \draw [ thin ] ( 1,12 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 3,2 ) circle [ radius=0.4 ] ; \node[above ] at ( 13,9.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture } \qquad\qquad \begin{tikzpicture}[scale=0.225,line join = round ] \draw [ very thick ] ( 2,17)--(1,12)--(4,16 ) ; \draw [ very thick ] ( 5,15)--(1,12)--(11,14 ) ; \draw [ very thick ] ( 12,13)--(1,12 ) ; \draw [ blue , very thick ] ( 6,11)--(3,2)--(7,10 ) ; \draw [ blue , very thick ] ( 8,9)--(3,2)--(9,8 ) ; \draw [ blue , very thick ] ( 10,7)--(3,2)--(13,6 ) ; \draw [ blue , very thick ] ( 14,5)--(3,2)--(15,4 ) ; \draw [ blue , very thick ] ( 16,3)--(3,2 ) ; \draw [ red , thin ] ( 5,15)--(3,2)--(4,16 ) ; \draw [ red , thin ] ( 11,14)--(6,11)--(12,13)--(7,10)--(11,14)--(8,9)--(12,13)--(9,8)--(11,14)--(10,7)--(12,13 ) ; \plotpermnobox{}{12,17 , 2,16,15,11,10 , 9 , 8 , 7,14,13 , 6 , 5 , 4 , 3 } \draw [ thin ] ( 1,12 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 3,2 ) circle [ radius=0.4 ] ; \node[above ] at ( 11,13.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture } \qquad\qquad \begin{tikzpicture}[scale=0.225,line join = round ] \draw [ very thick ] ( 2,17)--(1,12)--(4,16 ) ; \draw [ very thick ] ( 5,15)--(1,12)--(11,14 ) ; \draw [ very thick ] ( 12,13)--(1,12 ) ; \draw [ blue , very thick ] ( 6,11)--(3,2)--(7,9 ) ; \draw [ blue , very thick ] ( 8,8)--(3,2)--(9,5 ) ; \draw [ blue , very thick ] ( 10,4)--(3,2 ) ; \draw [ blue , very thick ] ( 16,3)--(3,2 ) ; \draw [ blue , very thick ] ( 7,9)--(13,10)--(8,8 ) ; \draw [ blue , very thick ] ( 9,5)--(13,10)--(10,4)--(14,7)--(9,5)--(15,6)--(10,4 ) ; \draw [ red , thin ] ( 5,15)--(3,2)--(4,16 ) ; \draw [ red , thin ] ( 11,14)--(6,11)--(12,13)--(7,9)--(11,14)--(8,8)--(12,13)--(9,5)--(11,14)--(10,4)--(12,13 ) ; \plotpermnobox{}{12,17 , 2,16,15,11 , 9 , 8 , 5 , 4,14,13,10 , 7 , 6 , 3 } \draw [ thin ] ( 1,12 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 3,2 ) circle [ radius=0.4 ] ; \node[above ] at ( 11,13.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture}\ ] ] one way to create a @xmath39permutation from an @xmath33permutation is to add a source graph from @xmath63 , positioning its root to the left of zero or more of the leaves of the bottom subtree of the @xmath33permutation and its non - root vertices to the right of the bottom subtree . in this case , the new permutation inherits its spike from the added source graph . this is illustrated in the left diagram in figure [ figabbuild ] . the generating function for this set of permutations is thus given by @xmath67 for simplicity , we choose not to present the expanded form of @xmath68 , or that of most subsequent expressions . they can all be represented in the form @xmath69 for appropriate polynomials @xmath70 , @xmath71 and @xmath72 . the other possibility for creating a @xmath39permutation from an @xmath33permutation involves the positioning of some non - root vertices of the source graph to the left of some of the leaves in the bottom subtree , making one of the vertices in the original bottom subtree the spike . the source graph may be drawn from either @xmath48 or @xmath63 , as illustrated in the centre and right diagrams in figure [ figabbuild ] . in this situation , if the source graph has a spike , it must be positioned to the right of all leaves in the bottom subtree , or else a @xmath46 would be created . furthermore , any source graph vertices placed to the left of leaves in the bottom subtree must occur at the same position in the bottom subtree , or else a @xmath42 would be created . this position may be chosen independently of where the root vertex is placed . from these considerations , it can be determined that the resulting set of permutations has the generating function defined by @xmath73 where the presence of the derivative @xmath74 is a consequence of the independent choice of two positions in the bottom tree . @xmath45 \draw [ very thick ] ( 2,12)--(1,5)--(3,10)--(10,11 ) ; \draw [ very thick ] ( 5,9)--(1,5)--(6,7 ) ; \draw [ very thick ] ( 1,5)--(12,6 ) ; \draw [ very thick ] ( 11,8)--(6,7)--(10,11)--(5,9 ) ; \draw [ blue , very thick ] ( 7,4)--(4,1)--(8,3 ) ; \draw [ blue , very thick ] ( 9,2)--(4,1 ) ; \draw [ red , thin ] ( 5,9)--(4,1)--(6,7 ) ; \draw [ red , thin ] ( 10,11)--(7,4)--(11,8)--(8,3)--(10,11)--(9,2)--(11,8 ) ; \draw [ red , thin ] ( 7,4)--(12,6)--(8,3 ) ; \draw [ red , thin ] ( 9,2)--(12,6 ) ; \plotpermnobox{}{5,12,10 , 1 , 9 , 7 , 4 , 3 , 2,11 , 8 , 6 } \draw [ thin ] ( 1,5 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 4,1 ) circle [ radius=0.4 ] ; \node[above ] at ( 10,10.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture}\ ] ] finally , we consider the addition of a source graph to a @xmath39permutation . as we noted in our key observation on page , no vertex of the source graph may be positioned to the right of the spike in the bottom subgraph . as a result , the new source graph may not contain a @xmath36 or else a @xmath46 would be created , so the source graph must be a member of @xmath48 ( a fan ) . moreover , the leaves of the source tree must be positioned _ immediately _ to the left of the spike , or else a @xmath46 would be created . see figure [ figbbbuild ] for an illustration . note that , as a consequence of these restrictions , it is impossible for the addition of a source graph to a @xmath39permutation to create a @xmath42 or @xmath43 . so it is not possible to extend a @xmath39permutation so as to create a @xmath12permutation . thus the bivariate generating function @xmath75 of set @xmath39 is defined by the following recursive functional equation : @xmath76 where the final term reflects the addition of a source tree to a @xmath39permutation . this equation is amenable to the kernel method . it can be rearranged to express @xmath75 in terms of @xmath77 . the kernel can then be cancelled by setting @xmath59 , which yields an expression for @xmath77 : @xmath78 substitution then results in an explicit algebraic expression for @xmath75 , which we refrain from presenting explicitly due to its size . @xmath45 \draw [ blue , very thick ] ( 2,15)--(1,1)--(3,14 ) ; \draw [ blue , very thick ] ( 1,1)--(4,10)--(6,13)--(5,9 ) ; \draw [ blue , very thick ] ( 5,9)--(1,1)--(7,7 ) ; \draw [ blue , very thick ] ( 8,4)--(1,1 ) ; \draw [ blue , very thick ] ( 12,8)--(11,3)--(1,1)--(15,2 ) ; \draw [ blue , very thick ] ( 9,12)--(4,10)--(10,11)--(5,9)--(9,12)--(7,7)--(10,11)--(8,4)--(9,12 ) ; \draw [ blue , very thick ] ( 7,7)--(12,8)--(8,4 ) ; \draw [ blue , very thick ] ( 13,6)--(8,4)--(14,5)--(11,3)--(13,6 ) ; \plotpermnobox{}{1,15,14,10 , 9,13 , 7 , 4,12,11 , 3 , 8 , 6 , 5 , 2 } \draw [ thin ] ( 1,1 ) circle [ radius=0.4 ] ; \node[above ] at ( 6,12.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture}\ ] ] we begin our enumeration of @xmath12 by counting its set of source graphs , which we denote @xmath79 . rather than doing this directly , we enumerate all the source graphs that contain a @xmath36 ( i.e. those in either @xmath63 or @xmath79 ) and then subtract those in @xmath63 . to begin , we consider how we might build an _ arbitrary _ source graph in class @xmath38 by adding vertices from left to right . suppose we have a partly formed source graph with at least one non - root vertex , whose rightmost vertex is @xmath80 , and we want to add further vertices to its right . what are the options ? if @xmath80 is not the lowest of the non - root vertices , then any subsequent vertices must be placed lower than @xmath80 . the only other restriction is that vertices must be positioned higher than the root . if we use @xmath81 to mark the number of positions in which a vertex may be inserted , then the action of adding a new vertex can be seen to be reflected by the following linear operator : @xmath82 \;=\ ; z\+y^2\+\frac{f(1)-f(y)}{1-y}.\ ] ] we choose to denote this operator @xmath83 because it corresponds to the action used in building a ukasiewicz path . now let us consider source graphs that have no vertices to the right of the spike . these are in @xmath63 , so let s call this set @xmath84 . as usual , we mark with @xmath49 those vertices to the left of the spike . if , in addition , we mark with @xmath81 those vertices not above the spike , then @xmath84 is defined by the structural equation @xmath85 it is readily seen that @xmath81 correctly marks the number of positions in which an additional vertex may be inserted to the right . let @xmath86 . since every member of @xmath87 is built from an element of @xmath84 by applying @xmath83 zero or more times , it follows that the generating function for @xmath87 is defined by the recursive functional equation @xmath88 this equation can be solved for @xmath89 by the kernel method , using @xmath90 to cancel the kernel . the generating function for @xmath79 is then defined by @xmath91 we now study the process of building a @xmath12permutation from a sequence of source graphs . as with set @xmath39 , there are two cases . a permutation in @xmath12 may result either from the addition of a source graph to an @xmath33permutation , or else from adding a source graph to another @xmath12permutation . ( as we observed above , it is not possible to create a @xmath12permutation by adding a source graph to a @xmath39permutation . ) we address the two cases in turn . @xmath45 \draw [ very thick ] ( 2,19)--(1,14)--(4,18 ) ; \draw [ very thick ] ( 5,17)--(1,14)--(6,16 ) ; \draw [ very thick ] ( 7,15)--(1,14 ) ; \draw [ blue , very thick ] ( 8,13)--(3,1)--(9,10 ) ; \draw [ blue , very thick ] ( 10,9)--(3,1)--(11,8 ) ; \draw [ blue , very thick ] ( 12,7)--(3,1)--(13,5 ) ; \draw [ blue , very thick ] ( 15,4)--(3,1)--(16,3 ) ; \draw [ blue , very thick ] ( 3,1)--(19,2 ) ; \draw [ blue , very thick ] ( 14,12)--(10,9)--(17,11)--(9,10)--(14,12)--(13,5 ) ; \draw [ blue , very thick ] ( 11,8)--(14,12)--(12,7)--(17,11)--(11,8 ) ; \draw [ blue , very thick ] ( 17,11)--(13,5)--(18,6)--(15,4)--(17,11)--(16,3)--(18,6 ) ; \draw [ red , thin ] ( 4,18)--(3,1)--(5,17 ) ; \draw [ red , thin ] ( 6,16)--(3,1)--(7,15 ) ; \plotpermnobox{}{14,19 , 1,18,17,16,15,13,10 , 9 , 8 , 7 , 5,12 , 4 , 3,11 , 6 , 2 } \draw [ thin ] ( 1,14 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 3,1 ) circle [ radius=0.4 ] ; \node[above ] at ( 14,11.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture } \qquad\qquad\qquad\quad \begin{tikzpicture}[scale=0.225,line join = round ] \draw [ very thick ] ( 2,19)--(1,14)--(4,18 ) ; \draw [ very thick ] ( 8,17)--(1,14)--(10,16 ) ; \draw [ very thick ] ( 13,15)--(1,14 ) ; \draw [ blue , very thick ] ( 5,13)--(3,1)--(6,10 ) ; \draw [ blue , very thick ] ( 7,9)--(3,1)--(9,8 ) ; \draw [ blue , very thick ] ( 11,7)--(3,1)--(12,5 ) ; \draw [ blue , very thick ] ( 15,4)--(3,1)--(16,3 ) ; \draw [ blue , very thick ] ( 3,1)--(19,2 ) ; \draw [ blue , very thick ] ( 14,12)--(7,9)--(17,11)--(6,10)--(14,12)--(12,5 ) ; \draw [ blue , very thick ] ( 9,8)--(14,12)--(11,7)--(17,11)--(9,8 ) ; \draw [ blue , very thick ] ( 17,11)--(12,5)--(18,6)--(15,4)--(17,11)--(16,3)--(18,6 ) ; \draw [ red , thin ] ( 4,18)--(3,1 ) ; \draw [ red , thin ] ( 8,17)--(5,13)--(10,16)--(6,10)--(8,17)--(7,9)--(10,16)--(9,8)--(13,15 ) ; \draw [ red , thin ] ( 5,13)--(13,15)--(6,10 ) ; \draw [ red , thin ] ( 7,9)--(13,15)--(11,7 ) ; \draw [ red , thin ] ( 12,5)--(13,15 ) ; \plotpermnobox{}{14,19 , 1,18,13,10 , 9,17 , 8,16 , 7 , 5,15,12 , 4 , 3,11 , 6 , 2 } \draw [ thin ] ( 1,14 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 3,1 ) circle [ radius=0.4 ] ; \node[above ] at ( 8,16.6 ) { $ { } ^\bigstar$ } ; \end{tikzpicture}\ ] ] one way to create a @xmath12permutation from an @xmath33permutation is to add a source graph from @xmath79 , positioning its root to the left of zero or more of the leaves of the bottom subtree of the @xmath33permutation and its non - root vertices to the right of the bottom subtree . this is illustrated in the left diagram in figure [ figacbuild ] . the generating function for this set of permutations is thus @xmath92 the other method for creating a @xmath12permutation from an @xmath33permutation involves the positioning of some non - root vertices of the source graph to the left of some of the leaves in the bottom subtree . this is illustrated in the right diagram in figure [ figacbuild ] . in analysing this method , it is more convenient to look , more generally , at how an @xmath33permutation can be extended to yield a permutation containing a @xmath36 , in either @xmath39 or @xmath12 . we can then subtract those members of @xmath39 that are enumerated by @xmath93 . we achieve the enumeration by adding vertices from left to right in four steps : the first step adds the root . the second step adds the first non - root vertex , which determines the position of the new spike , and also any other vertices positioned to the left of the spike . the third step adds any additional vertices to the right of the spike but to the left of some other leaves in the bottom subtree . the addition of such vertices creates occurrences of @xmath42 . finally , the fourth step adds any vertices to the right of the bottom subtree . step 1 : permutations that result from the addition of the root vertex are enumerated by @xmath94 step 2 : in this step , we insert the descending sequence of vertices that creates the new spike . in the generating function for permutations resulting from this action , we introduce two additional catalytic variables that we require for steps 3 and 4 . for use in step 3 , @xmath80 marks the number of source tree leaves to the right of the new spike . for step 4 , we use @xmath81 to mark valid positions for the insertion of subsequent vertices , as we did previously . the generating function is @xmath95 step 3 : the effect of adding additional vertices to the right of the spike but to the left of some other leaves in the bottom subtree is represented by the recursive functional equation @xmath96 again , the kernel method can be used to solve this for @xmath97 , the kernel being cancelled by setting @xmath98 . step 4 : finally , the addition of vertices to the right of the bottom subtree is reflected by the ukasiewicz operator @xmath83 , giving rise to the recursive functional equation @xmath99 which can be solved for @xmath100 by cancelling the kernel with @xmath90 . the generating function for the set of permutations resulting from the second way of creating a @xmath12permutation from an @xmath33permutation is then defined by @xmath101 our work is almost complete . we only have to consider how a source graph may be added to a @xmath12permutation . in fact , the situation is extremely constrained . first , as noted earlier , the source graph must be positioned to the left of the spike . furthermore , the presence of a @xmath42 or @xmath43 means that the addition of a source graph with even a single non - root vertex would create a @xmath46 . so the only possibility is the addition of a trivial ( single vertex ) source tree . thus the bivariate generating function @xmath102 of set @xmath12 is defined by the following recursive functional equation : @xmath103 where the final term reflects the addition of a trivial source tree to a @xmath12permutation . this equation can be solved to yield the following expression for @xmath104 by a sixth and final application of the kernel method , cancelling the kernel by setting @xmath105 : @xmath106 we now have all we need to prove theorem [ thmf ] by obtaining an explicit expression for the generating function that enumerates class @xmath38 . since @xmath38 is the disjoint union of @xmath33 , @xmath39 and @xmath12 , its generating function is given by @xmath107 . thus , by appropriate expansion and simplification , the generating function for @xmath35 can be shown to be equal to @xmath108 this has singularities at @xmath109 , @xmath110 and @xmath111 . hence , the growth rate of @xmath35 is equal to 4 , the reciprocal of the least of these . the first twelve terms of the sequence @xmath112 are 1 , 2 , 6 , 22 , 89 , 376 , 1611 , 6901 , 29375 , 123996 , 518971 , 2155145 . more values can be found at http://oeis.org/a165540[a165540 ] in oeis @xcite . @xmath113 \draw [ red , thin ] ( 6,15)--(7,23 ) ; \draw [ red , thin ] ( 6,15)--(11,20)--(9,4)--(10,21)--(6,15 ) ; \draw [ red , thin ] ( 13,14)--(17,16)--(16,13 ) ; \draw [ blue , very thick ] ( 8,24)--(7,23)--(1,19)--(2,26)--(4,27)--(3,25)--(1,19 ) ; \draw [ blue , very thick ] ( 4,27)--(5,28 ) ; \draw [ blue , very thick ] ( 1,19)--(10,21)--(12,22)--(11,20)--(1,19 ) ; \draw [ blue , very thick ] ( 6,15)--(17,16)--(19,18 ) ; \draw [ blue , very thick ] ( 13,14)--(9,4)--(14,12)--(16,13)--(15,11)--(9,4 ) ; \draw [ blue , very thick ] ( 26,10)--(25,9)--(21,8)--(9,4)--(22,7)--(25,9)--(24,6)--(23,5)--(9,4)--(20,11 ) ; \plotpermnobox{}{19,26,25,27,28,15,23,24,4,21,20,22,14,12,11,13,16,17,18,11,8,7,5,6,9,10 } \draw [ thin ] ( 1,19 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 6,15 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 9,4 ) circle [ radius=0.4 ] ; \end{tikzpicture}\ ] ] let us use @xmath114 to denote @xmath37 . what can we say about the structure of source graphs in @xmath114 ? firstly , since @xmath115 \draw [ ] ( 1,1)--(2,2 ) ; \draw [ ] ( 3,4)--(2,2)--(4,3 ) ; \plotpermnobox{}{1,2,4,3 } \end{tikzpicture } } $ ] may not occur as a subgraph , only the root of a source graph may fork towards the upper right . secondly , each source graph in @xmath114 is _ plane_. this is the case because every non - plane graph contains a @xmath116 \draw [ ] ( 1,2)--(3,4)--(2,1)--(4,3)--(1,2 ) ; \plotpermnobox{}{2,1,4,3 } \end{tikzpicture } } $ ] , and , furthermore , any @xmath117 in a source graph occurs as part of a @xmath118 ( where the @xmath119 is the root of the source graph ) . but this is impossible in @xmath114 , since @xmath118 does not avoid @xmath120 . @xmath121 \draw [ blue , very thick ] ( 2,18)--(1,1)--(3,17 ) ; \draw [ blue , very thick ] ( 4,15)--(1,1)--(5,14 ) ; \draw [ blue , very thick ] ( 10,11)--(1,1)--(13,10 ) ; \draw [ blue , very thick ] ( 14,8)--(1,1)--(15,5 ) ; \draw [ blue , very thick ] ( 16,4)--(1,1)--(17,2 ) ; \draw [ ultra thick , black!40!green ] ( 5,14)--(6,16)--(4,15 ) ; \draw [ ultra thick , black!40!green ] ( 6,16)--(7,19)--(3,17 ) ; \draw [ ultra thick , black!40!green ] ( 2,18)--(7,19)--(9,21 ) ; \draw [ ultra thick , black!40!green ] ( 10,11)--(12,13 ) ; \draw [ ultra thick , black!40!green ] ( 17,2)--(18,3)--(19,6)--(16,4 ) ; \draw [ ultra thick , black!40!green ] ( 15,5)--(19,6)--(20,7)--(21,9)--(14,8 ) ; \plotpermnobox{21}{1,18,17,15,14,16,19,20,21,11,12,13,10,8,5,4,2,3,6,7,9 } \draw [ thin ] ( 1,1 ) circle [ radius=0.4 ] ; \draw [ thick , black!40!green ] ( 9,21 ) circle [ radius=0.4 ] ; \draw [ thick , black!40!green ] ( 12,13 ) circle [ radius=0.4 ] ; \draw [ thick , black!40!green ] ( 13,10 ) circle [ radius=0.4 ] ; \draw [ thick , black!40!green ] ( 21,9 ) circle [ radius=0.4 ] ; \end{tikzpicture}\ ] ] if we combine these two observations , we see that the non - root vertices of a source graph consist of a sequence of inverted subtrees whose roots are right - to - left maxima . the avoidance of @xmath122 \draw [ ] ( 1,2)--(2,3)--(4,4)--(3,1 ) ; \plotpermnobox{}{2,3,1,4 } \end{tikzpicture } } $ ] places restrictions on the structure of the subtrees , so that they must consist of a path at the lower right , which we call the _ trunk _ , with pendant edges attached to its left . it is readily seen that these correspond to permutations in @xmath123 . we call trees of this form _ u - trees _ , short for _ unbalanced _ trees . see figure [ figesource ] for an illustration of a source graph constructed from u - trees . the class @xmath124 of u - trees satisfies the structural equation @xmath125 where the first term on the right represents the lowest leaf at the tip of the trunk and the second represents the remainder of the vertices in the trunk , each with a ( possibly empty ) sequence of pendant edges attached to the upper left . hence the generating function for @xmath124 is @xmath126 if we use @xmath49 to mark the number of u - trees , the class @xmath127 of source graphs satisfies the structural equation @xmath128 and thus has bivariate generating function @xmath129 let us now examine how a permutation in @xmath114 can be built from a sequence of source graphs . observe that , when a source graph is added , no vertex of the source graph can be positioned between two vertices of a u - tree in the bottom subgraph , because otherwise a @xmath130 would be created . in addition , there are strong constraints on when u - trees in the new source graph can be positioned to the left of a u - tree in the bottom subgraph . @xmath131 \draw [ very thick ] ( 2,19)--(1,8)--(3,18 ) ; \draw [ very thick ] ( 5,16)--(1,8)--(6,15 ) ; \draw [ very thick ] ( 9,13)--(1,8)--(10,12 ) ; \draw [ very thick ] ( 12,10)--(1,8)--(13,9 ) ; \draw [ blue , very thick ] ( 15,6)--(8,1)--(16,5 ) ; \draw [ blue , very thick ] ( 18,3)--(8,1)--(19,2 ) ; \draw [ red , thin ] ( 9,13)--(8,1)--(10,12 ) ; \draw [ red , thin ] ( 12,10)--(8,1)--(13,9 ) ; \plotpermnobox{20}{8,19,18,20,16,15,17,1,13,12,14,10,9,11,6,5,7,3,2,4 } \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 3,18)--(4,20)--(2,19 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 6,15)--(7,17)--(5,16 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 10,12)--(11,14)--(9,13 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 13,9)--(14,11)--(12,10 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 16,5)--(17,7)--(15,6 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 19,2)--(20,4)--(18,3 ) ; \draw [ thin ] ( 1,8 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 8,1 ) circle [ radius=0.4 ] ; \end{tikzpicture } } \qquad\qquad\qquad\quad \begin{tikzpicture}[scale=0.2,line join = round ] \draw [ very thick ] ( 2,24)--(1,13)--(3,23 ) ; \draw [ very thick ] ( 6,21)--(1,13)--(7,20 ) ; \draw [ very thick ] ( 9,18)--(1,13)--(10,17 ) ; \draw [ very thick ] ( 1,13)--(17,14 ) ; \draw [ ultra thick , black!40!green ] ( 19,16)--(17,14 ) ; \draw [ blue , very thick ] ( 12,12)--(5,1)--(13,11 ) ; \draw [ blue , very thick ] ( 14,9)--(5,1)--(15,8 ) ; \draw [ blue , very thick ] ( 20,6)--(5,1)--(21,5 ) ; \draw [ blue , very thick ] ( 23,3)--(5,1)--(24,2 ) ; \draw [ red , thin ] ( 6,21)--(5,1)--(7,20 ) ; \draw [ red , thin ] ( 9,18)--(5,1)--(10,17 ) ; \draw [ red , thin ] ( 12,12)--(17,14)--(13,11 ) ; \draw [ red , thin ] ( 16,10)--(17,14 ) ; \plotpermnobox{25}{13,24,23,25,1,21,20,22,18,17,19,12,11,9,8,10,14,15,16,6,5,7,3,2,4 } \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 3,23)--(4,25)--(2,24 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 7,20)--(8,22)--(6,21 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 10,17)--(11,19)--(9,18 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 15,8)--(16,10)--(14,9 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 21,5)--(22,7)--(20,6 ) ; \draw [ very thick , black!40!green , fill = black!40!green!65!white ] ( 24,2)--(25,4)--(23,3 ) ; \draw [ thin ] ( 1,13 ) circle [ radius=0.4 ] ; \draw [ thin ] ( 5,1 ) circle [ radius=0.4 ] ; \end{tikzpicture}\ ] ] these conditions result in there being two distinct ways in which a source graph may be added . these are illustrated in figure [ figebuild ] . in the first method , the root of the source graph is positioned to the left of zero or more u - trees in the bottom subgraph and the u - trees in the source graph are positioned to the right of the bottom subgraph . the second method is more subtle . it is only applicable if the rightmost u - tree of the bottom subgraph is a path . if that is the case , then an initial sequence of u - trees in the source graph can be positioned to the left of this path subtree , as long as each of them , except possibly the last , consists of a single vertex . if the rightmost u - tree of the bottom subgraph were not a path , then a @xmath120 would be created . similarly , if a non - final u - tree consisted of more than one vertex , then a @xmath130 would be created . in order to handle this second method , we need to keep track of those source graphs in which the rightmost u - tree is a path . let @xmath132 be the class of such graphs . it satisfies the structural equation @xmath133 where @xmath49 marks the number of u - trees as before . this class thus has bivariate generating function @xmath134 in order to distinguish between those situations when the second method of adding a source graph is applicable and those when it is nt , let us use @xmath135 to denote the set of those permutations in @xmath114 whose hasse graphs have bottom subgraphs in which the rightmost u - tree is a path . we are interested in determining the two bivariate generating functions @xmath136 and @xmath137 for @xmath114 and @xmath135 respectively , where @xmath49 marks the number of u - trees _ in the bottom subgraph_. to do this , we will establish four linear operators on these generating functions that reflect the different ways in which a source graph can be added . the action of adding a source graph using the first method is readily seen to be reflected by the following linear operator : @xmath138 \;=\ ; s(u)\+\frac{f(1)-u\+f(u)}{1-u}.\ ] ] the first method creates a member of @xmath135 from an arbitrary element of @xmath114 whenever the source graph is in @xmath132 ( i.e. its rightmost u - tree is a path ) . thus the appropriate linear operator is @xmath139 \;=\ ; s_\pp(u)\+\frac{f(1)-u\+f(u)}{1-u}.\ ] ] the set , @xmath140 , of source graphs that can be added using the second method satisfies the structural equation @xmath141 in which the third term on the right identifies the u - tree which is positioned immediately to the left of the rightmost ( path ) u - tree in the bottom subgraph . this specification thus counts multiple times those source graphs that can be added in more than one way due to the presence of a non - empty initial sequence of single - vertex u - trees . note also that we do nt mark the initial sequence of single - vertex u - trees with @xmath49 . the generating function for @xmath140 is @xmath142 the action of adding a source graph using the second method is then seen to be reflected by the following linear operator : @xmath143 \;=\ ; s^\star(u)\+\frac{f_\pp(1)-f_\pp(u)}{1-u } .\ ] ] finally , let us consider when adding a source graph to an arbitrary member of @xmath135 creates another permutation in @xmath135 . the second method creates an element of @xmath135 if the source graph is in @xmath132 and its rightmost ( path ) u - tree is added to the right of the bottom subgraph . an element of @xmath135 is also created if the source graph has a single path u - tree or consists of a single vertex ( the root ) . thus the set , @xmath144 , of source graphs , counted with multiplicity , that can be added to create an element of @xmath135 satisfies the structural equation @xmath145 its generating function is @xmath146 and the corresponding linear operator is @xmath147 \;=\ ; s_\pp^\star(u)\+\frac{f_\pp(1)-f_\pp(u)}{1-u}.\ ] ] we are now in a position to derive the generating function for @xmath114 and hence prove theorem [ thme ] . from the analysis above , we know that the bivariate generating function @xmath136 of class @xmath114 is defined by the following pair of mutually recursive functional equations : @xmath148 & \!+\ ! & \oper_{\pp\ee}\big[p(u)\big ] \\[3pt ] p(u ) & = & s_\pp(u ) & \!+\ ! & \oper_{\ee\pp}\big[e(u)\big ] & \!+\ ! & \oper_{\pp\pp}\big[p(u)\big ] \end{array } .\ ] ] these can be expanded to give the following : @xmath149 @xmath150 an unusual simultaneous double application of the kernel method can then be used to yield the algebraic generating function for class @xmath114 as follows . first , we eliminate @xmath151 from and , and express @xmath152 in terms of @xmath153 and @xmath154 as a rational function . cancelling the resulting kernel , @xmath155 with the appropriate root then gives us an equation relating @xmath153 and @xmath154 . secondly , we eliminate @xmath152 from and , and express @xmath151 in terms of @xmath153 and @xmath154 . cancelling the ( same ) kernel ( using a different root ) gives a second equation relating @xmath153 and @xmath154 . finally , we eliminate @xmath154 from these two equations to yield an extremely complicated explicit expression for @xmath153 . thus , using a computer algebra system to handle the details of the algebraic manipulation , it can be determined that the generating function @xmath156 for @xmath37 has the minimal polynomial @xmath157 the growth rate of the class is given by the reciprocal of the least positive real singularity of its generating function ( * ? ? ? * theorems iv.6 and iv.7 ) . hence , by determining the location of the singularities of @xmath153 , it is possible to establish that the growth rate of class @xmath114 is approximately 5.1955 , the greatest real root of the quintic polynomial @xmath158 as required . michael albert s permlab software @xcite was of particular benefit in helping to visualise and explore the structure of permutations in the two permutation classes . the author is also grateful to robert brignall and two anonymous referees for suggestions that led to improvements in the presentation of parts of the paper .

a free boundary problem arising from materials science is studied in the one - dimensional case . the problem studied here is an obstacle problem for the non - convex energy consisting of a bending energy , tension and an adhesion energy . if the bending energy , which is a higher order term , is deleted then `` edge '' singularities of the solutions ( surfaces ) may occur at the free boundary as alt - caffarelli type variational problems . the main result of this paper is to give a singular limit of the energy utilizing the notion of-convergence , when the bending energy can be regarded as a perturbation . this singular limit energy only depends on the state of surfaces at the free boundary as seen in singular perturbations for phase transition models .

1502.04212

let us consider a non - convex higher order variational problem in the one - dimensional case , which is the obstacle problem for the energy as proposed in @xcite : @xmath1=\varepsilon^2 \int \kappa^2 ds + \int ds - \int_{\{u=\psi\ } } ( 1-\alpha)\ ds.\end{aligned}\ ] ] here a smooth function @xmath2\rightarrow\mathbb{r}$ ] called an _ obstacle _ ( function ) , a constant coefficient @xmath3 , and a continuous function @xmath4\rightarrow(0,1)$ ] are given . the function @xmath5 is an admissible function constrained above the obstacle , and @xmath6 , @xmath7 denote the curvature and the arclength of the graph of @xmath5 respectively . the first term of the energy is called _ bending energy _ , the second term _ tension _ , the third term _ adhesion energy _ , and @xmath8 is called _ adhesion coefficient_. the multiple constant of the tension is normalized to one . the adhesion coefficient @xmath8 can be inhomogeneous so that it may depend on a space variable . according to @xcite , this problem is motivated to determine the shape of membranes , interfaces or filaments on rippled surfaces ( as figure [ membranes ] ) in certain mesoscopic or nearly mesoscopic settings . the coefficients @xmath9 and @xmath8 and the obstacle function @xmath10 depend on the setting of materials , scaling and so on . in this case the graph of @xmath5 is a membrane . in this paper we consider the one dimensional model so when one considers a membrane it depends on only one direction and invariant in other direction in our setting . in addition , we regard the bending energy which is a higher order term as a perturbation , that is , only consider for sufficiently small @xmath3 . cc , width=188 ] , width=188 ] a characteristic point of this energy is to contain the adhesion energy . by this energy , surfaces shall reasonably adhere to the obstacle in order to decrease the energy . accordingly , there may occur patterns as drawn in figure [ membranes ] in this model . however that law is complicated . one of physical and mathematical concerns is to perceive such pattern formation . we shall first take the simplest approximation @xmath11 in order to consider the case @xmath9 is sufficiently small . this approximation simplifies the problem ( [ problem1 ] ) thus we can obtain many fine properties for minimizers ( this is one of the important results of this paper , see theorem [ regezero ] ) . it is rigorously stated in theorem [ regezero ] , but roughly speaking the shape of any minimizer of @xmath12 is as drawn in figure [ membrane ] . there occur `` edge '' singularities at the free boundary and their angles are determined by the adhesion coefficient at the place of the free boundary , symbolically @xmath13 ( young s equation ) . this condition has been formally given in @xcite . however , even if @xmath11 , the energy is not convex and may admit that there exist multiple minimizers lacking consistency in their shape . that is to say , for instance , either of two different states as drawn in figure [ membrane ] might be a minimizer for the same energy . since this minimizing problem is considered to be a physical model , this non - uniqueness may be due to some small effect , perhaps , of higher order terms . therefore , restoring the effect of the bending energy , it is expected to ameliorate the approximation . cc [ membrane],width=188 ] [ membrane],width=188 ] the main goal of this paper is a formulation of a singular perturbation by the bending energy , that is , to characterize the limit of minimizers of @xmath14 as @xmath15 rigorously . to this end we utilize the notion of @xmath0-convergence established by de giorgi @xcite in the 70 s ( more precisely , see @xcite , @xcite , @xcite , @xcite ) which is a convergence of energy functionals for minimizing problems . the idea of this convergence is to identify the first nontrivial term in an asymptotic expansion for the energy of perturbed problems . our main singular limit result is rigorously stated in theorem [ gammathm ] , but roughly speaking as follows : [ mainthm1 ] if we define the singular limit energy @xmath16 by @xmath17:=\int_{\partial\{u>\psi\}}4(\sqrt{2}-\sqrt{1+\alpha})\ d\mathcal{h}^0,\ ] ] then the @xmath0-convergence holds with respect to @xmath18-norm : @xmath19 where @xmath20 is a certain space of admissible functions @xmath21 . from this theorem , the following asymptotic expansion is also valid for our problem in some sense ( see @xcite , @xcite ) : @xmath22 thus it means that @xmath16 is the main effect of the bending energy and its quantity only depends on the @xmath23-measure ( the number ) of the free boundary and the adhesion coefficient there . since the adhesion coefficient at the free boundary determines the contact angle there ( @xmath13 ) as mentioned above , it also means that the effect is determined by the number of `` edge '' singularities and their angles . thanks to our singular limit result and geometrical intuition for the minimizing problem of @xmath16 , we can easily find a key effect of the bending energy . for example , if @xmath24 and there are multiple minimizers of @xmath12 then the one which has less number of edges shall be a minimizer of @xmath14 when @xmath3 is sufficiently small . more precisely , by theorem [ mainthm1 ] and the fundamental theorem of @xmath0-convergence , we obtain the following : [ limitcoro ] if @xmath25 in @xmath18 and any @xmath26 is a minimizer of @xmath14 in ( [ problem1 ] ) , then @xmath5 minimizes @xmath16 among minimizers of @xmath12 in ( [ problem1 ] ) . it turns out that we can characterize the limit of minimizers of @xmath27 as @xmath15 by our theorem . an important point of our proof of theorem [ mainthm1 ] is to prove the liminf condition of @xmath0-convergence ( see definition [ gammaconv ] ) . the proof is mainly separated into two parts . the first part is to prove that it suffices to consider more regular sequences which are `` close '' to a minimizer of @xmath12 in some sense . to state it rigorously , we introduce a notion called _ @xmath28-associate _ which explains closeness of functions . especially , the part regarding how to coincide with @xmath10 is a key point because adhering or detaching leads to a discontinuous transition in our energy . we replace a general sequence by @xmath28-associate of @xmath29-regularity converging to a minimizer of @xmath12 so that all quantities in the energy are well - defined with no increase of the energy . the second part is to obtain a lower estimate for functions @xmath28-associated with a minimizer of @xmath12 . by this procedure , we are able to handle the energy geometrically and establish a modica - mortola type inequality to prove the liminf condition . now let us briefly survey some mathematical problems related to our problem from two viewpoints . in the non - adhesive case @xmath30 , the energy in ( [ problem1 ] ) is consisting of the total squared curvature functional and the length functional : @xmath31 so - called _ euler s elastic energy_. the critical points of the energy are called _ elastica _ ( usually under boundary conditions and the length constraint for a curve @xmath32 ) . this problem is first considered by euler in 1744 @xcite . numerous authors have considered this variational problem or related problems under various constraints in order to analyze the configuration of elastic bodies ( see @xcite , @xcite , or @xcite including a very well - written summary of the history of elastic problems by one section ) . however , there are still many unclear points in this problem because of difficulties of a higher order problem . our problem is the variational problem for the euler s elastic energy with the adhesion term under an obstacle - type constraint , thus it seems to have similar difficulties . to circumvent such difficulties , we regard the bending energy as a perturbation in this paper . obstacle problems , which are variational problems under obstacle - type constraints , are motivating problems invoking free boundaries and have been studied for a long time ( @xcite , @xcite , @xcite , @xcite ) . a typical model is for the area functional ( or the dirichlet energy , the linearized version ) : @xmath33 under a boundary condition . here @xmath5 is a function on a bounded smooth domain @xmath34 . this is so - called the unilateral plateaux problem . this problem invokes the free boundary @xmath35 and in the non - coincidence set @xmath36 the graph of a solution @xmath5 is a minimal surface ( or harmonic ) . generally , if an energy is convex , bounded and coercive in a sense as ( [ problem3 ] ) then the classical variational inequality approach can work . thus we can obtain many fine properties of solutions ( at least in the linearized case ) , for instance the uniquely existence , further the regularity of solution and of its free boundary @xcite . the problem ( [ problem3 ] ) corresponds to the case no bending @xmath11 and no adhesion @xmath30 in our problem ( [ problem1 ] ) . in the problem ( [ problem3 ] ) , we only obtain a natural singular limit even if the energy is perturbed by bending ( * ? ? ? * theorem 9.5 ) . therefore , in the non - adhesive case , it seems that higher order terms can be neglected when @xmath37 unlike our problem . our problem is closely related to the alt - caffarelli type variational problem @xcite which is a model of cavitation : @xmath38 under a boundary condition , where @xmath39 is a certain function . this is not an obstacle problem explicitly but the solutions are automatically constrained above zero , thus it is equivalent to the problem with the constraint @xmath40 ( flat obstacle problem ) . this problem is one of important interface models and has been generalized variously ( for example , see @xcite generalizing the first term , @xcite the second term , or their references ) . especially , yamaura @xcite considered a non - linearized case , namely the case that the first term is replaced to the area functional . our problem is a generalization of the ( one - dimensional ) non - linearized alt - caffarelli problem regarding obstacle . indeed , if @xmath11 and @xmath41 in ( [ problem1 ] ) then it is equivalent to @xmath42 since @xmath43 in @xmath44 . thus the result in this paper is particularly valid for ( [ problem4 ] ) with area functional instead of the dirichlet energy and with continuous @xmath45\rightarrow(0,1)$ ] . in view of singular perturbation by @xmath0-convergence for variational problems , there are some related results especially in phase transition models . one of the most celebrated results is , owing to modica - mortola @xcite ( and @xcite , @xcite ) , for the energy arising from the van der waals - cahn - hilliard theory of fluid - fluid phase transitions : @xmath46 where @xmath47 satisfying a volume constraint and @xmath48 is a double - well potential function , usually it is taken as @xmath49 ( @xmath50 ) . they proved that the singular limit ( @xmath0-limit ) of the energy is proportional to the area of a transition layer . these works are generalized to the vector - valued case , anisotropic cases and multi - well potential cases ( see @xcite and references cited there ) . furthernore , there are several higher order version results for the following energy : @xmath51 one of the earliest studies is for the functional arising from the theory of smectic liquid crystals introduced by aviles - giga @xcite . the authors considered the case that @xmath52 satisfies certain boundary conditions and @xmath48 is a `` single - circle - well '' potential @xmath53 ( @xmath54 ) . the energy ( [ problem5 ] ) or ( [ problem6 ] ) or similar one also arises from , for example the theory of solid - solid phase transitions , thin - films and magnetism . there are several singular limit results for them ( @xcite , @xcite , @xcite , @xcite , @xcite ) . however , the number of results of higher order singular perturbations are limited compared with the first order cases . our result is one of the higher order singular perturbations and means that the problem ( [ problem1 ] ) can be regarded as a phase transition model as above . the cost of transition between the phase @xmath55 and @xmath36 is determined by the place of a transition layer . as mentioned above , a key point in our proof of the liminf condition of @xmath0-convergence is to reduce a general sequence to a sequence which is easy to handle . we mention that this concept resembles the `` slicing '' technique used in @xcite . this paper is organized as follows : in [ presec ] we prepare some notations and definitions . in [ ezerosec ] we first consider the case @xmath11 and derive some properties of minimizers of @xmath12 . they are useful to prove our singular limit result . in [ gammaconvsec ] we state our main singular limit theorem ( theorem [ gammathm ] ) , and we prove it in [ liminfsec ] and [ limsupsec ] . in this section , we give an energy functional by introducing several notations . let @xmath34 be a smooth bounded domain , @xmath56 be a smooth function and @xmath57 be a smooth function satisfying @xmath58 on a given @xmath59-measurable subset @xmath60 of the boundary @xmath61 , where @xmath59 denotes the @xmath62-dimensional hausdorff measure . for any positive integer @xmath63 and @xmath64 , we define the space of admissible functions @xmath65 by @xmath66 the boundary condition is in the sense of trace . the space @xmath67 is a non - empty , convex and closed set in the sobolev space @xmath68 . usually it is assumed that the partial boundary @xmath60 is not empty , however we do not assume in this paper since our problem is not trivial even if @xmath69 . let @xmath4\rightarrow\mathbb{r}$ ] be a continuous function satisfying @xmath70 , where @xmath71 and @xmath72 . for @xmath73 , we define the energy functional @xmath14 by @xmath74:=\varepsilon^2\int_\omega h_u^2\sqrt{1+|\nabla u|^2}\ dx + \int_\omega \tilde{\alpha}[u ] \sqrt{1+|\nabla u|^2}\ dx,\end{aligned}\ ] ] where @xmath75 is the mean curvature of @xmath5 : @xmath76 the coefficient @xmath77 is the redefined adhesion coefficient : @xmath78(\cdot):=\vartheta_\alpha(\ \cdot\ , u(\cdot)-\psi(\cdot))\ ] ] and @xmath79\times\mathbb{r}\rightarrow[\underline{\alpha},1]$ ] such that @xmath80 @xmath14 is well - defined on @xmath81 for any @xmath73 and especially @xmath12 is well - defined on @xmath82 . throughout this paper , we fix @xmath83 , @xmath10 , @xmath84 , @xmath60 and @xmath8 . let @xmath85 . we set @xmath86 we call @xmath87 coincidence set , @xmath88 non - coincidence set and @xmath89 free boundary . note that the non - coincidence set @xmath88 is open in @xmath83 and the coincidence set @xmath87 and the free boundary @xmath89 are closed in @xmath83 . moreover @xmath90 holds . in this section , we derive some properties of minimizers of @xmath12 in the one - dimensional case . in [ lscsec ] we verify the lower semicontinuity of @xmath12 and in [ 1-dsec ] we derive properties about the shape of minimizers of @xmath12 . these are useful to prove theorem [ gammathm ] which is our main singular limit result . [ lscgeneral ] let @xmath56 be a smooth function , @xmath91 be a borel function and @xmath92:=\int_{\omega}h(x , u-\psi,\nabla u)dx$ ] . if @xmath93 is lower semicontinuous for @xmath94 @xmath95 , then @xmath96 $ ] is lower semicontinuous . fix any convergent sequence @xmath97 in @xmath18 . for any subsequence , there exist a subsequence such that @xmath98 then @xmath99 & = \liminf_{j \to \infty}\int_\omega h(x , u^{{\varepsilon}_j}-\psi,\nabla u^{{\varepsilon}_j})\ dx\\ & \geq \int_\omega\liminf_{j \to \infty } h(x , u^{{\varepsilon}_j}-\psi,\nabla u^{{\varepsilon}_j})\ dx\\ & \geq \int_\omega h(x , u-\psi,\nabla u)\ dx = e[u ] . \end{aligned}\ ] ] the first inequality follows by fatou s lemma , and the second one follows by the lower semicontinuity of @xmath100 . thus we get the consequence . [ lscezero ] @xmath101 is lower semicontinuous . by taking @xmath102 in lemma [ lscgeneral ] and the lower semicontinuity of @xmath103 . now we assume @xmath104 and @xmath105 , bounded open interval . note that @xmath106 is contained in @xmath107)$ ] in this case . to state the main theorem in this subsection , we introduce some terminologies . 1 . let @xmath108 be a nonnegative integer . we say that @xmath109 is a _ partition _ of an interval @xmath110 in @xmath111 if @xmath112 . 2 . let @xmath113 . we say that @xmath5 is _ partitional regular _ if there exist a nonnegative integer @xmath108 and a partition @xmath114 of @xmath110 such that : 1 . the interval @xmath115 is contained in either @xmath116 or @xmath117 , 2 . if @xmath118 then @xmath119 , 3 . if @xmath120 then @xmath121 . + note that @xmath122 and @xmath123 or @xmath124 in @xmath125 alternately . we denote such regularity by _ @xmath126$]-regular _ when we want to show the partition explicitly . 3 . let @xmath5 be a function . if the derivative @xmath127 exists at @xmath128 , we define the _ tangent angle _ of @xmath5 at @xmath128 by @xmath129 . 4 . let @xmath113 be @xmath126$]-regular and @xmath108 be positive . for @xmath130 with @xmath119 ( resp . @xmath131 ) we define the _ contact angle @xmath132 _ of @xmath5 with @xmath10 at @xmath133 by @xmath134 if @xmath135 can be defined . denote the set of minimizers of a functional @xmath136 $ ] by @xmath137=\inf_{x^{m , p}(\omega)}e\right\}\right .. \ ] ] the following is the main theorem in this subsection . [ regezero ] let @xmath105 . if @xmath138 then there exists a partition @xmath139 of @xmath110 such that @xmath140 is @xmath141$]-regular , the graph of @xmath140 is a segment on any interval @xmath142 and the contact angle @xmath143 of @xmath140 with @xmath10 at @xmath144 satisfies @xmath145 for any @xmath130 . [ regrem ] this theorem especially means that the regularity of any minimizer of @xmath12 in @xmath146 is up to @xmath147 ( further , piecewise smooth ) and this is optimal . moreover , it is remarkable that the number of the connected components of the non - coincidence set of any minimizer is finite . this is not valid without the adhesion energy . to prove theorem [ regezero ] , we prepare several lemmas . [ psidelta ] there exists @xmath148 depending on @xmath10 and @xmath8 such that for any interval @xmath149 whose width is not larger than @xmath150 the inequality holds : @xmath151 for any interval @xmath152 , we have @xmath153 since @xmath154 is uniformly continuous , if the width of @xmath152 is sufficiently small then @xmath155 thus we get the consequence . [ reglem1 ] let @xmath113 , @xmath156 and @xmath149 be an interval whose width is @xmath28 . suppose that @xmath157 on @xmath158 and @xmath124 at some interior point of @xmath152 . then the function @xmath159 belongs to @xmath146 and satisfies @xmath160>e_0[v]$ ] . the function @xmath161 obviously belongs to @xmath146 since @xmath157 on @xmath158 , thus we prove the inequality @xmath160>e_0[v]$ ] . we first prove the case that @xmath124 at any interior point of @xmath152 . denote the endpoints of @xmath152 by @xmath162 and @xmath163 . then we have @xmath164-e_0[v ] \\ = & \ \int_{y_0}^{y_1}\sqrt{1+{u_x}^2}\ dx-\int_{y_0}^{y_1}\alpha\sqrt{1+{\psi_x}^2}\ dx\\ \geq&\ \sqrt{|y_1-y_0|^2+|u(y_1)-u(y_0)|^2}-\overline{\alpha}|y_1-y_0|\max_{[y_0,y_1]}\sqrt{1+{\psi_x}^2}\\ = & \ \sqrt{|y_1-y_0|^2+|\psi(y_1)-\psi(y_0)|^2}-\overline{\alpha}|y_1-y_0|\max_{[y_0,y_1]}\sqrt{1+{\psi_x}^2}\\ \geq&\ |y_1-y_0|\left(\min_{[y_0,y_1]}\sqrt{1+{\psi_x}^2}-\overline{\alpha}\max_{[y_0,y_1]}\sqrt{1+{\psi_x}^2}\right)\\ \geq&\ \delta\cdot\frac{1-\overline{\alpha}}{2}. \end{aligned}\ ] ] the last inequality follows by lemma [ psidelta ] and this implies the consequence . next we prove the general case . since @xmath5 is continuous , we can decompose @xmath165 into at most countable disjoint intervals @xmath166 , where @xmath130 and @xmath108 is a positive integer or @xmath167 . note that @xmath124 in any @xmath166 and @xmath157 at any endpoint of @xmath166 . denote @xmath168 and for all @xmath130 define @xmath169 by @xmath170 by the above paragraph , we have @xmath171>e_0[v_i]$ ] for any @xmath130 . therefore , if @xmath108 is finite then we have @xmath172 and @xmath160>e_0[v]$ ] . if @xmath173 then we have @xmath174 in @xmath146 and , by proposition [ lscezero ] , @xmath175>\liminf_{i\to\infty}e[v_i]\geq e_0[v].\ ] ] the proof is completed . [ minilem ] if @xmath138 then @xmath176 in @xmath177 , that is , the graph of @xmath140 is a segment on any connected component of @xmath178 . since @xmath140 is continuous , the set @xmath177 is open , thus the graph of @xmath140 is a minimal surface in @xmath177 . we now prove theorem [ regezero ] . fix a minimizer @xmath140 . we first prove by contradiction that the numbers of the connected components of @xmath178 and @xmath179 are finite . if either of them is infinite , then so is the other , thus there exists a sufficiently small interval @xmath152 as in lemma [ reglem1 ] . this contradicts the minimality of @xmath140 . next we prove any connected component of @xmath179 is not a point ( but an interval ) by contradiction . if there is a connected component @xmath180 then there exists @xmath181 such that the intervals @xmath182 and @xmath183 $ ] are contained in @xmath178 . note that @xmath140 is straight on @xmath182 and @xmath183 $ ] by lemma [ minilem ] . moreover , since @xmath184 , the following inequalities hold : @xmath185 if @xmath186 holds then the function @xmath187 which is equal to @xmath140 out of @xmath188 $ ] and a segment in @xmath188 $ ] belongs to @xmath146 and @xmath189>e_0[v^{r^*}]$ ] holds . this contradicts the minimality of @xmath140 . thus we may assume that the equalities are attained in ( [ regezerocond ] ) . for @xmath190 we define @xmath191\cup[x^*+r^*,b),\\ \psi & \text{in } ( x^*,x^*+r],\\ \text{segment } & \text{in } ( x^*+r , x^*+r^ * ) . \end{cases } \end{aligned}\ ] ] then there exists a sequence @xmath192 such that @xmath193 and we see @xmath194-e_0[\bar{u}]&=\int_{x^*}^{x^*+r'}\alpha\sqrt{1+\psi_x^2}\ dx\\ & + \sqrt{|r^*-r'|^2+|\bar{u}(x^*+r^*)-\psi(x^*+r')|^2}\\ & -\sqrt{|r^*|^2+|\bar{u}(x^*+r^*)-\bar{u}(x^*)|^2}. \end{aligned}\ ] ] noting that @xmath195 by @xmath196 , the equality assumption of ( [ regezerocond ] ) implies @xmath197 and the relation @xmath198 follows since @xmath140 is straight on @xmath183 $ ] , we have @xmath199-e_0[\bar{u}]}{r'}=\alpha(x^*)\sqrt{1+\bar{u}_{x+}^2(x^*)}-\sqrt{1+\bar{u}_{x+}^2(x^*)}.\ ] ] by @xmath200 , the above is negative . this also contradicts the minimality of @xmath140 . therefore , we find that any connected component of @xmath179 is an interval . finally , we compute the contact angles of @xmath140 . from the above arguments , we know that @xmath140 is @xmath201$]-regular for some partition . assume @xmath202 and fix any @xmath130 . we only consider the case @xmath142 ( the case @xmath203 is similar ) . since the partition is finite , there exists @xmath204 such that @xmath205\subset(a , b)_0^{\bar{u}}$ ] and @xmath206\subset(a , b)_+^{\bar{u}}$ ] . note that @xmath140 is straight on @xmath206 $ ] and @xmath207 . notice that @xmath208 by considering as above paragraph . thus we can perturb @xmath140 near @xmath144 for small @xmath209 as above . denoting the perturbed @xmath140 by @xmath210 , we can similarly compute as @xmath211-e_0[\bar{u}]}{r}=\alpha(\bar{x}_i)\sqrt{1+\psi_x^2(\bar{x}_i)}-\frac{1+\psi_x(\bar{x}_i)\bar{u}_{x+}(\bar{x}_i)}{\sqrt{1+\bar{u}^2_{x+}(\bar{x}_i)}}.\ ] ] since @xmath140 is a minimizer , the above quantity has to be zero , thus we obtain @xmath212 this completes the proof . [ bdryangle ] if @xmath213 and @xmath214 ( @xmath215 ) , then we obtain @xmath216 by a similar consideration . we also obtain @xmath217 . [ locminrem ] since we only used local perturbations in the above arguments , the results in this section are also valid for local minimizers @xmath140 of @xmath12 , that is , there exists @xmath3 such that @xmath160\geq e_0[\bar{u}]$ ] for any @xmath218 . now we rigorously state our main @xmath0-convergence result in the one - dimensional setting . we set @xmath105 and define @xmath219 $ ] by @xmath220 : = \begin{cases } \begin{array}{l } \varepsilon\displaystyle\int_{a}^{b } \kappa_u^2\sqrt{1+{u_x}^2}\ dx \\ \qquad+\cfrac{1}{\varepsilon}\left(e_0[u]-\displaystyle\inf_{x^{1,1}}e_0\right ) \end{array } & ( u\in x^{2,1}(a , b)),\\ \infty & ( \text{otherwise } ) , \end{cases}\end{aligned}\ ] ] where @xmath221 is curvature , and @xmath222:=\begin{cases } \displaystyle\int_{\partial(a , b)_{+}^u } 4(\sqrt{2}-\sqrt{1+\alpha})\ d\mathcal{h}^0 & ( u\in{\rm argmin}_{x^{1,1}}e_0),\\ \infty & ( \text{otherwise } ) . \end{cases}\end{aligned}\ ] ] we begin by recalling the definition of @xmath0-convergence . [ gammaconv ] let @xmath20 be a metric space and @xmath223 $ ] . we say that @xmath224 _ @xmath225-converges _ to @xmath16 as @xmath226 if the following conditions hold : 1 . for any convergent sequence @xmath227 ( @xmath226 ) in @xmath20 , @xmath228\geq f[u].\ ] ] 2 . for any @xmath229 , there exists a convergent sequence @xmath230 ( @xmath226 ) in @xmath20 such that @xmath231\leq f[u].\ ] ] we denote such convergence by @xmath232 on @xmath20 . we are now in position to state our main result on @xmath0-convergence . [ gammathm ] let @xmath224 and @xmath16 be the functionals on @xmath146 defined by ( [ feps1 ] ) and ( [ fzero1 ] ) . then @xmath232 on @xmath146 . this theorem is proved in [ liminfsec ] and [ limsupsec ] . it is one of the important properties of @xmath0-convergence that if @xmath233 and @xmath234 is a minimizer of @xmath224 then any cluster point of @xmath235 is a minimizer of @xmath16 . in our setting , the minimizing problem of @xmath224 and of @xmath14 are equivalent , thus any cluster point of @xmath236 which is a sequence of minimizers of @xmath14 is a minimizer of @xmath16 ( as mentioned in corollary [ limitcoro ] ) . in this section , we prove the liminf condition of theorem [ gammathm ] , which is the first condition of definition [ gammaconv ] . the proof is separated into three parts : [ liminfsub1][liminfsub3 ] . in [ liminfsub1 ] we show that we only have to check special sequences . in [ liminfsub2 ] we impose an essential and stronger restriction for sequences in order to obtain a lower estimate . by this restriction , we are able to consider the energy @xmath224 geometrically . in [ liminfsub3 ] we obtain a lower bound for such restricted sequences . a summary of the overall proof is given in [ liminfsub4 ] . we first give some simple assumptions for sequences . [ assumpseq ] assume that a sequence @xmath97 in @xmath146 satisfies * @xmath237<\infty$ ] , * @xmath238<\infty$ ] for all @xmath3 ( especially @xmath239 ) , * @xmath240 ( especially @xmath241 ) , * @xmath140 is @xmath141$]-regular for some @xmath202 . [ restseq ] if the liminf condition of theorem [ gammathm ] is valid for sequences @xmath97 satisfying assumption [ assumpseq ] , then the liminf condition is fulfilled for any sequence . \(1 ) if @xmath237=\infty$ ] then the liminf condition is trivial . + ( 2 ) if there exists @xmath242 such that @xmath238=\infty$ ] for any @xmath243 , then @xmath237=\infty$ ] , thus it is trivial . if not , then there exists a subsequence @xmath244 such that @xmath245<\infty$ ] for all @xmath246 and @xmath247\leq\liminf_{{\varepsilon}\to 0}f_{\varepsilon}[u^{\varepsilon}],\ ] ] thus we can assume ( 2 ) without loss of generality . + ( 3 ) if @xmath97 and @xmath160>\inf_{x^{1,1}}e_0 $ ] , then @xmath237=\infty$ ] by proposition [ lscezero ] , thus it is trivial . + ( 4 ) if @xmath248 is @xmath126$]-regular then @xmath249 thus it is trivial in the case @xmath250 . here the sum is defined to be zero when @xmath250 . in this subsection , we reduce sequences in the liminf condition of theorem [ gammathm ] while not increasing the limit inferior . this reduction is important because it becomes easy to obtain a lower estimate for such reduced sequences . to state the reduction result , we introduce a notation . [ deltaasso ] let @xmath251 and @xmath113 . we say that @xmath252 is _ @xmath28-associated _ with @xmath5 if @xmath253 and there exists a continuous strictly increasing operator @xmath254\rightarrow[a , b]$ ] such that @xmath255 for all @xmath256 $ ] and @xmath257=(a , b)_+^v$ ] . [ deltas ] cc for any partitional regular function @xmath5 and function @xmath161 which is @xmath28-associated with @xmath5 , the numbers of the connected components of @xmath258 and @xmath259 are equal and @xmath161 is partitional regular . the following proposition is the main reduction result of this subsection . [ lbprop1 ] let @xmath260 in @xmath146 as in assumption [ assumpseq ] . then there exists @xmath261 satisfying the following : for any @xmath262 there exists a sequence @xmath263 such that , 1 . @xmath264\leq\liminf_{{\varepsilon}\to 0}f_{\varepsilon}[u^{\varepsilon}]$ ] , 2 . for any @xmath265 the function @xmath266 is @xmath28-associated with @xmath140 . this is a direct consequence of the following three lemmas : lemma [ redlem1 ] , lemma [ redlem2 ] and lemma [ redlem3 ] . [ redlem1 ] let @xmath260 in @xmath146 as in assumption [ assumpseq ] and @xmath148 in lemma [ psidelta ] . then for any @xmath267 there exists a subsequence @xmath268 such that , 1 . @xmath269=\liminf_{{\varepsilon}\to 0}f_{\varepsilon}[u^{\varepsilon}]$ ] , 2 . for any @xmath270 and interval @xmath271 with width @xmath28 there exists @xmath272 such that @xmath273 . fix any @xmath267 . by definition of @xmath274 , we can take a subsequence @xmath275 such that @xmath276=\liminf_{{\varepsilon}\to 0}f_{\varepsilon}[u^{\varepsilon}]<\infty.\ ] ] we prove that this sequence has a subsequence satisfying the second condition of lemma [ redlem1 ] by contradiction . note that it suffices to prove that , for any @xmath271 whose width is @xmath277 , there exists a subsequence ( depending on @xmath152 ) touching @xmath10 somewhere in @xmath152 . indeed , since @xmath179 is covered by a finite number of intervals in @xmath179 with width @xmath277 , if we proceed to take such subsequences repeatedly for each interval of the covering , then any function of the eventual subsequence touches @xmath10 in each interval of the covering . the consequence follows since any interval in @xmath179 with width @xmath28 contains some interval of the covering . thus we suppose that for a given @xmath152 there is no such subsequence . namely , suppose that there would exist @xmath278 and some interval @xmath271 with width @xmath277 such that @xmath279 in @xmath152 for any @xmath280 . then we would obtain @xmath281&\geq e_0[u^{\varepsilon''}]-e_0[\bar{u}]\\ & \geq \int_{i } \sqrt{1+(u_{x}^{\varepsilon''})^2}\ dx-\int_{i}\alpha\sqrt{1+\psi_{x}^2}\ dx\\ & + \int_{(a , b)\setminus i } \tilde{\alpha}[u^{\varepsilon '' } ] \sqrt{1+(u_x^{\varepsilon''})^2}\ dx - \int_{(a , b)\setminus i } \tilde{\alpha}[\bar{u } ] \sqrt{1+\bar{u}_x^2}\ dx . \end{aligned}\ ] ] by lemma [ lscgeneral ] , we get @xmath282 and @xmath283 \sqrt{1+(u_x^{\varepsilon''})^2}\ dx \geq \int_a^b \chi_{(a , b)\setminus i } \tilde{\alpha}[\bar{u } ] \sqrt{1+\bar{u}_x^2}\ dx.\ ] ] therefore , by @xmath284 and lemma [ psidelta ] , we have @xmath285 & \geq \int_{i}\sqrt{1+\psi_{x}^2}-\alpha\sqrt{1+\psi_{x}^2}\ dx\\ & \geq \frac{\delta}{3}\left(\inf_{i}\sqrt{1+{\psi_x}^2}-\overline{\alpha}\sup_{i}\sqrt{1+{\psi_x}^2}\right)\\ & \geq \frac{\delta}{3}\cdot\frac{1-\overline{\alpha}}{2 } > 0 , \end{aligned}\ ] ] that is , @xmath286=\infty$ ] . this is a contradiction . [ redlem2 ] let @xmath260 in @xmath146 as in assumption [ assumpseq ] . define @xmath287 then for any @xmath288 there exists a sequence @xmath289 such that , 1 . @xmath290 in @xmath291 , 2 . @xmath292\leq\liminf_{\varepsilon\to0}f_{\varepsilon}[u^\varepsilon]$ ] , 3 . for any @xmath270 and @xmath293 with @xmath203 the equality @xmath294 holds in @xmath295 $ ] . fix @xmath288 and denote the subsequence obtained in lemma [ redlem1 ] by @xmath296 . recall that , by lemma [ redlem1 ] , for any @xmath297 and @xmath293 with @xmath203 there exist @xmath298 $ ] and @xmath299 such that @xmath300 at @xmath301 and @xmath302 . moreover , since @xmath303 is continuously differentiable and @xmath304 , we also have @xmath305 at @xmath301 and @xmath302 . we define @xmath306 by replacing @xmath303 to be @xmath10 in @xmath307 for all @xmath308 such that @xmath203 . then @xmath306 belongs to @xmath309 since @xmath310 and @xmath305 at all @xmath301 . furthermore , the sequence @xmath311 satisfies all the conditions in this lemma . indeed , the first condition follows by @xmath312 the second one by @xmath313-e_0[v_\delta^{{\varepsilon}_j}]=e_0[u_\delta^{{\varepsilon}_j}-v_\delta^{{\varepsilon}_j}+\bar{u}]-e_0[\bar{u}]\geq0\ ] ] and @xmath314-f_{{\varepsilon}_j}[v_\delta^{{\varepsilon}_j}]\geq-\varepsilon_j\int_a^b\kappa_\psi^2\sqrt{1+\psi_x^2}\ dx\xrightarrow{j\to\infty}\geq0,\ ] ] and the third one by the definition of @xmath306 . [ redlem3 ] let @xmath260 and @xmath315 as in lemma [ redlem2 ] . then for any @xmath316 there exists a sequence @xmath317 such that , 1 . @xmath318\leq\liminf_{\varepsilon\to0}f_{\varepsilon}[u^{\varepsilon}]$ ] , 2 . there exists @xmath319 such that @xmath266 is @xmath28-associated with @xmath140 for any @xmath320 . we first assume that @xmath321 on @xmath322 . fix @xmath316 . define @xmath323 for @xmath130 and denote by @xmath324 the ( disjoint ) union of @xmath325 , where @xmath326 is taken so that @xmath327 take @xmath328 as in lemma [ redlem2 ] for @xmath329 . since @xmath330 and @xmath331 as @xmath332 uniformly ( by sobolev embedding ) , there exists @xmath319 such that @xmath333 and @xmath334 in @xmath335 for any @xmath336 . on the other hand , for any @xmath337 we find @xmath338 in @xmath339 by the third condition of lemma [ redlem2 ] . therefore , to prove this lemma , we only have to reduce @xmath340 in @xmath324 . define @xmath341 for @xmath308 with @xmath203 . then we find @xmath342 on @xmath343 $ ] . thus , by lemma [ reglem1 ] , the reduction replacing @xmath344 to be @xmath10 only in @xmath345 $ ] does not increase @xmath12 and retains the function in @xmath309 . similarly , for @xmath308 with @xmath142 , we can get the result reversed left and right . the consequence follows by defining @xmath346 as the function reduced as above for all @xmath130 . indeed , the function @xmath346 is @xmath28-associated with @xmath140 for any @xmath336 by the definition of @xmath347 . here @xmath348 . moreover , the first condition follows by for any @xmath336 we have @xmath349\geq e_0[w_\delta^{{\varepsilon}_j}]$ ] and @xmath350-f_{\varepsilon_j}[w_\delta^{{\varepsilon}_j}]\geq -\varepsilon_j\int_a^b\kappa_\psi^2\sqrt{1+\psi_x^2}\ dx\xrightarrow{j\to\infty}\geq0.\ ] ] finally , we mention the case @xmath351 at @xmath352 or @xmath353 or both . in addition to the above proof , we only need to reduce @xmath354 near @xmath355 or @xmath356 or both . let us only consider for @xmath355 since it is similar for @xmath356 . when we take @xmath329 , the similar condition for @xmath357 shall be added . if @xmath358 and @xmath359 , then the consequence follows by replacing @xmath354 by @xmath10 in @xmath360 $ ] . if @xmath358 and @xmath361 , then @xmath354 may touch @xmath10 near @xmath355 even if @xmath265 is sufficiently large . nevertheless , the consequence follows by replacing @xmath354 by @xmath10 in @xmath362 $ ] , where @xmath363 \mid v_{\delta_*}^{\varepsilon_j}(y)=\psi(y)\}\ ( = s\bar{x}_0).\ ] ] the proof is completed . in this subsection , we give a lower estimate for the functions as obtained in [ liminfsub2 ] . in proposition [ lbprop2 ] , for functions @xmath28-associated with a minimizer of @xmath12 , we rewrite the energy in order to consider geometrically , and obtain a key estimate in proposition [ lbprop3 ] . [ defcurve ] 1 . we say that @xmath364\rightarrow\mathbb{r}^2 $ ] is a _ ( regular ) @xmath29-curve _ if the two components @xmath365 and @xmath366 belong to @xmath367)$ ] and @xmath368.\ ] ] 2 . for a @xmath29-curve @xmath32 , we denote by @xmath369 the geodesic ( segment ) from @xmath370 to @xmath371 . 3 . for a @xmath29-curve @xmath32 , we denote the tension ( length ) by @xmath372:=\int_0 ^ 1|\dot{\gamma}(t)|dt.\ ] ] 4 . for a @xmath29-curve @xmath32 , we denote the bending energy by @xmath373:=\int_0 ^ 1|\kappa_\gamma(t)|^2|\dot{\gamma}(t)|dt,\ ] ] where @xmath374 is the curvature of @xmath32 defined by @xmath375 . 5 . for a @xmath29-curve @xmath32 , we define the _ boundary warp energy _ by @xmath376:=4\left(2\sqrt{2}-\sqrt{1+\cos{\theta_\gamma^0}}-\sqrt{1+\cos{\theta_\gamma^1}}\right),\ ] ] where @xmath377 and @xmath378 are the _ boundary warp angles _ of @xmath32 defined by @xmath379\ ] ] at @xmath370 and @xmath371 respectively . @xmath380 , @xmath381 , @xmath382 , @xmath377 and @xmath378 are well - defined for @xmath29-curves , i.e. they are invariant by @xmath29-reparameterization . they are also invariant with respect to translation , reflection and rotation . to be more precise , see appendix . moreover , for any @xmath383 on a bounded interval , the graph of @xmath5 is a @xmath29-curve . [ lbprop2 ] let @xmath138 be @xmath141$]-regular and @xmath384 be the number of the connected components of @xmath178 . then there exists @xmath385 such that for any @xmath3 , @xmath386 and @xmath387 which is @xmath28-associated with @xmath140 the inequality holds : @xmath388\geq\sum_{k=1}^{m}\varepsilon\mathcal{b}[\gamma^k]+\frac{1}{\varepsilon}(\mathcal{l}[\gamma^k]-\mathcal{l}[\bar{\gamma}^k ] ) , \end{aligned}\ ] ] where @xmath389 @xmath390 is a @xmath29-curve which is the graph of @xmath5 on the @xmath391-th connected component of @xmath117 . moreover , the difference between the boundary warp angle of @xmath389 at an endpoint and the contact angle of @xmath140 with @xmath10 at the corresponding endpoint of the @xmath391-th connected component of @xmath178 tends to be zero as @xmath392 independently of @xmath5 . fix @xmath251 and @xmath393 which is @xmath28-associated with @xmath140 . we denote @xmath394 for @xmath395 ( the shift operator @xmath396 is defined in definition [ deltaasso ] ) and define @xmath397 by @xmath398 then , since any contact angle of @xmath140 with @xmath10 is positive by theorem [ regezero ] , there exists @xmath385 such that @xmath399 is in @xmath146 for any @xmath386 . therefore , since @xmath400\geq e_0[\bar{u}]$ ] by the minimality of @xmath140 , we have @xmath401\geq \varepsilon\int_{a}^{b } { \kappa_u}^2\sqrt{1+u_x^2}\ dx+\frac{1}{\varepsilon}(e_0[u]-e_0[u_\delta]).\ ] ] moreover , we can restrict the domain of integration of the right - hand term to the disjoint union of @xmath402 thus we have @xmath403\geq\sum_{\substack{0\leq i\leq n\\ ( x_i , x_{i+1})\subset(a , b)_+^u}}\left(\varepsilon\int_{x_i}^{x_{i+1 } } \kappa_u^2\sqrt{1+u_x^2}\ dx\right.\\ \left.+\frac{1}{\varepsilon}\int_{x_i}^{x_{i+1 } } \sqrt{1+u_x^2 } - \sqrt{1+\tan^2{\tilde{\theta}_i}}\ dx \right ) , \end{gathered}\ ] ] where @xmath404 . by the above argument , if we take @xmath389 as the statement then ( [ lbrep1 ] ) holds . now we take any @xmath130 with @xmath119 . the boundary warp angle of @xmath389 at @xmath405 is @xmath406 , that is , @xmath407 and the contact angle of @xmath140 at @xmath144 is @xmath408 , that is , @xmath409 therefore , the difference of them tends to be zero as @xmath392 not depending on @xmath5 ( but only @xmath10 and @xmath140 ) since @xmath410 , @xmath411 , and the functions @xmath140 , @xmath412 and @xmath413 are uniformly continuous . we can similarly consider for @xmath130 with @xmath121 , thus the proof is completed . we now obtain a key estimate for the right - hand term of ( [ lbrep1 ] ) . [ lbprop3 ] for any @xmath3 and @xmath29-curve @xmath32 with @xmath414 , the inequality holds : @xmath415+\frac{1}{\varepsilon}(\mathcal{l}[\gamma]-\mathcal{l}[\bar{\gamma}])\geq\mathcal{w}[\gamma].\ ] ] this follows by lemma [ redlem5 ] and lemma [ lblem1 ] . [ redlem5 ] let @xmath3 . for any @xmath29-curve @xmath32 with @xmath414 , there exist @xmath416 and a function @xmath417 such that @xmath418 , @xmath419 , @xmath420 and @xmath421+\frac{1}{\varepsilon}(\mathcal{l}[\gamma]-\mathcal{l}[\bar{\gamma } ] ) \geq \int_{-l_0}^{l_1}\varepsilon \kappa_{\hat{u}}^2\sqrt{1+{\hat{u}_x}^2}+ \frac{1}{\varepsilon}\left(\sqrt{1+{\hat{u}_x}^2 } - 1\right)dx . \end{aligned}\ ] ] we can assume @xmath422<\infty$ ] , @xmath423 , @xmath424 with @xmath425 without loss of generality . note that @xmath426=\gamma_1(1),\quad\tan\theta_\gamma^0=|\gamma_2'(0)|/\gamma_1'(0),\quad\tan\theta_\gamma^1=|\gamma_2'(1)|/\gamma_1'(1),\ ] ] and there exists @xmath427 such that @xmath428 . now we fix @xmath364\rightarrow\mathbb{r}^2 $ ] . take an arbitrary angle @xmath429 and for @xmath430 $ ] define @xmath431 then @xmath432 is a @xmath29-curve such that @xmath432 is in the first quadrant of @xmath433 , @xmath434,\ ] ] and the following conditions hold : @xmath435=\mathcal{b}[\gamma],\quad\mathcal{l}[\hat{\gamma}]=\mathcal{l}[\gamma],\quad\hat{\gamma}(0)=(0,0),\quad\hat{\gamma}_1(1)\geq\gamma_1(1),\quad\hat{\gamma}_2'(t_\gamma)=0,\ ] ] @xmath436 these conditions hold since we only used translation or reflection partially in this transformation , and while @xmath437 only translation or one reflection @xmath438 . in addition , since @xmath439 in @xmath440 $ ] , we can define the inverse function of @xmath441 and it is in @xmath442 . thus , by taking @xmath443 we obtain a desired function . indeed , we have @xmath444 @xmath445 and @xmath446 similarly and the desired inequality follows by @xmath373=\mathcal{b}[\hat{\gamma}]=\int\kappa_{\hat{u}}^2\sqrt{1+{\hat{u}_x}^2},\quad\mathcal{l}[\gamma]=\mathcal{l}[\hat{\gamma}]=\int\sqrt{1+{\hat{u}_x}^2},\ ] ] @xmath447.\ ] ] the proof is completed . [ lblem1 ] let @xmath448 and @xmath449 satisfying @xmath450 and @xmath451 . then the inequality holds : @xmath452 for such @xmath448 , we have @xmath453 where @xmath454 is the odd function given by @xmath455 the first inequality follows by the easy trick @xmath456 . moreover , we have @xmath457 and @xmath458 thus the proof is completed . let @xmath32 be any @xmath29-curve and @xmath459 be the function obtained by lemma [ redlem5 ] . using lemma [ lblem1 ] for the @xmath29-functions @xmath460 on @xmath461 $ ] and @xmath462 on @xmath463 $ ] , then we have @xmath415+\frac{1}{\varepsilon}(\mathcal{l}[\gamma]-\mathcal{l}[\bar{\gamma}])\geq4\left(\sqrt{2}-\sqrt{1+\cos\theta_\gamma^0}\right)+4\left(\sqrt{2}-\sqrt{1+\cos\theta_\gamma^1}\right)\ ] ] and the right - hand term is nothing but @xmath464 $ ] . we shall prove the liminf condition of theorem [ gammathm ] . take any sequence @xmath465 in @xmath146 and fix it . we can assume assumption [ assumpseq ] by proposition [ restseq ] ( thus @xmath466 is @xmath141$]-regular ) . then for any sufficiently small @xmath251 , we have @xmath467\geq\sum_{i=1}^n 4\left(\sqrt{2}-\sqrt{1+\cos{\theta_{i,\delta}}}\right)\ ] ] by proposition [ lbprop1 ] , [ lbprop2 ] and [ lbprop3 ] , where @xmath468 are angles satisfying @xmath469 as @xmath392 and @xmath143 is the contact angle of @xmath140 with @xmath10 at @xmath144 . thus we get the consequence by taking @xmath470 since @xmath145 and ( [ repsinglim ] ) hold . finally , we prove the limsup condition of theorem [ gammathm ] . we construct sequences concretely by modifying singularities of minimizers of @xmath12 . to this end , for arbitrary @xmath3 and @xmath471 we consider the following ode : @xmath472,\\ u(0)=0,\ u'(0)=\tan{\theta}>0 . \end{cases } \end{aligned}\ ] ] this cauchy problem has the unique solution @xmath473\rightarrow\mathbb{r}$ ] satisfying @xmath474 , @xmath475 and @xmath476 in @xmath477 . moreover , @xmath478 for @xmath479 , @xmath480 and by definition we see @xmath481 where @xmath454 is defined in ( [ evenfunc ] ) . considering the above , for any two points in @xmath433 and angles @xmath482 , we can take a sequence of @xmath29-curves @xmath483 connecting the two points such that the boundary warp angles are @xmath484 , @xmath485 and the ( rescaled ) total energy of the curves converges to the boundary warp energy of them , that is , @xmath486+\frac{1}{\varepsilon}(\mathcal{l}[\gamma^\varepsilon_{\theta,\theta'}]-\mathcal{l}[\bar{\gamma}^\varepsilon_{\theta,\theta'}])\right)\\ = 4\left(2\sqrt{2}-\sqrt{1+\cos{\theta}}-\sqrt{1+\cos{\theta'}}\right ) . \end{aligned}\ ] ] this is achieved , for example , by rotating the graph of a @xmath29-function @xmath487 on some interval @xmath488 $ ] defined by @xmath489,\\ \text{suitable function}\quad\text{in } ( a+\varepsilon^{2/3},b-\varepsilon^{2/3}),\\ u_{\theta',\varepsilon}(x - b)\quad x\in[b-\varepsilon^{2/3},b ] . \end{cases } \end{aligned}\ ] ] in @xmath490\cup[b-\varepsilon^{2/3},b]$ ] the total energy converges to @xmath491 as @xmath15 . in @xmath492 , for example we use two arcs of the circle with radius @xmath9 and central angles @xmath493 , @xmath494 tending to zero as @xmath15 and a segment suitably ( as figure [ upperfig ] ) . we decompose @xmath492 into the circular parts @xmath495 $ ] and the segment part @xmath496 $ ] . then we see that the total energy of the circular parts tends to be zero by @xmath497 the segment part also tends to be zero by @xmath498 since @xmath499 as @xmath15 . therefore , for any @xmath500 which is @xmath501$]-regular , by modifying it as above in @xmath142 , we can take @xmath502 such that @xmath503 in @xmath504 and @xmath505=\sum_{i=1}^n 4\left(\sqrt{2}-\sqrt{1+\cos\theta_i}\right)=\sum_{i=1}^n 4\left(\sqrt{2}-\sqrt{1+\alpha(\bar{x}_i)}\right),\ ] ] where @xmath143 is the contact angle of @xmath140 with @xmath10 at @xmath144 . it is not trivial that the obtained functions @xmath26 belong to @xmath506 thus we should be carefully . note that the curves obtained by the modification can be represented as the graph of a @xmath29-function if @xmath3 is sufficiently small . thus @xmath507 . in addition , the second derivative of the modified parts of @xmath26 goes to infinity as @xmath15 near the free boundary of @xmath140 , thus we see @xmath508 for any sufficiently small @xmath3 . this implies @xmath509 , and the proof is completed . [ cvlem1 ] let @xmath16 be an @xmath510-integrable or non - negative measurable function on @xmath511 $ ] and @xmath512 be a @xmath513-diffeomorphism from @xmath488 $ ] to @xmath511 $ ] . then @xmath514 the case that @xmath515 is finite follows by ( * ? ? ? * , theorem 2 ) . if @xmath515 is infinite then so is the right hand term since @xmath512 is a @xmath513-diffeomorphism . [ cvlem2 ] let @xmath516 and @xmath512 be a @xmath513-diffeomorphism from @xmath488 $ ] to @xmath511 $ ] . then @xmath517 and @xmath518 by lemma [ cvlem1 ] . the following is nothing but ( * ? ? ? * corollary 8.11 ) . [ cvlem3 ] let @xmath516 and @xmath519 . then @xmath520 and @xmath521 the above lemmas lead to lemma [ cvlem4 ] , further lemma [ cvlem5 ] and [ cvlem6 ] . [ cvlem4 ] let @xmath393 , @xmath522 , @xmath523 in @xmath488 $ ] , @xmath524)=[a , b]$ ] and @xmath525 be the inverse function of @xmath512 . then @xmath526 and @xmath527 . moreover , @xmath528 @xmath529 [ cvlem5 ] let @xmath32 be a @xmath29-curve and @xmath512 be a @xmath29-reparameterization , i.e. @xmath530 , @xmath523 in @xmath440 $ ] and @xmath531)=[0,1]$ ] . then @xmath532 is a @xmath29-curve . moreover , @xmath381 , @xmath380 and @xmath382 are invariant by @xmath29-reparameterization . [ cvlem6 ] the energies @xmath381 , @xmath380 and @xmath382 are invariant with respect to translation , reflection and rotation . the author would like to thank his supervisor yoshikazu giga for suggesting this problem and fruitful discussions . the author also would like to thank oliver pierre - louis , who is one of the physicists considering this model , for his useful comments and remarks from a physical viewpoint . this work was supported by a grant - in - aid for jsps fellows 15j05166 and the program for leading graduate schools , mext , japan .

consider a-ary rooted tree ( ) where each edge is assigned an i.i.d . ( bounded ) random variable of negative mean . assign to each vertex the sum of over all edges connecting to the root , and assume that the maximum of over all vertices at distance from the root tends to infinity ( necessarily , linearly ) as tends to infinity . we analyze the metropolis algorithm on the tree and show that under these assumptions there always exists a temperature of the algorithm so that it achieves a linear ( positive ) growth rate in linear time . this confirms a conjecture of aldous [ _ algorithmica _ * 22 * ( 1998 ) 388412 ] . the proof is obtained by establishing an einstein relation for the metropolis algorithm on the tree

1304.0552

given a @xmath0-regular rooted tree , attach to each edge @xmath2 a random variable @xmath3 , such that the variables are independent and identically distributed . for a vertex @xmath4 in the tree , denote by @xmath5 the sum of the variables @xmath3 over all edges @xmath2 on the path from the root to @xmath4 . this defines a branching random walk , a basic model for a disordered tree . it is natural to ask for an efficient algorithm which explores the vertices of this tree in order to find vertices @xmath4 with a large value of @xmath5 . in fact , aldous @xcite proposed this problem as a benchmark problem for comparing different generic optimization algorithms , since the nave approach , which would be to simply explore all vertices down to the level @xmath7 in the tree and taking the one with the maximal value of @xmath5 , is a bad choice for an algorithm because the number of vertices grows exponentially in @xmath7 . the metropolis algorithm is a general recipe for constructing a discrete - time markov chain on a finite state space for which a given distribution @xmath9 is stationary and whose transitions respect a given graph structure of the state space . in the context of the comparison of algorithms discussed earlier , aldous @xcite suggested using the metropolis algorithm to `` sample '' a certain gibbs measure on the vertices of a branching random walk tree , namely the one which assigns mass @xmath10 to a vertex @xmath4 , for some parameter @xmath11 . in the case where this measure is infinite , for example when there is an infinite number of vertices @xmath4 with @xmath12 , this algorithm should `` walk down the tree '' and , for an appropriate choice of the parameter @xmath13 , find vertices @xmath4 with high values of @xmath5 . let @xmath14 denote the level of the vertex @xmath4 in the tree , and let @xmath15 be the vertex visited by the metropolis algorithm at the time @xmath16 . aldous raised the following natural question : if the maximum of the branching random walk has positive speed , that is , if @xmath17 , does there exist a choice of the parameter @xmath13 , such that @xmath18 ? we will answer this question in the affirmative for a certain class of laws of the variables @xmath3 , including the binomial distribution . in fact , we show more : let @xmath19 , which exists almost surely @xcite . we show that there exists a parameter @xmath20 , such that @xmath21 and @xmath22 , where @xmath23 is the asymptotic variance of @xmath24 , which we show to be positive and finite . this result was conjectured by aldous , who gave heuristic arguments and numerical evidence for it [ in the case where the variables @xmath3 only take the values @xmath25 and @xmath26 . results of this type are also known as _ einstein relations _ in the domain of random walks in random environments and our methods of proof will indeed rely on many techniques from this field , some of which have been obtained recently . we are given a @xmath0-regular infinite rooted tree , @xmath27 . the root is denoted by @xmath28 and the level / depth of a vertex @xmath4 in the tree by @xmath14 . the notation @xmath29 denotes that @xmath30 and @xmath4 are connected by an edge . the parent of a vertex @xmath4 is denoted by @xmath31 ( with the convention @xmath32 ) . we write @xmath33 if @xmath30 is an ancestor of @xmath4 and @xmath34 if @xmath33 but @xmath35 . we furthermore use the following handy notation : if @xmath33 , then @xmath36 $ ] denotes the set of vertices on the path from @xmath30 to @xmath4 , including @xmath30 and @xmath4 . the notation @xmath37 $ ] , @xmath38 and @xmath39 then has obvious meaning . to each edge @xmath40 , we then attach a random variable @xmath3 , such that the collection @xmath41 ) is i.i.d . according to the law of a random variable @xmath42 . here , orientation of the edges matters , and we will set @xmath43 for all @xmath4 . in what follows , we will introduce several assumptions , _ which we assume to hold throughout the paper_. we begin with the following assumptions on the law of @xmath42 . the law of @xmath42 is of compact support , that is , @xmath44 . there exists @xmath20 , such that @xmath45=e[f(-x)]$ ] for all bounded measurable functions @xmath46 . @xmath47 , where @xmath48 + \log(d-1)$ ] . note that ( xr ) is equivalent to the laplace transform @xmath49 $ ] being symmetric around @xmath50 . in particular , the constant @xmath51 is necessarily unique unless @xmath52 almost surely . an example for a law satisfying ( xs ) , ( xr ) and ( xm ) is the distribution of @xmath53 , where @xmath54 follows a binomial distribution of parameters @xmath7 and @xmath55 , with @xmath56 , where @xmath57 . in this case , @xmath58 . in general , in order to construct a law satisfying ( xs ) and ( xr ) , one can start from a symmetric random variable @xmath42 taking values in a compact interval @xmath59 $ ] and define a law with radon nikodym derivative proportional to @xmath60 with respect to the law of @xmath42 . this law will then satisfy ( xm ) for @xmath51 small enough . we remark that assumption ( xs ) seems not to be crucial , and the argument extends to certain distributions with non compact support , at the cost of more complicated technical arguments . to avoid this complication we chose to present the result under this simplifying assumption . on the other hand , assumption ( xr ) is essential for our treatment , as it ensures , at @xmath51 , the reversibility of the markov chain consisting of the environment viewed from the point of view of the particle ; see proposition [ prop : rev_erg ] . the reversibility will be crucial both in the application of the kipnis varadhan theory , as well as in the proof of validity of the einstein relation ( one may expect a correction term for non reversible chains ) . we now define the branching random walk by @xmath61 } x({\vec}u , u),\qquad s ( \root ) = 0.\ ] ] note that @xmath62 for every two vertices @xmath30 and @xmath4 with @xmath29 , by the above convention that @xmath63 . since @xmath64 is the log - laplace transform of this branching random walk , it is known @xcite that @xmath65 exists and is positive under assumption ( xm ) . note further that assumptions ( xm ) and ( xr ) together imply that @xmath66 for all @xmath67 , such that @xmath68 exists and is negative @xcite . in order to define the metropolis algorithm , we are given a function @xmath69 satisfying the following conditions [ examples are @xmath70 and @xmath71 : @xmath72 takes values in @xmath73 $ ] , is nondecreasing and satisfies @xmath74 and @xmath75 . it is lipschitz - continuous and continuously differentiable on @xmath76 . it satisfies the functional equation @xmath77 for all @xmath78 . for a given realization of the branching random walk and a parameter @xmath67 , the metropolis algorithm is then the markov chain @xmath79 on the vertices of the tree with the transition probabilities @xmath80 given by @xmath81 we denote the ( annealed , i.e. , averaged over the environment ) law of the metropolis algorithm on the branching random walk tree by @xmath82 and expectation with respect to this law by @xmath83 . our main theorem is the following : [ th : einstein ] set @xmath84 . the limit @xmath85 exists @xmath86-almost surely and is a strictly positive and finite constant . for each @xmath67 , the deterministic limit @xmath87 exists @xmath82-almost surely and satisfies @xmath88 we note that the existence of @xmath89 and the fact that it vanishes at @xmath90 were already shown in @xcite ( in a slightly more restrictive setup ) . the main novelty in theorem [ th : einstein ] is the proof of the einstein relation ( [ eq - einsteinfinal ] ) , as well as the fact that the right side is strictly positive . our main inspiration , as noted above , is aldous s work @xcite . in that paper , aldous makes the crucial observation that a reversible invariant measure for the environment viewed from the point of view of the particle exists at @xmath90 , and derives from this that @xmath91 , and the existence of the limit @xmath23 under @xmath86 ; he also completely analyzes a greedy algorithm and formulates a series of conjectures , some answered here . in the same paper , aldous also refers speculatively to @xcite as relevant to the analysis near @xmath90 ; indeed , the approach of the latter to proofs of the einstein relation forms the basis of the current paper , as well to recent advances in the analysis of the einstein relation for disordered systems , as we now discuss . the einstein relation ( er ) links the asymptotic variance of additive functionals of ( reversible ) markov chains in equilibrium to the chains response to small perturbations . in a weak limit ( where the time - scale is related to the strength of the perturbation ) , lebowitz and rost @xcite provide a general recipe ( based on the kipnis varadhan theory , see @xcite for a comprehensive account ) for the validity of a weak form of the er in disordered systems . for the tagged particle in the symmetric exclusion process , the er was proved by loulakis in @xmath27 @xcite by perturbative methods ( using transience in an essential way ) ; this approach was adapted to bond diffusion in @xmath92 in special environment distributions @xcite . for mixing dynamical random environments with spectral gap , a full perturbation expansion was proved in @xcite . significant recent progress was achieved by @xcite , where the lebowitz rost approach was combined with good _ uniform in the environment _ estimates on certain regeneration times in the transient regime , that are used to pass from a weak er to a full er . these uniform estimates are typically not available for random walks on ( random ) trees , and a completely different approach , based on explicit recursions , was taken in @xcite , where ( biased ) random walks on galton watson trees were analyzed . while we still consider walks on trees , the approach we take is closer to that of @xcite , while replacing their uniform regeneration estimates with probabilistic estimates , in the spirit of @xcite . see also @xcite for another approach to the proof of the er in the context of balanced random walks . as mentioned above , the starting point is aldous s observation that under @xmath86 , the environment viewed from the point of view of the particle forms a _ reversible _ markov chain . we begin by proving this ( proposition [ prop : rev_erg ] ) , and then apply the kipnis varadhan theory to deduce an invariance principle for anti - symmetric additive functionals ( lemma [ lem : antisym ] ) . this allows us to prove the weak er , theorem [ th : weak_er ] , following the lebowitz rost recipe . to handle the perturbation , estimates on regeneration times and distances are crucial . we work with _ level _ regeneration times that are introduced in section [ sec - reg ] ; these involve the random walk location @xmath93 , not the vertices values @xmath94 ; of course , the latter influence the transition probabilities of the random walk . in order to transform the weak er to a full er , we need uniform bounds on the moments of the regeneration times . these are obtained in proposition [ prop : tau_moments ] , where it is proved that the regeneration times exhibit uniform annealed stretched - exponential bounds . the proof has two main steps : first , exponential moments are proved for regeneration _ distances _ , using in a crucial way a structure lemma of grimmett and kesten ; see lemma [ lem : excursion_tail ] . then , the estimates for regeneration times are obtained , using that the walk must visit many well - separated fresh vertices , and between two such visits , the walk has a large enough probability to hit distant levels . in proving the last statement , an argument of aidkon @xcite is used ; see lemma [ lem : fresh ] . in this section , we show that the einstein relation holds for times of the order of @xmath95 specifically , we will prove the following result : [ th : weak_er ] @xmath96=0 $ ] for all @xmath97 . furthermore , the limit @xmath98/n$ ] exists and is a finite , nonnegative constant . set @xmath99 . then , @xmath100 \to\frac{\sigma^2 } 2 \qquad\mbox{as } \beta \to0.\ ] ] the proof of theorem [ th : weak_er ] uses a fairly generic and now classical change of measure argument in the spirit of lebowitz and rost @xcite . it uses the crucial concept of the _ environment seen from the particle _ , which we define as follows . let @xmath101 be the space of rooted @xmath0-regular unlabeled trees @xmath102 with marked edges , that is , to every two vertices @xmath30 and @xmath4 with @xmath29 we associate a real number @xmath103 with @xmath104 . note that `` unlabeled '' means that we do not distinguish between the neighbors of a vertex , of labeled rooted @xmath0-regular trees with marks , which is homeomorphic to @xmath105 . the space @xmath101 is then defined as the quotient space with respect to the group of graph automorphisms fixing the root . it is endowed with the borel @xmath106-algebra induced by the quotient topology . ] this will be crucial for what follows . the root of every tree @xmath107 is denoted by @xmath108 . for every vertex @xmath4 , we then define the shift operator @xmath109 , which yields the tree @xmath107 `` seen from the vertex @xmath4 . '' a bit more formally , ( see above ) and then show that it induces a well - defined operator on @xmath101 . ] if for a vertex @xmath30 in @xmath107 we denote by @xmath110 its corresponding vertex in @xmath111 , then @xmath112 . in particular , if @xmath113 , then the mark of the edge @xmath114 `` changes its sign upon passing from @xmath107 to @xmath115 . '' define @xmath116 for each vertex @xmath4 analogously to ( [ eq : s ] ) . for @xmath117 , we then define the operators @xmath118 and @xmath119 acting on functions @xmath120 by @xmath121 where @xmath122 ; see section [ sec : definition ] . let @xmath123 be the law on @xmath101 under which all edges @xmath124 are i.i.d . according to the law of @xmath42 ; see section [ sec : definition ] . we denote by @xmath125 the expectation with respect to @xmath123 . furthermore , let @xmath126 be the law of the metropolis algorithm @xmath79 with transition probabilities @xmath127 defined in section [ sec : definition ] and the underlying tree @xmath128 distributed according to @xmath123 . expectation w.r.t . @xmath82 is denoted by @xmath83 , and we also set @xmath129 and @xmath130 . setting @xmath131 then defines a markov chain @xmath132 on the space @xmath101 , which jumps from @xmath107 to @xmath111 with probability @xmath133 for every @xmath134 . let @xmath135 be the natural filtration of @xmath132 , augmented by sets of zero measure . the process @xmath136 is then adapted to @xmath135 , since it can be almost surely reconstructed from @xmath137 . the following result was already observed by aldous @xcite , who had a more complicated proof for it . [ prop : rev_erg ] the process @xmath132 is reversible and ergodic under @xmath138 . in order to show reversibility , since @xmath132 is a markov process , we only have to show that @xmath139 = \ee [ f(\omega_1,\omega_0)]$ ] for every bounded ( borel ) measurable functional @xmath140 . for this , it is obviously enough to show that @xmath141 = 0 $ ] . now we have @xmath142 & = & \sum_{v\sim \rho}\ee\bigl[p\bigl(x_{\omega_0 } ( \rho , v)\bigr)f(\omega_0,\theta_v\omega_0 ) \bigr ] \\ & = & \sum_{v\sim\rho}\e\bigl[p\bigl(-x_{\omega } ( \rho , v)\bigr)e^{\beta_0x_{\omega } ( \rho , v)}f(\omega,\theta_v\omega)\bigr],\end{aligned}\ ] ] where the last equality follows from assumption ( h3 ) . conditioned on@xmath143 , the environment @xmath111 is distributed as @xmath107 but with one edge pointing away from the root bearing the value @xmath144 ( remember that the vertices are unlabeled , such that `` it can be any one of them , '' which amounts to saying that `` we do not know where we came from '' ) . by assumption ( xr ) , the right - hand side of the last equation is therefore equal to @xmath145 = \ee\bigl[f ( \omega_1,\omega_0)\ind_{\omega_0 \ne\omega_1}\bigr],\ ] ] which finishes the proof of the reversibility . ergodicity follows from a classical ellipticity argument which we recall ( see also @xcite , corollary 2.1.25 , for a similar argument ) : let @xmath146 be a stationary probability measure of the markov chain @xmath147 with @xmath148 . we wish to show that @xmath149 , which will imply ergodicity since ergodic measures are the extremal points in the convex set of stationary probabilities . define the event @xmath150 . by invariance , @xmath151= e_{q}[\ind_e]=0 $ ] , where @xmath152 is the transition kernel of the markov chain . this further implies @xmath153 , @xmath123-almost surely . since the transition probabilities are strictly positive , we then have @xmath154 , because @xmath155 takes values in @xmath156 . fixing an infinite ray @xmath157 we then get by iteration of the previous inequality that @xmath158 for every @xmath159 , whence @xmath160 for every @xmath7 . but since @xmath161 is a product measure and therefore ergodic with respect to the shift along the ray , birkhoff s ergodic theorem now gives @xmath162 , @xmath123-almost surely , which implies @xmath163 but @xmath164 by hypothesis , whence @xmath165 . this finishes the proof . we recall the following basic fact about reversible processes . [ lem : rev_functional ] for any bounded measurable functionals @xmath166 and @xmath167 and every @xmath97 , we have @xmath168 = \ee\bigl[f(\omega_n,\ldots , \omega_0)g(\omega_0)\bigr].\ ] ] we will need the following result about anti - symmetric additive functionals of reversible ergodic markov processes . it is implicit in the proof of theorem 2.1 in @xcite and relies on a celebrated result from @xcite ; see also chapters 1 and 2 in @xcite for a comprehensive account of the theory . [ lem : antisym ] let @xmath140 be an anti - symmetric measurable functional , that is , @xmath169 for all @xmath170 , with @xmath171 < \infty$ ] . define a sequence of random variables by @xmath172 for all @xmath173 . then : the time - variance @xmath174 $ ] exists and is finite . there exists a square integrable martingale @xmath175 with stationary ergodic increments , such that @xmath176 converges to 0 in @xmath177 . in particular , @xmath178 converges in law to a centered gaussian variable with variance @xmath23 . the following lemma makes precise an expansion of @xmath179 around @xmath55 for small @xmath13 . it easily follows from assumptions ( h1)(h3 ) . [ lem : p_expansion ] there exist measurable functions @xmath180 and @xmath181 , such that : @xmath182 ; @xmath180 is uniformly bounded for all @xmath183 and small enough @xmath13 , and @xmath184 as @xmath185 , for all @xmath186 ; @xmath187 ; there exists a constant @xmath188 , such that for small enough @xmath13 , we have @xmath189 and @xmath190 . we are now ready for the proof of theorem [ th : weak_er ] . proof of theorem [ th : weak_er ] _ first part _ : we first note that there exists a measurable functional @xmath140 , such that @xmath191 , @xmath138-almost surely . this follows from the fact that @xmath123-almost surely , the @xmath0 shifted environments @xmath115 , @xmath134 are all different , otherwise there would be at least two identical subtrees of the vertices in the second generation which is an event of probability zero ( except if @xmath52 almost surely , in which case the lemma is trivial ) . by the definition of @xmath192 , we can furthermore choose @xmath166 to be anti - symmetric in the sense of lemma [ lem : antisym ] . in particular , @xmath193 = -\ee[s_1 ] = 0 $ ] by lemma [ lem : rev_functional ] , whence @xmath194=0 $ ] for all @xmath97 . the second statement follows from the first part of lemma [ lem : antisym ] . _ second part _ : we will use a change of measure argument as in @xcite . the basic idea is to write the radon nikodym derivative of @xmath82 with respect to @xmath138 as an exponential martingale of the form @xmath195 for some martingale @xmath196 and to show that the pair @xmath197 converges in law under @xmath138 to a centered gaussian vector @xmath198 with covariance @xmath199 = \frac{1 } 2 \ee[g_s^2 ] = \frac{1 } 2 \sigma ^2 $ ] . the theorem then follows from a standard change of measure argument for gaussian variables . here are the details : _ step _ 0 : set @xmath200 for all @xmath97 . the radon nikodym derivative of @xmath82 with respect to @xmath138 is given by @xmath201= : y_n.\ ] ] note that if @xmath202 , then @xmath203 with probability 1 , such that the second summand is well defined . note also that here and in what follows , the empty sum always has value 0 . if @xmath204 and @xmath205 are the functions from lemma [ lem : p_expansion ] , we can write the process @xmath206 as @xmath207 we then define the process @xmath208 by ( we suppress the dependence on @xmath13 from the notation ) @xmath209 \\ & = & -2\sum _ { k=0}^{n-1 } \biggl[\beta\lt_{\omega_k}q_\beta+ ( 1-\lc_{\omega_k}p)\log\biggl(1-\frac { \lt_{\omega_k}(e^{\beta q_\beta}-1)}{1-\lc_{\omega_k}p } \biggr ) \biggr].\end{aligned}\ ] ] ( consistent with the definitions we have that @xmath210 . ) note that by lemma [ lem : p_expansion ] , we have for small @xmath13 , @xmath211 \\[-8pt ] \nonumber & = & \beta^2 \bigl(\lt_{\omega_n } q^2 + ( \lt_{\omega_n } q)^2/(1-\lc_{\omega _ n } p ) + o(1 ) \bigr).\end{aligned}\ ] ] we further define the process @xmath212 ( again suppressing the dependence on @xmath13 ) by @xmath213 = y_n + \frac{1 } 2 a_n,\ ] ] such that @xmath212 is a martingale under @xmath138 with respect to the filtration @xmath135 . _ step _ 1 : we wish to show that the random variable @xmath214 converges in law under @xmath138 to a centered gaussian variable with variance @xmath215 . define the @xmath138-martingale @xmath216 by @xmath217 by ( [ eq : an ] ) and lemma [ lem : p_expansion ] , we then have @xmath218 & = & \ee\biggl[\sum_{k=0}^{n-1 } \bigl[y_{n+1 } - y_n - \beta(m_{n+1 } - m_n ) + o\bigl(\beta^2\bigr)\bigr]^2 \biggr ] \nonumber \\[-8pt ] \\[-8pt ] \nonumber & = & o\bigl(\beta^2 n\bigr),\end{aligned}\ ] ] whence @xmath219 \to0 $ ] as @xmath185 . note that by the fourth point of lemma [ lem : p_expansion ] , @xmath175 is square - integrable and by proposition [ prop : rev_erg ] , the sequence of its increments @xmath220 is stationary and ergodic . by the martingale central limit theorem for stationary ergodic sequences ( see , e.g. , @xcite , theorem 7.7.5 ) , the sequence @xmath221 then converges in law under @xmath138 to a centered gaussian variable with variance @xmath222 = \ee[\lt _ \omega q^2+(\lt_\omega q)^2/(1-\lc_\omega p)]$ ] . together with ( [ eq : zn_mn ] ) , this proves the above - mentioned convergence of @xmath223 . _ step _ 2 : we wish to show that the random variable @xmath224 converges in probability to @xmath225 under the law @xmath138 . define the process @xmath226 by @xmath227 . $ ] by proposition [ prop : rev_erg ] and the ergodic theorem , the sequence @xmath228 converges @xmath138-almost surely to @xmath229 = \sigma_z^2 $ ] . together with ( [ eq : an ] ) , this yields the above - mentioned convergence of @xmath230 as @xmath185 . _ step _ 3 : recall the definition @xmath231 . we wish to show that the pair @xmath232 converges in law under @xmath138 to a centered gaussian vector @xmath198 with covariance @xmath199 = \frac{1 } 2 \ee[g_s^2 ] = \frac{1 } 2 \sigma ^2 $ ] ( with @xmath23 from lemma [ lem : antisym ] ) . by ( [ eq : zn_mn ] ) , it is enough to show that this convergence holds for the pair @xmath233 as @xmath234 . by lemma [ lem : antisym ] , every linear combination @xmath235 is the sum of a square - integrable martingale with stationary and ergodic increments and a process @xmath236 with @xmath237 in @xmath177 , as @xmath238 . again by the martingale clt for stationary , ergodic sequences ( @xcite , theorem 7.7.5 ) , the pair @xmath239 then converges in law as @xmath234 to a centered gaussian vector @xmath198 with @xmath240 = \sigma^2 $ ] , @xmath241 = \sigma_z^2 $ ] and @xmath199 = \lim_{n\to\infty } \ee[s_nm_n]/n = \lim_{\beta\to 0}\ee [ s^\beta z^\beta]$ ] . it remains to show that @xmath242= \sigma ^2/2 $ ] . in order to prove this , recall the definition of @xmath243 and define @xmath244 . we have for every @xmath97 , @xmath245 = \ee\bigl[(\delta s_n)^2 + 2s_n\delta s_n\bigr ] = \ee\bigl[\lt_{\omega_n}\bigl(x^2\bigr ) + 2s_n \lt_{\omega_n } x\bigr],\ ] ] where @xmath246 is the identity function . by lemma [ lem : rev_functional ] , we have @xmath247 = \ee[(-s_n)\lt_{\omega_0 } x]$ ] , whence , summing ( [ eq:3 ] ) over @xmath248 we get , @xmath249 = \sum _ { k=1}^n \ee\biggl[\lt_{\omega_0 } \bigl(x^2\bigr ) - 2 \lt_{\omega _ 0 } x \times\sum _ { j=0}^{k-2 } \delta s_j \biggr].\ ] ] furthermore , since @xmath212 is a martingale , we have @xmath250 & = & \ee[z_{n+1}\delta s_n ] \nonumber \\[-8pt ] \\[-8pt ] \nonumber & = & \ee\bigl[\beta\delta s_n \bigl(q_{\beta}(\delta s_n)- \lt_{\omega_n } q_\beta\bigr)+(\lt_{\omega_n } x)z_n \bigr],\end{aligned}\ ] ] where we made use of the fact that @xmath251 on the event that @xmath252 . applying proposition [ prop : rev_erg ] to the term @xmath253 $ ] , we see that the terms corresponding to the second summand in the brackets of ( [ eq:2 ] ) cancel . summing ( [ eq:5 ] ) over @xmath7 , this yields @xmath254 & = & \beta\sum_{k=1}^n \ee\biggl[\lt_{\omega_0}(xq_{\beta } ) - \lt_{\omega_0 } x \times \sum_{j=0}^{k-2 } \bigl(q_\beta ( \delta s_j ) - q_\beta(-\delta s_j ) \bigr ) \biggr ] \\ & & { } + \ee[s_{n-1}\lt_{\omega_0 } q_\beta].\end{aligned}\ ] ] now , by assumption ( h3 ) we have @xmath255 for every @xmath246 ( this is the critical point ! ) . moreover , by reversibility , we have @xmath256 = \e[\lt_\omega\widebar f]$ ] for every function @xmath46 , where @xmath257 . this yields @xmath258 = \tfrac1 2 \times\e \bigl[\lt_{\omega}(xq_{\beta}-x\widebar{q}_\beta ) \bigr ] = \tfrac1 2 \times\e\bigl[\lt_{\omega}\bigl(x^2\bigr ) \bigr].\ ] ] altogether , the previous equations now yield @xmath259 = \ee\bigl [ \bigl(s^\beta\bigr)^2\bigr]/2 + \beta\ee[s_{\lfloor\beta ^{-2}-1\rfloor } \lt_{\omega_0}q_\beta].\ ] ] convergence of the first summand has been established above , and the second tends to 0 by lemma [ lem : antisym ] and the cauchy schwarz inequality . hence , we obtain @xmath242= \sigma^2/2 $ ] as claimed . _ step _ 4 : we claim that @xmath260 \to\ee[g_s\exp ( g_z - \frac{1 } 2 \sigma_z^2)]$ ] , as @xmath185 . since @xmath198 is a centered gaussian vector with @xmath199 = \frac{1 } 2 \ee[g_s^2 ] = \sigma^2/2 $ ] and @xmath241=\sigma_z^2 $ ] , this will finish the proof of the theorem . by ( [ eq:2 ] ) , @xmath260 = \ee[s^\beta \exp ( z^\beta- \frac{1 } 2 a^\beta)]$ ] and by the convergences in law established above , it suffices to show that this last expression is uniformly integrable . now , since @xmath261 and @xmath262 are martingales , @xmath263 is a submartingale . @xmath263 being @xmath264-measurable , it is therefore increasing in @xmath7 . it then remains to show that @xmath265 is uniformly integrable . by the fourth point of lemma [ lem : p_expansion ] , @xmath261 is a martingale with bounded increments for @xmath13 small enough . azuma s inequality @xcite then implies that all exponential moments of @xmath223 are uniformly bounded in @xmath13 , for small enough @xmath13 . furthermore , @xmath266 $ ] is uniformly bounded by the first part of this theorem . hlder s inequality then yields uniform boundedness of @xmath267 $ ] for some constant @xmath268 , which finishes the proof . in this section , we establish an estimate for the branching random walk ( lemma [ lem : brw_all_good ] below ) . we recall that for two vertices @xmath269 with @xmath33 , we denote by @xmath36 $ ] the set vertices on the path connecting @xmath30 and @xmath4 . similarly , if @xmath270 , then we define @xmath271 $ ] to be the set of vertices between levels / depths @xmath7 and @xmath272 . more generally , for a vertex @xmath30 , we let @xmath271_u$ ] denote the set of vertices between levels / depths @xmath7 and @xmath272 in the subtree rooted at @xmath30 ( which means that these vertices are between levels @xmath273 and @xmath274 in the original tree ) , such that @xmath275 = [ m , n]_\root$ ] . finally , we write @xmath276 $ ] for @xmath277 $ ] and @xmath276_u$ ] for @xmath277_u$ ] . [ lem : brw_root_great ] there exist @xmath278 and @xmath279 , such that for large @xmath280 , @xmath281\dvtx\max_{w\in[\root , v ] } \bigl|s(w)\bigr| > c\log l \bigr ) \le e^{-l^b}.\ ] ] we will first establish the following intermediate bound : [ lem : brw_root_good ] there exist constants @xmath282 such that for all large @xmath280 , @xmath283\dvtx\max_{w\in[\root , v ] } \bigl|s(w)\bigr| \le c_1\bigr ) > c_2.\ ] ] the proof is a standard first and second moment calculation . fix @xmath284 large . let @xmath285 be a random walk starting at @xmath286 with steps distributed according to the law of @xmath42 . for @xmath97 and @xmath287 $ ] , define the event @xmath288 and set @xmath289 . by assumption ( xm ) and standard large and small deviations estimates , there exists @xmath290 , such that for @xmath291 large enough , @xmath292 indeed , this is obtained , for example , by combining the change of measure in the mogulskii varadhan theorem @xcite , theorem 5.1.2 , with the fact that a centered random walk with bounded i.i.d . increments stays in a tube of width @xmath293 for time @xmath7 with probability at least @xmath294 for all @xmath7 and @xmath295 and some constant @xmath296 ; for finer estimates see , for example , @xcite . in the sequel , we fix such a @xmath291 once and for all . by an argument similar to the above , there exists a constant @xmath297 depending on @xmath291 only such that @xmath298 } \p\bigl(b^{(x)}_n\bigr ) \le \p(b_n ) \le c_1 ' \inf_{x\in[-c_1/2,c_1/2 ] } \p\bigl(b^{(x)}_n\bigr).\ ] ] to see ( [ eq : bnx ] ) , note from the above that @xmath299 is bounded by a constant depending on @xmath300 only , uniformly in @xmath301 , and then couple the walk started at @xmath246 with the walk started at @xmath286 by time @xmath300 , with a fixed positive probability . the second inequality in ( [ eq : bnx ] ) yields the existence of a constant @xmath302 ( depending on @xmath291 ) so that for every @xmath303 , @xmath304 because conditioned on @xmath305 , the probability that @xmath306 is bounded from below by a strictly positive constant uniformly in @xmath16 . for @xmath307 $ ] , let @xmath308 } |s(w)| \le c_1\}}.\ ] ] further let @xmath309 } a_v$ ] . then , @xmath310 = ( d-1)^l \p\bigl(\forall n \le l\dvtx|s_n| \le c_1\bigr ) = ( d-1)^l \p(b_l).\ ] ] as for the second moment , denote by @xmath311 the most recent common ancestor of @xmath30 and @xmath4 . we then have for large @xmath280 , @xmath312 & = & \sum_{u , v\in[l ] } \e[a_va_u ] \le\sum_{u , v\in[l ] } \p(b_l ) \sup_{x\in[-c_1,c_1]}\p\bigl(b^{(x)}_{l-|u\wedge v| } \bigr ) \\ & \le & c_1 ' c_1 '' \p(b_l)^2 \sum_{u , v\in[l ] } \p(b_{|u\wedge v|})^{-1},\end{aligned}\ ] ] where the last inequality follows from ( [ eq : bnx ] ) and ( [ eq : blk ] ) . equation ( [ eq - lsdev ] ) now yields @xmath313}\p(b_{|u\wedge v|})^{-1 } \le \sum_{u , v\in[l ] } c_0^{|u\wedge v| } \le c ( d-1)^{2l}\ ] ] for some @xmath296 . the lemma now follows from the previous three inequalities together with the paley zygmund bound @xmath314^{2}}/{\e[a^2]}$ ] . proof of lemma [ lem : brw_root_great ] let @xmath188 , and set @xmath315 . let @xmath291 be as in lemma [ lem : brw_root_good ] . let @xmath316 , which is finite by assumption ( xs ) . the branching random walks spawned by the vertices at level @xmath317 being independent , we have @xmath318\dvtx\max_{w\in[\root , v ] } \bigl|s(w)\bigr| > gh+c_1\bigr ) \\ & & \qquad\le\p\bigl(\forall v\in[l - h]\dvtx\max _ { w\in[\root , v]}\bigl |s(w)\bigr| > c_1\bigr)^{(d-1)^h}.\end{aligned}\ ] ] the lemma now follows from the last inequality together with lemma [ lem : brw_root_good ] , by choosing @xmath319 large enough . [ lem : brw_all_good ] there exist @xmath278 and @xmath279 , such that for large @xmath280 , @xmath320\ \forall v\in[l]\mbox { with } u\le v\dvtx \max _ { w\in[u , v ] } \bigl|s(w)-s(u)\bigr| > c\log l\bigr ) \le e^{-l^b}.\ ] ] let @xmath319 be as in the statement of lemma [ lem : brw_root_great ] . we say that a vertex @xmath30 is @xmath317-bad if for all @xmath321_u$ ] there exists @xmath322 $ ] , such that @xmath323 . note that if @xmath30 is @xmath317-bad , then it is @xmath324-bad for every @xmath325 . a simple union bound gives @xmath326\dvtx\mbox { $ u$ is $ \bigl(l-|u|\bigr)$-bad}\bigr ) & \le&\p\bigl ( \exists u\in[0,l]\dvtx\mbox { $ u$ is $ l$-bad}\bigr)\\ & \le&(d-1)^{l+1 } \p ( \root\mbox { is $ l$-bad}).\end{aligned}\ ] ] the statement then follows from lemma [ lem : brw_root_great ] . in this section , we establish a regeneration structure for the metropolis algorithm , which will permit us to prove theorem [ th : einstein ] from the previously established theorem [ th : weak_er ] . recall the definition of the metropolis algorithm @xmath327 from section [ sec : definition ] , which depends on a parameter @xmath328 . define the _ level regeneration _ times @xmath329 by @xmath330 and @xmath331 to be the first time after @xmath332 where the chain @xmath79 hits a level @xmath280 for the first time , then immediately jumps to level @xmath333 and never gets back to level @xmath280 again . as in sections [ sec : weak_er ] and [ sec : brw ] , we denote the law of the branching random walk by @xmath123 , which is a law on @xmath101 . we further denote the ( quenched ) law of the metropolis algorithm @xmath79 started from the vertex @xmath4 and given the branching random walk @xmath107 ( the _ environment _ ) by @xmath334 . the annealed law is denoted by @xmath335 . we also set @xmath336 , and note that this agrees with earlier notation . our goal is to show : [ prop : tau_moments ] for each @xmath337 , there exists @xmath338 and @xmath339 such that for all @xmath340 and @xmath341 , @xmath342 and @xmath343 . the main point in proposition [ prop : tau_moments ] is in _ uniformity _ ( in @xmath13 ) of the tail bounds for the regeneration times . this uniformity is in sharp contrast to other settings discussed in the literature , where the regeneration times usually blow up when the parameter approaches the critical value @xcite . we remark that we actually only need that @xmath344 $ ] is uniformly bounded for @xmath13 in a neighborhood of @xmath286 , for some @xmath345 . in order to prove proposition [ prop : tau_moments ] , we will make use of the relation between the markov chain @xmath79 and electrical networks @xcite : let @xmath346 be the set of neighbors of @xmath4 including @xmath4 . for @xmath347 , set @xmath348 one checks that for every @xmath347 , @xmath349 and that @xmath350 whence the markov chain @xmath79 has an interpretation as the random walk on the rooted @xmath0-regular tree with loops , induced by the edge conductances @xmath351 . by assumption ( h3 ) , one has for @xmath33 , @xmath352 by assumptions ( xs ) and ( h1)(h3 ) , this implies the existence of a constant @xmath188 , such that @xmath353 define @xmath354 to be the first strictly positive time the chain @xmath79 hits the level @xmath280 . furthermore , denote by @xmath355 and @xmath356 , respectively , the first nonnegative and strictly positive times the chain hits a vertex @xmath30 . the first lemma gives a uniform bound on the annealed probability that the metropolis algorithm started from a vertex @xmath4 escapes to infinity without coming back to its parent @xmath31 . it was essentially already observed by aldous @xcite , lemma 8 . [ lem : transience ] for each @xmath337 , there exists @xmath357 , such that for each vertex @xmath358 and for all @xmath359 , we have @xmath360 > c.\ ] ] fix @xmath358 and define @xmath361 $ ] . note that @xmath46 does not depend on @xmath4 by the definition of the measure @xmath123 . as mentioned in section [ sec : definition ] , under assumptions ( xm ) and ( xr ) , there exists almost surely two infinite rays @xmath362 and @xmath363 with @xmath364 and @xmath365 . by ( [ eq : conductance_ratio_est ] ) , the former has finite resistance if @xmath366 , and the latter if @xmath367 , whence @xmath368 for each @xmath369 . if @xmath370 , the metropolis algorithm is just a simple random walk on the @xmath0-regular tree and therefore @xmath371 as well . it follows that @xmath46 is positive for every @xmath67 . furthermore , @xmath372 is continuous because it is the decreasing limit as @xmath373 of @xmath374 $ ] and each of these quantities depends only on a finite portion of the tree and is therefore continuous in @xmath13 by assumption ( h2 ) . this immediately implies the lemma . the following important lemma controls quenched hitting probabilities and will be used in the evaluation of quenched escape probabilities . [ lem : level_hitting_prob ] for each @xmath337 , there exist @xmath375 , @xmath376 depending on @xmath324 , such that for @xmath341 and @xmath377 , @xmath378\dvtx p_{\omega,\beta}^v(t_l < t_{{\vec}v } ) < l^{-c } \bigr ) < e^{-l^b}.\ ] ] let @xmath379 $ ] , and let @xmath380_v$ ] , such that @xmath381 . by ( [ eq : conductance_ratio_est ] ) , we have for a fixed environment @xmath107 , for some @xmath188 , @xmath382 this gives for @xmath359 , @xmath383_v } p_{\omega,\beta}^v(t_u < t_{{\vec}v})\\ & > & \frac{c } l \max_{u\in [ l-|v|]_v } \min _ { w\in[v , u ) } e^{-(k+\beta_0)|s(w)-s(v)|}.\end{aligned}\ ] ] the statement now follows from lemma [ lem : brw_all_good ] . for a vertex @xmath4 , denote by @xmath384 the depth of the first excursion below @xmath4 after @xmath385 , that is , @xmath386 \\[-8pt ] \nonumber & \in&\n\cup\{\infty\}.\end{aligned}\ ] ] note that since the probability of jumping from @xmath4 to one of its children does not involve @xmath387 , the event @xmath388 is independent from @xmath387 ( conditioned on @xmath389 ) . [ lem : excursion_tail ] for each @xmath337 there exist @xmath390 , such that for @xmath359 and @xmath377 , @xmath391 fix @xmath392 . for a vertex @xmath4 , define the variable @xmath393 by @xmath394 if for one of @xmath4 s sisters @xmath395 , one has @xmath396 and @xmath397 otherwise . in words , @xmath398 if the ( quenched ) probability of a walk , started at an appropriate sister @xmath395 of @xmath4 , to escape to infinity through the subtree rooted at @xmath395 without visiting again @xmath395 is at least @xmath399 . by lemma [ lem : transience ] , we can choose @xmath399 such that @xmath400 > 1/2 $ ] for all @xmath401 . by a result due to grimmett and kesten ( see @xcite , lemma 2.2 , ( 2.1 ) for this version ) , there exist then @xmath402 , such that @xmath403 for large @xmath280 , where @xmath404 } \sum _ { w\in[2,v ] } a(w ) \ge\gamma l \biggr\}.\ ] ] now , let @xmath405 . we wish to bound @xmath406 . for this , define @xmath407 for @xmath408 to be the first time after @xmath354 that the markov chain @xmath79 hits level @xmath272 . if @xmath409 and @xmath410 , then by assumptions ( xs ) and ( h1)(h3 ) , the probability that from @xmath411 the chain reaches @xmath412 after two steps is bounded from below by @xmath413 for some @xmath414 sufficiently small . it follows that for @xmath405 , @xmath415 } a(w ) } \le(1-\delta)^{\gamma l}.\ ] ] this yields the lemma ( reducing the value of @xmath416 if necessary ) . [ lem - alpha ] for each @xmath337 there exist @xmath390 , such that for @xmath359 and @xmath377 , @xmath417 define a sequence of random numbers @xmath418 recursively as follows : * @xmath419 ; * for @xmath420 , let @xmath421 . if @xmath422 , then @xmath423 ; * otherwise , set @xmath424 for @xmath425 . let @xmath426 be the largest number @xmath7 , such that @xmath427 . then by construction , @xmath428 . furthermore , the differences @xmath429 are independent and identically distributed as @xmath430 conditioned on @xmath431 , and @xmath426 is geometrically distributed with success probability @xmath432 [ @xmath433 as in ( [ eq - starend ] ) ] . the lemma then follows from lemma [ lem : excursion_tail ] . [ lem : fresh ] let @xmath434 be the @xmath106-field generated by @xmath435 and let @xmath436 be a stopping time with respect to the filtration @xmath437 , such that @xmath438 for all @xmath439 . then for each @xmath337 there exists a constant @xmath440 , such that for @xmath359 and all @xmath441 , we have @xmath442 we follow the proof of @xcite , theorem 1.5 . throughout the proof , @xmath443 will denote positive constants which are uniform in @xmath341 . _ step _ 1 . for a vertex @xmath444 , define @xmath445 . note that the random variables @xmath446 , @xmath444 , are identically distributed under @xmath123 ( but not independent ) . let @xmath447 denote a random variable with this law . we wish to show that for some constant @xmath448 , @xmath449 \le c_0.\ ] ] denote by @xmath450 the children of the vertex @xmath4 . by assumptions ( xs ) and ( h1)(h3 ) , we have @xmath451 which yields @xmath452 . now , note that the variables @xmath453 , @xmath454 are independent under @xmath123 . the previous inequality then yields for every @xmath78 , @xmath455 \\[-8pt ] \nonumber & = & \p\bigl(\pi_{\omega,\beta}(v ) \le x / c_1 \bigr)^{d-1}.\end{aligned}\ ] ] furthermore , by lemma [ lem : transience ] , there exists a constant @xmath456 , such that @xmath457 together with ( [ eq : pi_omega_beta_1 ] ) , this now easily implies ( [ eq : pi_omega_beta ] ) . _ step _ 2 . for a vertex @xmath4 , let @xmath458 denote the number of times the vertex @xmath4 has been visited by the metropolis algorithm @xmath79 . we wish to show that for each @xmath459 , @xmath460 \le c_3.\ ] ] recall that @xmath461 denotes the first strictly positive hitting time of the vertex @xmath4 , such that @xmath462 = 1/p^v_{\omega , \beta}(t^*_v = \infty)$ ] . by ( [ eq : one_step ] ) , we have @xmath463 , such that @xmath464 & = & \e\bigl[p^\root_{\omega,\beta}(t_v < \infty)e^v_{\omega , \beta}[n_{v } ] \bigr ] \le c_1 \pp_\beta(t_v<\infty)\e\bigl[1/\pi _ { \omega , \beta}(v_1)\bigr ] \nonumber \\[-8pt ] \\[-8pt ] \nonumber & \le & c_4 \pp_\beta(t_v<\infty),\end{aligned}\ ] ] by ( [ eq : pi_omega_beta ] ) . furthermore , we have for every @xmath465 , @xmath466 \\[-8pt ] \nonumber & \ge & c_1 \sum_{|v|=k } \pp_\beta(t_v<\infty)\e\bigl[\pi_{\omega , \beta}(v_1 ) \bigr],\end{aligned}\ ] ] and by ( [ eq : pi_omega_beta_2 ] ) , we have @xmath467 = \e[\pi_{\omega,\beta } ] \ge c_2 $ ] . equations ( [ eq : tv ] ) and ( [ eq : nv1 ] ) now yield ( [ eq : nv ] ) . _ step _ 3 . recall the notation in the statement of the lemma , and let @xmath468 . define the event @xmath469 . a straightforward extension of the proof of the last step allows us to prove that for every constant @xmath296 , @xmath470 } n_{w}>cl \big| \mathcal g_t \biggr ) \le\frac{1 } { cl } \ee_\beta\biggl[\sum_{w\in[v_t , l ] } n_{w}\ind_{e_t } \big| \mathcal g_t \biggr ] < c_3c^{-1}.\ ] ] furthermore , by ( [ eq : one_step ] ) , we have @xmath471 \ge c_1c_2 $ ] . this now yields for every constant @xmath188 and every @xmath472 , @xmath473 } n_{w}\le n \big| \mathcal g_t \biggr ) > c_1c_2 - c_3 c.\ ] ] setting @xmath474 yields the proof . proof of proposition [ prop : tau_moments ] the proof follows an argument in the spirit of @xcite , proposition 3 . let @xmath337 . let @xmath475 go to infinity with @xmath7 ( we will later choose @xmath476 for some constant @xmath477 ) . we have , with @xmath354 denoting the hitting time of level @xmath280 , and with @xmath478 as in the statement of lemma [ lem - alpha ] , @xmath479 \\[-8pt ] \nonumber & \leq&\pp_\beta ( t_l > n)+e^{-\alpha l},\end{aligned}\ ] ] where in the last inequality we used the fact that @xmath480 for @xmath481 . let @xmath319 be the constant from lemma [ lem : level_hitting_prob ] and set @xmath482 . throughout the proof , all constants will be uniform in @xmath13 for @xmath483 . we write @xmath484 . define a vertex @xmath4 to be _ fresh _ if it is visited by the random walk for the first time before time @xmath485 . we will upper bound @xmath486 by showing that on the one hand , there can not be too few fresh points that are well separated and on the other hand , if there are many such fresh points , it is unlikely that @xmath354 is large . for a vertex @xmath4 , let @xmath458 denote the number of visits to @xmath4 by time @xmath485 , and let @xmath487 denote the number of times the random walk visits @xmath4 before time @xmath485 , and then , at the next step that it moves , it visits the ancestor of @xmath4 before time @xmath485 . clearly , for each @xmath4 , @xmath488 , while , using the markov property , there exist constants @xmath489 such that @xmath490 on the other hand , the event @xmath491 implies that the random walk visited @xmath4 and then hit @xmath31 at least @xmath492 times . by lemma [ lem : level_hitting_prob ] and the markov property , the probability that there exists a fresh vertex @xmath4 satisfying the last event is bounded above by @xmath493 combining this with ( [ eq - ofer3 ] ) , we conclude that for all large @xmath280 , @xmath494 on the event @xmath495 , there are at least @xmath496 fresh points , and therefore there are at least @xmath497 fresh points that are @xmath497-separated . let @xmath319 be the constant from lemma [ lem : fresh ] . at each arrival to such a fresh point , with ( annealed ) probability at least @xmath498 the walk hits level @xmath280 before time @xmath499 , for large @xmath280 . it follows that @xmath500 for some @xmath501 . by ( [ eq - ofer4 ] ) and ( [ eq - ofer5 ] ) , there exists now a constant @xmath502 , such that with @xmath503 , we have @xmath504 . together with ( [ eq - ofer1 ] ) , this completes the proof of the first statement of proposition [ prop : tau_moments ] . as for the second statement , let @xmath505 be the children of the root . the law of the subtree of the branching random walk tree rooted at @xmath506 is equal in law under @xmath82 to the subtree rooted at @xmath507 , conditioned on @xmath508 and on @xmath509 . in particular , @xmath510 where @xmath511 and @xmath456 are the constants from the proof of lemma [ lem : fresh ] . this finishes the proof . proof of theorem [ th : einstein ] existence and finiteness of the limit @xmath512 follows from the first part of theorem [ th : weak_er ] . in order to prove the remaining statements , we will use the regeneration structure established in this section . note that the random vectors @xmath513 are i.i.d . under the law @xmath82 and independent from @xmath514 . furthermore , by ( xs ) , @xmath515 , where @xmath516 . proposition [ prop : tau_moments ] and the @xmath138-almost sure convergence of @xmath517 to @xmath23 now yields @xmath518}{\ee[\tau_2-\tau_1 ] } > 0,\ ] ] which proves the first part of the theorem . as for the second part , by standard arguments ( see , e.g. , the proof of theorem 4.1 in @xcite ) , @xmath519 exists almost surely if @xmath520<\infty$ ] , which is the case for all @xmath13 by proposition [ prop : tau_moments ] . furthermore , we have @xmath521 /\ee _ \beta[\tau_2-\tau_1]$ ] , which implies in particular that @xmath522 . now let @xmath523 . by the optional stopping theorem , we have @xmath524 = \ee_\beta[s_{\tau_1 } ] + v_\beta\ee_\beta[\tau_{k_\beta}-\tau_1],\ ] ] such that by proposition [ prop : tau_moments ] and assumption ( xs ) , for some constant @xmath296 , @xmath525 - v_\beta\beta^{-2}\bigr| \le c \ee_\beta\bigl[\tau_{k_\beta}-\beta^{-2}\bigr].\ ] ] crude moment bounds using proposition [ prop : tau_moments ] yield that the right - hand side of the above equation is @xmath526 , which yields @xmath527 = \sigma^2/2 $ ] by theorem [ th : weak_er ] . this finishes the proof . we thank david aldous for bringing the problem of the analysis of the metropolis algorithm on the tree to our attention . we further thank an anonymous referee for detailed comments improving the exposition of the article .

we show that every plane graph with maximum face size four whose all faces of size four are vertex - disjoint is cyclically-colorable . this answers a question of albertson whether graphs drawn in the plane with all crossings independent are-colorable .

0811.2702

coloring of graphs embedded in surfaces , in the plane in particular , attracts a lot of attention of researchers in graph theory . the famous four color theorem @xcite asserts that every graph that can be drawn in the plane with no crossings is @xmath1-colorable . it is natural to ask what number of colors is needed to color graphs that can be embedded in the plane with a restricted number of crossings . if every edge is crossed by at most one edge ( such graphs are called @xmath2-embeddable and we restrict our attention solely to such graphs throughout this paper ) , ringel @xcite conjectured that six colors suffice . this conjecture was answered in affirmative by borodin @xcite . albertson @xcite considered graphs with even more restricted structure of crossings . two distinct crossings are independent if the end - vertices of the crossed pair of edges are mutually different . in particular , if all crossings are independent , then each edge is crossed by at most one other edge . albertson showed that every graph drawn in the plane with at most 3 crossings is @xmath0-colorable ( note that the complete graph of order five can be drawn in the plane with a single crossing ) and conjectured @xcite that every graph that can be drawn in the plane with all its crossings independent is @xmath0-colorable . in this paper , we prove his conjecture . the coloring problem that we study is closely related to the notion of cyclic coloring . a coloring of vertices of an embedded graph is _ cyclic _ if any two vertices incident with the same face receive distinct colors . let us show how the original problem can be expressed using this notion . let @xmath3 be a plane graph with all its crossings independent . we can assume ( by adding edges if necessary ) that all faces of @xmath3 that do not contain a crossing have size three and those that contain a crossing have size four . remove now all edges that are crossed by another edge . clearly , a cyclic coloring of the obtained graph @xmath4 is a proper coloring of the original graph @xmath3 and vice versa . the assumption that all crossings of @xmath3 are independent translates to the fact that all faces of @xmath4 with size four are vertex disjoint . hence , our main result can be stated as follows : [ thm - main ] let @xmath3 be a plane graph with faces of size three and four only . if all the faces of size four are vertex - disjoint , then @xmath3 is cyclically @xmath0-colorable . before we proceed with proving theorem [ thm - main ] , let us survey known results on cyclic colorings of plane graphs . since the maximum face size is a lower bound on the number of colors needed in a cyclic coloring , it is natural to study the number of colors needed to cyclically color a plane graph as a function of its maximum face size @xmath5 . if @xmath6 , then the graph is a triangulation and the optimal number of colors is four by the four color theorem . if @xmath7 , then the optimal number of colors six by results of borodin @xcite ; the optimality is witnessed by the prism over @xmath8 . for larger values of @xmath5 , the cyclic coloring conjecture of ore and plummer @xcite asserts that the optimal number of colors is equal to @xmath9 ( the optimality is witnessed by a drawing of @xmath10 with subdivided edges ) . after a series of papers @xcite on this problem , the best general bound of @xmath11 has been obtained by sanders and zhao @xcite . amini , esperet and van den heuvel @xcite cleverly used a result by havet , van den heuvel , mcdiarmid and reed @xcite on coloring squares of planar graphs and showed that the cyclic coloring conjecture is asymptotically true in the following sense : for every @xmath12 , there exists @xmath13 such that every plane graph of maximum face size @xmath14 admits a cyclic coloring with at most @xmath15 colors . there are two other conjectures related to the cyclic coloring conjecture of ore and plummer . a conjecture of plummer and toft @xcite asserts that every @xmath16-connected plane graph is cyclically @xmath17-colorable . this conjecture is known to be true for @xmath18 and @xmath19 , see @xcite . the restriction of the problems to plane graphs with a bounded maximum face size is removed in the facial coloring conjecture @xcite that asserts that vertices of every plane graph can be colored with at most @xmath20 colors in such a way that every two vertices joined by a facial walk of length at most @xmath21 receive distinct colors . this conjecture would imply the cyclic coloring conjecture for odd values of @xmath5 . partial results towards proving this conjecture can be found in @xcite . the proof of theorem [ thm - main ] is divided into several steps . we first identify configurations that can not appear in a counterexample with the smallest number of vertices . later , using a discharging argument , we show that a plane graph avoiding all these configurations can not exist . in particular , vertices and faces of a counterexample are assigned charge whose total sum is negative and which is redistributed preserving its total sum . lemmas [ lm - f55][lm - v6 ] claim that the final amount of charge of every vertex and every face is non - negative which excludes the existence of a counterexample and yields a proof of theorem [ thm - main ] . we now introduce notation used throughout the paper . let us start with some general notation . a vertex of degree @xmath22 is referred to as a _ @xmath22-vertex _ and a face of size @xmath22 as a _ @xmath22-face_. a _ cyclic neighbor _ of a vertex @xmath23 is a vertex lying on the same face as @xmath23 and the _ cyclic degree _ of @xmath23 is the number of its cyclic neighbors . our goal is to prove theorem [ thm - main ] . we assume that the statement of the theorem is false and consider a counterexample with the smallest number of vertices ; such a counterexample is referred to as _ minimal _ , i.e. , a minimal counterexample @xmath3 is a plane graph with faces of size three and four such that all @xmath1-faces of @xmath3 are vertex - disjoint , @xmath3 has no cyclic @xmath0-coloring and any graph @xmath4 satisfying assumptions of theorem [ thm - main ] with a smaller number of vertices than @xmath3 has a cyclic @xmath0-coloring . a vertex @xmath23 of a minimal counterexample @xmath3 is _ pentagonal _ if the degree of @xmath23 is five , @xmath23 is incident with no @xmath1-face and every neighbor of @xmath23 is incident with a @xmath1-face . a @xmath1-face incident with a neighbor of a pentagonal vertex @xmath23 is said to be _ close _ to @xmath23 if it contains an edge between two consecutive neighbors of @xmath23 ; a @xmath1-face incident with a neighbor of a pentagonal vertex that is not close is _ distant_. if @xmath24 is close / distant to a vertex @xmath23 , then we also say that @xmath23 is close / distant to @xmath24 . a pentagonal vertex is _ solitary _ if no @xmath1-face is close to it . let @xmath23 be a pentagonal vertex and @xmath25 a neighbor of it . let @xmath26 and @xmath27 be the common neighbors of @xmath25 and another neighbor of @xmath23 ( see figure [ fig - sided ] ) . if the @xmath1-face incident with @xmath25 contains both @xmath26 and @xmath27 , then the degree of @xmath25 is five . if the @xmath1-face contains one of the vertices @xmath26 and @xmath27 , then @xmath25 is said to be _ one - sided _ , and if the @xmath1-face incident with @xmath25 contains neither @xmath26 and @xmath27 , then @xmath25 is _ double - sided_. observe that if a pentagonal vertex is adjacent to a vertex of degree five , it must also be adjacent to a double - sided vertex ( otherwise , some of the @xmath1-faces incident with its neighbors would not be vertex - disjoint ) . in this section , we show that a minimal counterexample can not contain certain substructures which we refer to as _ configurations_. let us start with the following simple observation . [ lm - sep ] a minimal counterexample @xmath3 does not contain a separating cycle of length two or three . assume that @xmath3 contains a separating cycle @xmath28 of length two or three . let @xmath4 and @xmath29 be the subgraphs lying in the interior and the exterior of the cycle @xmath28 ( including the cycle @xmath28 itself ) . if @xmath28 is of length two , remove one of the two parallel edges bounding @xmath28 from @xmath4 and @xmath29 . by the minimality of @xmath3 , both @xmath4 and @xmath29 have a cyclic @xmath0-coloring . the colorings of @xmath4 and @xmath29 readily combine to a cyclic @xmath0-coloring of @xmath3 . we will use lemma [ lm - sep ] as follows : if we identify some vertices of a minimal counterexample , lemma [ lm - sep ] guarantees that the resulting graph is loopless as long as every pair of the identified vertices have a common neighbor . indeed , if a loop appeared , the two identified vertices with their common neighbor would form a separating cycle of length three . we next show that the minimum degree of a minimal counterexample is at least five . [ lm - mindeg ] a minimal counterexample @xmath3 does not contain a vertex @xmath23 of degree four or less . if the cyclic degree of @xmath23 is less than five , let @xmath4 be the graph obtained by removing @xmath23 from @xmath3 and triangulating the new face . by the minimality of @xmath3 , @xmath4 has a cyclic @xmath0-coloring . since the cyclic degree of @xmath23 is less than five , this coloring can be extended to a cyclic @xmath0-coloring of the original graph @xmath3 . hence , we can assume that the cyclic degree of @xmath23 is five . in particular , the degree of @xmath23 is four and @xmath23 is contained in a @xmath1-face ( see figure [ fig - mindeg ] ) . let @xmath30 be the neighbors of @xmath23 . by symmetry we can assume that the @xmath1-face incident with @xmath23 is @xmath31 . let @xmath4 be the graph obtained from @xmath3 by removing the vertex @xmath23 and identifying the vertices @xmath32 and @xmath33 to a new vertex @xmath34 , see figure [ fig - mindeg ] . note that the vertex @xmath34 is contained in at most one @xmath1-face since the @xmath1-face incident with @xmath32 becomes a @xmath16-face in @xmath4 . since the vertices @xmath32 and @xmath33 have a common neighbor , the graph @xmath4 is loopless by lemma [ lm - sep ] . by the minimality of @xmath3 , @xmath4 has a cyclic @xmath0-coloring . since two of the neighbors of @xmath23 ( the vertices @xmath32 and @xmath33 ) are assigned the same color and the cyclic degree of @xmath23 is five , the coloring can be extended to a cyclic @xmath0-coloring of @xmath3 . our next step is to show that all vertices of degree five that appear in a minimal counterexample must be pentagonal or incident with a @xmath1-face . [ lm - pentagonal ] every vertex @xmath23 of degree five in a minimal counterexample @xmath3 is either pentagonal or incident with a @xmath1-face . we proceed as in the proof of lemma [ lm - mindeg ] . consider a @xmath0-vertex @xmath23 incident with @xmath16-faces only such that one of its neighbors is not incident with a @xmath1-face . let @xmath30 be the neighbors of @xmath23 and @xmath32 a neighbor not incident with a @xmath1-face . remove @xmath23 and identify vertices @xmath32 and @xmath35 ( see figure [ fig - pentagonal ] ) . since the vertex @xmath32 is not incident with a @xmath1-face in @xmath3 , the new vertex is contained in at most one @xmath1-face . by the minimality of @xmath3 , the new graph can be cyclically @xmath0-colored and this coloring readily yields a coloring of @xmath3 . in the next lemma , we show that no @xmath1-face of a minimal counterexample contains two adjacent vertices of degree five . [ lm - square ] a minimal counterexample @xmath3 does not contain a @xmath1-face with two adjacent vertices of degree five . assume that @xmath3 contains a @xmath1-face @xmath36 such that the degrees of @xmath32 and @xmath37 are five . let @xmath34 be the common neighbor of @xmath32 and @xmath37 , @xmath38 and @xmath39 the other neighbors of @xmath32 ( named in such a way that @xmath39 is a neighbor of @xmath33 ) and @xmath40 and @xmath41 the other neighbors of @xmath37 . see figure [ fig - square ] . let @xmath4 be the graph obtained by removing the vertices @xmath32 and @xmath37 and identifying the vertices @xmath34 and @xmath35 and the vertices @xmath38 and @xmath33 . clearly , the graph @xmath4 is loopless ( as the graph @xmath3 has no separating @xmath16-cycles by lemma [ lm - sep ] ) and all its @xmath1-faces are vertex - disjoint . by the minimality of @xmath3 , @xmath4 has a cyclic @xmath0-coloring . assign the vertices of @xmath3 the colors of their counterparts in @xmath4 . next , color the vertex @xmath37 : observe that two of its @xmath42 cyclic neighbors have the same color and one is uncolored . hence , @xmath37 can be colored . since the vertex @xmath32 has @xmath42 cyclic neighbors and two pairs of its cyclic neighbors have the same color , the coloring can also be extended to @xmath32 . in the next two lemmas , we show that a @xmath1-face of a minimal counterexample can not contain a vertex of degree at most six adjacent to a close pentagonal vertex . [ lm - pentagon5 ] a minimal counterexample @xmath3 does not contain a vertex of degree five contained in a @xmath1-face that is adjacent to a close pentagonal vertex . assume that @xmath3 contains a @xmath1-face @xmath36 such that @xmath32 has degree five and is adjacent to a close pentagonal vertex @xmath23 . let @xmath43 be the neighbors of @xmath23 ( see figure [ fig - pentagon5 ] ) . let @xmath4 be the graph obtained by removing the vertices @xmath23 and @xmath32 and identifying the vertices @xmath37 and @xmath44 and the vertices @xmath33 and @xmath45 . since every pair of identified vertices has a common neighbor , @xmath4 is loopless by lemma [ lm - sep ] . the @xmath1-faces of @xmath4 are also vertex - disjoint . by the minimality of @xmath3 , the graph @xmath4 has a cyclic @xmath0-coloring . assign the vertices of @xmath3 the colors of their counterparts in @xmath4 . we next color the vertex @xmath32 with an available color ( the cyclic degree of @xmath32 is six , it has a pair of neighbors colored with the same color and an uncolored neighbor ) and then the vertex @xmath23 ( its cyclic degree is five and it has a pair of neighbors colored with the same color ) . the existence of this coloring contradicts that @xmath3 is a counterexample . [ lm - pentagon6 ] a minimal counterexample does not contain a vertex of degree six contained in a @xmath1-face that is adjacent to a close pentagonal vertex . assume that @xmath3 contains a @xmath1-face @xmath36 such that @xmath32 has degree six and is adjacent to a close pentagonal vertex @xmath23 . let @xmath43 be the neighbors of @xmath23 and @xmath34 the common neighbor of @xmath32 and @xmath45 ( since all @xmath1-faces are vertex disjoint , both faces containing the edge @xmath46 have size three and the vertex @xmath34 must exist ) . also see figure [ fig - pentagon6 ] . let @xmath4 be the graph obtained from @xmath3 by removing the vertices @xmath23 and @xmath32 and identifying the vertices @xmath37 and @xmath45 and the vertices @xmath33 and @xmath34 . since every pair of identified vertices has a common neighbor , @xmath4 is loopless by lemma [ lm - sep ] . the @xmath1-faces of @xmath4 are also vertex - disjoint . by the minimality of @xmath3 , the graph @xmath4 has a cyclic @xmath0-coloring . assign the vertices of @xmath3 the colors of their counterparts in @xmath4 . we next color the vertex @xmath32 with an available color ( the cyclic degree of @xmath32 is seven , it has two pairs of neighbors colored with the same color and an uncolored neighbor ) and then the vertex @xmath23 ( its cyclic degree is five and it has a pair of neighbors colored with the same color ) . again , the existence of this coloring contradicts that @xmath3 is a counterexample . by lemmas [ lm - pentagon5 ] and [ lm - pentagon6 ] , we have : [ lm - pentagons ] let @xmath3 be a minimal counterexample and @xmath23 a pentagonal vertex with neighbors @xmath32 , @xmath37 , @xmath35 , @xmath33 and @xmath47 in @xmath3 . if the edge @xmath48 is contained in a @xmath1-face , then the degrees of @xmath49 and @xmath50 are at least seven . at the end of this section , we exclude two more complex configurations from appearing around a pentagonal vertex in a minimal counterexample . the configurations described in lemmas [ lm - pentagon65 ] and [ lm - pentagon66 ] are depicted in figures [ fig - pentagon65 ] and [ fig - pentagon66 ] , respectively . [ lm - pentagon65 ] no minimal counterexample contains a pentagonal vertex @xmath23 with neighbors @xmath51 such that for some @xmath52 1 . the degree of @xmath49 is six , 2 . the vertices @xmath49 and @xmath50 have a common neighbor @xmath34 of degree five , 3 . the vertices @xmath49 and @xmath34 have a common neighbor @xmath26 , and 4 . the edges @xmath53 and @xmath54 lie in @xmath1-faces . we can assume that @xmath55 . let @xmath27 be the neighbor of @xmath34 distinct from @xmath37 that lies on the @xmath1-face incident with @xmath34 . remove the vertices @xmath23 , @xmath32 and @xmath34 from @xmath3 , identify the vertices @xmath37 , @xmath47 and @xmath26 to a new vertex @xmath56 , and add an edge @xmath57 . let @xmath4 be the resulting graph . as any pair of identified vertices have a common neighbor , the graph @xmath58 is loopless by lemma [ lm - sep ] . if the edge @xmath57 were a loop , then the vertices @xmath47 and @xmath27 would coincide in @xmath4 which would yield a separating @xmath16-cycle @xmath59 in @xmath3 . we conclude that @xmath4 is loopless . similarly , all @xmath1-faces of @xmath4 are vertex - disjoint . by the minimality of @xmath3 , the graph @xmath4 has a cyclic @xmath0-coloring . assign vertices of @xmath3 the colors of their counterparts in @xmath4 . the only vertices without a color are the vertices @xmath34 , @xmath32 and @xmath23 which we color in this order . let us verify that each of these vertices is cyclically adjacent to vertices of at most four distinct colors when we want to color it . at the beginning , the vertex @xmath34 has six cyclic neighbors , out of which two have the same color ( @xmath37 and @xmath26 ) and one is uncolored . next , the vertex @xmath32 has cyclic degree seven but it is adjacent to a triple of vertices with the same color and an uncolored vertex . finally , the cyclic degree of @xmath23 is five and two of its neighbors have the same color . the constructed coloring violates our assumption that @xmath3 is a counterexample . [ lm - pentagon66 ] no minimal counterexample contains a pentagonal vertex @xmath23 with neighbors @xmath51 such that for some @xmath52 1 . the degree of @xmath49 is six , 2 . the vertices @xmath49 and @xmath50 have a common neighbor @xmath34 of degree six , 3 . the vertices @xmath49 and @xmath34 have a common neighbor @xmath26 , and 4 . the edges @xmath53 and @xmath54 lie in @xmath1-faces . we can assume that @xmath55 . let @xmath27 , @xmath60 and @xmath61 be the neighbors of @xmath34 as depicted in figure [ fig - pentagon66 ] . remove the vertices @xmath23 , @xmath32 and @xmath34 from @xmath3 , identify the vertices @xmath37 , @xmath47 and @xmath26 to a new vertex @xmath56 and identify the vertices @xmath27 and @xmath61 . let @xmath4 be the resulting graph . as any pair of identified vertices have a common neighbor , the graph @xmath4 is loopless by lemma [ lm - sep ] . moreover , all @xmath1-faces of @xmath4 are vertex - disjoint . by the minimality of @xmath3 , the graph @xmath4 has a cyclic @xmath0-coloring . now assign vertices of @xmath3 the colors of their counterparts in @xmath4 . the only vertices without a color are the vertices @xmath34 , @xmath32 and @xmath23 which we color in this order . let us verify that each of these vertices is cyclically adjacent to vertices of at most four distinct colors when we want to color it . at the beginning , the vertex @xmath34 has seven cyclic neighbors , out of which two pairs have the same color ( the pair @xmath37 and @xmath26 , and the pair @xmath27 and @xmath61 ) and one neighbor is uncolored . next , the vertex @xmath32 has also cyclic degree seven but it is adjacent to a triple of vertices with the same color and an uncolored vertex . finally , the cyclic degree of @xmath23 is five and two of its neighbors have the same color . finally , the obtainec coloring contradicts that @xmath3 is a counterexample . the core of the proof is an application of the standard discharging method . we fix a minimal counterexample and assign each vertex and each face initial charge as follows : each @xmath22-vertex receives @xmath62 units of charge and each @xmath22-face receives @xmath63 units of charge . an easy application of euler formula yields that the sum of initial amounts of charge is @xmath64 . the amount of charge is then redistributed using the rules introduced in this section in such a way that all vertices and faces have non - negative amount of charge at the end . since the redistribution preserves the total amount of charge , this will eventually contradict the existence of a minimal counterexample . let us start presenting the rules for charge redistribution . rules s1 and s2 guarantee that the amount of final charge of every vertex incident with a @xmath1-face is zero ( vertices not incident with a @xmath1-face are not affected by rules s1 and s2 ) . rule s1 : : every @xmath0-vertex receives @xmath2 unit of charge from its ( unique ) incident @xmath1-face . rule s2 : : every @xmath22-vertex , @xmath65 , sends @xmath62 units of charge to its incident @xmath1-face . a more complex set of rules is needed to guarantee that the amount of final charge of pentagonal vertices is non - negative . the following notation is used in rules p5a p8 + : @xmath23 is a pentagonal vertex adjacent to a vertex @xmath34 incident with a @xmath1-face @xmath24 distant from @xmath23 ; the neighbors of @xmath34 incident with @xmath24 are denoted @xmath26 and @xmath27 . a vertex @xmath34 is understood to be one - sided or double - sided with respect to @xmath23 . rules p5a p7c are illustrated in figure [ fig - rules ] . rule pc : : every pentagonal vertex receives @xmath2 unit of charge from each close @xmath1-face . rule p5a : : if @xmath34 has degree five and exactly one of the vertices @xmath26 and @xmath27 have degree six , then @xmath23 receives @xmath66 units of charge from @xmath24 . rule p5b : : if @xmath34 has degree five and both @xmath26 and @xmath27 have degree at least seven , then @xmath23 receives @xmath67 units of charge from @xmath24 . rule p6a : : if @xmath34 has degree six , exactly one of the vertices @xmath26 and @xmath27 have degree five and the other has degree six , then @xmath23 receives @xmath68 units of charge from @xmath24 . rule p6b : : if @xmath34 has degree six and the sum of the degrees of @xmath26 and @xmath27 is at least twelve , then @xmath23 receives @xmath69 units of charge from @xmath24 . rule p7a : : if @xmath34 is a one - sided vertex of degree seven and both @xmath26 and @xmath27 have degree five , then @xmath23 receives @xmath70 units of charge from @xmath24 . rule p7b : : if @xmath34 is a one - sided vertex of degree seven and at most one of the vertices @xmath26 and @xmath27 has degree five , then @xmath23 receives @xmath69 units of charge from @xmath24 . rule p7c : : if @xmath34 is a double - sided vertex of degree seven , then @xmath23 receives @xmath69 units of charge from @xmath24 . rule p8 + : : if the degree of @xmath34 is eight or more , then @xmath23 receives @xmath69 units of charge from @xmath24 . the amount of final charge of faces and vertices after redistributing charge based on the above rules is analyzed in the next two sections . in this section , we analyze the final amount of charge of faces in a minimal counterexample . since @xmath16-faces do not receive or send out any charge , it is enough to analyze the final charge of @xmath1-faces . we break down the analysis into four lemmas that cover all possible cases how a @xmath1-face can look like ( up to symmetry ) . we start with @xmath1-faces incident with two vertices of degree five . [ lm - f55 ] let @xmath71 be a @xmath1-face of a minimal counterexample . if the degrees of @xmath32 and @xmath35 are five , then the final amount of charge of @xmath24 is non - negative . by lemma [ lm - square ] , the degree of @xmath37 and @xmath33 is at least six , and by lemma [ lm - pentagons ] , no pentagonal vertex is close to @xmath24 . let @xmath72 be the number of vertices of degree seven or more incident with @xmath24 . by rules p5a or p5b , the face @xmath24 sends pentagonal vertices adjacent to @xmath32 or @xmath35 at most @xmath73 units of charge . let @xmath74 be the degree of a vertex @xmath49 , @xmath75 . if @xmath76 for @xmath75 , then @xmath24 sends out no charge to pentagonal vertices adjacent to @xmath49 . if @xmath77 for @xmath75 , then the face @xmath24 sends either @xmath70 units of charge to at most two pentagonal vertices adjacent to @xmath49 by rule p7a or @xmath69 units of charge to a single vertex by rule p7c ; this follows from the fact no two adjacent neighbors of a vertex @xmath49 can be both pentagonal and the common neighbors of @xmath49 and @xmath32 or @xmath35 are not pentagonal by lemma [ lm - pentagon5 ] . these two facts also imply for @xmath78 that @xmath49 sends to each of at most @xmath79 pentagonal vertices adjacent to @xmath49 @xmath69 units of charge by rule p8 + . let us summarize . after rules s1 and s2 apply , the amount of charge of @xmath24 is equal to @xmath80 . we next distinguish several cases based on @xmath81 and @xmath82 : * if @xmath83 and @xmath84 , no further charge is sent out and the final charge of @xmath24 is zero . * if @xmath83 and @xmath85 ( or vice versa ) , @xmath24 sends out at most @xmath67 units of charge to pentagonal vertices adjacent to @xmath32 or @xmath35 and at most @xmath86 units of charge to such vertices adjacent to @xmath33 . hence , its final charge is again non - negative . * if @xmath83 and @xmath87 ( or vice versa ) , @xmath24 sends out at most @xmath67 units of charge to pentagonal vertices adjacent to @xmath32 or @xmath35 and at most @xmath88 units of charge to such vertices adjacent to @xmath33 . hence , its final charge is again non - negative . * if @xmath89 and @xmath85 , @xmath24 sends out at most @xmath90 units of charge to pentagonal vertices adjacent to @xmath32 or @xmath35 , at most @xmath86 units of charge to pentagonal vertices adjacent to @xmath37 and at most @xmath86 units of charge to pentagonal vertices adjacent to @xmath33 . its final charge is again non - negative . * if @xmath89 and @xmath87 ( or vice versa ) , @xmath24 sends out at most @xmath90 units of charge to pentagonal vertices adjacent to @xmath32 or @xmath35 , at most @xmath86 units of charge to such vertices adjacent to @xmath37 and at most @xmath88 units of charge to pentagonal vertices adjacent to @xmath33 . hence , its final charge is again non - negative . * if @xmath91 and @xmath87 , the face @xmath24 sends out at most @xmath90 units of charge to pentagonal vertices adjacent to @xmath32 or @xmath35 , and at most @xmath92 units of charge to such vertices adjacent to @xmath37 or @xmath33 . hence , its final charge is again non - negative . next , we analyze @xmath1-faces incident with vertices of degree seven or more only . note that the bound on the number of pentagonal neighbors of vertices of a @xmath1-face is also used in lemmas [ lm - f5][lm - f6 ] without giving so much details on its derivation as in the proof of lemma [ lm - f7 ] . [ lm - f7 ] let @xmath71 be a @xmath1-face of a minimal counterexample . if the degrees of @xmath32 , @xmath37 , @xmath35 and @xmath33 are at least seven , then the final amount of charge of @xmath24 is non - negative . let @xmath93 be the sum of the degrees of the vertices @xmath32 , @xmath37 , @xmath35 and @xmath33 . after rule s2 applies to each of these four vertices , the face @xmath24 has charge @xmath94 . rules pc , p7a , p7b , p7c and p8 + apply at most @xmath95 vertices . the vertices @xmath32 , @xmath37 , @xmath35 and @xmath33 have @xmath96 neighbors not incident with the face @xmath24 counting the common neighbors of them twice . hence , if the common neighbors of @xmath49 and @xmath50 are counted once , there are at most @xmath97 neighbors not incident with @xmath24 and since no two adjacent vertices can be both pentagonal , the number of pentagonal neighbors is at most @xmath95 . rule pc can apply at most @xmath1 times since a single @xmath1-face can be close to at most @xmath1 pentagonal vertices . since @xmath24 can send out at most @xmath69 units of charge by rules p7a , p7b , p7c and p8 + , and it can send out at most @xmath2 unit of charge by rule pc , the @xmath1-face @xmath24 sends out at most the following amount of charge : @xmath98 by the assumptions of the lemma , the degree of each vertex @xmath49 is at least @xmath99 and thus @xmath100 . since @xmath101 for @xmath100 , the final amount of charge of @xmath24 is non - negative . we next analyze @xmath1-faces incident with a single vertex of degree five . [ lm - f5 ] let @xmath71 be a @xmath1-face of a minimal counterexample . if the degree of @xmath32 is five and the degree of @xmath35 is at least six , then the final amount of charge of @xmath24 is non - negative . if all vertices @xmath37 , @xmath35 and @xmath33 have degree six , then @xmath24 can send out @xmath68 units of charge by rule p6a to pentagonal neighbors of @xmath37 and @xmath33 ( note that each of these two vertices has at most one such pentagonal neighbor ) and @xmath69 units of charge by rule p6b to a pentagonal neighbor of @xmath35 . observe that no pentagonal vertex is close to @xmath24 by lemma [ lm - pentagons ] . altogether , @xmath24 receives no charge and sends out at most @xmath102 units of charge ( one unit by rule s1 to @xmath32 ) . consequently , its final charge is non - negative . if two of the vertices @xmath37 , @xmath35 and @xmath33 have degree six and one has degree @xmath103 , then @xmath24 can send out at most @xmath66 units of charge to a pentagonal neighbor of @xmath32 , at most @xmath69 units charge to a pentagonal neighbor of each vertex of degree six , at most @xmath69 to at most @xmath104 pentagonal neighbors of the vertex of degree @xmath22 and @xmath2 unit of charge to @xmath32 . altogether , it sends out at most @xmath105 units of charge . since the initial charge of @xmath24 amounts to @xmath102 units and @xmath24 receives @xmath62 units by rule s2 , its final charge is non - negative if @xmath106 ( observe that @xmath107 for @xmath106 ) . if @xmath108 and the vertex of degree @xmath22 is @xmath37 , then @xmath24 can send @xmath2 unit of charge to @xmath32 by rule s1 , @xmath66 units of charge to a pentagonal neighbor of @xmath32 by rule p5b , @xmath69 units of charge to each of at most two pentagonal neighbors of @xmath37 by rule p7b or p7c , @xmath69 units of charge to a pentagonal neighbor of @xmath35 by rule p6b and @xmath68 units of charge to a pentagonal neighbor of @xmath33 by rule p6a . in total , @xmath24 sends out at most @xmath109 units of charge . the case that the vertex of degree @xmath108 is @xmath33 is symmetric to this one . finally , if the vertex of degree @xmath108 is @xmath35 , then @xmath24 can send @xmath2 unit of charge to @xmath32 by rule s1 and @xmath69 units of charge to at most four pentagonal neighbors of @xmath37 , @xmath35 and @xmath33 . the face @xmath24 sends no charge to a pentagonal neighbor of @xmath32 since neither rule p5a nor p5b can apply . again , the final charge of @xmath24 is non - negative . we now assume that only one of the vertices @xmath37 , @xmath35 and @xmath33 have degree six and the remaining two vertices have degrees @xmath22 and @xmath110 , @xmath103 and @xmath111 . the face @xmath24 sends out @xmath2 unit of charge to @xmath32 by rule s1 , at most @xmath112 units of charge to a pentagonal neighbor of @xmath32 , at most @xmath113 units of charge to a pentagonal neighbor of the vertex of degree six , and at most @xmath113 units of charge to each of at most @xmath114 pentagonal neighbors of vertices of degree @xmath22 and @xmath110 unless rule pc applies . rule pc can apply at most once by lemma [ lm - pentagons ] . since the initial amount charge of @xmath24 is @xmath102 , @xmath24 receives @xmath115 units by rule s2 and sends out at most @xmath116 units of charge and at most @xmath117 if rule pc does not apply , the final charge of @xmath24 is non - negative ( note that @xmath118 for @xmath119 ) unless @xmath120 and rule pc also applies . if @xmath120 and rule pc applies , lemma [ lm - pentagon6 ] implies that the vertices of degree seven are adjacent . by symmetry , @xmath32 has degree five , @xmath37 has degree six and @xmath35 and @xmath33 have degree seven . hence , @xmath24 can send @xmath2 unit of charge to @xmath32 by rule s1 , @xmath66 units of charge to a pentagonal neighbor of @xmath32 by rule p5a , at most @xmath69 units of charge to each of at most three pentagonal neighbors of @xmath37 , @xmath35 and @xmath33 that are not close and @xmath2 unit of charge to the close pentagonal neighbor by rule pc . we conclude that @xmath24 sends out at most @xmath121 units of charge while it receives @xmath102 units of charge by rule s2 in addition to @xmath102 units of its initial charge , i.e. , its final charge is non - negative . it remains to consider the case when all the vertices @xmath37 , @xmath35 and @xmath33 have degree at least seven . let @xmath74 be the degree of the vertex @xmath49 , @xmath122 . there are at most @xmath123 pentagonal neighbors of the vertices @xmath37 , @xmath35 and @xmath33 and rule pc can apply at most twice . in addition , the face @xmath24 can send out @xmath67 units of charge to a pentagonal neighbor of a vertex @xmath32 and @xmath2 unit of charge to @xmath32 by rule s1 . altogether , the amount of charge sent out by @xmath24 is at most : @xmath124 the initial amount of charge of @xmath24 is @xmath102 units and @xmath24 receives @xmath125 units of charge by rule s2 from the vertices @xmath37 , @xmath35 and @xmath33 . hence , if @xmath126 , then the final charge of the face @xmath24 is clearly non - negative . if @xmath127 , then all the degrees @xmath81 , @xmath128 and @xmath82 must be equal to @xmath99 . if the vertices @xmath37 , @xmath35 and @xmath33 have six pentagonal neighbors , then none of them is close to @xmath24 . hence , rule pc never applies . we conclude that @xmath24 sends out at most the following amount of charge : @xmath129 on the other hand , if there are at most five pentagonal neighbors of @xmath37 , @xmath35 and @xmath33 , rule pc can apply ( at most twice ) . hence , the charge sent out by @xmath24 is at most : @xmath130 since the initial amount of charge of @xmath24 is @xmath102 units and @xmath24 receives @xmath16 units of charge from the vertices @xmath37 , @xmath35 and @xmath33 , its final charge is non - negative . finally , we analyze @xmath1-faces incident with vertices of degree six but no vertices of degree five . [ lm - f6 ] let @xmath71 be a @xmath1-face of a minimal counterexample . if the degree of @xmath32 is six and the degrees of @xmath37 , @xmath35 and @xmath33 are at least six , then the final amount of charge of @xmath24 is non - negative . let @xmath93 be the sum of the degrees of the vertices @xmath32 , @xmath37 , @xmath35 and @xmath33 . after rule s2 applies to each of these four vertices , the face @xmath24 has charge @xmath94 . we now distinguish several cases based on which vertices @xmath49 , @xmath131 , have degree six : * if all vertices @xmath49 have degree six , then there is no pentagonal vertex close to @xmath24 by lemma [ lm - pentagon6 ] . hence , each @xmath49 is adjacent to at most one pentagonal vertex and @xmath24 sends @xmath69 units of charge by rule p6b at most four times . this implies that the final amount of charge of @xmath24 is non - negative . * if three vertices @xmath49 have degree six , then there is again no pentagonal vertex close to @xmath24 by lemma [ lm - pentagon6 ] . let @xmath22 be the degree of the vertex with degree seven or more . such vertex is adjacent to at most @xmath104 pentagonal vertices and each other vertex to at most one pentagonal vertex . hence , @xmath24 sends out at most @xmath132 units of charge . since its charge after applying rule s2 was @xmath133 and @xmath103 , its final amount of charge is non - negative . * it two vertices @xmath49 have degree six , then there is at most one pentagonal vertex close to @xmath24 . the charge is sent by @xmath24 to at most @xmath95 pentagonal vertices and at most once by rule pc . hence , the total amount of charge sent out is at most @xmath134 since @xmath135 and the charge of @xmath24 after applying rule s2 is at @xmath94 , the final amount of charge of @xmath24 is non - negative . * if @xmath32 is the only vertex @xmath49 with degree six , the charge is sent by @xmath24 to at most @xmath95 pentagonal vertices and at most twice by rule pc . hence , the total amount of charge sent out is at most @xmath136 since @xmath137 and the charge of @xmath24 after applying rule s2 is at @xmath94 , the final amount of charge of @xmath24 is non - negative . a minimal counterexample has no vertices of degree four or less by lemma [ lm - mindeg ] . the amount of final charge of vertices that are not pentagonal is non - negative : vertices incident with a @xmath1-face have zero final charge since only rule s1 or s2 can apply to them and other non - pentagonal vertices keep their original ( non - negative ) charge since none of the rules applies to them ( note that every vertex of degree five is either pentagonal or incident with a @xmath1-face by lemma [ lm - pentagonal ] ) . hence , we can focus on the amount of final charge of pentagonal vertices . pentagonal vertices that are not solitary receive @xmath2 unit of charge from a close @xmath1-face by rule pc and thus their final charge is non - negative . we now analyze the amount of charge of solitary pentagonal vertices and start with those adjacent to a vertex of degree five . [ lm - v5 ] every solitary pentagonal vertex @xmath23 adjacent to a vertex of degree five has non - negative final charge . let @xmath30 be the neighbors of @xmath23 and @xmath138 the @xmath1-face containing the vertex @xmath49 , @xmath139 . by symmetry , we can assume that the degree of @xmath37 is five . since no two @xmath1-faces share a vertex , @xmath23 has a double - sided neighbor @xmath140 . note that @xmath141 and the @xmath1-face @xmath142 sends @xmath69 units of charge to @xmath23 ( either by rule p7c or rule p8 + ) . let @xmath143 be the common neighbor of @xmath32 and @xmath37 and @xmath144 the common neighbor of @xmath37 and @xmath35 ( see figure [ fig - v5 ] ) . since the degree of @xmath37 is five , the degrees of @xmath143 and @xmath144 are at least six by lemma [ lm - square ] . if the degree of @xmath143 is six , then the degree of @xmath32 is at least seven by lemma [ lm - pentagon66 ] and the @xmath1-face @xmath145 sends @xmath23 at least @xmath70 units of charge . similarly , if the degree of @xmath144 is six , then the @xmath1-face @xmath146 sends @xmath23 at least @xmath70 units of charge . on the other hand , if the degree of at least one of the vertices @xmath143 and @xmath144 is bigger than six , then @xmath23 receives at least @xmath66 units of charge from the @xmath1-face @xmath147 , and if the degrees of both @xmath143 and @xmath144 are bigger than six , then @xmath23 receives at least @xmath67 units of charge from @xmath147 . we conclude that if @xmath148 , then @xmath23 receives @xmath69 units of charge from @xmath142 and at least @xmath67 units of charge from the faces @xmath145 , @xmath147 and @xmath146 . in particular , the final charge of @xmath23 is non - negative unless @xmath23 receives exactly @xmath67 units of charge from the faces @xmath145 , @xmath147 and @xmath146 altogether . in such case , @xmath23 receives @xmath67 units of charge from @xmath147 , which implies that the degrees of @xmath143 and @xmath144 are more than six , and no charge is sent from @xmath145 or @xmath146 , which implies that the degrees of @xmath32 and @xmath35 are six and the degrees of their neighbors on @xmath145 and @xmath146 are five . let us analyze this case in more detail . by symmetry , we can assume that @xmath149 . let @xmath150 be the common neighbor of @xmath35 and @xmath33 . since @xmath146 sends no charge , the degree of @xmath150 is five . hence , the degree of @xmath33 is at least seven by lemma [ lm - pentagon65 ] . consequently , the face @xmath151 sends @xmath23 at least @xmath70 units of charge . altogether , @xmath23 receives @xmath67 units of charge from @xmath147 , at least @xmath70 units of charge from @xmath151 and @xmath69 units of charge from @xmath152 and its final charge is non - negative . we have just shown that if @xmath148 , then the final charge of @xmath23 is non - negative . in the rest , we assume that @xmath153 and one of the following two cases applies ( otherwise , the faces @xmath147 and @xmath146 send together at least @xmath69 units of charge to @xmath23 and the final charge of @xmath23 is non - negative ) . the other cases are excluded by lemmas [ lm - square ] and [ lm - pentagon66 ] . * * the @xmath1-face @xmath147 sends @xmath23 no charge and the @xmath1-face @xmath146 sends @xmath70 units of charge . * + in this case , the degrees of both @xmath143 and @xmath144 are six and @xmath35 is a one - sided vertex with degree seven with both neighbors on @xmath146 of degree five . in particular , the common neighbor @xmath150 of @xmath35 and @xmath33 lies in the face @xmath146 and it has degree five . by lemma [ lm - pentagon65 ] , the degree of @xmath33 is at least seven and thus the @xmath1-face @xmath151 sends at least @xmath70 units of charge to @xmath23 . in total , @xmath23 receives @xmath69 units of charge from @xmath145 , @xmath70 units of charge from @xmath146 and at least @xmath70 units of charge from @xmath151 . we conclude that the final charge of @xmath23 is non - negative . * * the @xmath1-face @xmath147 sends @xmath66 or @xmath67 units of charge and the @xmath1-face @xmath146 sends no charge . * + in this case , @xmath35 has degree six and its common neighbor @xmath150 with the vertex @xmath33 has degree five and lies on the face @xmath146 . lemma [ lm - pentagon65 ] now implies that the degree of @xmath33 is at least seven . hence , the face @xmath151 sends at least @xmath70 units of charge to @xmath23 . summarizing , @xmath23 receives @xmath69 units of charge from @xmath145 , at least @xmath66 units of charge from @xmath147 and at least @xmath70 units of charge from @xmath151 which makes its final charge non - negative . * * the @xmath1-face @xmath147 sends @xmath66 units of charge and the @xmath1-face @xmath146 sends @xmath68 units of charge . * + in this case , @xmath35 has degree six and its common neighbor @xmath150 with the vertex @xmath33 has degree five or six and lies on the face @xmath146 . lemmas [ lm - pentagon65 ] and [ lm - pentagon66 ] yield that the degree of @xmath33 is at least seven . this implies that the face @xmath151 sends at least @xmath70 units of charge to @xmath23 . we conclude that @xmath23 receives @xmath69 units of charge from @xmath145 , @xmath66 units of charge from @xmath147 , @xmath68 units of charge from @xmath146 and at least @xmath70 units of charge from @xmath151 , and the final charge of @xmath23 is non - negative . it remains to analyze solitary pentagonal vertices adjacent to no vertices of degree five . [ lm - v6 ] every solitary pentagonal vertex @xmath23 adjacent to no vertex of degree five has non - negative final charge . let @xmath30 be the neighbors of @xmath23 and @xmath154 the @xmath1-faces incident with the neighbors of @xmath23 as in the proof of lemma [ lm - v5 ] . if @xmath23 receives charge from at least four of the faces @xmath154 , then it receives at least @xmath2 unit of charge in total and its final charge is non - negative . hence , we can assume that @xmath23 does not receive charge from two of the faces , by symmetry , from the face @xmath145 and the face @xmath147 or @xmath146 . note that if @xmath23 receives no charge from the face @xmath138 , then @xmath49 has degree six and both its neighbors on @xmath138 must have degree five . let us first assume that the vertex @xmath23 receives no charge from the faces @xmath145 and @xmath147 . the situation is depicted in figure [ fig - v6a ] ; note that the vertices @xmath32 and @xmath37 can not have a common neighbor of degree five on a face @xmath145 or @xmath147 by lemma [ lm - pentagon65 ] . observe that there must be a double - sided vertex @xmath140 , @xmath155 . by lemma [ lm - pentagon65 ] , the degrees of the vertices @xmath35 and @xmath47 are at least seven . hence , if @xmath156 , @xmath23 receives at least @xmath70 units of charge from the faces @xmath146 and @xmath152 and @xmath69 units of charge from @xmath151 , and its final charge is non - negative . we now assume that @xmath149 and the face @xmath146 sends only @xmath70 units of charge to @xmath23 ( otherwise , @xmath23 receives @xmath69 units of charge from @xmath146 and its final charge is non - negative ) . hence , @xmath35 is a one - sided vertex of degree seven and the common neighbor @xmath34 of @xmath35 and @xmath33 has degree five and lies on @xmath146 . consequently , the degree of @xmath33 is at least seven by lemma [ lm - pentagon65 ] . we conclude that @xmath23 receives @xmath70 units of charge from @xmath146 , at least @xmath70 units of charge from @xmath151 and @xmath69 units of charge from @xmath152 . again , the final charge of @xmath23 is non - negative . we have ruled out the case that there would be two adjacent neighbors of @xmath23 whose @xmath1-faces sent no charge to @xmath23 . hence , it remains to analyze the case when the faces @xmath145 and @xmath146 send no charge to @xmath23 . we claim that the face @xmath147 sends @xmath69 units of charge to @xmath23 . this clearly holds if @xmath37 is double - sided or its degree is at least eight . if the degree of @xmath37 is six , then @xmath147 sends @xmath69 units of charge unless the neighbors of @xmath37 on @xmath147 have degrees five and six . such configurations are excluded by lemmas [ lm - pentagon65 ] and [ lm - pentagon66 ] . finally , if @xmath37 is one - sided and its degree is seven , then @xmath147 sends @xmath69 units of charge to @xmath23 unless both the neighbors of @xmath37 on @xmath147 have degree five . one of these neighbors is also a neighbor of @xmath32 or @xmath35 which is impossible by lemma [ lm - pentagon66 ] . we have shown that @xmath23 receives @xmath69 units of charge from @xmath147 . since @xmath23 receives in addition at least @xmath68 units of charge from each of the faces @xmath151 and @xmath152 , its final charge is non - negative . lemmas [ lm - f55][lm - v6 ] now yield theorem [ thm - main ] as explained in section [ sect - notation ] . if @xmath3 is a plane graph with faces of size three only , then four color theorem implies that @xmath3 is cyclically @xmath1-colorable . our theorem asserts that every plane graph with faces of size three and four such that all faces of size four are vertex - disjoint is cyclically @xmath0-colorable . it is natural to ask whether the following might be true : let us remark that it is quite easy to see that such graphs @xmath3 are @xmath158-colorable . indeed , adding a clique to every face of size four or more results in a graph with average degree less than @xmath159 . after removing a vertex from @xmath3 that has degree less than @xmath159 in the modified graph and adding edges to @xmath3 in such a way that big faces are still vertex - disjoint and all vertices lying on a common face in @xmath3 lie on a common face in the new graph , induction can be applied to the new graph which yields the proof of the claimed bound . 99 m. albertson : chromatic number , independence ratio , and crossing number , ars math . contemporanea 1 ( 2008 ) , 16 . m. albertson : colorings and crossings , presentation at siam conference on discrete mathematics 2008 , burlington , vt . o. amini , l. esperet , j. van den heuvel : a unified approach to distance - two colouring of planar graphs , manuscript . k. appel , w. haken : every planar map is four colorable , bull . am . ( 1976 ) , 449456 . o. borodin : solution of ringel s problems on vertex - face coloring of plane graphs and coloring of @xmath2-planar graphs , met . novosibirsk 41 ( 1984 ) , 1226 ( in russian ) . o. borodin : cyclic coloring of plane graphs , discrete math . 100 ( 1992 ) , 281289 . o. borodin : a new proof of the 6 color theorem , j. graph theory 19 ( 1995 ) , 507521 . o. borodin , d. p. sanders , y. zhao : on cyclic colorings and their generalizations , discrete math . 203 ( 1999 ) , 2340 . h. enomoto , m. hork , s. jendrol : cyclic chromatic number of 3-connected plane graphs , siam . j. discrete math . 14 ( 2001 ) , 121137 . f. havet , j. van den heuvel , c. mcdiarmid , b. reed : list colouring squares of planar graphs , in : electronic notes in discrete mathematics 29 ( 2007 ) , 515519 ( proceedings of eurocomb07 ) . f. havet , j. van den heuvel , c. mcdiarmid , b. reed : list colouring squares of planar graphs , manuscript . f. havet , d. krl , j .- s . sereni , r. krekovski : facial colorings using hall s theorem , submitted . f. havet , j .- s . sereni , r. krekovski : @xmath16-facial colouring of plane graphs , siam . j. discrete math . 22 ( 2008 ) , 231247 . m. hork , s. jendrol : on a conjecture by plummer and toft , j. graph theory 30 ( 1999 ) , 177189 . m. hork , s. jendrol : on vertex types and cyclic colourings of 3-connected plane graphs , discrete math . 212 ( 2000 ) , 101109 . m. hork , j. zlmalov : another step towards proving a conjecture of plummer and toft , submitted . d. krl , t. madaras , r. krekovski : cyclic , diagonal and facial coloring , european j. combin . 26 ( 2005 ) , 473490 . d. krl , t. madaras , r. krekovski : cyclic , diagonal and facial coloring a missing case , european j. combin . 28 ( 2007 ) , 16371639 . o. ore , m. d. plummer : cyclic coloration of plane graphs , in : recent progress in combinatorics ( proceedings of the third waterloo conference on combinatorics , may 1968 ) ( w. t. tutte , ed . ) , academic press , 1969 . m. d. plummer , b. toft : cyclic coloration of 3-polytopes , j. graph theory 11 ( 1987 ) , 507515 . g. ringel : ein sechsfarbenproblem auf der kugel , abh . hamburg 29 ( 1965 ) , 107117 ( in german ) . n. robertson , d. sanders , d. seymour , r. thomas : the four color theorem , j. combin . theory ser . b 70 ( 1997 ) , 244 . d. p. sanders , y. zhao : a new bound on the cyclic chromatic number , j. combin . theory ser . b 83 ( 2001 ) , 102111 .

a simple quantitative model is presented for the history of galaxies to explain galaxy number counts , redshift distributions and some other related observations . we first infer that irregular galaxies and the disks of spiral galaxies are young , probably formed at from a simultaneous consideration of colours and gas content under a moderate assumption on the star formation history . assuming that elliptical galaxies and bulges of spiral galaxies , both called spheroids in the discussion , had formed early in the universe , the resulting scenario is that spiral galaxies formed as intergalactic gas accreting onto pre - existing bulges mostly at ; irregular galaxies as seen today formed by aggregation of clouds at . taking the formation epochs thus estimated into account , we construct a model for the history of galaxies employing a stellar population synthesis model . we assume that the number of galaxies does not change except that some of them ( irregulars ) were newly born , and use a morphology - dependent local luminosity function to constrain the number of galaxies . we represent the galaxies by e / s0 , sab , sc and irr ; low luminosity dwarfs or any objects unobservable today do not play a role in our considerations . in our model , spheroids follow passive evolution and the luminosity of spiral galaxies evolves only very slowly for a wide redshift interval due to a counterbalance between fading stars and new star formation from the gas replenished from intergalactic space . irregular galaxies evolve moderately fast for . the predictions of the model are compared with the observation of galaxy number counts and redshift distributions for the , and colour bands . we show that band observations are largely controlled by spheroids , which make them particularly suitable to study cosmology . we argue that models are disfavoured , unless the basic assumptions of the present model are abandoned . the band observations reach quite high redshift : for instance observations at=23 mag may explore the formation epoch , which could be as high as . on the other hand , galaxies observed in the band are dominated by disks and irregulars , spheroids making a very small contribution . it is shown that young irregular galaxies cause the steep slope of the counts . the fraction of irregular galaxies increases with decreasing brightness : at mag , they contribute as much as spiral galaxies . thus , `` the faint blue galaxy problem '' is solved by invoking young galaxies . this interpretation is corroborated by a comparison of our prediction with the morphologically - classified galaxy counts in the band . we do not invoke sporadic star bursting : star formation takes place steadily as does today , but galaxies ( especially irregulars ) are gaseous at higher redshift , and hence star formation is much more active than today . consistency is also shown with the constraint on the luminosity evolution from a mg ii quasar - absorption - line selected sample . we estimate that 2/3 of the baryons in stars are stored in spheroids and 1/3 in disks , only% being in irregular galaxies . the amount of baryons in disk stars is increasing , as they form to , which just offsets the decrease of neutral gas towards the present epoch , as inferred from quasar absorption line surveys . 814i_f814w

astro-ph9705057

when galaxies formed and how they evolved are among the central issues of cosmology today . the traditional clue to this problem has been provided by the number count of galaxies as a function of apparent magnitude . early observations in the blue band showed that the number of galaxies per unit angular area increases faster than is expected in simple models for fainter magnitude ( koo & kron 1992 , for a review ) . it has become clear , however , that the number count alone does not allow an unambiguous interpretation . over the last decade , much effort has been expended to obtain redshift of faint galaxies ( broadhurst , ellis & shanks 1988 , hereafter bes ; colless et al . 1990 ; 1993 ; lilly , cowie & gardner 1991 ; koo & kron 1992 ; songaila et al . 1994 ; glazebrook et al . 1995a ; lilly et al . 1995 ; ellis et al . these redshift surveys have revealed , when combined with the count data , that the distribution of galaxies in the universe is much more puzzling than it looked : the excessive number of galaxies observed in the number count seemed to be most simply explained by assuming luminosity evolution of galaxies ( e.g. , koo & kron 1992 ) , but the redshift surveys have indicated that the shape of the redshift distribution appears almost as is predicted with a no - evolution model , with the normalization , however , being larger by a factor of two . simple models with all galaxies undergoing substantial luminosity evolution , which account for the steep slope of the number count , predict too many galaxies having high redshift to be consistent with the observations ( e.g. , ellis 1990 ; lilly 1993 ) ( though this seems to have been somewhat overstated ) . there have been a number of attempts to solve this problem : some authors speculated that the excess count is due to a new population of galaxies which have undergone star burst at rather low redshift ( bes ; cowie , songaila and hu 1991 ) . babul & rees ( 1992 ) and babul & ferguson ( 1996 ) interpreted this new population as dwarf galaxies from collapsed lyman @xmath12 clouds , eventually faded into galaxies making up an invisible population . some other authors have assumed a heuristic morphology - dependent luminosity evolution so that subluminous galaxies evolve fast to explain high counts while keeping the shape of redshift distribution as is in the no - evolution model , rather than introducing a new population ( bes ; lilly 1993 ; phillips & driver 1995 ) . another class of explanation invokes mergers of galaxies with a very high merging rate ( guiderdoni & rocca - volmerange 1991 ; broadhurst , ellis & glazebrook 1992 ; carlberg 1992 ) . the required rate to explain the steep slope of the @xmath4 band count is so high that a typical galaxy gains 50% of mass over the past 5 gyr . more recently , gronwall & koo ( 1995 ) suggested that the presence of an abundant non - evolving very blue dwarf population accounts for both counts and redshift data , although the meaning of this blue population is not clear to us . observational data have now accumulated for both counts and redshifts over a wide range of colours from @xmath4 to near infrared @xmath6 band , and the deepest redshift reaches beyond @xmath13 ( glazebrook et al . 1995a ; songaila et al . 1994 ; lilly et al . a large number of redshifts contained in the survey have enabled a direct construction of the @xmath4 band luminosity function as a function of redshift ( lilly et al . 1995 ; ellis et al . 1996 ; see also eales 1993 ) . this prompts us to think that we already have enough data to understand the evolution of galaxies below @xmath14 . in addition , more varieties of clues have become available to understand the nature of faint galaxies . sharp images of galaxies obtained with the hst have enabled a study of morphology to @xmath5=24 mag , giving us a direct probe concerning how the faint galaxies look like ( driver , windhorst & griffiths 1995 ; glazebrook et al . 1995b ; abraham et al . 1996 ; driver et al . 1996 ) . the work has shown that the fraction of irregular galaxies increases sharply with magnitude beyond @xmath15 mag . the same observations indicate that giant galaxies evolve only slowly , although they do indicate some evolution in colour , brightness and the detail of morphology . another piece of evidence that giant galaxies evolve only slowly , if at all , comes from a mg ii quasar - absorption - line selected sample ( steidel , dickinson & persson 1994 ) . the luminosity of galaxies that yield mg ii absorption lines for a quasar in their vicinity is remarkably constant both in the rest - frame @xmath4 and @xmath6 bands from @xmath16 to @xmath171 . there are also a number of observations that tell us about the content of galaxies . the information that would directly constrain the model for evolution of galaxies is the estimate of the global star formation rate at @xmath18 ( gallego et al . 1995 ) and high @xmath19 ( cowie , hu & songaila 1995 ; see also madau et al . 1996 for a more recent work ) from the strength of h@xmath12 or [ o ii]@xmath20 emission lines . this quantifies the inference from early observations concerning the increase of star formation activity in the past ( bes ) . yet another useful information is given by the measurement of the neutral hydrogen abundance in the universe as a function of redshift ( lanzetta , wolfe & turnshek 1995 ; storrie - lombardi , mcmahon & irwin 1996 ) . the data show clear depletion of hi gas with time , suggesting that this gas has been used for star formation . this information can be used to constrain the total volume emissivity of galaxies ( see , pei & fall 1995 ) , on which direct information is also available from galaxy observations ( lilly et al . 1996 ) . from accumulating observations we now have a reasonable picture for the evolution of galaxies ( fukugita , hogan & peebles 1996 ) : giant galaxies were already mature at @xmath21 and evolve only slowly lower than this redshift , whereas subluminous galaxies evolve rapidly in the same redshift range . the latter are probably young galaxies formed close to @xmath21 . it has also been speculated that most of elliptical galaxies and bulges of the spiral galaxies , which altogether we call spheroids , formed before @xmath22 , and spiral disks were assembled in the redshift range from @xmath23 to 1 . in this paper we present a simple but quantitative model for the history of galaxies , which is broadly consistent with the observations . the model employs a population synthesis model of stars . in the traditional attempts , galaxies are assumed to have fully assembled at an early epoch of the universe ( searle , sargent & bagnuolo 1973 ; sandage 1986 ; yoshii & takahara 1988 ) , and the star formation rates are adjusted to reproduce galaxy colours today . in our attempt it is crucial to properly build in the epoch of the formation of galaxies in the stellar population synthesis model . for this purpose we first show that one can estimate the age of the galaxies as seen today from a simultaneous consideration of the colour and gas content of galaxies with the aid of moderate assumptions on the star formation history , and infer the age of irregular galaxies and also of disks of spiral galaxies . for elliptical galaxies , and also bulges of spiral galaxies from family resemblance ( both are called spheroids ) , we adopt the conventional view that they formed at high @xmath19 as argued by eggen , lynden - bell and sandage ( 1962 ) . we assume that the number of galaxies is conserved since they were born , and it is constrained by the local type - dependent luminosity function . we calculate the evolution of bulges and disks separately and assume delayed formation of disks and irregular galaxies . the spheroids are supposed to follow the passive evolution to now . galaxies are grouped into four morphology types : e / s0 , sab , sc , and sdm - irr which is simply referred to as irr in this paper ; low luminosity dwarf galaxies do not play an essential role in our argument . the mix of morphological types of galaxies is not constant , but depends on absolute luminosity . the predictions of this model are compared with the observation of the galaxy number count and the redshift distribution in @xmath4 , @xmath5 and @xmath6 bands . we do not discuss the @xmath24 band count , since we can not model reliable @xmath6- and @xmath25-corrections for this band , and @xmath24 band data are too sensitive to the detail of galaxy activity . we discuss the evolution of gas and star content of galaxies predicted in the model . through these comparisons with the observations we may conclude that the present model is reasonably well constrained for @xmath26 . the prediction for @xmath27 , however , is basically an extrapolation , and ill - constrained . a number of checkpoints are also discussed to prove or disprove the validity of the model and , more importantly , underlying assumptions taken in this paper . we do not consider the no - evolution model seriously , since it is clearly unphysical . we know that elliptical galaxies _ must _ evolve even in the absence of new star formation after the burst , since the stars in the main sequence branch continuously evolve to giants , and the total luminosity should decrease with time ( tinsley 1972 ) . this basic feature is a solid prediction of the stellar population synthesis model , although quantitative details depend on models , in particular on assumed initial mass functions . for spiral galaxies , there are more uncertainties : our model corresponds to the case where luminosity changes very little with time : the decrease of light due to evolution of stars is compensated by newly born stars formed from the gas replenished into the disk . we note that the present model is quite different from `` cdm cosmologies '' , where morphologies of galaxies change from time to time due to mergers down to low redshift ( kauffmann , guiderdoni & white 1994 ; cole et al . that is , elliptical galaxies are not quite old , but formed by collision of , for instance , two spiral galaxies at rather low redshift ( toomre 1977 ) . one of the original reasons for interest in faint galaxy counts is that it may offer a chance to test world geometry ( sandage 1961 ) . a number of attempts have been made aiming at this goal ( e.g. , yoshii & takahara 1988 ; fukugita et al . 1990 ; gardner , cowie & wainscoat 1993 ; gronwall & koo 1995 ) . the reliability of the results , however , is strongly affected by the uncertainties of the model . in this paper we discuss where the predictions are reasonably robust . in particular we discuss the possibility to test the presence or absence of the cosmological constant . in the next section we discuss how to estimate the ages of galaxies from their colours and gas content . the details of the model and input assumptions are discussed in section 3 . in section 4 we present the predictions of the model and compare them with the observations in @xmath6 , @xmath4 and @xmath5 bands , and with the constraint on evolution from a mg ii quasar absorption line selected sample . the evolution of baryons in stars and in gas is also discussed . section 5 is devoted to discussion and summary of the problems . the two fundamental parameters that control the evolution of the stellar population are the age of the galaxy @xmath28 and the star formation rate @xmath29 ( gyr@xmath30 ) , provided that the initial mass function is fixed . we can obtain useful information concerning these parameters by plotting galaxies in the two dimensional plane , integrated galaxy colour versus gas . in fig . 1 we show a diagram for irregular galaxies plotted in @xmath31 versus @xmath32 , where @xmath33 is mass of the neutral atomic gas estimated from the integrated 21 cm hi flux , corrected for helium abundance of 25% in mass , with the aid of the formula given in the _ third reference catalogue of bright galaxies _ ( de vaucouleurs et al . 1991 , hereafter rc3 ) . molecular hydrogen is ignored . the data are taken from buta et al . ( 1994 ) , but a very similar figure can be obtained using the data given in rc3 . the bars indicate one standard deviation of the distribution . evolutionary tracks are presented in the figure for a given @xmath29 using a conventional evolution model of galaxies , where star formation is assumed to take place at a constant rate against gas mass . for the present calculation , we adopt the closed model of arimoto & yoshii ( 1986 ) , but other models also yield similar results . the track represented by a solid line corresponds to @xmath34 gyr@xmath30 , which is the star formation rate in the solar neighbourhood , and is also consistent with the estimate for sb - sbc galaxies ( roberts 1963 ; smith , biermann & mezger 1978 ; kennicutt 1983 ; sandage 1986 ) . the ticks on the track indicate the age of galaxies , 1 , 5 , 10 and 15 gyr after star formation began , so that dashed curves represent isochronal contours . we note that the traditional model , where irregular galaxies formed as old as the universe , assumes a very low @xmath29 ( @xmath35 gyr@xmath30 , say ) in order to obtain their blue colour ( @xmath36 ) . this terminating point of the @xmath37 track is indicated by an open circle . the family of evolutionary tracks indicates that one can obtain the same colour with a larger @xmath29 if the age of galaxies is younger ; the gas fraction determines the age . the data point for irregular galaxies shows that they are as young as @xmath38310 gyr and @xmath39 gyr@xmath30 , which is close to the solar neighbourhood value . the prediction of the traditional model , as presented by the open circle , is clearly inconsistent with the observation . we emphasize the large `` error bars '' , which indicate that the irregular galaxies observed today have not had a coeval formation at all , with an age spread of well over 4 to 12 gyr . one attractive feature with a larger @xmath29 , which is consistent with the value for spiral galaxies , is that it supports the validity of the schmidt law , which states that the star formation rate is proportional to the hi gas amount to some power ( schmidt 1959 ) . the approximate validity of the schmidt law with power law index close to unity is supported by an analysis of emission line features ( kennicutt 1989 ) . the traditional value @xmath37 gyr@xmath30 requires an unusually strong suppression of star formation in irregular galaxies , the surface brightness of which is not necessarily very low . we do not consider intermittent star bursts , which are often invoked in the literature to account for faint blue galaxies . we consider that the star formation activity is steady and continuous at least for the majority of galaxies . the galaxies are highly gaseous shortly after their birth , and the star formation rate _ per galaxy _ can be 4 times than it is today . those galaxies are seen as if they are undergoing star burst activity . we conclude that irregular galaxies observed today formed at around @xmath21 , although some irregulars are formed as low as @xmath40 or less , and some others as high as @xmath41 . we do not mean , however , that the irregular galaxies do not form at high @xmath19 . most of the galaxies formed at higher @xmath19 , might already have been faded or accreted onto giant galaxies and they do not show up in our local sample of irregulars . a similar analysis can be applied to disk components of spiral galaxies . figure 2 shows @xmath31 versus @xmath42 plot for spiral disks , where both @xmath31 and @xmath43 are corrected for the light from bulge component by assuming average disk / bulge ratios that depend on morphology ( see section 3.1 below ) . the data are taken from rc3 . a plot is also made for irr and sd for comparison . the tracks are calculated with a closed model . this figure shows that the disks of late - type spiral galaxies are also young with ages around @xmath175 - 12 gyr allowing for a large scatter , if a slight upward shift of the data due to the neglect of h@xmath44 gas is taken into account . the plot for sa indicates that some of these might be as old as the spheroids . the age of the disk of sb - sc galaxies inferred this way is consistent with the age of the oldest disk stars ( 7.511 gyr ) estimated from nearby white dwarfs ( winget et al . 1987 ; wood 1992 ) . for bulges of spiral galaxies it is reasonable to suppose that they formed very early , @xmath45 , from family resemblance with elliptical galaxies , if they indeed formed so . the resulting picture is that spiral disks formed as intergalactic gas accreted onto preexisting bulges and star formation began in the disk at @xmath46 . we assume that spheroids ( ellipticals and bulges of spiral galaxies ) formed at high @xmath19 ( @xmath8 ) by a single short burst event . the stars have to be formed from the initially collapsing gas faster than the collapse time to avoid disk formation ( eggen et al . the duration of the burst is taken to be 0.125 gyr , i.e. , the star formation rate @xmath47 ( hereinafter the star formation rate is denoted as @xmath47 ) being effectively 40 times the solar neighbourhood value ( e.g. , sandage 1986 ) . these galaxies evolve passively after the burst . the problem inherent in the presently available stellar population synthesis models is that they do not reproduce the uv spectrum of average elliptical galaxies today . it shows a significant uv component and frequently an upturn shortward of 2000 ( bertola , capaccioli & oke 1982 ; burstein et al . 1988 ) . in the models of passive evolution the uv component rapidly dies off a few gyr after the burst . the uv component of elliptical galaxies , however , plays very little role in the quantities that we discuss in the present paper . no effect is expected in the @xmath6 band until one samples galaxies above @xmath48 . in this pass band the effect is visible only in the high redshift tail of the redshift distribution at @xmath49 mag . for the @xmath5 band the effect appears only above @xmath50 , and a minor change is expected only for @xmath51 . for the @xmath4 band we expect some effect for redshifts is as low as 0.5 , but the spheroids play only a minor role in this pass band , and the presence or absence of the uv component does not change any predictions , unless elliptical galaxies are specifically selected from the sample . for this reason we can ignore the problem of the disagreement of the passive evolution model with the uv observations . distribution in the optical pass bands . a typical example is the `` lyman continuum break galaxies '' recently discovered by steidel and collaborators ( 1996 ) in a sample selected in the @xmath52 band . such observations just sample those galaxies that are dominantly controlled by the uv components of spheroids . ] we represent spiral galaxies with two morphological types , sab and sc . we assume that they consist of old spheroids and disks which formed typically at @xmath53 . we adopt for the disk components a model where gas continuously accretes onto spheroids , making spiral disks with @xmath54-folding time @xmath55 gyr ( @xmath56 is the inverse of the infall rate ) , which corresponds to @xmath53 . the star formation rate @xmath47 is set equal to the value in the solar neighbourhood , @xmath57 gyr@xmath30 , irrespective of galaxy types . we use the stellar population synthesis model of arimoto , yoshii & takahara ( 1992 ) to accommodate gas infall . this infall model gives evolutionary tracks significantly different from the ones based on a closed model shown in fig . 2 for a high @xmath47 case , but the self - consistency between the input and output is maintained for ( age , star formation rate ) versus ( colour , age ) . the bulge fraction in the blue light is taken to be 0.3 for sab and 0.1 for sc in agreement with the analysis of kent ( 1985 ) after the colour transformation . we remark that we adopt here a model of disk formation exponential in time with @xmath54-folding time @xmath58 for simplicity . the formation epoch of irregular galaxies is taken to be 8 gyr back from the present for simplicity of the model , although the actual formation epoch spreads over a wide range of age . we assume a universal star formation rate , the same value as that for disks . we take the age of the universe to be 15 gyr , but only minor modifications are caused if we take it to be 13.5 gyr ( e.g. , e galaxies are slightly bluer , and evolution is a little stronger etc . ) . the hubble constant enters into our scenario only through the evolution time - scale of stars . we assume the hubble constant to have the value specified by the assumed age of the universe and the specific cosmology we consider . .assumed history of star formation in galaxy . [ cols="^,^,^,^",options="header " , ] [ tab2 ] the parameters of our model are summarized in table 1 . some basic predictions for galaxy properties , bulge fraction ( input ) , @xmath59 , gas fraction , @xmath31 , @xmath60 , @xmath61 and @xmath62 colours at @xmath18 , are presented in table 2 ( discussion is deferred to later sections ) . we use the standard johnson - morgan - cousins colour band system , unless otherwise explicitly denoted . the evolution of the light emitted by e , sc and irr galaxies is shown in fig . 3 for the rest frame @xmath4 and @xmath6 bands . spheroids follow the standard passive evolution for both pass bands . the brightening of 0.3 mag in @xmath4 between @xmath16 and @xmath63 is consistent with the observation for red galaxies in the lilly et al . ( 1995 ) sample , in particular for @xmath64 cosmology . sc galaxies hardly evolve in the @xmath6 band . in the @xmath4 band , weak evolution is visible only between @xmath18 and 0.5 , only in the epoch after gas infall has ceased . the evolution between @xmath18 and 0.5 is about a half mag . ( this is compared to lilly et al . 1995 , who reported 1 mag evolution of @xmath65 from @xmath18 to @xmath66 ; they have not seen a change of @xmath67 above @xmath68 , which is consistent with our model ) . below @xmath69 , the galaxies that have undergone the fastest evolution are irr , but the total increment in the @xmath4 band luminosity is only 0.8 mag between @xmath18 and 0.8 . of course , the evolution is much faster in the uv light , and the @xmath25 correction for the @xmath4 count amounts to @xmath172 mag in the same redshift range . this is almost the maximum evolution that can be attained with a conventional stellar population synthesis model . lilly ( 1993 ) and phillips & driver ( 1995 ) assumed significantly faster evolution ( @xmath70 mag between @xmath18 and 0.4 , and 2.5 mag between @xmath18 and 0.5 , respectively ) . another input to the model that deserves discussion is the local luminosity function ( lf ) . we need the morphology - type dependent luminosity function . the only case where we know such lf reasonably well is that for the virgo cluster ( bingelli , sandage & tammann 1988 ) . the lfs , when galaxies are classified into morphological types , show cutoffs in both bright and faint ends . on the other hand , we know reasonably well that the lf summed over all morphological types behaves like the schechter function ( loveday et al . 1992 ; lin et al . 1996 , ellis et al . 1996 ) , although its faint end is uncertain . we adopt as a fiducial lf the one reconstructed by lilly ( 1993 ) , which takes account of all these features ( reproduced in fig . 4 ) . in this figure sdm and dirr are plotted separately , but our results do not depend on whether we keep the dirr component or drop it . hereinafter , we simply use terminology of irr for the sum of sdm and dirr . the total lf is represented by a schechter function with @xmath71 , @xmath72 mag , and @xmath73 ( mpc)@xmath74 , where an upward shift by 0.1 dex is made for the normalization . this corresponds to local luminosity density @xmath75(mpc)@xmath74 . we remark that the normalization of the lf has a substantial uncertainty . our adopted normalization is consistent with those of efstathiou , ellis & peterson ( 1988 ) and of ellis et al . ( 1996 ) with the luminosity density 1.93 and 2.05@xmath76(mpc)@xmath74 , respectively , but is higher than that of loveday et al . ( 1992 ) , @xmath77(mpc)@xmath74 . we see a 50% uncertainty in @xmath78 among the widely accepted lfs .. the determination of @xmath79 is correlated with @xmath67 , and the resulting uncertainty in @xmath78 is smaller than that in @xmath79 ] as we discuss later the interpretation of the number count is affected by this uncertainty . we remark that the slope of the faint end is not very important to the overall shape of the number counts , unless the faint tail of the lf rises so sharply that it leads to a substantial enhancement of the integral , i.e. , @xmath81 . the contribution to the number count ( see eq . [ 1 ] below ) is luminosity weighted , as @xmath82 , so that the behaviour of the number count is basically determined by galaxies with the characteristic luminosity . on the other hand , the lf of irregular galaxies may well be higher by a factor of two or more , which increases the contribution from irregular galaxies by this amount . in this paper we adopt the above lf as a fiducial choice and then discuss whether the modifications necessary to explain the observation , if any , are within the uncertainties of the current lf , rather than modifying the adopted lf . the differential number count of galaxies is given by @xmath83 where @xmath84 is solid angle , @xmath85 is the comoving volume , @xmath86 is the lf , and the relation between @xmath87 and @xmath88 in the @xmath89-pass band is given by @xmath90 with @xmath91 the luminosity distance . the @xmath6 correction for the optical band is taken from fukugita , shimasaku & ichikawa ( 1995 ) . we adopt the @xmath6 correction of cowie et al . ( 1994 ) for the @xmath6 band . there is little difference among authors for @xmath6 corrections for any colour bands other than far uv . we also refer to fukugita et al . ( 1995 ) for the transformation of magnitudes between the different colour band systems , which is extensively used in this paper . the @xmath25 correction is calculated according to the prescription given in section 3.1 above.3000 did not change in shape but scaled only in intensity with the point at @xmath92 to avoid too large an @xmath25 correction , which is predicted by the pure passive evolution model ( yoshii & takahara 1988 ) . as discussed previously , however , we only consider the predictions which are insensitive to the uv sed of the spheroids . ] the prediction is compared with the observation by integrating ( 1 ) over @xmath93 and/or @xmath94 for appropriate ranges . we consider a number of cosmological models when we discuss the @xmath6 band count , which shows the largest power in discriminating among cosmological models . for other colour bands we are confined mainly to two cases : ( i ) open universe with @xmath95 , and ( ii ) @xmath89-dominated universe with @xmath95 and @xmath96 , as an extreme case , although such high @xmath89 seems to be excluded from the statistical lensing . @xmath7 models fail to reproduce the data for the @xmath6 band count , and hence are omitted from most of our considerations . _ number count _ figure 5 shows galaxy number counts in the @xmath6 band . the data are taken from djorgovski et al . ( 1995 ) , gardner et al . ( 1993 ) and soifer et al . ( 1994 ) for faint counts , and glazebrook et al . ( 1994 ) and mobasher , ellis & sharples ( 1986 ) for brighter counts . the curves are shown for four cosmology models : @xmath7 ; @xmath95 open model ; ( @xmath95 , @xmath96 ) ; and ( @xmath97 , @xmath98 ) . we can see gross agreement between the models and the observation for open and @xmath99 models . the predicted count for the @xmath7 model falls short of the observed count beyond 20 mag , by a factor of 4 at 23 mag , showing that @xmath7 model is disfavoured . this conclusion would not be modified unless there are unexpectedly many red dwarfs in the local lf ( so that @xmath100 ) at very faint magnitudes , or the basic assumptions of the present model ( conserved number of galaxies ; early spheroid formation etc . ) are abandoned ; see the discussion section below . looking at details we note a few features : a good agreement at bright magnitudes indicates that the normalization of the @xmath6 band count is given correctly with the @xmath4 band lf and the colour transformation without further adjustment . the observed counts are slightly higher at @xmath101 ; the data look as if there is a mild change in the slope at this magnitude , while such change is not accounted for by the usual @xmath6 or @xmath25 correction . another feature is that the prediction of the @xmath102 model overshoots the data of djorgovski et al . for the range 20 - 23 mag by a factor of @xmath103 . on the other hand , this @xmath89-dominated model just fits the data of soifer et al . the survey areas for these deep counts are only 1 - 2 square arcminutes for the above two surveys , and 16 square arcminutes in the gardner et al . survey . it is likely that the data suffer from the effect of large - scale inhomogeneity beyond @xmath104 error estimates . for this reason we can not yet rule out the @xmath89-dominated model from the analysis of these counts . a deep survey of larger area is highly awaited in this regard . we find that some small wiggle is induced to the @xmath105 curve if the epoch of burst - like spheroidal formation is as low as @xmath106 . the effect is more clearly visible in the redshift distribution . nevertheless , the gross feature of @xmath6 band count is fairly insensitive to the model details ; it is more sensitive to cosmology . it is generally expected that the @xmath6 band light is sensitive to the old population . we show in fig . 6 the fraction of light in the @xmath6 band arising from spheroidal components for an @xmath95 universe . ( a similar plot is also shown for the blue light for comparison . ) this figure shows that @xmath107 of light comes from spheroidal components in the entire magnitude range that concerns us , confirming the expectation that @xmath6 band counts are a good probe for old population . since we understand the evolution of spheroids reasonably well , there is no much difference among predictions of different authors . this makes @xmath6 counts suitable to study cosmology , especially for a test for the cosmological constant . the uncertainty present in the data , however , does not allow us to make a conclusion . for the blue light the fraction of spheroid contribution is about 10% for @xmath108 mag . _ redshift distribution _ three panels in fig . 7 show the redshift distribution for the magnitude range ( a ) @xmath101 , ( b ) @xmath109 and ( c ) @xmath110 . the prediction is compared in ( a ) with the @xmath19 survey data of songaila et al . ( 1994 ) after normalizing them to the count data with the sample incompleteness taken into account . the figure shows good agreement for the global shape , although the data are higher than the prediction by 20% ( note that this is the magnitude where the disagreement between prediction and data becomes maximum , as we noted above ) and they lack a high @xmath19 tail above @xmath111 . at this magnitude 15% of galaxies are not given redshift data , and it is conceivable that they are missed by redshifting the characteristic [ o ii ] line out of the spectroscopy window , as it is more clearly seen in their fainter magnitude sample , where basically no galaxies above @xmath13 are catalogued . we remark that shoulders or wiggles seen in summed redshift distribution are artefact of representing galaxies with only four distinct types . since morphology is a continuous class , these curves must be sufficiently smeared when we compare them with the observation . the calculation indicates that spiral galaxies dominate at lower redshift , whereas galaxies at higher redshift are predominantly of early - type . for @xmath101 mag , the distribution consists of spiral galaxies with @xmath112 and early type galaxies with @xmath113 . although the morphologies of the @xmath6 selected sample are not available , one can use emission features to show consistency with this prediction . we show in fig . 8 equivalent widths of [ oii]@xmath114 emission lines for galaxies in the same magnitude range . the figure shows a trend of decreasing equivalent widths with increasing @xmath19 . though the evidence is not compelling , these data are consistent with increasing fraction of early type galaxies with increasing @xmath19 , as the model calculation predicts . this may appear to be in contrast to the claim that emission lines become stronger as @xmath19 increases . of course , these two statements are not contradictory : in our case we confine ourselves to a fixed magnitude range . a general trend of strengthening emission features to higher @xmath19 is understood as star formation being more active in spiral disks , as in our model . the maximum redshift reaches @xmath115 even at @xmath101 . this increases to @xmath116 at @xmath109 and to @xmath8 at @xmath110 . these maximum redshifts are so high that they allow exploration of the formation and early evolution history of spheroids , if these high redshift tails can be sampled . at the deepest magnitude , @xmath110 , @xmath117 of elliptical galaxies is 3.5 , although 60% of galaxies are spiral galaxies at around @xmath118 . the high redshift tail of @xmath119 picks up uv emission from elliptical galaxies , and is particularly sensitive to the formation history . conversely , the uv from spheroids does not contribute until this very high @xmath19 is reached at the faintest magnitudes ; the prediction depends solely on optical emission , the evolution of which is reasonably well understood within a passive evolution model . another noticeable feature is a small fraction of irregular galaxies , which increases only a little with magnitude . the dominance of giant galaxies means that the effect of evolution is very small as a whole , as often said is consistent with no evolution , at least for @xmath120 mag ( e.g. , songaila et al . 1994 ) . if the burst - like formation epoch of elliptical galaxies is lower than @xmath106 , we expect an appreciable fraction ( 15% ) of galaxies having this redshift even in a @xmath101 mag sample ( see fig . this fraction increases to 30% at @xmath109 mag . though this looks somewhat unusual , we are not able to exclude this possibility immediately because of the sample incompleteness of the redshift survey of songaila et al . ( 1994 ) , which amounts to 15% at @xmath101 mag . _ number counts _ the astrophysics of the @xmath4 band count is very different from that in the @xmath6 band , as expected from figure 6 shown above . the light , especially that from faint objects , is dominated by young components , disks and irregular galaxies . we present in fig . 10 the prediction of the @xmath121 band count for the two cosmological models : ( a ) @xmath95 , and ( b ) ( @xmath95 , @xmath96 ) . the data are taken from metcalfe et al . ( 1995 ) , tyson ( 1988 ) , maddox et al . ( 1990 ) , jones et al . ( 1991 ) and lilly et al . we do not calculate the counts beyond @xmath122 mag ( predicted median redshift is 0.8 ) because of the lack of reliable @xmath6 and @xmath25 corrections for redshift above @xmath123 . the thick line represents the total count , and the thin solid line and the dotted lines are e / s0 and spiral galaxies , respectively . the agreement between the calculation and the data is reasonable , although the former is short of the latter by 0.3 dex at the faintest magnitude in ( a ) . the contribution from spiral galaxies remains significant even at faint magnitude , but the significance of e / s0 galaxies diminishes for fainter magnitudes , due to a large @xmath6 correction . the most conspicuous in the figure is a rapid increase of irr ( dashed line ) as we go to fainter magnitudes : the contribution to the counts relative to spiral is 10% at 17 mag , but they contribute equally at 24 mag . this is explained by a combination of three effects : ( i ) irr s are located at relatively near distances , and so the reduction of spatial volume in a relativistic cosmology is only modest ; ( ii ) the blue - weighted nature of spectrum gives only small @xmath6-corrections ; and most importantly , ( iii ) these galaxies formed at low redshift and a large evolutionary effect is present . these effects together make the slope of irr counts quite steep . we can conclude that these irregular galaxies are the agent that yields a steep slope of the @xmath4 count . we note that the component that gives an important contribution at 22 - 25 mag is not `` dwarf '' galaxies , but is those which are on the immediate extension of late - type spiral galaxies to the fainter side . these galaxies have sub @xmath67 luminosity today , but were nearly as bright as giant spiral galaxies for a few gyr after their births . this explains the `` faint blue galaxy problem '' . in the above figure , we see a better fit with a high @xmath89 model , but we do not claim that this model is favoured . the discrepancy between the prediction and data seen in @xmath95 model almost disappears , if the normalization of the lf for irr galaxies is increased by a factor of @xmath1242 - 3 , which is within the observational uncertainties , but also is indicated by a few analyses of local lf including blue galaxies ( metcalfe et al . 1991 ; marzke et al . 1994 ; lilly et al . 1995 ; subbarao et al . irregulars and these two observations are mutually consistent . ] alternatively , the possibility also exists that the lf of irregulars is actually lifted as @xmath19 increases ( ellis et al . 1996 ) , and the normalization of the local lf for irregulars used in this paper is correct . this requires that the number of irregular galaxies decreases with decreasing redshift , for instance by these galaxies being accreted onto nearby giant galaxies . these `` minor '' mergers modify the property of giant galaxies very little , and this possibility seems perfectly viable , as a minimum modification of the scenario presented in this paper . on the other hand , it is unlikely that these population has simply faded away . bes have discovered that [ oii ] emission increases rapidly with increasing magnitude . the fraction of galaxies that show equivalent widths larger than 20 increases from 15% at @xmath125 mag to as much as 55% at @xmath126 mag . koo & kron ( 1992 ) ascribed this increase to the increase of the fraction of subluminous late type galaxies , which generally are strong [ oii ] emitters , as one goes to fainter magnitude . we agree that this is one cause , but not the major one . in our model , this increase is caused by both an increase of irr fractions as a result of a steep slope of @xmath105 for irr , _ and _ an increase of gas in spiral galaxies . straightforward reading of the [ oii ] equivalent width distribution of kennicutt ( 1992 ) ( see his figure 9 ) gives 13% of normal galaxies having @xmath127 [ oii ] equivalent widths , consistent with the bes @xmath128 mag sample . the disks at @xmath129 are two times more gaseous than at @xmath18 . a simple calculation using the [ o ii ] distribution in the kennicutt figure leads to the result that the fraction of galaxies having ew([oii])@xmath13020 is 45% at @xmath126 . this is consistent with the result of bes . in our explanation of the b band count , we do not particularly invoke incidental star burst activity . star formation takes place steadily as it does today , but those galaxies at @xmath131 are much more gaseous : the gas in irr 1 gyr after their birth is 3 times that of today , and so the global star formation rate per galaxy is 3 times higher , which might give the appearance that those galaxies are undergoing star bursting activity . _ redshift distribution _ the redshift distribution of the sample selected in the @xmath4 band corroborates the view given above . in fig . 11 , we compare the prediction with the observations of glazebrook et al . ( 1995a ; solid histogram ) and of cowie et al . ( 1996 ; dotted histogram ) for @xmath132 mag . we can see a good agreement between the two for the dominant feature around the peak . the redshift data are normalized to the observed count . an interesting fact is that the observed distribution looks quite similar to that of spiral ( plus e / s0 ) galaxies , while the normalization of the former is higher by a factor of two . this is the feature that has puzzled many cosmologists after the discovery by british groups ( bes ; colless et al . 1990 ) . in our model this factor two difference is explained by irregular galaxies which were almost as luminous as giant spiral galaxies in the blue band . the survey selected in the blue light preferentially samples actively star forming components , which were newly formed in the near past . at this magnitude the contribution from e / s0 is quite small . the presence of high @xmath19 tail has been a matter of debate . earlier observations ( bes ; colless et al . 1990 ; lilly , cowie & gardner 1991 ; glazebrook et al . 1995a ) have claimed the absence of the high @xmath19 tail . on the contrary , some recent observations ( cowie et al . 1996 ; koo 1996 ) indicate the presence of the tail . our model predicts a small high redshift tail which arises from luminous spiral galaxies ; we can not eliminate entirely the high @xmath19 tail . the size of tail depends largely on the amount of evolution of spiral galaxies . a large high @xmath19 tail of cowie et al . , if confirmed , suggests that the evolution should be stronger than is in the present model . _ number count _ the novel feature for the @xmath5 band is that morphologically classified number counts are available ( glazebrook et al . 1995b ; driver et al . 1996 ; abraham et al . 1996 ) to @xmath133 mag . at this magnitude we expect the median redshift to be unity . we compare the prediction of the @xmath95 model with the observation in fig . 12 for e / s0 ( panel b ) , spiral ( panel c ) , irr ( panel d ) and the total sample ( panel a ) . we see a good agreement for spiral galaxies to @xmath134 mag . for e / s0 galaxies the predicted curve is somewhat lower by @xmath135 dex to the same magnitude , but the statistics are poor for this type . the disagreement is significant for irr . the important feature here , however , is that the steep slope of the irr counts ( index=@xmath136 for @xmath137 mag ) is correctly reproduced by a sub @xmath67 population of young , quickly evolving galaxies . that the predicted counts of irregulars are lower by a factor of three is similar to the problem we encountered with the blue count , and can be accounted for if the lf of irr is increased by a factor of 3 . glazebrook et al . ( 1995b ) have noted that the normalization of their @xmath5 count is higher than predicted from the local lf in the @xmath4 band . in our calculation this discrepancy is not so conspicuous : a half of the discrepancy is accounted for by our higher normalization of the local lf , which is larger than that of loveday et al . ( 1992 ) by 0.15 dex . we still see a discrepancy for the amount of 0.12 dex at @xmath134 . this is reduced to 0.05 dex if the normalization of the irr - lf is multiplied by a factor of 3 . at fainter magnitudes the observed counts are somewhat higher than calculated . this is particularly true for e / s0 galaxies . we note that the shape of the prediction of e / s0 galaxies is rather generic and it is not easy to modify the prediction . on the other hand , somewhat higher observed counts of spiral galaxies may be explained by increasing evolution of spiral galaxies . it may also be possible that there is a misclassification between spirals and irr galaxies , since the classification becomes increasingly ambiguous towards the faintest magnitude : the number count of irr may not flatten but increase linearly towards the fainter magnitude , while the number of spiral galaxies may rise close to the predicted curve . it is also possible that the counts from @xmath138 mag survey are largely affected by large - scale inhomogeneity of the galaxy distribution , since the survey fields are quite small . driver et al . ( 1996 ) concluded that high @xmath89 models ( @xmath139 in a flat cosmology ) are excluded from a comparison of their e / s0 counts with their model prediction , as the predicted count with high @xmath89 overshoots the observation . our model calculation , however , does not confirm their conclusion . we present in fig . 13 the predicted counts for @xmath140 ) model , where the discrepancy between the prediction and the data almost disappears to the faintest magnitude . in particular , the prediction of e / s0 is consistent with data of driver et al . we note that the main difference between our model and driver et al.s lies in the normalization of e / s0 galaxy counts , i.e. , a much higher normalization adopted by the latter work . _ redshift distribution _ lilly et al . ( 1995 ) have carried out a large redshift survey to @xmath5=22 . their data are compared with the prediction in fig . the predicted redshift distribution agrees with the data very well , with the normalization , however , smaller by 20% for the open cosmology . the case of @xmath89 dominated universe @xmath102 is displayed in fig . 14 ( b ) . unlike the case for the @xmath4 band , the distribution is dominated by spiral galaxies , and the contribution from e / s0 and irr is quite small . the small contribution from irr is understood by the fact that the very active star formation for @xmath141 largely affect the @xmath4 band observations , whereas @xmath5 band observations are more sensitive to the total number of stars in the galaxy , and stars are not yet accumulated before @xmath142 . we note that the evolution of spiral galaxies is less than 0.5 mag to @xmath13 , and the evolutionary effect is not manifest in the @xmath19 distribution . the prediction for the redshift distribution is shown in fig . 15 for @xmath143 mag . we note that e / s0 counts are always lower than the contribution from spiral galaxies for all @xmath19 . this is different from what we saw for the @xmath6 counts , where the high redshift part was dominated by early type galaxies . the colour distribution is shown in fig . we take the @xmath144 data from a compilation of koo & kron ( 1992 ) for the magnitude between @xmath145 and 27 . the data for the brightest magnitudes ( @xmath146 ) are @xmath31 from gronwall & koo ( 1995 ) . the prediction , denoted with dotted lines , agrees well with the data to @xmath9 mag . we see that the data are slightly bluer for fainter magnitudes . better agreement is achieved by increasing the irregular population by a factor of 2 , although the prediction for @xmath147 mag is still somewhat ( 0.2 mag ) redder . we do not calculate the colour distribution for @xmath148 for the reason given in section 4.2 above . * 4.5 the mg ii absorption - line selected sample and the constraint on the luminosity evolution * steidel et al . ( 1994 ) have studied properties of 58 galaxies that yield mg ii absorption lines near the line of sight to quasars . giant galaxies are always found close to the quasar sightline along which mg ii absorption lines are detected , and conversely all giant galaxies located close to the quasar sightline cause mg ii absorption in the quasar spectrum . their conclusion is that those galaxies that cause mg ii absorption lines are giant galaxies and these galaxies hardly evolve at least between @xmath16 and 1.0 : the allowed amount of evolution is 0.1@xmath149 mag in the rest frame @xmath4 band for this redshift interval . this puts a very strong constraint on the model . in our model , e / s0 undergo evolution by a half magnitude and sc galaxies by 0.2 mag ( see fig . while our evolution of e / s0 galaxies looks faster , the fraction of galaxies that show e / s0 colours is small in the steidel et al . sample , and their data do not give a strong constraint on the evolution of e / s0 galaxies . irregular galaxies also evolve faster , but such galaxies do not yield mg ii absorption lines and are excluded from the sample ; hence no constraint is obtained on the evolution of irregulars . therefore , the prediction of our model is consistent with the result from the mg ii sample . we note that sc galaxies undergo evolution , after gas infall has ceased , by 0.4 mag between @xmath18 and @xmath16 , which is out of the range surveyed by mg ii absorption lines . it is interesting to note that this causes brightening of @xmath67 below @xmath16 , and hence increases the number of super @xmath67 galaxies observed at higher redshift by a factor of 1.7 times . sc is the median colour of mg ii absorbers , and this brightening may account for the offset in the normalization ( by a factor of 1.5 , if the normalization of the local lf used in the present paper is adopted ) between the mg ii sample and the local lf as found by steidel et al . in order to summarize the basic prediction of our model , we present in fig . 18 the redshift as a function of magnitude for @xmath4 , @xmath5 and @xmath6 colours for the @xmath95 model . the plotted curves are median ( solid curve ) , quartile ( dashed curve ) and 90 percentile ( dotted curve ) . we calculate the amount of baryons frozen into stars as a function of look - back time in fig . the solid , dotted and dashed curves indicate spheroids , spheroids + disks , and total = spheroids + disks + irregulars , respectively . the mass of spheroids is constant in this redshift range as assumed . the mass in disks increases to lower redshift . the apparent rapid increase towards @xmath18 in this figure , however , is an artefact of the compression of the time scale of the abscissa towards @xmath18 ; the increase close to @xmath18 is actually quite slow if the diagram is represented as a function of time rather than redshift . the prediction of @xmath150 for the spheroids with the population synthesis model is far from robust , since it depends significantly on the lower mass cutoff , even if the initial mass function is fixed . charlot , worthey & bressan ( 1996 ) have given a compilation of @xmath151 for widely accepted models , which give @xmath152 for the age 13 - 15 gyr . we adopt @xmath153 , which agrees with the value obtained from kinematics of elliptical galaxies by van der marel ( 1991 ) , if @xmath154 . we note that about 2/3 of baryons in stars today are in spheroids , and only 1/3 in disks . this agrees with the estimates obtained with different assumptions ( fukugita et al . 1996 ; see also persic & salucci 1992 ) . it is interesting to note that the contribution from irregulars to the stellar mass density is very small in spite of their drastic contribution to the @xmath121 counts . we note that @xmath155 and 1.8 for disks and irregular galaxies in our model . @xmath150 for galaxies of various morphology types are given in table 2 above . we also plot in fig . 19 the evolution of the neutral atomic gas density in units of the critical mass density obtained from surveys of damped lyman @xmath12 systems ( wolfe et al . 1986 ; lanzetta et al . we adopt for @xmath156 the data of storrie - lombardi et al . ( 1996 ) , which update those of lanzetta et al . ( 1995 ) : the data at high @xmath19 have gone down by a factor of two , and this is ascribed to small statistics and an underestimate of statistical errors of the lanzetta et al . sample . the point at @xmath18 is taken from an hi survey of rao & briggs ( 1993 ) ( see also fall & pei 1993 ) . it is interesting to observe that the amount of decrement in gas from @xmath157 to @xmath18 is just consistent with the increment of baryons in disk stars in the same redshift range . this is somewhat different from the claim by lanzetta et al . ( 1995 ) who regarded the consumption of neutral gas as being used to form all stars in the universe . accepting the new result of storrie - lombardi et al . , the baryon budget calculated here indicates that the decrement of the neutral gas is significantly smaller ( by a factor of 3 ) than the total amount of baryons frozen in stars . this favours the picture that most spheroids existed prior to @xmath157 , and the neutral gas observed at this epoch basically gone into stars in disks and irregulars . in our model the residual gas left in galaxies is @xmath158 . this is compared with @xmath159 from an hi survey of rao & briggs ( 1993 ) . this residual gas still causes star formation with a rate of @xmath160 mpc@xmath74 yr@xmath30 which is compared to @xmath161 mpc@xmath74 yr@xmath30 obtained for the local sample by gallego et al . ( 1995 ) from an h@xmath12 survey . the star formation rate in our model increases rapidly to @xmath21 ( by a factor of 2 - 3 ) . such an increase is observed by cowie , hu & songaila ( 1995 ) and by lilly et al . ( 1996 ) , but quantitatively , our increase between @xmath18 and 1 is about 1/21/3 that reported by lilly et al . we have presented a simple , but quantitative model for the history of galaxies . most of the ingredients are not particularly new , known for a while from studies of many authors , although the interpretations may differ from author to author . here , we have assembled them into a single model in a consistent way . the model is reasonably successful to reproduce many different features observationally found . many aspects are checked with observations for @xmath162 , and hence the model is well constrained . on the other hand , the model is not well constrained for higher @xmath19 , due to the lack of observations other than the number count data . an almost inevitable consequence of the present scenario is that morphology of galaxies becomes manifest only in a later epoch . a likely control parameter is spheroid mass ( meisels & ostriker 1984 ) , or the virial temperature of spheroids , besides environment . we might speculate that cooling is not sufficient to make disks when the virial temperature is high , leading to elliptical galaxies , and disks only form when associated with lower mass spheroids . it is likely that disk formation is hindered in cluster environments , since the virial temperature is quite high in clusters . elliptical galaxies generally form from high @xmath163 peaks of the density fluctuations , which makes the two point correlation of elliptical galaxies automatically high . in this scenario the cutoff of the elliptical lf in its faint end is obvious . above @xmath164 disk galaxies would be observed as `` naked bulges '' ; fuzzy cool cloud may eventually be assembled around the bulges as time goes on , and finally show up the full disk structure by @xmath21 . this implies that the size of disk galaxies was substantially smaller at @xmath165 . in our model mergers do not play a major role , although the presence of the modest amount of mergers is not precluded . there is much evidence that the merger rate can not be as high as needed to explain the sharp increase of the @xmath4 count data . the disk of our galaxies can not acquire more than 4% of its mass in the past 5 gyr , otherwise the disk scale height and toomre @xmath29 parameter would exceed the observed values ( tth & ostriker 1992 ) . much evidence from colour and emission properties of normal galaxies suggests that the star formation has persisted over the hubble time without much change ( kennicutt 1983 ; gallagher , hunter & tutukov 1984 ) . this supports the view that galaxy properties have not undergone much change , as expected in the `` major mergers '' , at least after @xmath21 . ( `` minor mergers '' , e.g. , accretion of small galaxies onto the disks might have taken place though ; see below . ) a high merger rate ( guiderdoni & rocca - volmerange 1991 ; broadhurst et al . 1992 ) leads to a trend of decreasing redshift for fainter magnitude for redder colour bands , which however is not supported by the recent deep redshift survey ( @xmath166 ) from the keck telescope ( koo 1996 ) . more directly , the mg ii absorption - line selected sample indicates that properties of giant galaxies change very little after @xmath21 , indicating no evidence for bulk merging after this redshift . no data , however , can constrain the importance of mergers for @xmath167 . where only the count data are available , an increase in numbers and a decrease in luminosity compensate each other and modest merging does not modify the predictions . therefore , the success of the present model does not necessarily mean the absence of merging for @xmath168 . this question is still not settled . the epoch of the spheroidal formation is also a long standing problem . in our model , we simply assumed that the spheroid formation took place at a high redshift . an underlying idea for the early formation of spheroids is that stars must have formed within the collapse time and hence the first collapse of massive objects may be a suitable place for spheroid formation . the rise of the quasar population at @xmath169 to 3 may also imply that spheroids formed at least earlier than quasars get active ( rees 1995 ) . if , indeed , most of spheroids formed below @xmath170 , the light from the star formation epoch should be visible even in a @xmath171 sample . therefore , it will be particularly interesting to find spectroscopic features of those galaxies in the @xmath6 selected sample ( songaila et al . 1994 ) which have been missed in their spectroscopic follow - up . perhaps , the technique similar to that developed by steidel et al . ( 1996 ) can most efficiently be applied to such a sample . if the result excludes the possibility of these galaxies having @xmath172 , the remaining possibilities are either ( i ) they formed at very high @xmath19 , or ( ii ) the formation redshift of spheroids is not so high , and they formed at around the same time as galactic disks , i.e. , @xmath173 . we note that the baryons in spheroids amount to 2/3 ( somewhat less in case of late spheroid formation ) of total baryons frozen into stars . then , if ( ii ) is true , the integrated light from @xmath174 must be more than @xmath175 times higher than is predicted here , since star formation must have been stopped at some redshift above @xmath13 . such activity can easily be detected , e.g. , especially with use of narrow band filters selecting for h@xmath12 emission . of course , this modifies the basic assumptions taken here and hence most of the predictions . the evolution of spheroids is reasonably well understood , although quantitative details depend on models . the poorly - understood part is the uv spectrum . there are a number of speculations about the origin of the uv component observed in elliptical galaxies today , but there seems to be no consensus . in this paper we have passed over this problem . what we have discussed in this paper is the prediction that is not sensitive to the presence or absence of the uv component of spheroids . one specific case where such uv is important is seen in the high @xmath19 tails of the redshift distributions of optically selected samples , the most notable example being `` lyman continuum break galaxies '' having @xmath176 in the @xmath177 mag sample , as discovered by steidel et al . ( 1996 ) . in this case the observability of these galaxies is completely controlled by the uv light emitted from spheroids . in the present model , we predict that the tail of the redshift distribution of @xmath177 sample does not extend beyond @xmath178 . the initial burst of spheroids does not give enough a number of lyman break galaxies , unless the formation epoch is lower than @xmath169 . if , however , some star formation activities persist , e.g. , those with recycled gas , it is easy to give a sufficient number of such galaxies , without modifying any other predictions discussed in this paper . the evolution of spiral galaxies is less well understood . there are , however , a number of constraints from observations . among them , the presence and strength of high redshift tail in the redshift distribution of a @xmath4 selected sample serve as an interesting indicator for the amount of evolution of giant spiral galaxies . while it is generally agreed that the tail is not very large , it is still not settled observationally to what degree such a tail is present ( see cowie et al . 1996 ; koo 1996 ) . a very strong constraint comes from the mg ii sample , which precludes any appreciable luminosity evolution of spiral galaxies in the relevant redshift interval . on the other hand , the cfrs survey ( lilly et al . 1995 ) does see evolution of the characteristic luminosity @xmath67 for blue , presumably spiral , galaxies by one magnitude ( such evolution is not claimed in ellis et al . 1996 work , though ) . in our model these two constraints are somehow reconciled in the way that evolution is faster between @xmath18 to @xmath179 , but very slow above this redshift . this is because gas infall has almost ceased by @xmath142 , and the galaxies started to show more evolution after this redshift . such evolution constraints , if taken literally , indicate that a spiral galaxy is not a closed system , but supplied with fresh gas down to @xmath13 or less . one of the main points of the present paper is to estimate the age of irregular galaxies and introduce them as being newly formed at low redshift . they are the agent that is responsible for the steep slope of the number count in the blue bands . these galaxies are quite gaseous , and undergo very active star formation , which may be observed as star bursts . a similar idea has already been advocated by a number of authors , in particular by cowie et al . ( 1991 ) and babul & rees ( 1992 ) . the difference is in the point that we have not introduced any population unobservable today and that we assume continuous star formation that is proportional to the gas abundance rather than the `` burst '' . the dwarf population , which is less luminous than ordinary irregular galaxies , plays no role in the results presented in this paper . there may be more irregular galaxies formed at higher @xmath19 , than are considered in this paper , but they should follow the same track as the disk component of spiral galaxies , and may constitute a low - luminosity tail of the lf of spiral galaxies , which is not important in galaxy number counts . the detailed shape of the subluminous part of the luminosity function is not important , unless the lf is very steep . there is much indication that correlation of galaxies seen in blue bands is weak ( efstathiou et al . 1991 ; couch , jurcevic & boyle 1993 ; infante & pritchet 1995 ) , the clustering being at most comparable to that of late type galaxies . this is easy to understand : the @xmath4 sample at faint magnitudes is dominated by spiral galaxies and irregular galaxies which form from the peaks of low-@xmath163 fluctuations and are thus least biased . the local lf is clearly one of the most important sources of uncertainties , and it often hinders us from making unambiguous predictions from our model . the uncertainty in the local lf is not only in the faint end behaviour , but also in its normalization . an upward shift of the normalization by 0.15 dex would bring a number of predictions of the galaxy counts in much better agreement with observations , and also it would decrease substantially the amount of evolution required by the high redshift data . our normalization of the local lf , however , is on the high side of what is allowed by current surveys , so we can not justify a further increase by 0.15 dex . more important is the problem of the faint end ( @xmath180 mag ) of the lf . if the total lf has a flat slope ( @xmath181 ) as in , e.g. , loveday et al . ( 1992 ) and lin et al . ( 1996 ) , that is compelling evidence for galaxy number evolution . luminosity evolution does not have enough power to explain the abundance of irregular galaxies required in our analysis . this view is consistent with the lf at various redshifts derived by ellis et al . ( 1996 ) : the luminosity evolution does not have power to lift the faint end of the lf . on the other hand , if the local luminosity function shows an increase towards the faint end ( metcalfe et al . 1991 ; marzke et al . 1994 ; subbarao et al . 1996 ) , we do not particularly need number evolution of galaxies , but luminosity evolution is sufficient to explain the observations . it is also possible that irregular galaxies have evolved into dwarf spheroidals near giant galaxies or in a galaxy rich environment . in this case the rise of the lf may take place @xmath1821 mag fainter than the typical luminosity of irregular galaxies . we have argued that @xmath6 band observations provide a good tool to study cosmology . this is because spheroids always contribute significantly to the @xmath6 band counts and the evolution of spheroids is reasonably well known . we have concluded that @xmath7 is disfavoured by the count data . this conclusion remains true , as long as we keep our basic assumptions . if , however , mergers play an important role at high @xmath19 , say @xmath183 , the @xmath7 model could be allowed . the @xmath19 distribution for @xmath7 cosmology has a component substantially extended to higher @xmath19 than in @xmath95 model . the median redshift of the spheroids in a @xmath184 sample is about 3 ( for an @xmath95 model ) , and an observation with @xmath24 and @xmath185 bands ( as in steidel et al . 1996 ) searching for lyman break galaxies just hits the middle of the @xmath19-distribution at this magnitude . the possibility that not all spheroids are in place by @xmath186 may also modify the conclusion about cosmology , and it may bring a confusion as to the interpretation of the cosmological models . we should like to thank nobuo arimoto , richard ellis , roger davies , tadayuki kodama , simon lilly , richard macmahon and jim peebles for useful discussions , and jarle brinchman , richard ellis and ofer lahav for reading manuscripts and many comments improving the manuscript . one of us ( mf ) thanks the hospitality of the institute of astronomy of the university of cambridge , where this paper is made into the final form . he also wishes to acknowledge generous support from the fuji xerox corporation at princeton . this research was supported in part by the grants - in - aid by the ministry of education , science , sports and culture of japan ( 07ce2002 ) to resceu ( research center for the early universe ) .

we show that dark matter halos , in n body simulations , have a _ boundary layer _ ( bl ) , which neatly separates dynamically bound mass from unbound materials . we define and as the differential kinetic and potential energy of halos and evaluate them in spherical shells . we notice that , in simulated halos , such differential quantities fulfill the following properties : ( i ) the differential virial ratio has at least one _ persistent _ ( resolution independent ) minimum , such that , close to , ( ii ) the function has a maximum , while ( iii ) the relation holds . bl s are set where these three properties are fulfilled , in halos found in simulations of tcdm andcdm models , run _ ad hoc _ , using the art and gadget codes ; their presence is confirmed in larger simulations of the same models , with a lower level of resolution . here we find that of the clusters ( per model ) we have with , owns a bl . those clusters which appear not to have a bl are seen to be undergoing major merging processes and to grossly violate spherical symmetry . the radius has significant properties . first of all , the mass it encloses almost coincides with the mass , evaluated from the velocities of all particles within , according to the virial theorem . also , materials at are shown not be in virial equilibrium . using we can then determine an individual density contrast for each virialized halo , that we compare with the `` virial '' density contrast ( where is the matter density parameter ) obtained assuming a spherically symmetric and unperturbed fluctuation growth . as expected , for each mass scale , is within the range of s . however the spread in is wide , while the average is smaller than the corresponding . we argue that the matching of properties derived under the assumption of spherical symmetry must be a consequence of an approximate sphericity , after violent relaxation destroyed features related to ellipsoidal non linear growth . on the contrary , the spread of final is an imprint of the different initial 3d geometries of fluctuations and of the variable environment during their collapse , as suggested by a comparison of our results with sheth & tormen analysis .

astro-ph0301247

the inner region of dark matter halos in n body simulations is certainly virialized . a precise detection of the boundaries of such a region , however , is a long lasting and not yet completely solved problem . if we try just to identify a `` collapsed '' region , @xmath20 a region detached from the cosmological expansion , the task is easier ( see , e.g. , monaco & murante 1999 and references therein ) . however , in general , a collapsed region is not yet in virial equilibrium . the point is that , in a cosmological context , each structure is embedded in a more or less homogeneous background , and we pass gradually from the inner halo to external materials . the aim of this work is to show a precise criterion able to find virialized regions in dark matter halos . this criterion will be tested on simulations and found to be more efficient than previously adopted rules . as a matter of fact , several global properties of galaxy clusters , such as mass ( @xmath21 ) , radius ( @xmath22 ) and , also , temperature ( @xmath23 ) or luminosity ( @xmath24 ) are strictly linked to the achievement of equilibrium . analytical predictions on such global quantities were performed assuming that structures evolve from uniform , spherically symmetric fluctuations , preserve such symmetries during their growth and feel no influence from their environment ( gott & rees 1975 , peebles 1980 ) . however , even under such simplifying conditions , a cosmological constant is a significant complication . lahav et al ( 1991 ) studied the growth of uniform spherical fluctuations within such context . they also outlined that the most direct way through which observations can distinguish among flat models with different @xmath17 ( matter density parameter ) is comparing cluster abundances at low and high redshift . the dynamics of the spherically symmetric growth of a uniform density fluctuation , in a model with @xmath25 ( where @xmath26 is the vacuum density parameter ) , was further explored by eke et al ( 1996 ) . brian and norman ( 1998 ) gave a polynomial expression for the expected density contrast of virialized clusters , which reads @xmath27 , with @xmath28 , and @xmath29 , @xmath30 , @xmath31 , and holds at any redshift @xmath32 ; this relation is well approximated by the expression : @xmath33 under the same assumptions , using a press & schechter ( 1974 ) approximation , one can predict the expected mass function of clusters . the impact of this approximation on the estimate of cluster global quantities could be appreciated only through numerical simulations . much work was therefore devoted to this aim ( see , e.g. , lacey & cole , 1993,1994 ; cole & lacey 1996 ; carlberg et al . , 1996 ; eke et al . 1996 ; eke et al . 1998 ; brian & norman , 1998 ; gardini et al . 1999 ) . however , in the very analysis of the numerical work , the above approximations were often implicit . an example is the identification of dark halos based on spherical overdensity ( so ) algorithms , with reference to the density contrast @xmath18 given in eq . ( 1.1 ) . in this work we shall also use an so algorithm ( described below ) as a selection criterion . however , using eq . ( 1.1 ) , to work out values for the cluster radius ( @xmath34 ) or mass ( @xmath35 ) , implies a bias . the values ( 1.1 ) are obtained under precise restrictions and numerical work should also verify their impact . in a sense , the problem is even more severe when the values ( 1.1 ) are used to define virialized halos without requiring explicitly that they are spherical . in this work we plan to find a rule , suitable for defining virialized systems , taking into account that a variety of growth histories led to different final features of virialized structures . a number of recent papers ( sheth & tormen ( 1999 , 2002 ) , sheth , mo & tormen , 2001 ) had a similar motivation . they tried to study the growth of primeval fluctuations taking into account the effect of an ellipsoidal collapse . then , in order to avoid having the collapse on some axis going to zero , they set up a recipe to freeze it out once it has shrunk by some critical factor . this freeze out radius is explicitly chosen so that the density contrast at virialization for the whole halo is again @xmath18 as given in eq . ( 1.1 ) , @xmath20 in spherical growth models . in contrast , in this work the use of @xmath18 to define virialized systems is completely overcome . the point is that the detailed growth history of individual halos , in which different ellipticities and velocity anisotropies played an important role , is not expected to leave major traces on final halo shapes after violent relaxation has occurred during the virialization process ( navarro , frenk & white 1997 ) . on the other hand , it may result in a significant spread of the equilibrium density contrasts @xmath15 of individual halos . therefore , when the virial radius of relaxed objects is sought , deviations from spherical symmetry and velocity anisotropies may not be as important as deviations of @xmath15 from @xmath18 . we expect that the value @xmath18 , given by eq . ( 1.1 ) , is within the range of the actual @xmath15 values , but we think that it is important to first test how extended this range is , and how @xmath18 is set with respect to actual values . furthermore , a different @xmath15 implies a different radius and mass ( @xmath13 and @xmath11 ) for each object . accordingly , we expect that some individual object masses increase and others decrease . we will explore the extent of these changes on galaxy clusters in numerical simulations of critical cdm models with and without a cosmological constant . such changes may have an impact on the mass function ( see section 5 and fig . 13 and 14 ) , while some scaling relations might be significantly altered , for instance the relation between @xmath21 and @xmath23 . in order to implement our program , we need a simple and effective rule to identify virialized material in clusters . in trying to find such rule we discovered a precise regularity in the transition region between material belonging to clusters and the surrounding medium . in fact , the volume occupied by virialized materials is confined by a _ boundary layer _ ( bl ) , whose depth @xmath36 ranges around 50100@xmath37kpc , and whose properties fulfill precise analytical prescriptions . such a bl has been found in @xmath7 of the clusters in our simulated models and is one of the findings of this work . the bl is not a physical confinement barrier , and particles can travel outwards and inwards through it . on the contrary , it confines a volume , where material fulfills precise equilibrium conditions , which cease to hold outside it . inside the bl itself , the formal condition for virial equilibrium of an isolated system holds , as pressure forces at its upper and lower borders cancel each other ; on the bl , kinetic energy has a minimum with respect to potential energy , and matter density is rapidly decreasing . these conditions follow from a precise analytical requirement ( @xmath38 requirement ) derived under the assumption of spherical symmetry . this @xmath38 requirement will be only approximately satisfied by halos in simulations , where the halo geometries apparently violate spherical symmetry . we show , however , that normal deviations from a spherical shape in dm halos do not prevent them from having a bl . we explicitly note that we do not _ force _ the bl to exist , but we apply the @xmath38 requirement and _ find _ it . for this reason , finding the bl is different from finding a given spherical overdensity ; the latter always exists when average densities of halos decrease with increasing radii , and this does not imply even an approximately spherical symmetry . on the contrary , we could have bound and relaxed regions which are not confined by a bl : our analysis of numerical simulations shows that this is not the case . there is also a major difference between the @xmath38 requirement we are defining here , and other prescriptions , such as skid , which defines a halo as the set of all particles _ gravitationally bound _ to the halo itself . in fact , bound particles are not guaranteed to belong to virialized zones and may belong to regions which are bound but still unrelaxed . in this paper we shall first discuss when the bl should exist and introduce the @xmath38 requirement used to detect it . to have significant statistics based on this rule and to confirm its validity requires simulations performed with a large dynamic range , as can only be done by parallel computing . finding and situating bl s allows a better evaluation of individual cluster masses , radii and density contrasts , as well as an analysis of their possible dependence on various physical parameters . the study of the structure of dark matter halos in numerical simulations can also be put on a sounder footing . in previous analyses ( see , e.g. , cole & lacey 1996 ) the @xmath22 dependence of the integral virial ratio @xmath39 was discussed and found not to be `` useful for defining the boundaries of the virialized region '' . one of the findings of our analysis is precisely the possibility of detecting such a boundary , using the virial ratio for shells instead of spheres . 1 compares integral and differential virial ratios , showing how much more information appears to be contained in the latter ( see sec . 3 for details on how @xmath40 points are calculated ) . previous analyses have also stressed halo properties apparently depending `` on how groups are identified '' ( sheth et al . , 2001 ) . the existence of a border , with precise physical properties , secluding cluster material lets one overcome any such ambiguity . it may be also important to remember that according to the definitions given above , in this work @xmath18 , @xmath34 , @xmath35 are quantities worked out starting from the fixed density contrast given by ( 1.1 ) , while @xmath15 , @xmath13 , @xmath11 are the same quantities when worked out from the setting of the bl around each cluster . let us also point out that chb the procedure discussed in this work is not the basis for an che alternative cluster finding algorithm , such as , e.g. , friends - of - friends , skid or so . on the contrary , our analysis assumes that cluster locations are given . here we give a different prescription for defining the physical dimension of the cluster , not its position . the plan of the paper is as follows . in section 2 we give suitable information on the simulations used , which were partially run _ ad , for this work . in section 3 and 4 we define the boundary layer ( bl ) and discuss how it can be found in simulated clusters . in section 5 we show the results of our work , and in section 6 we discuss them and some future perspectives . we use two main sets of simulations , performed with different codes : the parallel ap3 m n body code described in gardini et al . ( 1999 ) , the parallel pm art code developed by kratshov & klypin ( 1997 ) , and the parallel tree code gadget ( springel et al . 2001 ) . the code of gardini et al . ( 1999 ) was developed from the serial public ap3 m code of couchman ( 1991 ) , extending it to different cosmological models and allowing suitable flexibility in particle masses . using this code , two simulations were performed . the first of them , dealing with a `` tilted '' einstein - de sitter model ( hereafter tcdm ) , is the same already considered by gardini et al . ( 1999 ) ; the normalization of the run was , however , rescaled to yield @xmath41 at the final step , instead of @xmath42 , so that the final abundance of rich clusters is the same for both models at the final epoch ( as ususal , @xmath43 is the m.s . density fluctuation on the scale of @xmath44mpc , @xmath45 being the hubble parameter in units of 100 km / s / mpc ) . the latter simulation dealt with a @xmath46cdm model , i.e. a flat cdm universe with non zero cosmological constant with @xmath47 . owing to recent observational results ( see , e.g. , schuecker et al , 2002 ) , these normalizations are both rather high , as a reasonable value for @xmath43 in @xmath46cdm models is @xmath48 ; however , this has no impact when studying inner cluster dynamical properties . the above simulations ( denoted a and b , respectively ) are performed in cubes with sides of 360@xmath37mpc . cdm+baryons are represented by @xmath49 particles , whose individual mass is @xmath50 for tcdm and @xmath51 for @xmath46cdm . we use a 256@xmath52 grid to compute the fft s needed to evaluate the long range contribution to the force ( pm ) and we allow for mesh refinement where the particle density attains or exceeds @xmath53 times the mean value . the starting redshifts are @xmath54 for tcdm and @xmath55 for @xmath46cdm . the particle sampling of the density field is obtained by applying the zeldovich approximation ( zeldovich 1970 ; doroshkevich et al . 1980 ) starting from a regular grid . we adopt the same random phases in both a and b. the number of steps were 1000 equal @xmath56time steps ( the time parameter is @xmath57 , where @xmath58 is the expansion factor ) . the comoving force resolution , given by the softening length , is @xmath59kpc , yielding a plummer equivalent softening @xmath60kpc . these simulations were described in more detail in gardini et al . ( 1999 ) , and were also used in macci et al . the parameters of the models are reported in detail in table 1 . in order to see how resolution affects the results of the @xmath38 requirement , we performed further simulations with the art code ( courtesy of a. klypin ) . this allows us to select regions inhabited by clusters and to re run them with increasing particle mass resolution : from @xmath61 to @xmath62 . the art code ( adaptive refinement tree : kravtsov et al . 1997 , knebe et al . 2000 ) starts with a uniform grid , which covers the whole computational box . this grid defines the lowest ( zeroth ) level of resolution of the simulation . the standard particles - mesh algorithms are used to compute the density and gravitational potential on the zeroth - level mesh . the code then reaches high force resolution by refining all high density regions using an automated refinement algorithm . the refinements are recursive : the refined regions can also be refined , each subsequent refinement having half of the previous level s cell size . this creates a hierarchy of refinement meshes of different resolution , size , and geometry covering regions of interest . because each individual cubic cell can be refined , the shape of the refinement mesh can be arbitrary and effectively match the geometry of the region of interest . this algorithm is well suited for simulations of a selected region within a large computational box . the criterion for refinement is the local density of particles : if the number of particles in a mesh cell ( as estimated by the cloud - in - cell method ) exceeds the level @xmath63 , the cell is split ( `` refined '' ) into 8 cells of the next refinement level . the refinement threshold may depend on the refinement level . the code uses the expansion parameter @xmath58 as the time variable . besides spatial refinements , during the integration time refinements are also performed . the art code can handle particles of different masses . this lets us increase the mass ( and correspondingly the force ) resolution of regions centered around the highest mass clusters . the multiple mass resolution is implemented in the following way . we first set up a realization of the initial spectrum of perturbations , so that initial conditions for our largest number ( @xmath64 ) of particles can be generated in the simulation box . coordinates and velocities of all the particles are then calculated using all waves ranging from the fundamental mode @xmath65 to the nyquist frequency @xmath66 , where @xmath67 is the box size and @xmath68 is the number of particles in the simulation . we obtained lower resolution regions by merging high resolution particles into particles of larger mass where needed . this process can be repeated to get still lower resolution . the larger mass ( merged ) particles have velocities and displacements equal to the average of the velocities and displacements of the smaller - mass particles . using this technique we firstly generated initial conditions for low resolution simulations . these simulations were run from their initial redshift to @xmath69 and they were used to locate the lagrangian regions forming the most massive clusters . then , we were able to obtain initial conditions at high resolution ( both in force and mass ) in the selected lagrangian zones , with the same random fourier phases . such zones were surrounded by lower and lower resolution shells taking into account the tidal field which acts on them . finally , these ics were used to run the high resolution cluster simulations . the simulations employed here were performed using @xmath70 zeroth - level grid in a computational box of @xmath71 . the threshold for cell refinement ( see above ) was low on the zeroth level : @xmath72 . thus , every zeroth - level cell containing two or more particles was refined . the threshold was higher on deeper levels of refinement : @xmath73 and @xmath74 for the first level and higher levels , respectively . for the low resolution runs the step in the expansion parameter was chosen to be @xmath75 on the zeroth level of resolution . this gives about 500 steps for particles located in the zeroth level for an entire run to @xmath69 and 128.000 for particles at the highest level of resolution . using the art code we performed two simulations : a @xmath46cdm model and a tcdm model , that we will indicate as c and d , respectively . these models are similar , although not identical , to those of gardini et al . 1999 ; their parameters are listed in table 2 . we identified the clusters in these zero level simulations and selected the 6 largest mass clusters for each cosmological model . they were re run with increased mass ( and force ) resolution , using the method described above . for every selected cluster we have three runs with resolutions of @xmath70 ( particles mass @xmath76 ) , @xmath77 ( @xmath78 ) and @xmath64 ( @xmath79 ) with more than 350.000 particles within a radius of 2.0 mpc h@xmath80 . we then run 3 out of the 6 high resolution @xmath46cdm clusters with the public parallel tree - code gadget ( springel et al . 2001 ) , to verify that our results do not depend upon the peculiarities of one n - body code . one of the clusters was magnified with both art and gadget . the initial conditions were set as described above . we used a plummer - equivalent softening length @xmath81 ; this is also the linear dimension or our highest resolution art cell . we chose the time step criterion 3 , based on the local dynamical time , with a tolerance of the integration error @xmath82 , which gave us more than 100000 time steps from @xmath83 to @xmath69 . the criterion for opening the tree cells is based on the absolute truncation error in the multipole expansion , with a tolerance on the error on the force of @xmath84 ( see springel et al . 2001 and the code user manual for more details ) . in all of our numerical simulations , clusters were identified using a so algorithm . as a first step , candidate clusters are located by a fof algorithm , with linking length @xmath85 ( here @xmath86 is the average particle separation ) , keeping groups with more than @xmath87 particles . we then perform two operations : ( i ) we find the point , @xmath88 , where the gravitational potential , due to the group of particle , is minimum ; ( ii ) we determine the radius , @xmath89 , of a sphere centered in @xmath88 where the density contrast is @xmath90 . using all particles in this sphere we perform again the operations ( i ) and ( ii ) . the procedure is iterated until we converge onto a stable particle set . the set is discarded if , at some stage , we have less than @xmath87 particles . if a particle is a potential member of two groups it is assigned to the more massive one . ( gardini et al . 1999 also describe this so algorithm in more detail and compare it with other group identification algorithms , e.g. governato et al . in this work we set @xmath91 and take @xmath87 so to have a mass threshold @xmath92 . above this mass threshold there are @xmath93 _ halos _ in the simulations a , b and @xmath94 _ halos _ in the simulations c , d . tests performed by gardini et al . ( 1999 ) show that no cluster is missed above @xmath95 . we have 303 such clusters in a , 316 in b , 40 in c , 36 in d ; 6 clusters of c and d where then blown up to various resolution levels . finding the volume where cluster materials are in virial equilibrium may seem a tough and somewhat ambiguous task . if kinetic and potential energies , within a radius @xmath22 , are defined according to : @xmath96@xmath97 the virial ratio @xmath98 should be unity , in a virialized sphere of radius @xmath22 , provided that it is fully isolated . on the contrary , in the real world , as well as in simulations , the virialized volume is bordered by infalling and outgoing material , possibly traveling through a partially depleted area . for this reason , the convergence of @xmath99 to unity , at a precise radius @xmath22 , should not even be expected . in principle , one can take into account such perturbations by fixing a radius @xmath22 well inside the cluster volume and adding a _ pressure term _ to take into account external material , as has been done in observational work aiming to estimate the mass of optical clusters ( see , e.g. , girardi et al . , 1998 ) . accordingly , the virial equilibrium condition reads @xmath100 ( @xmath101 is the volume occupied by virialized material and @xmath56 is the pressure at its borders ) . ( 3.4 ) translates the problem into finding a reasonable estimate for @xmath56 . the problem can also be considered from a slightly different point of view : if we find , around the cluster , a _ boundary layer _ ( bl ) , which is in virial equilibrium by itself , and above which virial equilibrium is lost , its very material and all material it borders , down to the cluster center , need to be in stationary equilibrium . particles may pass through this bl in both directions , but the overall virial balance shall keep constant in time . we will now assume that this bl is spherical . the conditions found under this restriction are , however , well approximated by simulated halos . this seems to indicate that , even after a strongly asymmetric growth , the final virialization process tends to partially suppress elliptical features . the @xmath22 dependence of the potential energy can be derived by performing the sums in eq . ( 3.2 ) starting with the outside : @xmath102@xmath103 where the last term is the very definition of @xmath104 . in the continuous limit ( and still assuming spherical symmetry ) @xmath105 with a suitable density , @xmath106 . quite in general , a volume integral of @xmath107 increases with @xmath22 . henceforth , @xmath108 where c is a normalization constant evaluated at an arbitrary position @xmath89 and @xmath109 will , however , depend on @xmath22 . accordingly , it must be that @xmath110{\cal z}(r ) ~. \eqno ( 3.8)\ ] ] let us now suppose that there exists an interval @xmath111 , which satisfies the following conditions : * it is in virial equilibrium , * the @xmath22 dependence of pressure is such that the r.h.s . of eq . ( 3.4 ) , yielding a term of the form @xmath112 , can be neglected in the virial balance , * inside @xmath36 , @xmath113 is constant . further comments on the ( ii ) requirement can be found in appendix a , where we show that halo profiles are expected to approach it . together with ( i ) , it implies that @xmath114 in the layer . let us then take into account the condition that @xmath113 is constant and use eq . ( 3.8 ) with @xmath115 and the _ virial ratio _ @xmath116 to obtain that @xmath117 all along the interval @xmath36 . this equation coincides with eq . ( 3.29 ) in goldstein , pole & safko ( 2002 ) ( in turn , the latter equation yields eq . ( 3.9 ) hereabove , in the limit @xmath118 ) . requiring that @xmath115 , then , also yields that @xmath119 in @xmath36 . _ vice versa _ , if the eqs . ( 3.11)(3.12 ) are both fulfilled in a layer of depth @xmath36 , this layer is at rest and in virial equilibrium . we thus define _ boundary layer ( bl ) _ a region of depth @xmath36 satisfying eqs . ( 3.11)(3.12 ) . let us stress that the condition ( iii ) is essential to set the @xmath38 requirement , defined by the eqs . ( 3.11)(3.12 ) . unless @xmath113 is constant along a suitable interval , we can neither replace @xmath120 by @xmath121 in eq . ( 3.9 ) , nor factorize @xmath113 out of the sum , so to obtain eq . ( 3.11 ) . moreover , should eq . ( 3.11 ) hold on a single @xmath22 , we could have @xmath115 there , with @xmath122 . finding a layer fulfilling the @xmath38 requirement , goes therefore beyond finding a layer in virial equilibrium . as we shall see , however , once the @xmath38 requirement is fulfilled , we have found a layer bounding a virialized region . let us also stress that the @xmath38 requirement does not imply that particles can not pass through the bl ; it only prescribes that such ( possible ) passages do not violate its stationary conditions . however , if we find about a cluster a boundary layer , it will act as a confinement barrier to inner kinetic and potential energies and mass . notice that , in principle , inner materials might not be in virial equilibrium themselves ; even in this case , however , there can be no net exchange of mass between inside and outside and the possible energy exchange is limited to the work done by tidal torques due to anisotropies . let us now show that no further material , outside the bl , can be in virial equilibrium . we will show that this is certainly forbidden if the values , taken by @xmath123 and @xmath113 at @xmath124 , are a minimum and a maximum , respectively , when compared with values at @xmath125 . this simple condition , however , is not strictly necessary ; it is sufficent that the difference @xmath126 increases , becoming positive , at @xmath127 . this will be shown here below . let us however anticipate that , while the @xmath38 requirement can be fairly easily tested , it is not easy to deal with such further requirement . the tests performed on a large number of clusters and described in the next sections , however , let us conjecture that , when the @xmath38 requirement is fulfilled , this further requirement is somehow inescapable . we now show why , at @xmath127 , the equilibrium conditions are violated if a boundary layer is found . notice that , at @xmath125 , we can still assume that @xmath128 , but @xmath113 , starting from the value it had in @xmath36 , will now vary with @xmath22 and , according to eq . ( 3.8 ) , @xmath129 this equation can not be fulfilled , if the difference @xmath126 , vanishing up to @xmath124 , increases at @xmath127 , becoming positive . in fact we required that @xmath113 is maximum and therefore @xmath130 . let us finally notice that if a layer is characterized by higher @xmath113 values , this means that it is emptier than inner or outer layers . a maximum of @xmath113 , therefore , indicates a low density layer . in turn , a minimum of @xmath123 indicates a low kinetic energy , with respect to potential energy . bls , therefore , are partially depleted @xmath22intervals , where particles , on average , are particularly slow . the mathematical definition must now be applied to numerical simulations . a failure to find bls could be attributed to insufficient numerical resolution or to bad violations of spherical symmetry . however , in principle , even if the resolution is enough and no substantial departures from sphericity are visible , bls could still be absent . let us now describe our procedure . after finding clusters in simulations and their centers @xmath88 , as described in sec . 2 , we evaluated @xmath1 and @xmath40 , for spherical @xmath22intervals containing a fixed number @xmath131 of particles ( here @xmath132 of the total mass within 4.5@xmath37mpc from @xmath88 and @xmath133 is the particle mass ) . however , successive points along @xmath22 are obtained by shifting outwards by @xmath134 particles . in the relevant @xmath22 range , the above criterion yields @xmath135 for tcdm ( simulation a ) and @xmath136 for @xmath46cdm ( simulation b ) , while successive points lay at a distance @xmath137kpc . this is below ( but not much below ) the force resolution ( @xmath138kpc ) , as desired . we do not expect bl s to be much thicker than our force resolution ; accordingly , we first seek the minima of @xmath123 . owing to the procedure by which @xmath123 points are worked out , a minimum is considered _ significant _ only if it is such with respect to 8 neighbors . finding the maxima of @xmath113 is harder , as @xmath139 , obtained by differentiating @xmath1 , should then be further differentiated to find its maxima . in practice , however , this second order differentiation is not needed . quite in general , there will be just a few points @xmath140 where @xmath123 has _ significant _ minima . let then @xmath141 be the virial ratio at such points and let us fit @xmath1 with the expression @xmath142 in @xmath143 points @xmath144 , covering an interval @xmath145@xmath146kpc . the fit is performed in two steps : we first minimize the function @xmath147 just using @xmath148 as a parameter ; then , in eq . ( 4.2 ) , we allow @xmath141 to take values different from the value of @xmath123 in @xmath140 , and perform a 2parameter fit . we considered the fit to be satisfactory , if two conditions hold : ( i ) when performing the first fit , the residual @xmath149 ( i.e. , a mean square discrepancy @xmath150 , between the numerical values of @xmath1 and its fitting expression , was our limit of tolerance ) . this detects r intervals where eqs . ( 3.11)-(3.12 ) are both fulfilled . ( ii ) when performing the second fit , the best fit value of @xmath141 is within 1@xmath151 from the actual @xmath141 value . here we test that also the slope of @xmath152 is not too far from what is required . such constraints should therefore select points where @xmath123 is minimum and ( nearly ) intersects @xmath113 . as outlined above , one might reasonably expect that a large fraction of clusters , at their boundaries , is far enough from sphericity that the expected system of maxima and minima , as well as their coincidences , is disrupted . the numerical noise in calculating @xmath113 is also quite large and can be expected to create severe problems . on the contrary , we found that _ all clusters with mass @xmath153 have at least one value of @xmath140 fulfilling the above requirements _ , while some clusters have 2 or 3 points where the fitting criteria are fulfilled . in the few cases , when the fitting criteria were satisfied at more than one @xmath140 , we discarded @xmath140 values yielding @xmath15 outside an interval @xmath154@xmath155 . this selection criterion left us with @xmath7 of the initial clusters . when more than one @xmath140 was still available , we selected the one corresponding to the smallest @xmath156 ; this last criterion was needed for @xmath157 of the clusters only . let us recall again that the requirement of intersection between a minimum of @xmath123 and a maximum of @xmath113 was found for spherical systems . when we deal with the halos of our simulations , sphericity is likely to be an approximation . when a single intersection exists , however , it is natural to guess that this is the point where the @xmath38 requirement is best approached . however , both in this case and when several @xmath140 points are found , it is natural to perform a closer inspection to make sure that no unexpected feature may bias our understanding of the physical situation . this closer inspection requires an increased resolution . as described in sec . 2 , a tcdm model and a @xmath46cdm model were then reconsidered within a smaller box with sides of 180@xmath37mpc , making use of the art and gadget codes . this gave us @xmath158 of the clusters provided by the ap3 m simulations , but enabled us to focus on particular clusters , making use of the art package facilities to increase the resolution there while still keeping identical boundary conditions . in order to test that the regularities found are not linked to peculiarities of the art package , some clusters were magnified using gadget , instead of art . altogether we have 12 clusters magnified , 6 for each model . 4 of the latter were obtained using art and 3 using gadget ; hence , one cluster was magnified with both codes showing that our results are independent of the n body code used . in fact , the final positions of particles have just minor differences , while @xmath123 and @xmath152 coincide . a number of subsequent tests also concerned the stability of the minima of @xmath123 . in general , the number of local minima @xmath140 decreases when the resolution is increased , but the point where @xmath152 fits the expression ( 4.1 ) with the worst resolution is a minimum also with greater resolutions . in just a few cases , new minima arise when the resolution is improved ; but in no case is a minimum yielding the bl erased when changing resolution . the `` worst '' cases for each model are shown in fig . 2 . here we have a tcdm cluster , taken from the simulation c , which passes from 3 minima ( @xmath70 particles ) , to 1 minimum ( @xmath77 particles ) and , again , to 3 minima ( 512@xmath52 particles ) . the bl is displaced outwards by @xmath159kpc , when we pass from 128 to 256 ; then , when passing from 256 to 512 , no appreciable shift occurs . out of the 12 clusters considered , only this one presents a shift slightly above the resolution of the initial simulation . the @xmath46cdm cluster of simulation d shown in fig . 2 , shows 4 minima with @xmath70 particles , 2 minima with @xmath77 particles and , again , to 3 minima with 512@xmath52 particles ; but when changing resolution , the shifts are @xmath160kpc . for these 12 clusters we follow in detail the joint behavior of @xmath123 and @xmath113 . we show them in fig . 3 for the best and worst cases we found . of course , a precise coincidence between a minimum of @xmath123 and a maximum of @xmath113 would indicate that , within the resolution limits , the system is spherical . seeking this kind of coincidence , therefore is somehow fatuous . furthermore , in order to work out @xmath113 , we ought to differentiate @xmath152 , which is obtained from a discrete point distribution . hence , although the nice coincidence of the _ best _ case is quite appealing , this case is not _ safer _ than the _ worst _ case we show . by tolerating a 3@xmath161 disagreement in @xmath152 , for @xmath162 , we allow for a tolerance @xmath163 on the exponent @xmath113 . in the worst case , we find a @xmath113@xmath123 discrepancy of this order , but in most cases , @xmath164@xmath165 turns out to be @xmath166 . we tentatively conclude that this resolution is sufficient to show physical features and that the level of the @xmath113@xmath123 ( dis)agreement is a measure of the @xmath58sphericity of actual systems . the important issue is that we _ always _ find a maximum of @xmath113 to match the minimum of @xmath123 , quite close to the @xmath140 selected at the initial resolution level . this point can be further illustrated by fig . 4 , which shows a plot similar to fig . 3 , for a typical case at the initial resolution level . both the @xmath123 and the @xmath113 curves are much noisier here . @xmath123 , in particular , shows a wide set of local minima ; the only _ persistent _ minima , however , are those marked by thick filled circles . as expected , the behavior of @xmath113 is still noisier and most of the oscillations shown are clearly numerical . the role of @xmath113 , however , just amounts to choosing among the points marked by a thick filled circle and selects the point marked by a vertical arrow . even at this resolution level , the presence of a maximum of @xmath113 , close to such point , can be suspected . when the resolution is increased , this maximum is more clearly visible , while no @xmath113 maxima exist close to the other @xmath123 minima . a further preliminary test of our technique , concerning galactic - sized and unvirialized dm halos , is given in appendix b. before concluding this section , let us draw the reader s attention on the setting of the bl around particle agglomeration , as shown in fig . 5 for a @xmath46cdm cluster . this figure shows that the bl is actually set outside of the main matter agglomeration ; its spherical shape is somehow in contrast with the apparent @xmath58sphericity of the cluster materials and easily justifies minor discrepancies from results based on an assumption of full spherical symmetry . once the sphere confining cluster materials is set , we can evaluate the density contrast @xmath15 and the mass @xmath11 inside it . in figs . 6 and 7 , points give @xmath15 and @xmath11 for all clusters in simulations a and b , respectively . they show a rather wide spread of @xmath15 values . by subdividing the @xmath11 abscissa in intervals of constant logarithmic width , we evaluate the average density contrast in each of them , to seek systematic trends with mass . averages are still subject to a significant uncertainty owing to the spread of @xmath15 . they are shown , at the 1@xmath151 level , in the plots . at high masses , some logarithmic intervals of fig . 7 are empty or contain a single object . in the latter cases no error bar is given . there seems to be no evidence of any peculiar trend of density contrasts with mass apart from , perhaps , a modest indication of an increasing density contrast at very high scales in @xmath46cdm . it is therefore possible to consider the overall average among @xmath15 s . this average is indicated by the continuous horizontal line and compared with the `` virial '' density contrast @xmath18 , as given by eq . ( 1.1 ) . in both cases , @xmath18 ( dotted line ) is well inside the range of the density contrasts we found ; the average @xmath15 , however , in both cases , is smaller than @xmath18 by @xmath19 . it is , however , clear that gravitationally evolved , stationary objects are _ not _ uniquely identified with a given , fixed density contrast . in further work , we plan to deal with the impact of variable @xmath15 on the physical characteristics of dm halos , such as concentration , average angular momentum , density and velocity profiles . it may also be significant to test our results against the effects of different particle selection criteria , e.g. using the skid algorithm instead of the so one to pre - select the dm halos . in figs . 8 and 9 we plot @xmath13 against @xmath11 , for the simulations a and b , respectively . here , error bars account for the variance of @xmath13 around its average , which is wider than the error of the average @xmath13 , shown in fig . continuous and dotted curves show the expected behavior of @xmath13 against @xmath11 , when the density contrast is set either at the average @xmath15 value or at @xmath18 . it is also significant to compare the plot of @xmath13 against @xmath11 with the behaviour of the radii @xmath13 against the masses @xmath35 of the corresponding clusters . this comparison is shown in fig . 10 , for the simulation a only ; in order to avoid confusion , we omit error bars which , however , are approximately of the same size in both cases . the highest mass points , for which error bars can not be evaluated , are also omitted . filled circles yield @xmath13 @xmath167 . @xmath11 , empty circles yield @xmath13 @xmath167 . @xmath35 . filled circles nicely fit the line worked out from the average density contrast @xmath15 ; only at high masses , where the statistics is poorer , some of the circles are far from this line . the same line is also a reasonable fit for the open circles . these , however , are systematically farther at all mass scales . in the figure , however , we also plot an horizontal line showing the average of the @xmath13 values of all clusters with mass @xmath168 . even this average can be considered a reasonable fit for open circles ( an unweighted @xmath156 evaluation gives similar probabilities to both fitting lines ) . this plot shows that there are actual dangers , if quantities worked out assuming a constant @xmath18 are compared with real data . in this case , one might tentatively suggest that there is a typical , mass independent , cluster radius @xmath169mpc , quite close to the abell value . our procedure , instead , outlines that this arises because some clusters were attributed a biased mass and the corresponding points were then set at a wrong abscissa . although , altogether , this abscissa reshuffling is substantially casual , the danger that an characteristic scale erroneously appears is far from absent . a critical result of our analysis is however shown in figs . 11 and 12 ( for simulations a and b , respectively ) . in simulations , numerical " cluster masses can be easily evaluated by summing up particle masses . the actual approach , in the real world , ought to be quite different and is based , first of all , on an analysis of galaxy velocities or x - ray emitting gas temperature . these analysis make use of the virial hypothesis ; therefore , measured masses of galaxy clusters are estimates of their _ dynamical _ masses . it is therefore important to test how numerical " cluster masses in simulations agree with the masses @xmath170 evaluated averaging over the velocities of the particles within @xmath171 . this comparison has been often performed , showing a reasonable coincidence between @xmath35 and @xmath12 . the results of such fit are shown in figs . 11 and 12 by the dashed histograms , which confirm the slight excess of @xmath12 @xmath167 . @xmath35 already noticed by previous authors . in the same plots we also show the results of a fit between @xmath12 and @xmath11 , obtained on the basis of the setting of the bl . there seems to be little doubt that the latter fit is better . the average values of @xmath172 are both @xmath173 , quite close to unity and consistent with it at the 1@xmath151 level . we have also compared the shapes of the cluster ( integral ) mass functions @xmath174 and @xmath175 . in fig . 13 and 14 , upper and lower plots show their behaviors , respectively . in the same plots , we also show press & schechter and sheth & tormen theoretical curves . they can be obtained from two different expressions of @xmath176 here @xmath177 is matter density , the bias factor @xmath178 is the ratio of the critical overdensity in the spherical growth model to the rms density fluctuation on the length scale corresponding to the mass @xmath21 , and @xmath179 is the differential mass function , which shall then be integrated to obtain @xmath180 . in the press & schechter formulation , assuming spherical unperturbed fluctuation growth , @xmath181 while sheth & tormen , taking into account the ellipsoidal nature of the collapse , argue that @xmath182 with @xmath183 ; this expression introduces the parameters @xmath148 , @xmath184 and @xmath58 , however constrained by the requirement that all matter is bound in objects of some mass @xmath21 , however small , and therefore the integral of @xmath185 must be unity . in our plots , @xmath148 is normalized to yield the number of clusters found in simulations , and this constraint is automatically satisfied . sheth & tormen ( 1999 ) showed that , in the gif simulations for an scdm , 0cdm and @xmath46cdm models ( see kauffmann et al 1999 ) , a good fit to the numerical mass function was obtained if @xmath186 , @xmath187 . in fig . 13 and 14 we report the mass function obtained from eq . ( 5.4 ) for these values , for tcdm and @xmath46cdm , respectively . the fit with the @xmath175 is surely better than using eq . we also tried to modify the parameter @xmath58 , that , according to sheth & tormen arguments , is related to the density contrast of virialized structures . as a matter of fact @xmath174 is derived with a variable density contrast , as explained in previous sections , and , however , with an average final density contrast which is @xmath19 below @xmath18 . it is therefore reasonable to expect that a lower value of @xmath58 is needed to compare theoretical predictions with the numerical behavior of @xmath174 . in fig . 13 and 14 , we show the expression ( 5.4 ) , both for @xmath188 and 0.5 . it is also evident that , while @xmath187 provides a good fit for @xmath35mass functions , the fit obtained provided by @xmath189 for the @xmath11mass function is perhaps better . the results of these comparisons , altogether , are that : ( i ) no major variations in the numerical mass function arises when @xmath35 is replaced by @xmath11 . ( ii ) variations however exist , which are of the same order of the differences between the standard theoretical expression ( 5.3 ; press & schechter ) and the improved theoretical expression ( 5.4 ; sheth & tormen ) . ( iii ) when the latter expression is used , the fits to the numerical mass functions are best if different values of the parameter @xmath58 are used for @xmath35 and @xmath11 ( @xmath1900.7 and 0.5 , respectively ) . the differences between the mass functions @xmath175 and @xmath174 are more evident in the intermediate mass range ( @xmath191 ) . they arise from a widespread re - shuffling " among the masses of dm halos , when the two definitions are used , causing thereby different numbers of halos in different mass intervals . this can also be seen from fig . 10 , where the mean cluster radii ( @xmath13 , @xmath34 ) of halos in this mass range show appreciable differences . observational work on galaxy clusters made a giant leap forward thanks to x ray observations and , in turn , cluster data are providing more and more cosmological information . the rosat all sky survey ( rass ) allowed the extraction of large samples of galaxy clusters , like reflex ( boehringer et al . 2001 ) or noras ( viana et al . 2001 ) , tracing the large scale structure ( lss ) of the universe up to @xmath192mpc of distance ( @xmath45 is the hubble constant in units of 100 km / s / mpc ) . these data are consistent with the spectrum of a low density cdm model , with matter density parameter @xmath193 . also an analysis of the redshift dependence of the x ray cluster luminosity function , based on a press & schechter approximation , sets a relation between @xmath17 and @xmath43 , which requires again @xmath194 , 0.6 ( @xmath17 : matter density parameter ; @xmath43 : squared mean density contrast , on the scale of 8@xmath37mpc ) ( borgani et al . such constraints on the model will become more stringent , in the forthcoming years , thanks to xmm newton and chandra data , providing high@xmath32 cluster samples with tenths of thousands objects ( see , e.g. , bartlett et al . 2001 ) . in our opinion , such fast improvements of the observational materials deserve to be matched by a suitable effort of the theoretical analysis . in this work we try to contribute to it , by providing an improved pattern to measure individual cluster densities , radii and masses in numerical simulations , without reference to a standard density contrast . although , both in real samples as well as in simulations , cluster members are directly observable together with their environs , a definition of cluster radii , masses , and density contrasts is needed both for statistical aims , when treating a large set of halos , and for individual objects , e.g. , to compare the observed mass distribution with analytical expressions of radial profiles . in the standard approach , a value for the density contrast is usually borrowed from analytical results , which are obtained under assumptions which can be incorrect in the real world , as well as in simulations . such assumptions , however , were explicitly considered to be the best possible choice . in particular , cole & lacey ( 1996 ) had explored the possibility to make use of the virial ratio , but considered its value in spheres , instead of layers . here we see that , passing from the integral to the differential virial ratio @xmath195 , allows the detection of a neat boundary , where a transition from halo to surrounding material occurs . in fact , first of all , ( i ) this inspection allows the detection of minimum points of @xmath123 in all halos , which could be easily missed in an integral inspection . some of them are safely shown not to be numerical features , as they persist without appreciable shifts when we compare different resolution simulations of a given halo . furthermore , ( ii ) we found that , chb quite close to the radius @xmath89 , where one of the persistent minima occurs , there is always a maximum of the function @xmath139 ; finally , chb ( iii ) at the radius @xmath89 , we have also that @xmath196 let us recall that , in general , @xmath139 is defined with reference to the behavior of the potential energy around @xmath89 , by setting : @xmath197 and , therefore , its @xmath22 dependence accounts for the radial dependence of @xmath1 . in this paper we have shown that the above properties ( i ) , ( ii ) , ( iii ) are consistent with the requirements that : ( a ) @xmath89 sets a neat boundary between bound and unbound material . ( b ) a significant interval ( depth @xmath198100@xmath37kpc ) , where @xmath139 is constant , lies about @xmath89 . we called it a boundary layer ( bl ) . accordingly , the @xmath22 dependence of @xmath1 , inside of the bl , is expressed by eq . ( 6.1 ) , by replacing the exponent @xmath139 with its constant value @xmath199 . the relation between the observed features ( i ) , ( ii ) , ( iii ) and the properties ( a ) , ( b ) , was shown under the assumption of spherical symmetry . the fact that the conditions ( i ) , ( ii ) , ( iii ) are not satisfied exactly , but with a ( slight ) approximation , can be easily attributed to deviations of numerical structures from sphericity . however , such deviations ought to be quite mild , owing to the statistics we performed on a large number of halos , and this indicates that violent relaxation erased most previous @xmath58spherical features . spherical collapses , however , leave a substantial imprint on the final density contrasts @xmath15 found for virialized halos whose spread accounts for different initial 3d geometries of fluctuations and different interactions with the environment during the collapsing stages . the simulations used to focus the above properties , at different levels of resolution , in tcdm and @xmath46cdm modes , were run both with the gadget and the art codes . results do not depend on the code . we could also extend the analysis , at the lowest resolution level , to simulations run in quite large boxes ( 360@xmath37 mpc aside ) , where we had a large statistics of clusters ( @xmath200 with @xmath201 , for each model ) . we applied to all of them an @xmath38requirement , meant to detect the bl , and found it in 97@xmath161 of the clusters considered . _ a detailed inspection of these clusters shows that the mass @xmath11 within the bl substantially coincides with the mass @xmath12 , evaluated from the velocities of all particles within @xmath202 , according to the virial theorem . _ furthermore , a direct inspection of the clusters without a bl , showed they were exceptional structures , mostly due to undergoing major merging , which certainly justified a dynamically unsettled condition and a badly @xmath58spherical geometry . let us however notice ( see fig . 5 ) that `` standard '' @xmath58sphericity of halo structures does not appear to be a serious obstacle to bl detection . at present , model clusters to fit x ray data are mostly built using hydro codes . this allows substantial improvements with respect to initial analyses based on pure newtonian dynamics . there are , however , some features in the interpretation of both kinds of simulations , which rely on a sound definition of the virialization radius @xmath203 . this radius , whose setting immediately follows from the detection of the bl position , corresponds to a wide spread of density contrasts @xmath15 , and , in general , is significantly different from the radius @xmath204 , obtained using a given density contrast @xmath18 . even keeping to models without hydrodynamics , it is clear that using different radii @xmath203 and density contrast @xmath15 could lead to different relations between @xmath24 , @xmath23 and @xmath11 . as a consequence the one to one correspondence between @xmath11 and @xmath23 could be lost and it would no longer be granted that a mass function and a temperature function are substantially equivalent quantities . we thank inaf for allowing us the cpu time to perform the art simulation c and d at the cineca consortium ( grant cnami44a on the sgi origin 3800 machine ) . gadget simulations of high resolution clusters have been run on the 16 linux pc beowulf cluster at the osservatorio astronomico di torino . alessandro gardini is thanked for allowing us to make use of the simulations a and b ; we also wish to thank him , loris colombo and giuseppe tormen for useful discussions during the preparation of this paper . bartlett j. , et al . , 2001 `` galaxy clusters and the high redshift universe observed in x - rays '' proceedings of the xxi moriond conference , ( preprint astro - ph/0106098 ) boehringer , h. , et al . , 2000 , apjs , 129 , 435 b borgani , s. , et al . , 2001 , apj , 561 , 13 b brian , g. & norman , m. , 1998 , apj , 495 , 80 carlberg , r.g . , et al . , 1996 , apj , 462 , 32 cole , s. & lacey , c. , 1996 , mnras , 281 , 716 couchman , h. m. p. , 1991 , apj , 268 , l23 doroshkevivh , a.g . , kotok , e. , novikov , i.,polyudov , a.n . , shandarin , s.f . & schctman , s. , 1980 , mnras , 192 , 321 eke , v.r . , cole , s. , @xmath205 frenk , c.s . , 1996 , mnras , 282 , 263 eke , v. , navarro , j. & frenk , c. , 1998 , apj , 503 , 569e gardini , a. , murante , g. @xmath205 bonometto , s. a. , 1999 , apj , 524 , 510 ghigna s. , moore b. , governato f. , lake g. , quinn t. & stadel j. , 2000 , apj 544 , 616 girardi m. , borgani s. , giuricin g. , mardirossian f. & mezzetti m. , 1998 , apj , 506 , 45 g goldstein h. , poole c. & safko , 2002 , classical mechanics , 3rd ed . addison wesley pub . gott , r. & rees , m. 1975 , a&a , 45 , 365 g governato f. , babul a. , quinn t. , tozzi p. , baugh c.m . , katz n. , @xmath205 lake g. , 1999 , mnras , 307 , 949 kauffmann g. , colberg j. , diaferio a. & white s.d.m . 1999 , mnras , 303 , 188 klypin a. , kravtsov a. , bullock j. & primack j. , 2001 . apj 554 , 903 knebe a. , kravtsov a. , gottlber s. & klypin a. , 2000 mnras , 317 , 630 kravtsov a. , klypin a. & khokhlov a. , 1997 apj , 111 , 73 k lacey , c. & cole , s. , 1993 , mnras , 262 , 627 lacey , c. & cole , s. , 1994 , mnras , 271 , 676 lahav , o. , lilje , p.r . , primack , j.r . & rees , m. , 1991 , mnras , 282 , 263e macci a. v. , gardini a. , ghigna s. & bonometto s. a. , 2002 , apj , 564 , 1 m monaco , p & murante , g. , physical review d , 60/10 , i d . 103502 , 1999 navarro j. , frenk c. & white s.d.m . 1996 , apj , 462 , 563 navarro j. , frenk c. & white s.d.m . 1997 , apj , 490 , 493 peebles p.j.e . , 1980 , the large scale structure of the universe , princeton university press , princeton power c. , navarro j. , jenkins a. , frenk c. , white s.d.m . , springel v. , stadel j. & quinn t. , 2002 , mnras ( submitted ) astro ph/0201544 press w.h . & schechter p. , 1974 , apj , 187 , 425 sheth r.k . , mo h.j . & tormen g. , 2001 mnras , 323 , 1 sheth r.k . & tormen g. , 1999 mnras , 308 , 119 sheth r.k . & tormen g. , 2002 mnras 329 , 61 schueker p. , caldwell r.r . , boehringer h. , collins c.a . & guzzo l. , astro - ph/0211480 and a&a ( in press ) springel v. , yoshida n. & white s.d.m . , 2001 newa , 6 , 79 s viana p. , nichol r. & liddle a. , 2002 , apj , 569l , 75 , v zeldovich , ya . b. , 1970 a&a , 5,84 let us compare this term with the potential energy of the layer @xmath217 also evaluated assuming spherical symmetry and neglecting terms of higher order in @xmath210 . the ratio between @xmath218 terms and potential energy is @xmath219 and would vanish for @xmath220 . let us now assume to be dealing with halos whose profile is @xmath221 with @xmath222^{-1 } ~ , \eqno ( a4)\ ] ] where @xmath223 is the scale radius of the profile and the concentration of the halo is @xmath224 . this profile is found for most halos in our simulations , up to @xmath225 , with a value of @xmath226 in the interval 47 and values of @xmath227 ; this agrees with previous numerical analyses ( see , e.g. , navarro et al . 1996 , 1997 , ghigna et al 2000 , klypin et al 2001 , power at al 2002 ) , also extended to smaller values of @xmath22 , finding values of @xmath228 ranging between 0.8 and 1.2 . using this profile we can evaluate the @xmath22 dependence of @xmath213 for @xmath228 values in the above interval . such values are shown in fig . 15 and indicate that for @xmath2297 , @xmath2300.3 . this shows that the contribution of the pressure terms to the virial balance is expected to be significantly smaller than the contribution of the potential term . let us however add some further comments : \(i ) using the profile in eq . ( a4 ) for @xmath231 is an approximation . an inspection of numerical halos , already significantly inhomogeneous at such radii ( see , e.g. , fig . 5 ) , shows that the slope of the profile is often smoother there . \(ii ) accordingly , owing to eqs . ( a2)(a3 ) , the @xmath218 terms can be absorbed in a factor ( @xmath232 ) , set in front of the potential term of the virial balance . this correction does not cause any displacement of the point where @xmath123 has its minima . on the contrary , it may interfere with the condition @xmath233 , which , in principle , can be suitably improved . \(iii ) deviations from spherical symmetry , however , are far from being fully negligible and the discrepancies between @xmath123 and @xmath113 , found in halo analysis , show that a correction of @xmath123 by @xmath234 , would not improve the efficiency of the method . we tried to find a bl , i.e. to apply the @xmath38 requirement , also in two n body simulations aimed at studying the evolution of a galactic stellar disk in a cosmological context . these simulations , performed with gadget , will be discussed in a forthcoming paper ( murante , curir & mazzei , in preparation ) . here we used simulations with dm only ( no galactic disk is present ) . in the first one , initial conditions were set up with the same multi mass technique described in sec . 2 , using the art package . a dm halo , whose mass is @xmath235 at redshift @xmath69 , was simulated at high resolution using dm particles of mass @xmath236 . according to eq . ( 1.1 ) , the virial radius @xmath237 mpc . external forces were taken into account by setting heavier and heavier particles in 3 concentric shells . we checked that no intruders ( heavier mass particles ) were ever closer than 0.5 @xmath238mpc from the center of mass of the halo . the @xmath38 requirement was applied to this halo , at redshifts @xmath69 and @xmath239 . at @xmath69 a bl was found , with a radius @xmath240 mpc , yielding a ratio @xmath241 . on the contrary , at z=2 , we could not find a good fit for the @xmath156 parameter used to implement the @xmath38 requirement . it should be noticed that these results held , in spite of the fact that the halo concentration is @xmath242 . this suggests that bl s are found by the @xmath38 requirement also when the condition @xmath243 is ( mildly ) violated ; on the contrary , no bl is found when the _ the halo is not yet virialized _ , as it was at z=2 and is also confirmed by direct inspection . apparently , the presence of a bl however distinguishes gravitationally relaxed objects from those still in an evolutionary phase . a partial confirm of these two results comes from the application of the @xmath38 requirement to a second simulation , following the evolution of an isolated halo of nominal mass equal to the previous one , with a nominal radius of @xmath244mpc , generated with a nfw density profile ( see eq . a4 ) and with initial particle velocities picked up from a multivariate maxwellian distribution ; the nfw density profile was cut off at the nominal radius ; the concentration @xmath226 was set to the value measured in the previous case ( using @xmath34 to evaluate it , as in navarro et al , 1997 ) . the halo was evolved for @xmath245gyr , to reach the age of the universe at @xmath239 in a @xmath46cdm cosmology where @xmath246 and @xmath247 . we applied the @xmath38 algorithm , to find a bl at the beginning ( @xmath248gyr ) and at the end ( @xmath249gyr ) of the simulation . no bl could be found at @xmath250 , when the halo was _ not _ in equilibrium . apparently , therefore , the presence of a bl is characteristic of gravitationally stable objects , independently of the geometrical distribution of particles . at @xmath250 , a so algorithm , based on a purely geometrical prescription , would comfortably detect a halo . the @xmath38 requirement , instead , which is based on dynamical prescriptions , appears to be more selective than purely geometrical recipes . at @xmath251 gyr , we find a layer fulfilling the @xmath38 requirement , with @xmath252mpc , yielding @xmath253 . this happens in spite of the value of the density contrast at @xmath13 , which is quite large ( @xmath254 ) . in this ( quite peculiar ) case , the definition of virialization based on the density contrast fails , in spite of the fact that gravitational equilibrium is established . these arguments seem to indicate that a technique based on the @xmath38 requirement is applicable on a wide range of dm halo masses , even when @xmath226 is fairly large ; mild deviations from the requirement that @xmath255 is constant do not seem to destroy its effectiveness . further analysis will be performed to put precise limits to the cases when the @xmath38 criterion is a useful tool for telling apart virialized halos from still strongly evolving ones .

we present commissioning data for the polish instrument obtained on the hale 5-m telescope . the goal of this high precision polarimeter is to constrain orbital inclination of high mass x - ray binaries and to therefore obtain independent mass estimates for their black hole companions . we have obtained photon shot noise limited precision on standard stars , and we have measured the polarization of bright stars at the part per million level on a nightly basis . systematic effects have been reduced to less than of the measured polarization for polarized sources and to the part per million level for weakly polarized sources . the high sensitivity of this instrument to asymmetry suggests that valuable contributions will be made in many other fields , including studies of extrasolar planets , debris disks , and stellar astrophysics .

0810.5561

while most astrophysical objects require many parameters in order to be fully described , black holes are unique in that only three parameters suffice : mass , spin , and charge . mass and spin describe the black hole s gravitational field and event horizon location . therefore , black holes provide a rare opportunity for theory and observation to jointly pursue a few quantities to completely describe one of the most exotic kinds of objects in the universe . + though modeling of the x - ray spectral energy distribution is somewhat effective in constraining black hole spin ( mcclintock et al . 2006 ) , important constraints on black hole mass exist in the case of high mass x - ray binaries ( hereafter hmxbs ) . these binaries consist of an o or b type supergiant and a black hole or neutron star . the most well - studied of these , cygnus x-1 , is thought to consist of a @xmath1 @xmath2 , o9.7iab star and a @xmath3 @xmath2 black hole at a distance of @xmath4 kpc ( zilkowski 2005 ) . while the constraints on the mass of the compact object are tight enough to declare that it is a black hole , they are insufficient to permit precise modeling of the progenitor star s mass . we have commissioned a polarimeter on the hale 5-m telescope at palomar observatory in california to provide an independent method for determining black hole mass . this polarimeter has the potential to constrain the mass of the cygnus x-1 black hole to a few solar masses a wealth of radial velocity data exists for cygnus x-1 ( gies et al . 2003 ) and other hmxbs . however , in the same way that precise masses are elusive for non - transiting extrasolar planets , determination of precise black hole mass is hindered by unknown orbital inclination . this is evidenced by the mass function , which is the end product of radial velocity observations . for cygnus x-1 , gies et al . ( 2003 ) quote the following value for the mass function : @xmath5 here , @xmath6 is the mass of the black hole , @xmath7 is the mass of the visible binary component , @xmath8 is the system inclination , and @xmath9 is the mass ratio of the visible component to the black hole . thus , an observational technique that constrains orbital inclination can take advantage of radial velocity data and offer an estimate of black hole mass . + since system polarization is a geometric effect , the polarization of an hmxb system can be used to determine geometric information about the system , such as orbital inclination . the effective temperature of the supergiant in an hmxb is @xmath10 30,000 k , which is hot enough to ionize photospheric hydrogen . this generates a high density of free electrons that thomson - scatter emitted light from the supergiant . while net linear polarization from a spherical cloud of free electrons is zero , asymmetry in the system causes net polarization ( brown & mclean 1977 ) . the tidal effects of the black hole cause such an asymmetry in the circumbinary envelope , and the orbital modulation of polarization is the key to determining orbital inclination . for instance , consider a face - on hmxb with zero eccentricity and an optically thin circumbinary envelope ( figure [ 2myfiga]a ) . the total amount of observed , polarized light is independent of orbital phase , and the degree of polarization is therefore constant . however , the angle of net polarization , which is perpendicular to the scattering plane , rotates as the binary progresses in its orbit . + in contrast , for a nearly edge - on geometry the degree of polarization varies significantly throughout the orbit , while the angle of net polarization is roughly constant ( figure [ 2myfiga]b ) . therefore , the modulation of the degree and angle of net polarization is a unique measure of orbital inclination for synchronously rotating hmxbs . combining equations 1 and 2 from friend & cassinelli ( 1986 ) and equation 2 from brown et al . ( 2000 ) , the polarization of an optically thin , axisymmetric envelope due to thomson scattering is @xmath11 here , @xmath12 is the thomson scattering cross section , @xmath13 is the cosine of the polar angle , the electron number density as a function of radial coordinate @xmath14 is @xmath15 , @xmath16 is a scaling constant , stellar radius is @xmath17 , system inclination is @xmath8 , and orbital phase in radians is @xmath18 . two polarization periods occur per orbital period because of the @xmath19 term . + this technique has been utilized by a few groups ( kemp et al . 1979 ; dolan & tapia 1989 ; wolinski et al . 1996 ) , and cygnus x-1 has been found to have variable polarization of order @xmath20 with respect to its unpolarized flux . however , measurement precision from the above groups is of order one part in @xmath21 . the derived inclination estimates were questioned by the community on the grounds that significant underestimation of error occurred because of limited measurement precision ( milgrom 1979 ; aspin et al . measuring inclination to @xmath22 requires polarimetric precision of one part in @xmath23 to one part in @xmath24 , depending on system inclination @xcite . this requires at least @xmath25 to @xmath26 detected photons , which necessitates the use of large telescopes . since our system combines a high precision instrument with a 5-m telescope , we aim to measure the polarimetric variability of cygnus x-1 to better than one part in @xmath21 . + polarimeters require the following fundamental components : a polarization modulator , analyzer , detector , and demodulator . the modulator induces a known , periodic characteristic to the unknown polarization of the input beam . the analyzer converts modulation in polarization to modulation in the beam s intensity , since most detectors are sensitive to intensity and not polarization . finally , the demodulator extracts the component of the detector s output that varies at the known frequency of the modulator to reject noise . see figure [ 2myfigb ] for a block diagram , figure [ 2myfigc ] for a ray trace diagram , and figure [ 2myfigz ] for photographs of the instrument . + traditionally , the modulator is a rotating halfwave plate that rotates the linear polarization of the incident beam . the highest modulation frequency attainable with this component is of order 100 hz , which is not fast enough to freeze out atmospheric turbulence or to mitigate electronic @xmath27 noise . additionally , spatial inhomogeneity in both the retardance and cleanliness of the plate can introduce spurious signals , because the beam samples different sections of the halfwave plate at different times . the polarization goal for our instrument , polish ( polarimeter for inclination studies of high mass x - ray binaries / hot jupiters ) , is one part per million on bright , unpolarized stars , which necessitates the use of a photoelastic modulator ( hereafter pem , kemp 1969 ) . + a pem is a bar of optical materials ( fused silica for use in optical light ) in which a resonant acoustic signal at a frequency of tens of kilohertz is induced . the stress - birefringent property of the silica causes time - varying , sinusoidal retardance of the component of polarization oriented at @xmath28 with respect to the compression / extension axis , while the polarization components at @xmath29 and @xmath30 experience no retardance . that is , a pem at @xmath31 with respect to the stokes @xmath32 axis will cause retardance to the @xmath33 stokes parameter but not to the @xmath34 parameter . since peak retardance is a function of the amplitude of the acoustic signal , both peak retardance and wavelength of peak retardance can be easily modified . we set the retardance amplitude to @xmath35 waves at 500 nm wavelength to give uniform pem efficiency for both low and high linear polarizations . this also causes linear polarization to be directly proportional to the amplitude of the ac signal divided by the mean dc signal , which is derived in appendix a. + the high frequency pem modulation strongly reduces both atmospheric turbulence and electronic @xmath27 noise . additionally , the beam always samples the same optical path during the modulation , because the strain on the modulator element is only of order ten parts per million ( kemp 1969 ) . operationally , a pem is the opposite of a rotating halfwave plate : while the plate has a constant retardance and time - variable fast axis orientation , a pem has constant fast axis orientation but time - variable retardance . since the absolute value of the pem s retardance determines the polarization of the beam at any instant , compression and extension of the fused silica bar affect the beam identically . therefore , linear polarization is modulated at twice the frequency of the pem modulation . we use the i / fs50 pem and pem100 controller from hinds instruments , inc . the modulation frequency of this pem is 50 khz , and modulation of linearly polarized light occurs at 100 khz . + directly downstream from the pem is a 95r/5 t beamsplitter at @xmath36 incidence that allows @xmath37 of the stellar flux to fall on a xybion ccd camera for on - axis guiding , while the remaining @xmath38 is reflected into a wollaston prism toward the detectors . the beamsplitter has a 50-mm diameter , fused silica substrate from edmund optics with a custom 400 to 700 nm coating from opticorp , inc . the surface accuracy on the substrate is one - tenth of a wavelength . + we utilize a two - wedge , calcite wollaston prism from karl lambrecht , inc . , as the analyzer . this prism separates each component of a single stokes parameter into two beams . that is , the @xmath39 ( or @xmath40 ) component is split into one beam , and the @xmath41 ( or @xmath42 ) component is in the other beam . both beams have equal deviation of @xmath43 from the optical axis , which allows the optical layout to be symmetric with respect to the optical axis . a two - wedge prism is used because the larger beam deviation of a three - wedge design would cause the instrument package to be larger than necessary . the surfaces of the wollaston prism have an antireflection coating in the wavelength range of 400 to 700 nm , and the transmission in @xmath44 band is @xmath45 per surface . by injecting light through a linear polarizer with known fast axis orientation and then through the wollaston prism , we find that the left beam seen from the detectors is vertically polarized ( @xmath41 when projected onto the sky for a cassegrain ring angle of @xmath29 ) . the right beam is horizontally polarized ( @xmath39 at @xmath29 ring angle ) . + each wollaston beam then impinges on an @xmath46 , mgf@xmath47 antireflection - coated field lens from melles griot . these lenses image the telescope secondary mirror onto the detectors , and they ensure starlight is uniformly spread over the detector active area even in the presence of image wander . field stops are located in the image plane , after the field lenses , but these are not currently used because contamination of stellar polarization from the sky field is not significant . the beams reach the detectors with a diameter of @xmath48 mm , which underfills all detectors . + since cygnus x-1 has @xmath44 magnitude @xmath49 , but the polarization standard stars observed can be as bright as @xmath50 , polish has two interchangeable pairs of detectors . stars fainter than @xmath51 are detected at a higher signal to noise ratio with the pair of hamamatsu h9307 - 04 photomultiplier tubes ( pmts ) , while objects brighter than this will destroy the pmts . brighter stars are observed with custom - made advanced photonix sd197 - 70 - 72 - 661 ( red enhanced ) and sd197 - 70 - 74 - 661 ( blue enhanced ) avalanche photodiode modules ( apds ) . the high quantum efficiency of apds is desirable on bright stars to minimize photon shot noise , while the low dark current of pmts is desirable on fainter stars to minimize detector noise . + the detectors are not downstream from spectral filters , and they are integrated light detectors in both spatial and spectral senses . spatial resolution is unnecessary , as the angular size of the cygnus x-1 system is much smaller than the atmospheric seeing disk . spectral resolution , while desirable , would seriously degrade the precision attainable with this instrument . such resolution must be left for future generations of polish . neglecting detector quantum efficiency , the instrumental throughput is calculated to be @xmath52 , @xmath53 , @xmath54 , and @xmath55 in @xmath56 , @xmath44 , @xmath17 , and @xmath57 bands . the throughput of the telescope / instrument system , again neglecting detector response , is calculated to be @xmath58 , @xmath59 , @xmath60 , and @xmath61 in those bands . + the pmts are identical , side - on modules with active area dimensions @xmath62 mm . their quantum efficiencies are quoted from the manufacturer to be @xmath63 , @xmath64 , @xmath65 , and @xmath66 in @xmath56 , @xmath44 , @xmath17 , and @xmath57 bands . the fwhm of the pmt bandpass is 475 nm ( 200 to 675 nm ) , but atmospheric transparency is of course low shortward of about 400 nm . thus , the pmts detect broadband @xmath67 light . the pmt gain can be set by a potentiometer , and we set this gain to @xmath68 for all observations . the modules also have a @xmath69 khz bandwidth amplifier with transimpedance @xmath70 v / a . the quoted output noise voltage resulting from dark current is @xmath71 ( typical ) to @xmath72 @xmath13v ( maximum ) , which implies a noise equivalent power of @xmath73 to 0.13 fw@xmath74 . dark current is @xmath75 na . + the apds are optimized for 100 khz operation and have 5 mm diameter , circular active areas . the apds are not identical , as one beam is sampled by the red enhanced module while the other is sampled by the blue enhanced one . the quantum efficiencies for the red enhanced module are quoted as @xmath54 , @xmath76 , @xmath77 , and @xmath78 in @xmath56 , @xmath44 , @xmath17 , and @xmath57 bands ( figure [ 2myfigq ] ) . the fwhm of the bandpass for the red enhanced apd is 550 nm ( 350 to 900 nm ) , which corresponds to broadband @xmath79 light . the blue enhanced module is quoted to have @xmath80 , @xmath81 , @xmath82 , and @xmath83 quantum efficiencies . the fwhm of the bandpass for the blue enhanced apd is about 500 nm ( 350 to 850 nm ) , which corresponds to broadband @xmath84 light . the red module operates with an observed gain of @xmath85 , and the blue module operates at a quoted gain of @xmath86 . transimpedance is @xmath87 v / a for both modules , and amplifier bandwidth is @xmath88 khz for the red enhanced and @xmath89 khz for the blue enhanced modules . after the apd chip is thermoelectrically cooled to @xmath29 c , dark current is measured to be @xmath90 na and 3.5 na at the output of the red and blue modules , respectively . therefore , the noise equivalent power for each module is @xmath91 fw@xmath74 and 9.7 fw@xmath74 , respectively . each detector is supplied @xmath92v and @xmath93v , and a 12v case fan blows heat from the apd heat sinks . + the demodulator picks out the component of the detected signal that varies at the reference frequency and ideally rejects signals at all other frequencies . the demodulator can either be software or hardware ; polish makes use of one stanford research sr830 digital , dual - phase lock - in amplifier for each detector . the pem controller sends a square wave reference signal to the lock - in amplifiers at twice the frequency of the pem modulation , and each lock - in amplifier recovers @xmath94 ( in phase with reference signal ) and @xmath95 ( @xmath30 out of phase with the reference signal ) components of the detector signal . together , @xmath94 and @xmath95 determine amplitude @xmath17 and phase @xmath96 according to the following : @xmath97 @xmath98 the lock - in amplifiers record the rms components of the in - phase and quadrature phase signals , so multiplication by a factor of @xmath99 is necessary to determine the amplitude of the ac signal . the notation of the argument of the arctangent is meant to account for the signs of @xmath94 and @xmath95 when determining phase . + signal phase allows direct measurement of the sign of each stokes parameter ( e.g. , @xmath39 versus @xmath41 ) . this is important , because insensitivity to sign would preclude direct measurement of more than @xmath30 of rotation of the cygnus x-1 system . both ac and dc signals from the detector must be recorded to measure polarization ( appendix a ) . the ac signals are recorded by the lock - in amplifiers , and each detector s dc signal is recorded by a separate hp 34401a digital voltmeter . to reject 60 hz noise and its harmonics , each dc reading by the voltmeters consists of an integration over 10 power line cycles . thus , the voltmeters record data at 6 hz . the lock - in amplifiers may only sample the ac signal at powers of two in frequency , so we choose to record the ac data at 8 hz . the discrepancy in sampling rates between ac and dc data is not important , because ac data should be normalized by mean dc data and not in a point - by - point fashion . + in order to nyquist sample the ac data , we set the lock - in time constants to 30 ms . this gives an effective noise bandwidth of enbw @xmath100 hz . therefore , we sample the ac data @xmath101 times per effective time constant , which satisfies the nyquist criterion and reduces aliasing . the lock - in amplifiers thus measure the components of the ac signal with frequencies bounded by @xmath102 , where @xmath103 is the reference frequency in hz . the auxiliary dc output of one lock - in amplifier is connected to a chopping motor on the telescope secondary mirror , which allows the secondary mirror to chop north to a sky field for sky subtraction of both ac and dc data . + polish is located at cassegrain focus to ensure beam reflections of about @xmath104 , and it samples the @xmath105 beam . these qualities minimize telescope polarization . to minimize instrument polarization , the pem is the first instrumental optic encountered by the beam . the lock - in amplifiers and voltmeters are controlled by a laptop , which is mounted to the instrument , via the gpib interface . matlab r2006a from the mathworks , inc . , is used to control the voltmeters and lock - in amplifiers , chop the secondary mirror , and rotate the cassegrain ring to allow access to both linear stokes parameters . a similar , albeit larger , instrument named planetpol is mounted on the 4.2-m william herschel telescope in la palma , spain ( hough et al . 2006 , hereafter hlb06 ) . the goal of this instrument is to detect the modulation of linear polarization caused by stellar flux scattered by hot jupiters . this observation requires polarimetric precision of one part per million to one part in ten million ( seager et al . 2000 , stam et al . 2004 ) , which is a precision nearly achievable with planetpol on the brightest stars . we observed many of the polarized and unpolarized standard stars from hlb06 in addition to others from the combined polarimetric catalog of heiles ( 2000 ) . a list of the stars observed is given in table [ 2tablec ] , and polarization values in parenthesis represent the @xmath106 uncertainty in the last two digits of the mean value . polarization reference is given in the ref " column . @xmath44 band magnitude and spectral type of the @xmath107 cepheid star hd 187929 are from bastien et al . ( 1988 ) and oke ( 1961 ) , respectively . the spectral type of hd 212311 is from schmidt et al . all other positional and spectral information is from the simbad database . observations of cygnus x-1 itself will be given in a forthcoming paper . + c c c c c c c c c algenib & @xmath108 peg & 00 13 14.23 & + 15 11 00.9 & @xmath109 & 118.1(5 ) & 1 & 2.83 & b2iv + hd 7927 & @xmath110 cas & 01 20 04.92 & + 58 13 53.8 & @xmath111 & 94.0(5 ) & 1 & 5.01 & f0ia + hd 9270 & @xmath112 psc & 01 31 29.07 & + 15 20 44.8 & @xmath113 & 158(14 ) & 1 & 3.63 & g7iia + hr 5854 & @xmath114 ser & 15 44 16.07 & + 06 25 32.3 & @xmath115 & @xmath116 & 2 & 2.64 & k2iiib + hd 147084 & @xmath117 sco & 16 20 38.18 & @xmath11624 10 09.6 & @xmath118 & 32.1(3 ) & 1 & 4.55 & a4ii / iii + hd 154445 & sao 141513 & 17 05 32.24 & @xmath11600 53 31.7 & @xmath119 & 90.2(2 ) & 1 & 5.64 & b1v + u her & hd 156633 & 17 17 19.57 & + 33 06 00.4 & @xmath120 & @xmath116 & 1 & 4.80 & b1.5vp+ + @xmath108 oph & hd 161868 & 17 47 53.56 & + 02 42 26.3 & @xmath121 & 33.3(3.6 ) & 1 & 3.75 & a0v + hd 157999 & @xmath122 oph & 17 26 30.98 & + 04 08 25.1 & @xmath123 & 85.9(1.0 ) & 1 & 4.34 & k3iab + hd 187929 & @xmath112 aql & 19 52 28.37 & + 01 00 20.4 & @xmath124 & 94.2(1 ) & 2 & @xmath125 & ( f6.5@xmath116g2)ib + hd 204827 & sao 33461 & 21 28 57.70 & + 58 44 24.0 & @xmath126 & 59.0(1.1 ) & 1 & 8.00 & o9.5v + hd 212311 & sao 34361 & 22 21 58.55 & + 56 31 52.8 & @xmath127 & @xmath116 & 1 & 8.12 & a0v + hr 8974 & @xmath108 cep & 23 39 20.85 & + 77 37 56.2 & @xmath128 & @xmath116 & 2 & 3.23 & k1iv + [ 2tablec ] after the target star is acquired , a measurement with 15 seconds duration is initiated on both voltmeters simultaneously with a measurement for about 30 seconds on both lock - in amplifiers . for the second half of the 30 second lock - in amplifier measurements , the voltmeters transmit data to the laptop . the secondary mirror is chopped 25 arcsec north with respect to the target star for sky subtraction , and another set of voltmeter and lock - in amplifier measurements is begun . after this has completed , the target star is returned to the field of view and another set of measurements begins . an integration triplet " is defined to be an on - source , 30 second ac integration and a 15 second dc integration both before and after the same measurement on a sky field . sky levels in the optical are small compared to target star levels , so sky fields are observed with an asymmetric , 2:1 source to sky chop . see figure [ 2myfigo ] for a typical ac and dc measurement of the unpolarized hr 5854 , and see figure [ 2myfigp ] for the strongly polarized hd 204827 . the increase in ac level , and therefore in polarization , is striking between these stars . + since two light beams with perpendicular polarization orientations experience the same retardance when passed through the pem , the same polarization should be observed for the pem at @xmath31 with respect to the optical axis of the instrument . we rotate the pem between these two positions to investigate the systematics of the pem . the pem is mounted to a gear driven by a stepper motor with an 8:1 step ratio , where the center of the pem aperture is coincident with the rotation axis of the gear . each motor step corresponds to a rotation of the pem by @xmath129 . for a cassegrain ring angle of @xmath130 , the home position of the pem projects its compression / extension axis northeast onto the sky ( the @xmath131 direction ) this will be referred to as the pem @xmath132 " position . the pem @xmath133 " position causes this projection to be northwest on the sky ( the @xmath134 direction ) . + we rotate the cassegrain ring and instrument through @xmath135 to investigate rotation systematics , and pem systematics are investigated by rotating the pem to the @xmath28 positions ( sections 6.2 and 6.3 ) . the precision of the cassegrain ring angle is @xmath136 . a standard observing sequence begins with the cassegrain ring at @xmath137 and an integration triplet at the pem @xmath132 position followed by a triplet at the pem @xmath133 position . the ring angle is then decremented by @xmath138 , at which point pem @xmath133 and pem @xmath132 triplets are taken . this process occurs for each target star for cassegrain ring angles of @xmath139 in @xmath138 increments to sample all @xmath140 and @xmath141 stokes components . + the next star will see the ring angles begin at @xmath142 and end at @xmath137 . the endpoints of @xmath143 ensure that the ring will not wind up " and be forced to de - rotate during an observing sequence , wasting observing time . after each triplet , either the pem or cassegrain ring is rotated but not both . each standard star is generally given eight cassegrain ring rotations ( @xmath135 ) at two pem positions each ( @xmath144 ) , and two chop integrations are taken at each pem position . thus , most standard stars receive about 16 minutes of ac data and about 8 minutes of dc data per night . mean @xmath94 , @xmath95 , and dc values for each detector are found for all on - source and sky measurements . the mean on - source values are then subtracted by the mean sky values . assuming stokes @xmath33 is observed , the polarization is calculated by the following ( appendix a ) : @xmath145 the efficiency of the pem , @xmath146 , is the strength of the ac signal based on the choice of pem peak retardance , and it is derived in appendix a. for polish , this efficiency is @xmath147 . the sign of the final polarization of each on - source measurement is multiplied by the sign of the stokes parameter measured . that is , the sign of the calculated stokes parameter is calibrated by the phase from the lock - in amplifiers ( equation [ 2eql ] ) . + to calibrate absolute polarization measured with polish , we compare our measured polarization degree with the values of heiles ( 2000 ) for all stars ( section 8) . gain factors of @xmath148 and @xmath149 are therefore multiplied to apd and pmt derived polarizations , respectively . since the bandpasses of the red and blue enhanced apds are nearly identical , the polarization gain factors for the apds are not significantly different . however , the large difference in bandpass between the apds and pmts causes the significant difference in the polarization gain factors between the apds and pmts . position angle of net polarization is not calibrated for , because our observations are generally consistent with those in the literature for strongly polarized targets . + nightly mean and run - averaged @xmath33 and @xmath34 for each source are determined by taking the weighted mean polarization of all data over the requested timescale . the weighting for each measurement is its sky - subtracted dc level multiplied by integration time . this value is proportional to the total number of detected photons and ensures that all detected photons , rather than all measurements , are weighted equally . this is important for data taken in partly cloudy conditions . the polarimetric uncertainty is the square root of the weighted variance divided by the square root of the number of measurements . it is important to note that precision calculated in this manner is only applicable to stars with no intrinsic polarimetric variability . analyses of the variability of the observed stars , including cygnus x-1 , will be made in a future paper . + expected photon shot noise , detector noise , and observed noise are derived in appendix b as equations [ 2eqau ] through [ 2eqat ] : @xmath150 \right\ } ^{1/2 } \label{2eqau}\ ] ] @xmath151 \right\ } ^{1/2}\ ] ] @xmath152 \right\ } ^{1/2 } \label{2eqat}\ ] ] @xmath153 here , @xmath154 , @xmath155 , @xmath156 is the electron charge , @xmath157 hz is the bandwidth of the lock - in amplifiers , and @xmath158 is the integration time of the lock - in amplifiers ( figures [ 2myfigo ] and [ 2myfigp ] ) . the values @xmath159 , @xmath160 , @xmath56 , @xmath161 , and @xmath162 are the detector s output dark current , excess noise factor , bandwidth , gain , and transimpedance , which are listed in table [ 2tablea ] . a perfect detector will have noiseless gain , @xmath163 , and dark current @xmath164 . consultation of @xmath160 for both detector types shows that gain from the pmts is an order of magnitude less noisy than from the apds , which is why pmts are preferred over apds for faint objects . each of the uncertainties @xmath165 represents the sample standard deviation of @xmath94 , @xmath95 , or dc of the source added in quadrature to that of the sky . the values @xmath94 , @xmath95 , and dc in equation [ 2eqat ] are sky - subtracted . + c c c c c c blue apd & 300 & 0.138 & @xmath166 & 90 & 3.5 + red apd & 220 & 0.138 & @xmath166 & 100 & 5.6 + pmt & @xmath167 & 0.013 & @xmath168 & 200 & 0.1 + [ 2tablea ] from table [ 2tablec ] , the polarizations of both hr 5854 ( hd 140573 ) and hr 8974 ( hd 222404 ) are close to zero , which makes them candidates for being truly unpolarized sources . individual , shot noise limited measurements of hr 5854 and hr 8974 , before subtraction of telescope polarization , are plotted in figure [ 2myfigh ] . individual measurements for both stars lie within error of each other , and nightly mean stokes parameters for both stars do not generally differ significantly . the equatorial mount of the hale 5-m inhibits traditional telescope polarization measurement , which involves allowing the field to rotate and determining the center of the @xmath169 locus . however , since hlb06 performed this analysis and claim part per million polarization of hr 5854 and hr 8974 , we therefore assume that these stars are indeed unpolarized . lucas et al . ( 2008 ) report stability of hr 5854 after one year at the level of @xmath170 parts per million , indicating only weak evidence for part per million variability . + nightly and run - averaged telescope polarization are calculated by the weighted mean polarization from hr 5854 and hr 8974 , and they are listed in table [ 2tablee ] as tp " . uncertainties in these estimates are given as the square root of the weighted variance of the individual measurements divided by the square root of the number of measurements . the light beam from the secondary mirror impinges directly on the pem , so we assume that instrumental polarization is negligible . indeed , the similar instrument planetpol has an intrinsic polarization of only a few parts per million ( hlb06 ) . + c c c c c c 2007 aug 3 & hr 5854 & @xmath116195.1(1.5 ) & + 161.9(2.5 ) & 253.5(2.0 ) & 70.16(25 ) + 2007 aug 4 & @xmath171 & @xmath116196.6(2.3 ) & + 156.9(2.9 ) & 251.6(2.5 ) & 70.70(31 ) + 2007 aug 5 & @xmath171 & @xmath116198.8(2.5 ) & + 159.4(2.8 ) & 254.8(2.6 ) & 70.65(30 ) + 2007 aug 6 & @xmath171 & @xmath116185.2(4.9 ) & + 142.8(3.2 ) & 233.9(4.3 ) & 71.17(48 ) + overall & @xmath171 & @xmath116195.3(1.5 ) & + 156.5(1.7 ) & 250.2(1.6 ) & 70.64(19 ) + 2007 aug 3 & hr 8974 & @xmath116202.9(2.7 ) & + 163.7(1.9 ) & 260.8(2.4 ) & 70.55(25 ) + 2007 aug 4 & @xmath171 & @xmath116173(31 ) & + 149.7(2.9 ) & 229(24 ) & 69.6(2.6 ) + 2007 aug 5 & @xmath171 & @xmath116170.0(5.8 ) & + 140.2(5.0 ) & 220.3(5.5 ) & 70.24(69 ) + 2007 aug 6 & @xmath171 & @xmath116189.1(4.7 ) & + 170.2(4.5 ) & 254.4(4.6 ) & 69.00(52 ) + overall & @xmath171 & @xmath116189.3(3.1 ) & + 156.2(2.4 ) & 245.5(2.9 ) & 70.24(32 ) + 2007 aug 3 & tp & @xmath116198.8(1.8 ) & + 162.6(1.7 ) & 256.8(1.8 ) & 70.37(19 ) + 2007 aug 4 & @xmath171 & @xmath116195.8(2.4 ) & + 155.7(2.4 ) & 250.1(2.4 ) & 70.75(27 ) + 2007 aug 5 & @xmath171 & @xmath116191.6(3.4 ) & + 154.6(2.9 ) & 246.2(3.2 ) & 70.55(36 ) + 2007 aug 6 & @xmath171 & @xmath116186.6(3.5 ) & + 149.0(3.2 ) & 238.9(3.4 ) & 71.70(40 ) + overall & @xmath171 & @xmath116193.5(1.5 ) & + 156.4(1.3 ) & 248.9(1.4 ) & 70.53(16 ) + [ 2tablee ] the cause of the large telescope polarization of the hale 5-m , @xmath172 , may be due to inhomogeneities in the coating of the primary and/or secondary mirrors . net stellar polarizations are shown in tables [ 2tableh ] and [ 2tablei ] after nightly telescope polarization is subtracted . weakly polarized stars ( @xmath173 ) are given in table [ 2tableh ] , while strongly polarized stars ( @xmath174 ) are listed in table [ 2tablei ] . since hr 5854 and hr 8974 are effectively unpolarized , uncertainty in polarimetric position angle @xmath175 is so large as to preclude meaningful estimates on @xmath175 . + c c c c c c 2007 aug 3 & hr 5854 & + 3.7(2.4 ) & @xmath1160.8(3.0 ) & 3.8(2.4 ) & @xmath116 + 2007 aug 4 & @xmath171 & @xmath1160.7(3.3 ) & + 1.9(3.8 ) & 2.0(3.8 ) & @xmath116 + 2007 aug 5 & @xmath171 & @xmath1167.7(4.2 ) & + 4.2(4.0 ) & 8.8(4.1 ) & @xmath116 + 2007 aug 6 & @xmath171 & + 1.7(6.0 ) & @xmath1166.8(4.5 ) & 7.1(4.6 ) & @xmath116 + overall & @xmath171 & @xmath1162.0(2.2 ) & + 0.3(2.2 ) & 2.0(2.2 ) & @xmath116 + 2007 aug 3 & hr 8974 & @xmath1163.2(3.3 ) & + 1.3(2.5 ) & 3.5(3.1 ) & @xmath116 + 2007 aug 4 & @xmath171 & + 22.6(2.4 ) & @xmath1166.1(3.8 ) & 23.3(2.5 ) & @xmath116 + 2007 aug 5 & @xmath171 & + 21.6(6.7 ) & @xmath11614.4(5.7 ) & 25.9(6.4 ) & @xmath116 + 2007 aug 6 & @xmath171 & @xmath1162.5(5.9 ) & + 21.6(5.5 ) & 21.7(5.5 ) & @xmath116 + overall & @xmath171 & + 4.7(3.5 ) & @xmath1160.1(2.8 ) & 4.7(3.5 ) & @xmath116 + 2007 aug 4 & hd 9270 & @xmath11633.9(5.1 ) & @xmath11676.0(4.0 ) & 83.1(4.3 ) & 123.0(1.7 ) + 2007 aug 5 & @xmath171 & @xmath11637.6(7.1 ) & @xmath11682.2(4.2 ) & 90.4(4.8 ) & 122.7(2.1 ) + 2007 aug 6 & @xmath171 & @xmath11643.0(8.4 ) & @xmath11681.6(6.9 ) & 92.1(7.3 ) & 121.2(2.6 ) + overall & @xmath171 & @xmath11635.0(3.5 ) & @xmath11681.8(2.7 ) & 88.9(2.8 ) & 123.4(1.1 ) + 2007 aug 4 & @xmath108 oph & @xmath11686.5(5.6 ) & + 134.9(4.3 ) & 160.3(4.7 ) & 61.33(94 ) + 2007 aug 5 & @xmath171 & @xmath11681.9(4.0 ) & + 136.9(4.9 ) & 159.5(4.7 ) & 60.45(76 ) + 2007 aug 6 & @xmath171 & @xmath11665.8(7.7 ) & + 118.5(7.1 ) & 135.5(7.2 ) & 59.5(1.6 ) + overall & @xmath171 & @xmath11677.6(3.5 ) & + 128.4(3.4 ) & 150.1(3.4 ) & 60.56(65 ) + 2007 aug 3 & hd 212311 & + 330(63 ) & @xmath1164(60 ) & 330(63 ) & 179.6(5.2 ) + 2007 aug 4 & @xmath171 & + 260(48 ) & + 15(60 ) & 261(48 ) & 1.6(6.6 ) + 2007 aug 5 & @xmath171 & + 300(52 ) & @xmath11661(68 ) & 306(52 ) & 174.5(6.4 ) + 2007 aug 6 & @xmath171 & + 178(58 ) & @xmath116136(68 ) & 226(65 ) & 161.1(8.9 ) + overall & @xmath171 & + 273(29 ) & @xmath11633(34 ) & 275(29 ) & 176.5(3.6 ) + 2007 aug 5 & algenib & @xmath116526(10 ) & @xmath116528(11 ) & 745(11 ) & 112.54(41 ) + 2007 aug 6 & @xmath171 & @xmath116579.2(7.3 ) & @xmath116512.7(7.2 ) & 773.5(7.2 ) & 110.76(27 ) + overall & @xmath171 & @xmath116557.1(6.8 ) & @xmath116521.8(5.8 ) & 763.3(6.4 ) & 111.56(24 ) + [ 2tableh ] c c c c c c 2007 aug 4 & u her & + 0.1309(13 ) & @xmath1160.03803(81 ) & 0.1363(13 ) & 171.90(18 ) + 2007 aug 5 & hd 157999 & @xmath1160.8763(12 ) & + 0.1531(20 ) & 0.8895(12 ) & 85.045(65 ) + 2007 aug 6 & @xmath171 & @xmath1160.86422(93 ) & + 0.1492(24 ) & 0.8770(10 ) & 85.103(75 ) + overall & @xmath171 & @xmath1160.8699(13 ) & + 0.1509(16 ) & 0.8828(13 ) & 85.079(51 ) + 2007 aug 3 & hd 187929 & @xmath1161.5949(10 ) & @xmath1160.1610(56 ) & 1.6030(12 ) & 92.883(99 ) + 2007 aug 5 & @xmath171 & @xmath1161.5551(27 ) & @xmath1160.1758(41 ) & 1.5651(28 ) & 93.224(75 ) + 2007 aug 6 & @xmath171 & @xmath1161.5262(27 ) & @xmath1160.1727(40 ) & 1.5360(28 ) & 93.229(76 ) + overall & @xmath171 & @xmath1161.5784(37 ) & @xmath1160.1659(35 ) & 1.5872(37 ) & 93.001(65 ) + 2007 aug 4 & hd 7927 & @xmath1163.0658(40 ) & @xmath1160.2510(93 ) & 3.0761(40 ) & 92.342(87 ) + 2007 aug 3 & hd 147084 & @xmath116 & + 3.3972(47 ) & @xmath116 & @xmath116 + 2007 aug 4 & @xmath171 & + 1.658(14 ) & + 3.4386(25 ) & 3.8171(65 ) & 32.134(95 ) + overall & @xmath171 & @xmath116 & + 3.4050(56 ) & @xmath116 & @xmath116 + 2007 aug 4 & hd 154445 & @xmath1163.8076(67 ) & @xmath1160.0405(12 ) & 3.8078(67 ) & 90.3047(86 ) + 2007 aug 3 & hd 204827 & @xmath1162.6413(82 ) & + 4.733(89 ) & 5.4202(89 ) & 59.582(46 ) + 2007 aug 4 & @xmath171 & @xmath116 & + 4.819(12 ) & @xmath116 & @xmath116 + 2007 aug 5 & @xmath171 & @xmath1162.638(11 ) & + 4.742(10 ) & 5.427(10 ) & 59.542(56 ) + 2007 aug 6 & @xmath171 & @xmath1162.6638(89 ) & + 4.7970(95 ) & 5.4869(95 ) & 59.521(47 ) + overall & @xmath171 & @xmath1162.6463(56 ) & + 4.7575(68 ) & 5.4440(66 ) & 59.542(31 ) + [ 2tablei ] even though the bandpasses differ between the apds and pmts , we do not determine telescope polarization with the pmts for many reasons . first , it is difficult to identify unpolarized stars with @xmath176 . second , differences in telescope polarization derived from apd and pmt observations will only be detected after long pmt observations . third , we aim to detect small scale changes in polarization of target stars , so constant offsets in telescope polarization between apd and pmt observations is not our goal . therefore , we choose to quickly measure telescope polarization at the part per million level using bright stars and apds to minimize overhead due to calibration . + the pem is rotated to positions of @xmath144 with respect to the wollaston axis , and the cassegrain ring is rotated through @xmath135 . this gives independent measures of the pem and ring rotation systematics . while planetpol can be rotated to positions of @xmath177 with respect to their pem , this @xmath30 rotation of the instrument also causes the stokes parameter of opposite sign to be observed . thus , pem and instrument rotation systematics are coupled for planetpol , while they can be independently measured for polish . in addition , planetpol can only rotate through @xmath178 , but the @xmath135 rotation of polish enables more thorough measurement of the instrument rotation systematics . + to investigate the pem systematics , we subtract the offset due to ring rotation systematics . we first find the weighted mean polarization of each stokes parameter for each star separately and at each of the @xmath144 positions . in increments of @xmath179 , we average ring angles @xmath130 to @xmath180 for the @xmath33 parameter and @xmath181 to @xmath182 for the @xmath34 parameter . therefore , the mean polarization at each pem position contains the same offset due to ring rotation systematics . the sign of the polarization taken at the @xmath183 position is reversed , and the unweighted mean is taken across both stokes parameters and both pem positions . the unweighted mean is employed so neither stokes parameter nor pem position dominates . this value is the pem offset , given by @xmath184 ( equation [ 2eqa ] ) . the uncertainty in this offset is half the difference between the results for stokes @xmath33 and @xmath34 ( equation [ 2eqb ] ) . here , the index @xmath8 represents the pem position , where @xmath185 indicates @xmath186 and @xmath187 indicates @xmath183 . this process is duplicated for each detector and star separately . + @xmath188 @xmath189 to investigate the systematics when rotating the cassegrain ring by @xmath30 , i.e. the differences between @xmath190 as well as between @xmath191 , we subtract the offset due to pem systematics . while this value has been calculated above , we prefer to combine the data in such a way as to cause it to cancel . we find the weighted mean value of each stokes parameter separately using both @xmath144 positions . we average ring angles @xmath130 and @xmath192 for the @xmath39 parameter , @xmath193 and @xmath180 for @xmath41 , @xmath181 and @xmath194 for @xmath40 , and @xmath195 and @xmath182 for @xmath42 . we then reverse the signs of polarization for the negative stokes parameters . taking the unweighted mean for @xmath33 and @xmath34 separately , we find the offsets for both stokes parameters , @xmath196 ( equation [ 2eqa ] ) . the uncertainty is half the difference between the offsets for the positive and negative stokes parameters ( equation [ 2eqb ] ) . here , the index @xmath8 represents the sign of the measured stokes parameter , where @xmath185 indicates @xmath197 and @xmath187 indicates @xmath198 . + both systematics are calculated for each detector combination . a weighted mean is taken across all stars to determine the mean systematics of the instrument / telescope system . weighting is given as the inverse square of the individual uncertainties , as opposed to weighting by detected photons . this is because mean values are taken across all stars as opposed to across all measurements of a particular star . the mean systematic across all stars when rotating the pem by @xmath199 is @xmath200 , while the mean systematic between positive and negative stokes parameters is @xmath201 . the quadrature addition of @xmath184 and @xmath196 divided by stellar polarization has a mean value of @xmath202 . therefore , systematic effects are consistent with zero and are only important at less than @xmath0 of the measured polarization . the flux of net polarized light from a star is proportional to the polarization , @xmath203 . since the photon shot noise on this quantity scales as @xmath204 , one would expect that the run - averaged , polarimetric precision attainable on stars of similar brightness would also be proportional to @xmath204 . in addition , we expect the instrument to have a noise floor that becomes noticeable for unpolarized stars . as can be seen in figure [ 2myfign ] , we find good agreement by fitting the data from the stars observed with apds to the model @xmath205^{1/2}\ ] ] here , @xmath206 is a scaling factor and @xmath207 parts per million is the noise floor of the instrument . the fitting was performed using a least - squares approach . however , since the data span five orders of magnitude in polarization , the residuals to be minimized are given by @xmath208 the stars observed with apds are all roughly the same visual magnitude . however , the precision achieved on the weakly polarized hd 212311 , observed with pmts , is about an order of magnitude worse than for the stars observed with apds . this is expected , since hd 212311 is roughly 5 magnitudes fainter than the apd stars ( table [ 2tablec ] ) . thus , the scaling factor @xmath206 determined for the apd stars will be an order of magnitude different from the scaling factor for the pmt stars . this is why the pmt stars were excluded from the above fit . however , precision on the strongly polarized hd 204827 is surprisingly consistent with the slope for the bright stars . the four stars with polarization @xmath209 , hd 7927 , hd 147084 , hd 154445 , and hd 204827 , are known to be variable ( dolan & tapia 1986 ; bastien et al . 1988 , 2007 ) . thus , differences in @xmath210 and the model fit are likely to be related to stellar variability . as stated in section 5 , it should be noted that run - averaged precision assumes nonvariable polarization of the target . the validity of this assumption will be investigated in a future paper . we compare our results to the agglomerated polarization catalog of heiles ( 2000 ) and to hlb06 in figure [ 2myfige ] . the degree of polarization measured by polish is plotted as open stars , and stellar polarization increases toward the bottom of the plot . precision in the degree of polarization from this work is plotted as filled black circles , precision values computed from hlb06 are light grey diamonds , and heiles ( 2000 ) precision values are dark grey squares . uncertainty in degree of polarization is listed nightly for strongly polarized stars in table 3 of hlb06 ( hd 7927 , hd 147084 , hd 154445 , and hd 187929 ) . to determine run - averaged uncertainty for these stars , we first convert their degree , position angle , and uncertainties to @xmath33 , @xmath34 , and associated uncertainties . we then perform a weighted mean for each stokes parameter separately , where the weights are the inverse square of the nightly uncertainties in those parameters . to determine polarization for each star , heiles ( 2000 ) take the weighted mean polarization from different authors . the weights are the inverse square of the uncertainties from each author . uncertainty in stellar polarization in heiles ( 2000 ) is the square root of the sum of squares of residuals between each author s polarization and the heiles ( 2000 ) mean polarization . + the rightmost column of the figure shows the position angle of net polarization , where north is at the top and east is at the left of the plots . orientations of the black lines indicate position angles measured with polish , hlb06 position angles are light grey lines , and heiles ( 2000 ) position angles are dark grey lines . agreement between the data sets for stars with low polarization is of course poor , because position angle of net polarization is meaningless for these stars . as stellar polarization increases , however , agreement in position angle also increases . since agreement between our measurements and the literature regarding degree of polarization is determined by choice of polarimetric gain factor ( section 5 ) , accuracy in our observations is assessed by agreement in position angle of polarization . + the unpolarized standard stars observed in order to determine telescope polarization , hr 5854 and hr 8974 , have run - averaged polarimetric precision of order one part per million . this is our precision goal for bright , unpolarized stars , and it is comparable to the precision achieved by planetpol . however , we have improved the precision on these stars by nearly three orders of magnitude with respect to the heiles ( 2000 ) catalog . it should be noted that hr 8974 is known to harbor an extrasolar planet with a minimum mass of @xmath211 jupiter masses , a period of @xmath212 days , and a semimajor axis of @xmath213 au ( neuhuser et al . however , we expect the amplitude of the planetary polarimetric signal to be of order @xmath214 or less and consequently undetectable . + we have improved the polarimetric precision achieved on hd 9270 by an order of magnitude with respect to heiles ( 2000 ) . precision on @xmath108 oph and algenib ( hd 886 ) , however , is only slightly better than tabulated in heiles ( 2000 ) . it is expected that more observations of these stars will improve this precision . there is an order of magnitude improvement in precision on hd 212311 with respect to heiles ( 2000 ) . the precision achieved on u her has been improved by two orders of magnitude from heiles ( 2000 ) . + precision on hd 157999 has been improved by an order of magnitude , while precision on hd 187929 is comparable between our measurements , those from planetpol , and those from heiles ( 2000 ) . this may indicate intrinsic variability of the source . precision on hd 7927 from polish is of the same magnitude as that from planetpol , and we improve the precision with respect to heiles ( 2000 ) by an order of magnitude . we assume that the discrepancy between the hlb06 position angle ( @xmath215 ) and our measurements for hd 7927 is simply a typographical error in their paper . we expect that their intended value is @xmath216 , which is close to our value . + precision on hd 147084 is comparable between our measurements and those from planetpol . we improve on the heiles ( 2000 ) results by almost an order of magnitude . precision on hd 154445 between our measurements , planetpol , and heiles ( 2000 ) lie within a factor of @xmath48 of each other , which may suggest intrinsic variability of the source . we only have one night of data on this star , so this possibility will not be investigated in our forthcoming paper . finally , we improve the precision on hd 204827 by over an order of magnitude with respect to heiles ( 2000 ) . + we have commissioned a high precision , integrated light polarimeter in order to detect variability in the optical , linear polarization of high mass x - ray binaries . this variability should be indicative of system inclination , and high precision monitoring is hoped to constrain the mass of the black holes in these systems . while results from observations of cygnus x-1 will be forthcoming , we observe polarization of standard stars to high precision . individual measurements of most stars are photon shot noise limited ( figure [ 2myfigh ] ) . nightly mean and run - averaged polarizations of most stars are observed at precision levels comparable to planetpol , a similar instrument mounted on the william herschel telescope . + precision achieved on unpolarized stars is of order one part per million , which is up to three orders of magnitude more precise than the combined polarimetric catalog of heiles ( 2000 ) . we find night - to - night precision of three to ten parts per million on bright , weakly polarized standard stars @xmath217 . this precision increases as @xmath218 and is of order one part in @xmath168 for stars with @xmath219 . precision on strongly polarized stars is improved by up to an order of magnitude with respect to heiles ( 2000 ) . thus , precision achieved in a night - to - night sense scales as expected from photon shot noise statistics . systematic effects reveal themselves at the level of less than @xmath0 of the measured polarization for polarized stars . the night - to - night noise floor of the instrument appears to be @xmath0 of the telescope polarization , or about two parts per million . this noise floor is comparable to that of planetpol , which is about one part per million ( lucas et al . the large improvement in polarimetric precision arises from the combination of large telescope aperture , a high - quality polarization modulator , and high frequency modulation . + in addition to observations of high mass x - ray binaries , the high precision of polish is well suited to polarimetric observations of extrasolar planet host stars , debris disks , and evolved stars . while detection of scattered light from hot jupiters requires higher precision than currently attainable , knowledge of host star baseline variability is necessary . in addition , observation of planetary transits in polarized light gives information on transit geometry and the scattering properties of the host star , which are poorly known . polarimetric observations of debris disks may constrain the size distribution of dust grains as well as the scattering properties of the disk . evolved stars represent the intermediate stage between roughly spherical stars and strikingly asymmetric planetary nebulae . the growth of asymmetry in these systems is poorly understood , and high precision polarimetry may make significant contributions to this field . indeed , inexpensive , high precision polarimeters are poised to impart their mark in many fields of astronomy . + sjw would like to thank shrinivas kulkarni , michael ireland , and ian mcewan for valuable discussions on instrumentation , polarimetry , and signal processing . in addition , the help of the palomar observatory staff , especially bruce baker , karl dunscombe , john henning , greg van idsinga , and steve kunsman , has been essential . we would like to thank the anonymous referee for helpful comments on this manuscript . we would like to acknowledge support from the moore foundation . this research has made use of the simbad database , operated at cds , strasbourg , france . + _ facilities : _ . the mueller matrix of polish is given by @xmath220 and observed polarization is related to incident polarization according to @xmath221 starting from the left hand side of equation [ 2eqp ] , the matrices are the rotation matrix for cassegrain ring angle @xmath110 , the mueller matrices for the detector window , field lenses , wollaston prism , and beamsplitter , the rotation matrix for pem angle @xmath222 , the mueller matrix for the pem , and the mueller matrix for the telescope . subtraction of telescope polarization is necessary to calibrate for @xmath223 . + since the pem and wollaston prism convert the polarization of the incident light into intensity modulation , the polarization state of light past the wollaston is not our concern . the @xmath30 reflection of light off the beamsplitter is stable during observations , so any polarization imparted to the light by the beamsplitter is just a constant offset to the polarization . therefore , we only consider the throughput of the detector window , field lenses , and beamsplitter , as opposed to their polarizing properties . we denote the throughput of the instrument as @xmath224 , and it is given in section 3 . the polish mueller matrix becomes @xmath225 the top sign in the wollaston matrix ( + ) indicates the left beam which reaches detector 2 , and the bottom sign ( @xmath116 ) represents the right beam which reaches detector 1 . + the instantaneous retardance of the pem is given by @xmath226 , and expanding the retardance in terms of bessel functions gives @xmath227 \label{2eqg}\ ] ] @xmath228 the @xmath229 expansion in equation [ 2eqg ] generates odd harmonics of the pem reference frequency , while the @xmath230 expansion in equation [ 2eqh ] generates even harmonics . when propagating incident light through polish as per equation [ 2eqq ] , one can see that the linear stokes parameters @xmath33 and @xmath34 will be modulated at even harmonics of the pem modulation , while the circular stokes parameter @xmath231 will be modulated at odd harmonics . thus , we choose to set our lock - in amplifiers to record the second harmonic of modulated intensity , which is @xmath232 q \nonumber \label{2eqaa } \\ & \pm & \ ! \ ! \left [ \sin^2 2\theta \cos 2\phi + \frac{1}{2 } \sin 4\theta \sin 2\phi\right ] \big[2j_2 \left ( \beta_0\right ) \cos 2 \omega t \big ] q \\ & \pm & \ ! \ ! \ ! \ ! \ ! \left[\cos^2 2\theta \sin 2\phi + \frac{1}{2 } \sin 4\theta \cos 2\phi + \left ( \sin^2 2\theta \sin 2\phi \mp \frac{1}{2 } \sin 4\theta \cos 2\phi \right ) j_0 \left(\beta_0 \right ) \right ] u \nonumber \\ & \pm & \ ! \ ! \ ! \ ! \ ! \left [ \sin^2 2\theta \sin 2\phi \mp \frac{1}{2 } \sin 4\theta \cos 2\phi\right ] \big[2j_2 \left ( \beta_0\right ) \cos 2 \omega t \big ] u \nonumber % \end{split } % \end{equation}\end{aligned}\ ] ] the lock - in amplifiers record the rms value of the ac component of the intensity , given by @xmath17 . the amplitude of the ac signal is therefore @xmath233 \right\ } \label{2eqr}\ ] ] the mean intensity , or dc level , is given by @xmath234 q \label{2eqs } \\ & \pm & \ ! \ ! \ ! \ ! \ ! \left[\cos^2 2\theta \sin 2\phi + \frac{1}{2 } \sin 4\theta \cos 2\phi + j_0 \left(\beta_0 \right ) \left ( \sin^2 2\theta \sin 2\phi \mp \frac{1}{2 } \sin 4\theta \cos 2\phi \right ) \right ] u \nonumber % \end{split } % \end{equation } \end{aligned}\ ] ] for @xmath235 , equations [ 2eqr ] and [ 2eqs ] reduce to @xmath236\ ] ] @xmath237\ ] ] two integrations with the cassegrain ring rotated @xmath138 apart are therefore required to determine both linear stokes parameters @xmath33 and @xmath34 . + for cassegrain ring angle @xmath130 , the normalized polarization in terms of the observables @xmath17 and dc is given by @xmath238 it can be seen that choice of @xmath239 radians , the first zero of @xmath240 , ensures that polarization is proportional to the ratio of @xmath17 and dc . in this case , equation [ 2eqt ] reduces to equation [ 2eqj ] , and the pem efficiency is @xmath241 . however , choosing halfwave retardance , @xmath242 radians , increases pem efficiency to @xmath243 at the expense of additional calibration . that is , the dc level is a function of source polarization for halfwave retardance , but it is independent for retardance of @xmath239 radians . since we plan to observe unpolarized well as strongly polarized sources , we choose a peak retardance of @xmath239 radians in our observations . given pre - gain signal current @xmath244 and pre - gain dark current @xmath245 , the number of pre - gain photoelectrons during an integration of duration @xmath246 , and the shot noise on this quantity , will be given by where @xmath56 is detector bandwidth , @xmath161 is detector gain , and @xmath250 is the gain noise factor ( post - gain quantities are primed ) . this factor arises because the gain process itself has statistical fluctuations . the excess noise factor , @xmath160 , is a constant . + since voltage is measured at the output of the detectors , we convert signal and dark current to signal and dark voltage . the pre - gain signal and dark current are @xmath251 and @xmath252 , where @xmath162 is the amplifier transimpedance in v / a and the post - gain dark current is @xmath159 . output noise voltage is related to noise current by @xmath253 , so this is the expected voltage noise at the output of imperfect detectors , where @xmath255 . the bandwidth @xmath56 must be chosen with care . photon shot noise is white noise , but it can not have infinite bandwidth because integrated power would also be infinite . the maximum frequency at which photon shot noise can occur is the count rate of noise photons . thus , we determine maximum bandwidth by the square root of the number of detected photons , the detectors have bandwidths between about 100 khz and 200 khz . however , the lock - in amplifiers only admit noise in a bandwidth of @xmath157 hz , while the voltmeters have bandwidth of @xmath257 mhz . thus , the bandwidth of shot noise on @xmath94 and @xmath95 will be @xmath258 for all stars . the bandwidth of shot noise on dc will be @xmath259 , which will depend on stellar intensity . + using equation [ 2eqao ] , we can determine photon shot noise ( @xmath163 and @xmath164 ) and detector noise ( @xmath255 and @xmath260 ) on @xmath94 , @xmath95 , and dc . we can also compare these values to the observed fluctuations during each measurement . @xmath261_{\rm{obs } } = \left [ \begin{array}{c } { \rm std}(x_{\rm{src}})^2 + { \rm std}(x_{\rm{sky}})^2 \\ { \rm std}(y_{\rm{src}})^2 + { \rm std}(y_{\rm{sky}})^2\\ { \rm std}({\rm dc}_{\rm{src}})^2 + { \rm std}({\rm dc}_{\rm{sky}})^2 \end{array } \right]^{1/2 } \label{2eqas}\ ] ] here , @xmath154 , @xmath155 , and std " indicates the sample standard deviation . we now relate this quantity to fluctuations in the observables @xmath94 , @xmath95 , and dc . by propagating error through equation [ 2eqj ] , we find polarimetric uncertainty of a measurement to be related to uncertainty in @xmath94 , @xmath95 , and dc according to finally , inserting equations [ 2eqaq ] through [ 2eqas ] into equation [ 2eqap ] yields the uncertainty in polarization from photon shot noise , detector noise , and observed fluctuations , given as equations [ 2eqau ] through [ 2eqat ] .

the leading correction from interactions to the transition temperature for bose - einstein condensation can be obtained from a nonperturbative calculation in the critical scalar field theory in 3 dimensions with . we show that the linear expansion can be applied to this problem in such a way that in the large- limit it converges to the exact analytic result . if the principal of minimal sensitivity is used to optimize the convergence rate , the errors seem to decrease exponentially with the order in the expansion . for we calculate the shift in to fourth order in . the results are consistent with slow convergence to the results of recent lattice monte carlo calculations .

cond-mat0206186

the critical temperature @xmath0 for bose - einstein condensation in an ideal gas of identical bosons was deduced in 1924 by albert einstein @xcite : @xmath7 where @xmath8 is the number density of the bosons . an obvious next question is the shift @xmath9 in the critical temperature due to the interaction between the bosons . if the potential between two bosons is short - ranged , their scattering amplitude at sufficiently low energy is a constant proportional to the @xmath10-wave scattering length @xmath11 the bose gas with such an interaction has been studied extensively since the pioneering work @xcite of lee and yang in 1957 on bosons with a hard - sphere potential . however the shift in @xmath0 from such an interaction remained an unsolved problem until recently . a major breakthrough was made by baym et al . @xcite , who showed that the problem could be reduced to a nonperturbative calculation in an effective @xmath12-dimensional statistical field theory with @xmath13 symmetry . this demonstrated conclusively that the leading order shift was linear in @xmath14 : @xmath15 where @xmath16 is a numerical constant . using a self - consistent calculation of the propagator , they estimated the coefficient to be @xmath17 @xcite . baym , blaizot and zinn - justin calculated the coefficient analytically in the large-@xmath4 limit and obtained @xmath18 @xcite . the first correction in the @xmath19 expansion was calculated by arnold and tomasik and it reduces @xmath16 to 1.72 @xcite . definitive calculations of the coefficient for the physical case @xmath2 were carried out by kashurnikov , prokofev , and svistunov @xcite and by arnold and moore @xcite by applying monte carlo methods to a lattice formulation of the effective field theory with the result @xmath20 . the second order correction to @xmath21 proportional to @xmath22 has since been calculated by arnold , moore and tomasik @xcite . before the recent breakthrough , there had been many previous attempts to calculate the leading shift in @xmath0 @xcite . the results for the coefficient @xmath16 in ( [ c - def ] ) ranged from @xmath23 to @xmath24 . the definitive solution of this problem provides a way to assess the reliability of the various calculational methods . one particularly simple method that can be applied to this problem is the _ linear @xmath5 expansion _ @xcite , also known as _ optimized perturbation theory _ @xcite or _ variational perturbation theory _ @xcite . with this method , an arbitrary parameter @xmath25 is introduced into the theory and calculations are carried out using perturbation theory in a formal expansion parameter @xmath5 whose value is 1 . the rate of convergence can be improved by adjusting the parameter @xmath25 at each order in @xmath5 by some criterion such as the principal of minimal sensitivity . it was first applied to the problem of calculating @xmath26 by de souza cruz , pinto and ramos @xcite . at second , third , and fourth orders in @xmath5 , the predictions are @xmath27 , @xmath28 , and @xmath29 , respectively @xcite , which seem to be converging to the lattice monte carlo result . in this paper , we revisit the application of the linear @xmath5 expansion to the calculation of the shift in @xmath30 . in section ii , we review the reduction of the calculation of @xmath26 to a problem in a 3-dimensional statistical field theory . in section iii , we introduce the linear @xmath5 expansion and review the results of refs . we present two alternative precriptions for applying this method to the shift in @xmath30 . in section iv , we show that in the large-@xmath4 limit , these alternative prescriptions converge to the analytic result of ref . @xcite for some range of @xmath25 . we study the improvement in the convergence rate that is obtained by using the principal of minimal sensitivity to choose a value of @xmath25 that depends on the order in @xmath5 . we show that there is a family of solutions to this equation that seems to improve the convergence rate from a power law in the order of the @xmath5 expansion to exponential . in section v , we calculate the result for @xmath2 at @xmath31 and @xmath32 orders in @xmath5 . in section vi , we compare our results with the previous calculations using the linear @xmath5 expansion of refs . @xcite . our results are consistent with slow convergence to the monte carlo results of refs . @xcite . we consider a system of identical bosons of mass @xmath25 at a temperature @xmath33 and number density @xmath8 close to the critical temperature @xmath0 for bose - einstein condensation . if the effects of interactions are small , einstein s formula ( [ einstein ] ) implies that the interparticle spacing @xmath34 and the thermal wavelength @xmath35 are comparable . we assume that they are both large compared to the range @xmath36 of the two - body potential . in this case , the interaction between the bosons can be characterized entirely by the s - wave scattering length @xmath14 , which we assume to have magnitude of order @xmath36 . thus we require @xmath37 . this system can be described by a second - quantized schrdinger equation . the corresponding imaginary - time lagrangian is @xmath38 where @xmath39 is the chemical potential associated with the particle number density . in this paper , we generally use units such that @xmath40 . treating the system at finite temperature using the imaginary time formalism , the field @xmath41 can be decomposed into fourier modes @xmath42 with matsubara frequencies @xmath43 the zero frequency mode @xmath44 is defined by @xmath45 if the temperature is near the critical temperature @xmath30 , the chemical potential @xmath46 is small in magnitude compared to @xmath33 . for momenta small compared to @xmath47 , the nonzero frequency modes have euclidean energies of order @xmath33 , while the zero frequency modes have euclidean energies of order @xmath48 . thus at large distances much greater than @xmath49 , the nonzero modes decouple from the dynamics , leaving an effective theory of only the zero modes @xcite . a naive derivation of the effective lagrangian is obtained simply by substituting @xmath50 in ( [ l - it ] ) . the action then reduces to @xmath51 \longrightarrow \int d^{3}x\text { } { \cal l}_{{\rm eff}}[\psi_{0}({\bf x})],\ ] ] where the effective lagrangian is @xmath52 it is convenient to replace the complex field @xmath53 by the 2-component real field @xmath54 defined by @xmath55 the effective lagrangian density for the dimensionally reduced theory can then be written as @xmath56 where the parameters are @xmath57 this @xmath13 field theory can be generalized to an @xmath1 field theory by replacing @xmath58 by an @xmath4-component real field @xmath59 . this naive derivation of the effective lagrangian is supported by a more sophisticated analysis within the effective field theory framework @xcite . this analysis reveals that there are also higher dimension operators in the effective lagrangian ( [ l - eff ] ) , such as @xmath60 and @xmath61 , but they are not needed to calculate the shift in @xmath30 to leading order in @xmath62 . it also reveals that there are corrections to the coefficients @xmath63 and @xmath64 that can be calculated perturbatively as expansions in @xmath65 . again , they are not needed to calculate @xmath26 to leading order in @xmath66 . finally , since the scale @xmath33 has been removed from the problem , the effective field theory defined by the effective lagrangian ( [ l - eff ] ) requires an ultraviolet cutoff . fortunately , the leading order shift in @xmath30 is dominated by the infrared region of momentum space and is insensitive to the ultraviolet cutoff . from einstein s formula ( [ einstein ] ) , the leading order shift @xmath67 in the critical temperature at fixed number density is related to the leading order shift @xmath68 in the critical density at fixed temperature by the simple relation @xmath69 the number density in the original theory is @xmath70 where the angular brackets indicate the expectation value in the state corresponding to a homogeneous gas . the number density in the effective theory , to the accuracy required to calculate the shift in @xmath30 to leading order in @xmath66 , is @xmath71 where @xmath72 is the short - distance contribution to the number density , which depends on the ultraviolet cutoff . the angular brackets indicate the statistical average of the operator @xmath73 . it can be expressed as @xmath74 ^{-1 } , \label{fi2}\ ] ] where @xmath75 is the self - energy of the @xmath76 field . the integral over @xmath77 is divergent and requires an ultraviolet cutoff . the number density for an ideal gas is given by the same expression with @xmath78 which implies @xmath79 : @xmath80 ^{-1}. \label{na0 } \label{n0}\ ] ] again , the integral over @xmath77 is divergent and requires an ultraviolet cutoff . to the accuracy required to calculate the shift in @xmath30 to leading order in @xmath66 , the constant @xmath72 is the same as in ( [ n ] ) . the critical point is determined by the condition that the correlation length associated with the propagator in ( [ fi2 ] ) is infinite : @xmath81 this condition reduces to @xmath82 if @xmath83 . the change @xmath84 in the critical density at fixed temperature @xmath33 due to interactions is the difference between ( [ n ] ) with @xmath85 and ( [ n0 ] ) with @xmath86 . the expression ( [ deltc-1 ] ) for the fractional shift in @xmath30 can therefore be written as @xmath87 where @xmath88 is a quantity that can be calculated within the effective field theory : @xmath89 ^{-1}-\left ( p^{2}\right ) ^{-1}\right ] . \label{delta - gen}\ ] ] the integral over @xmath77 is convergent and therefore no longer requires an ultraviolet cutoff . we have replaced the prefactor 2 in ( [ delta - gen ] ) by @xmath4 to generalize it to the @xmath1 case . comparing ( [ delta - def ] ) with ( [ c - def ] ) and using ( [ u - a ] ) and ( [ einstein ] ) , we find that the coefficient @xmath16 is @xmath90 the effective theory defined by the lagrangian ( [ l - eff ] ) is superrenormalizable . although the theory has ultraviolet divergences , @xmath91 is dominated by infrared physics and it is insensitive to the ultraviolet cutoff . the relevant length scales are therefore those set by the parameters @xmath63 and @xmath64 of the effective field theory . the critical point can be reached by tuning the parameter @xmath63 . at the critical point , there is only one relevant length scale and it is set by the parameter @xmath64 . since @xmath91 has dimensions of length , it must be proportional to @xmath64 by simple dimensional analysis . the determination of the coefficient of @xmath64 requires a nonperturbative calculation . since @xmath64 is linear in @xmath14 as @xmath92 , this demonstrates that the leading term in the shift in @xmath30 is linear in @xmath14 . the calculation of the first order shift @xmath9 in the critical temperature has been reduced to the calculation of the quantity @xmath91 in ( [ delta - gen ] ) for @xmath93 the first calculation of this quantity within a systematically improvable framework was a calculation to leading order in the @xmath19 expansion by baym , blaizot , and zinn - justin @xcite : @xmath94 the next - to - leading order ( nlo ) correction in the @xmath19 expansion was calculated by arnold and tomasik @xcite . the first two terms in the @xmath19 expansion are @xmath95 for the physical case @xmath2 , @xmath91 has been calculated nonperturbatively by kashurnikov , prokofev , and svistunov @xcite and by arnold and moore @xcite using monte carlo simulations of a lattice formulation of the @xmath13 field theory . the two results are consistent , but ref . @xcite quotes a smaller error : @xmath96 if we set @xmath2 in the @xmath19 expansion in ( [ delta-1/n ] ) , the coefficient of @xmath97 is @xmath98 at leading order and @xmath99 at next - to - leading order . these coefficients seem to be approaching the lattice monte carlo result @xmath100 . any attempt to calculate @xmath9 using ordinary perturbation theory in @xmath64 is doomed to failure . in the symmetric phase where @xmath102 , perturbation theory is an expansion in powers of @xmath103 . if we naively let @xmath104 to find the critical behavior , the perturbative corrections are infrared divergent , with the degree of divergence becoming more severe at each successive order . it is useful to think of perturbation theory as an expansion in powers of @xmath105 , where @xmath106 is the correlation length . perturbation theory breaks down in the critical region , because @xmath105 is of order 1 and every order of perturbation theory gives a leading order contribution . the _ linear @xmath5 expansion _ ( lde ) is a particularly simple method for obtaining nonperturbative results using perturbative techniques . calculations are carried out as in ordinary perturbation theory by evaluating a finite number of feynman diagrams . the name `` linear @xmath5 expansion '' comes from the fact that it can be defined by a lagrangian whose coefficients are linear in a formal expansion parameter @xmath5 . if @xmath107 is the lagrangian for the system of interest , the lagrangian that generates the lde has the form @xmath108 where @xmath109 is the lagrangian for an exactly solvable theory , the lagrangian @xmath110 interpolates between @xmath109 when @xmath111 and @xmath107 when @xmath112 . the linear @xmath5 expansion is much easier to apply than the _ nonlinear @xmath5 expansion _ in which the lagrangian that interpolates between @xmath109 and @xmath107 involves an operator that is a nonlinear function of @xmath5 @xcite . the lagrangian @xmath109 for the exactly solvable theory typically involves an arbitrary parameter @xmath25 . at @xmath112 , the lagrangian ( [ l - delta ] ) is independent of @xmath25 . if a perturbation series in @xmath5 converges at @xmath112 , it must converge to a value that is independent of @xmath25 . however at any finite order in the @xmath5 expansion , results depend on @xmath25 . as @xmath25 varies over its physical range , the prediction for the observable often extends out to @xmath113 . thus some prescription for @xmath25 is required to obtain a definite prediction . a simple prescription for @xmath25 is the _ principle of minimal sensitivity _ ( pms ) that the derivative with respect to @xmath25 should vanish . the lde supplemented by the pms prescription is often referred to as _ optimized perturbation theory _ @xcite or _ variational perturbation theory _ @xcite . a prescription such as pms may improve the convergence rate of the lde , because it allows the variable @xmath25 to change with the order @xmath8 of the lde . the convergence properties of the lde have been studied extensively for the quantum mechanics problem of the anharmonic oscillator . the hamiltonian is @xmath114 the lde is generated by the hamiltonian @xmath115 , which interpolates between @xmath116 at @xmath117 and @xmath118 at @xmath119 . it can be expressed as @xmath120 , where @xmath121 the lde has been proven to converge for appropriate order - dependent choices of the variable @xmath25 that include the pms criterion as a special case . duncan and jones showed that the finite temperature partition function converges exponentially , with the errors at @xmath122 order decreasing as @xmath123 where @xmath124 is a numerical constant @xcite . guida , konishi , and suzuki proved that the energy eigenvalues @xmath125 converge uniformly in @xmath126 as @xmath127 @xcite . janke and kleinert showed that the leading coefficient in the strong - coupling expansion for the ground state energy @xmath128 converges exponentially , with the errors decreasing like @xmath129 where @xmath16 is a numerical constant @xcite . bellet , garcia , and neveu have made numerical studies of the lde optimized by the pms criterion for the anharmonic oscillator @xcite . the pms criterion is a polynomial equation in the variable @xmath25 , so its solutions are in general complex - valued . the authors of ref . @xcite found that the solutions at different orders @xmath8 in the lde can be organized into distinct families . the most accurate predictions for the ground state energy @xmath128 and for the quantity @xmath130 are obtained by using families of complex solutions . the resulting values of @xmath128 and @xmath130 are of course complex - valued , but the imaginary parts are small . the authors of ref . @xcite suggested that the real parts could be taken as the predictions of the lde and the imaginary parts could be used as an indication of the error . the anharmonic oscillator can be regarded as a euclidean field theory with a single real - valued field in 1 space dimension . the statistical field theory defined by the lagrangian ( [ l - eff ] ) is a generalization to a multicomponent field in 3 space dimensions . the most serious obstacles to generalizing the convergence proofs for the anharmonic oscillator to this more complicated problem come from the infrared and ultraviolet regions of momentum space . it is reasonable to expect the convergence behavior to be similar if appropriate infrared and ultraviolet cutoffs are imposed on the field theory . the quantity @xmath88 defined in ( [ delta - gen ] ) is insensitive to the ultraviolet region , so we do not expect any complications from taking the ultraviolet momentum cutoff to @xmath131 . however , @xmath88 is very sensitive to the infrared region , so convergence in the presence of an infrared cutoff gives no information about the behavior of the lde in the limit when the infrared momentum cutoff goes to 0 . we proceed to apply the lde to the @xmath1 statistical field theory defined by the lagrangian @xmath132 in ( [ l - eff ] ) . our choice for the solvable field theory is the free field theory with mass @xmath25 . the lagrangian @xmath110 defined by setting @xmath133 in ( [ l - delta ] ) can be written @xmath134 , where @xmath135 calculations are carried out by using @xmath5 as a formal expansion parameter , expanding to a given order in @xmath5 , and then setting @xmath112 . thus the feynman diagrams generated by the lde will involve a propagator with mass parameter @xmath136 , a 4-point vertex proportional to @xmath137 , and a mass counterterm proportional to @xmath138 . in order to apply the lde to the shift in @xmath30 , we need a prescription for generalizing the quantity @xmath88 defined in ( [ delta - gen ] ) to the field theory whose lagrangian is @xmath139 . the prescription must have a well - defined expansion in powers of @xmath5 , and it must reduce to ( [ delta - gen ] ) when @xmath140 . one possibility , which we will refer to below as prescription ii , is to use the expression ( [ delta - gen ] ) , where @xmath141 is the self - energy for the field theory defined by @xmath142 . any such prescription defines @xmath143 as a function of three variables . the @xmath122 order approximation in the lde is then obtained by truncating the expansion in powers of @xmath5 at @xmath122 order to obtain a function @xmath144 and then setting @xmath140 : @xmath145 this sequence of approximations may not converge as @xmath127 , but if it does converge , we expect it to converge to the correct value @xmath88 . at any finite order @xmath8 , the approximation ( [ delta - n ] ) depends on @xmath25 . we will see that as @xmath25 varies over its physical range from 0 to @xmath146 , the range of the predictions for @xmath88 extends out to @xmath146 for @xmath8 even and out to @xmath147 for @xmath8 odd . thus some prescription for @xmath25 is required to obtain a definite prediction . one prescription is the _ principle of minimal sensitivity _ ( pms ) : @xmath148 after setting @xmath112 , @xmath149 is a function of @xmath64 and @xmath25 only . by dimensional analysis , a solution @xmath25 to the pms criterion must be proportional to @xmath64 and the value of @xmath149 at such a point must also be proportional to @xmath64 . an alternative prescription is the _ fastest apparent convergence _ ( fac ) criterion : @xmath150 by allowing the variable @xmath25 to change with the order @xmath8 , prescriptions such as pms or fac may improve the convergence rate of the lde . the lde was first applied to the problem of calculating the shift in @xmath30 by de souza cruz , pinto and ramos @xcite . their prescription for the quantity @xmath88 defined in ( [ delta - gen ] ) is @xmath151 ^{-1}-\left ( p^{2}\right ) ^{-1}\right ] , \label{d - i}\ ] ] where @xmath141 is the self - energy for the field theory whose lagrangian is @xmath142 . this prescription agrees with ( [ delta - gen ] ) at @xmath112 . note that in addition to the explicit dependence on @xmath5 in ( [ d - i ] ) , there is also implicit dependence on @xmath5 through @xmath141 , which can be calculated as a power series in @xmath5 . the subscript i on @xmath88 in ( [ d - i ] ) distinguishes this prescription from alternative prescriptions for @xmath88 that we will label ii and iii . in the feynman diagrams for evaluating @xmath141 in the lde , all the propagators have mass @xmath25 . thus when the prescription i for @xmath88 is expanded in powers of @xmath5 , all the integrals involve only propagators of mass @xmath25 , with the exception of the subtraction term in ( [ d - i ] ) , which involves a massless propagator . the quantity @xmath88 was calculated to @xmath152 order in @xmath3 in ref . @xcite and to @xmath31 and @xmath32 order in ref . after truncating the expansion in powers of @xmath5 and setting @xmath112 , they determined @xmath25 by applying the pms prescription ( [ pms ] ) . at @xmath122 order in the @xmath3 expansion , this reduces to a polynomial equation in @xmath25 with @xmath8 roots , most of which are complex - valued . for a particular choice of the roots at @xmath153 @xmath31 , and @xmath32 order in @xmath5 , the real parts of @xmath154 are @xmath155 @xmath156 , and @xmath157 respectively this sequence seems to be approaching the lattice monte carlo result @xmath100 . in ref . @xcite , the authors also applied the lde to the second order shift in @xmath0 with encouraging results . since the result for @xmath88 is known analytically in the large @xmath4 limit , one would also like to compare the predictions of the lde to this analytic result . unfortunately , as we shall see , the prescription i of ref . @xcite is not well - behaved in this limit . we therefore present two alternative prescriptions for @xmath88 that are well - behaved in the large-@xmath4 limit : @xmath158 , \label{d - ii } \\ \delta_{\rm iii } & = & n\int_{p}\left [ \left ( p^{2}+m^{2}\left ( 1-\delta \right ) + \sigma(p)-\sigma(0)\right ) ^{-1}-\left ( p^{2}+m^{2}\left ( 1-\delta \right ) \right ) ^{-1}\right ] . \label{d - iii}\end{aligned}\ ] ] these both agree with ( [ delta - gen ] ) at @xmath112 . with prescription iii , the lde for @xmath88 generates integrals involving only propagators of mass @xmath25 . with prescription ii , the integrals involve both propagators of mass @xmath25 and massless propagators . our prescription iii in ( [ d - iii ] ) is closely related to the prescription i in ref . the difference between them is given by a simple integral that is evaluated in ( [ j0 ] ) : @xmath159 because of the additional term in @xmath160 proportional to @xmath161 , the limit @xmath162 does not commute with the large-@xmath4 limit . at @xmath122 order in the @xmath5 expansion , the terms that dominate as @xmath163 are the @xmath164 term and the highest order term in the @xmath3 expansion which is proportional to @xmath165 . the pms criterion gives a value of @xmath25 that scales like @xmath166 and the corresponding value of @xmath91 scales like @xmath167 the correct result is that @xmath88 scales like @xmath168 in the large @xmath4 limit . thus prescription i has the wrong behavior in the large-@xmath4 limit at any order in the lde . in this section , we study the behavior of the linear @xmath5 expansion ( lde ) in the large-@xmath4 limit defined by @xmath170 , @xmath171 with @xmath172 fixed . we calculate @xmath88 to all orders in @xmath5 using prescriptions ii and iii . we show that the lde converges to the correct analytic result for @xmath173 greater than some minimum value , with errors that seem to decrease roughly like @xmath34 . we also study the improvement in the convergence obtained by using the pms criterion for @xmath25 . if we use the family of real solutions of the pms criterion , the errors in @xmath88 seem to decrease like @xmath174 . if we use a particular family of complex solutions , the errors seem to decrease exponentially with @xmath8 . at the critical point . the black blob represents the @xmath73 vertex , while the shaded blob represents the complete propagator with subtracted self - energy @xmath175.,width=113 ] at order @xmath176.,width=151 ] the quantity @xmath88 defined in ( [ delta - gen ] ) has the diagrammatic representation shown in fig . [ fig : delta ] , where the line with the shaded blob represents the propagator @xmath177^{-1}$ ] . in the large-@xmath4 limit , the leading contribution to the subtracted self - energy @xmath175 comes from the series of bubble - chain diagrams whose @xmath178 and @xmath32 members are shown in figs . [ fig:2loop ] and [ fig : largen ] . since these diagrams are of order @xmath19 , the leading contribution at large @xmath4 is obtained by expanding the expression ( [ delta - gen ] ) for @xmath88 to first order in @xmath175 . for the prescription ii given in ( [ d - ii ] ) , the resulting expression is @xmath179 the corresponding expression for the prescription iii is obtained by replacing @xmath180 in the denominator of ( [ delta - largen ] ) . although @xmath141 is ultraviolet divergent , the difference @xmath175 is finite . at any finite order in the lde , @xmath175 scales as @xmath181 at small @xmath182 and as @xmath183 for large @xmath182 , so the integral in ( [ delta - largen ] ) is convergent . in addition to the diagrams for @xmath141 generated by the interaction term @xmath184 in ( [ l - int ] ) , we must also take into account insertions of @xmath185 and insertions of @xmath186 . the effect of the @xmath187 insertions is to replace each 1pi self - energy subdiagram in @xmath141 by the subdiagram with the zero - momentum limit subtracted . the effect of the @xmath186 insertions is to replace @xmath136 by @xmath188 . in the series of diagrams for @xmath141 that survive in the large-@xmath4 limit.,width=188 ] we first consider the prescription ii for @xmath88 . the expression for the large-@xmath4 diagram for @xmath88 with @xmath189 loops is given in ( [ a21 ] ) . the integral @xmath190 is a constant multiplied by @xmath191 . after taking into account the @xmath186 insertions , the sum of the large-@xmath4 diagrams is @xmath192 the integrals @xmath190 are expressed as 1-dimensional integrals in ( [ b24 ] ) . inserting these expressions into ( [ 23 ] ) , we obtain @xmath193 ^{n-1 } , \label{d}\ ] ] where @xmath39 is the dimensionless mass parameter @xmath194 and @xmath195 is the function @xmath196 the prediction for @xmath88 at @xmath122 order in the @xmath3 expansion is obtained by expanding the expression ( [ d ] ) as a power series in @xmath5 , truncating after order @xmath197 , and then setting @xmath112 . it is easy to show that if the lde for ( [ d ] ) converges , it converges to the correct analytic result ( [ largen - exact ] ) . after interchanging the order of the sum and the integral , the sum can be evaluated and we obtain @xmath198 upon taking the limit @xmath199 all dependence on the mass parameter @xmath39 disappears : @xmath200 after evaluating the integral over @xmath201 , our final result is ( [ largen - exact ] ) . note that no prescription for @xmath46 was required to get this result . thus , if the @xmath3 expansion converges at some value of @xmath46 , it should converge to ( [ largen - exact ] ) . in the large-@xmath4 limit as a function of @xmath46 at @xmath122 order in the lde for @xmath202 , 4 , 5 , 7 , 11 , 19 , 35 . the curves for @xmath203 , 11 , and 19 appear in order between those labelled @xmath204 and 35.,width=453 ] the manipulations that were used to show the convergence to ( [ largen - exact ] ) involved several interchanges of limits . first , we summed up some terms in the @xmath3 expansion to all orders to obtain the sum over diagrams in ( [ 23 ] ) . next , we interchanged the order of the sum over diagrams and the integral over the last momentum for each diagram . next we took the limit @xmath205 . finally we evaluated the last momentum integral . it is difficult to translate the conditions for the validity of each of these steps into a statement about the convergence of the lde . however , the convergence can be easily studied numerically . in fig . [ fig : d - mun ] , we show @xmath88 as a function of the dimensionless variational parameter @xmath39 for several orders in the @xmath5 expansion : @xmath202 and @xmath206 @xmath207 . the horizontal line is the analytic result ( [ largen - exact ] ) . the results are consistent with convergence to ( [ largen - exact ] ) for all @xmath46 greater than a critical value @xmath208 which is close to 0.7 . our best estimate for the endpoint of the convergence region is @xmath209 . if @xmath210 , @xmath91 seems to diverge to @xmath211 for @xmath8 even and @xmath212 for @xmath8 odd . for any fixed value of @xmath213 , the convergence with @xmath8 is very slow . in column 1 of table [ tab1 ] , we give the values of @xmath88 at @xmath214 , which is the location of the minimum of the @xmath202 curve in fig . [ fig : d - mun ] . we define the fractional error @xmath215 by @xmath216 ( taking the real part of @xmath88 is superfluous for real values of @xmath46 , but it will be important later when we consider complex values of @xmath46 . ) in fig . [ fig : err - n ] , we show a log - log plot of @xmath215 as a function of @xmath8 . the points lie close to a straight line , indicating that the errors decrease like a power of @xmath8 . the dotted line is @xmath217 , which goes through the points for @xmath218 and 35 . thus the errors seems to decrease roughly like @xmath34 . as a function of the order @xmath8 in the lde . the squares , diamonds , and circles are for @xmath214 , the real solution to the pms criterion , and the imaginary solution with maximal @xmath219 , respectively . the lines are simple curves that pass through the last 2 points.,width=453 ] the rate of convergence can be improved by changing the value of @xmath39 at each order in the lde . one possibility is to use the pms criterion ( [ pms ] ) to determine @xmath46 . at @xmath122 order in the @xmath5 expansion , this condition reduces to a polynomial in @xmath39 of order @xmath220 for @xmath8 even , there are no real roots . for @xmath8 odd , there is always one real root . these roots correspond to the minima of the curves in fig . [ fig : d - mun ] . the values of @xmath221 are given in column 2 of table i. there is slight improvement compared to column 1 in the convergence toward the correct value 1 . the fractional errors defined in ( [ eps - n ] ) are shown as a function of @xmath8 in fig . [ fig : err - n ] . the points lie close to a straight line , indicating that the errors decrease like a power of @xmath8 . the dashed line is @xmath222 , which goes through the points for @xmath218 and 35 . thus the errors seem to decrease roughly as @xmath174 . the pms criterion has either zero or one real solution , depending on whether @xmath8 is even or odd . however there are always @xmath223 complex - valued solutions . in the case of the anharmonic oscillator , there are families of complex solutions with much better convergence properties than families consisting of purely real solutions @xcite . in our problem , at any odd order @xmath224 the real solution always has the largest value of re@xmath88 , which is the value farthest from the correct result ( [ largen - exact ] ) . thus this family of solutions gives the slowest possible convergence rate . however there is a strong anticorrelation between the errors in re@xmath88 and @xmath225 . this is illustrated in fig . [ fig : im - re ] , which is a scatter plot of @xmath219 vs. re@xmath88 for the solutions to the pms criterion for @xmath226 . the solutions that give the most accurate values for re@xmath88 are those with the largest values for @xmath227 . thus we can define a nearly optimal family of solutions by choosing those with the maximal values of @xmath227 . the values of @xmath91 for this family of solutions are given in column 3 of table i. note that im@xmath88 for these solutions shows no sign of converging to @xmath228 , and in fact iseems to be converging to a value close to @xmath229 . however the improvement in the convergence of re@xmath88 compared to the real values of @xmath88 in column 2 is evident . a log - log plot of the fractional errors in re@xmath88 is shown in fig . [ fig : d - mun ] . the downward curvature of the points indicates that the errors decrease faster than any power of @xmath8 . the solid line is @xmath230 , which goes through the points for @xmath218 and 35 . up to the order to which we have calculated , the errors are decreasing faster than this exponential . vs. re@xmath88 for the solutions @xmath46 of the pms criterion at @xmath231 order in the lde.,width=453 ] we obtain similar results if we use the prescription iii for @xmath88 given in ( [ d - iii ] ) . the sum of the large-@xmath4 diagrams is the same as in ( [ 23 ] ) , except that the integrals @xmath232 are replaced by @xmath233 . the integrals @xmath234 are expressed as 1-dimensional integrals in ( [ b25 ] ) . interchanging the order of the sum and the integral and then evaluating the sum , we obtain @xmath235 taking the limit @xmath199 all dependence on @xmath46 disappears and after evaluating the integral we again obtain ( [ largen - exact ] ) . thus , if the @xmath3 expansion converges , it should converge to the correct result . the convergence can be again studied numerically . we find that it seems to converge to ( [ largen - exact ] ) for all @xmath213 and it seems to diverge for @xmath210 , with @xmath236 . in column 4 of table [ tab1 ] , we give the values of @xmath88 for various orders @xmath8 at @xmath237 , which is the location of the minimum of @xmath88 for @xmath202 . the convergence at a fixed value of @xmath46 is very slow . as in the case of prescription ii , the errors seem to decrease roughly as @xmath34 . more rapid convergence can again be obtained by using the pms criterion for @xmath39 . the pms criterion has no real solutions for @xmath8 even and 1 real solution for @xmath8 odd . the values of @xmath91 for the real solution at various orders @xmath8 are given in column 5 of table i. this is the family of solutions with the slowest convergence rate for re@xmath88 . the errors seem to decrease roughly as @xmath174 . a family of solutions with a nearly optimal convergence rate for re@xmath88 consists of those with the maximal value of @xmath227 . the values of @xmath91 for this family are shown in column 6 of table i. the errors in re@xmath88 are numerically larger than those with prescription ii , but they also seem to decrease exponentially with @xmath8 . in this section , we use the linear @xmath5 expansion ( lde ) to calculate @xmath88 for finite @xmath4 . we calculate @xmath88 to @xmath32 order in @xmath5 using both of the prescriptions ii and iii . setting @xmath2 and using the pms criterion , we obtain the predictions relevant to the shifts in @xmath30 for bose - einstein condensation . at order @xmath238.,width=453 ] the first term in the lde for @xmath88 is second order in @xmath3 . it is obtained by calculating the 3-loop diagram involving the 2-loop self - energy diagram in fig . [ fig:2loop ] . the contribution to @xmath91 of order @xmath239 is obtained by calculating the 4-loop diagram involving the 3-loop propagator correction in fig . [ fig:3loop ] . the contribution of order @xmath240 is obtained by calculating the 5-loop diagram involving the 4-loop propagator corrections in fig . [ fig:4loop ] . the diagrams are reduced to momentum integrals in appendix [ app : diag ] and the momentum integrals are evaluated in appendix [ app : int ] . the complete expression for @xmath91 for the @xmath1 theory through order @xmath240 is @xmath241 . \label{delta - ii}\end{aligned}\ ] ] inserting the values for the integrals from appendix b , truncating at successively higher orders in @xmath3 and then setting @xmath112 , we obtain the results at @xmath152 , @xmath31 and @xmath32 orders in the lde : @xmath242 \frac{1}{\mu ^{3 } } \right ) , \label{d4-ii } \end{aligned}\ ] ] where @xmath46 is the dimensionless mass parameter @xmath243 in the large-@xmath4 limit , this reduces to the mass parameter introduced in ( [ mu - largen ] ) . the accuracy of the @xmath32 order result in ( [ d4-ii ] ) is limited by the accuracy to which we have evaluated the integrals @xmath244 and @xmath245 . in fig . [ fig : d - mu2 ] , we show @xmath88 for @xmath2 as a function of @xmath46 for @xmath246 . the horizontal lines are the edges of the upper and lower error bars of the lattice monte carlo result given in ( [ delta - mc ] ) . at order @xmath247.,width=680 ] at order @xmath248 the pms criterion has no solutions , because ( [ d2-ii ] ) is a monotonic function of @xmath46 . at order @xmath249 the pms criterion has a single real solution @xmath250 and the resulting value for @xmath91 is @xmath251 at 4@xmath252 order in @xmath3 , @xmath91 is a monotonic function for real - valued @xmath39 , so the pms criterion has no real solutions for @xmath39 . it does however have a conjugate pair of complex solutions whose values at @xmath2 are @xmath253 . the corresponding values of @xmath91 are complex , but we can take the real part as the prediction for @xmath88 at 4@xmath252 order in the lde : @xmath254 for @xmath2 , the @xmath31 and @xmath32 order approximations in ( [ delta3 ] ) and ( [ delta4:2 ] ) differ from the lattice monte carlo result of ( [ delta - mc ] ) by about 66% and 62% , respectively . these predictions are not as accurate as in the large @xmath4 limit , where the errors in the @xmath31 and @xmath32 order approximations are about 52% and 44% , respectively . furthermore , the percentage decrease in the error in going from @xmath31 order to @xmath32 order is only about half as large for @xmath2 as it is for large @xmath4 . for @xmath2 as a function of @xmath46 at @xmath122 order in the lde using prescriptions i ( upper 3 curves on the right ) and ii ( lower 3 curves on the right ) . the dotted , dashed , and solid curves are for @xmath255 , 3 , and 4 , respectively.,width=453 ] with the prescription iii , the expression for @xmath88 to @xmath32 order in @xmath5 is the same as ( [ delta - ii ] ) , except that the integrals @xmath232 are replaced by @xmath233 . inserting the value for the integrals @xmath233 from appendix b , the results at @xmath152 , @xmath31 and @xmath32 orders in the lde are @xmath256 \frac{1}{\mu ^{3 } } \right ) , \label{d4-iii } \end{aligned}\ ] ] the accuracy of the @xmath32 order result in ( [ d4-iii ] ) is limited by the accuracy to which we have evaluated the integrals @xmath257 and @xmath258 . at order @xmath259 the pms criterion has no solution . at order @xmath249 it has one real solution for @xmath39 . at 4@xmath252 order in @xmath3 , it has a conjugate pair of complex solutions . the corresponding values of @xmath88 are given in table [ tab2 ] . the values of @xmath88 at @xmath31 order and of re @xmath88 at @xmath32 order are close to the corresponding values from prescription ii , but a little farther from the lattice monte carlo result . the results of the lde for prescription i can be obtained from those for prescription iii by using ( [ i - iii ] ) : @xmath260 in fig . [ fig : d - mu2 ] , we show @xmath149 as a function of @xmath46 for @xmath255 , 3 , and 4 . the values of @xmath149 at the solutions to the pms criterion are given in column 1 of table [ tab2 ] . at orders 3 and 4 , the pms criterion has a conjugate pair of complex solutions . as the predictions of the lde , the authors of ref . @xcite chose the positive value for @xmath261 and the real parts of the complex values for @xmath262 and @xmath263 . these predictions differ from the lattice monte carlo results by about 131% , 85% , and 20% , respectively . the calculation of the shift in @xmath30 to leading order in @xmath62 can be reduced to a nonperturbative calculation of a quantity @xmath88 in a 3-dimensional @xmath13 field theory at its critical point . this quantity can be generalized to an @xmath1 field theory and it can be calculated systematically using the @xmath19 expansion . we have shown that the linear @xmath5 expansion ( lde ) can be applied to the calculation of @xmath88 in such a way that the large @xmath4 and @xmath264 limits commute . the prescriptions ii and iii defined in ( [ d - ii ] ) and ( [ d - iii ] ) have this property , but the prescription i introduced in ref . @xcite does not . for prescription ii and iii , we calculated @xmath88 in the large-@xmath4 limit to all orders in the lde , and studied its convergence properties . we found that it seems to converge to the correct analytic result for @xmath213 , where @xmath236 . the convergence for fixed @xmath46 is extremely slow : the errors seem to decrease roughly like @xmath34 . the convergence can be accelerated by using the pms criterion to choose a value of @xmath46 that depends on the order in @xmath5 . the pms criterion is a polynomial equation with multiple roots , so there are various families of predictions for @xmath88 . the family consisting of the real solutions of the pms criterion gives the slowest possible convergence : the errors in @xmath88 seem to decrease roughly like @xmath174 . the family consisting of the complex solutions of the pms criterion with the maximal value of @xmath219 have nearly optimal convergence : the errors in re@xmath88 seem to decrease exponentially with @xmath8 . unfortunately , accurate predictions still require calculations to high orders in the @xmath5 expansion . to achieve 10% accuracy requires computing to about @xmath265 order in @xmath5 . for the case @xmath2 relevant to bose - einstein condensation , we calculated @xmath88 to @xmath32 order in the lde using both prescriptions ii and iii . the quantity @xmath88 was previously calculated to @xmath32 order using prescription i in ref . the pms criterion has complex solutions beginning at @xmath31 order for prescription i and at @xmath32 order for prescriptions ii and iii . we take the values of re@xmath149 for these solutions to be the predictions for @xmath88 . the @xmath31 and 4@xmath252 order predictions from prescription ii are smaller than the lattice monte carlo result by about 66% and 62% , respectively . the @xmath152 , @xmath31 and 4@xmath252 order predictions from prescription i are larger than the lattice monte carlo result by about 131% , 85% , and 20% , respectively . we argue that the 20% error in the 4@xmath252 order prediction from prescription i is fortuitously small . prescription i differs from prescription iii by the additive term given in ( [ i - iii ] ) . the expansion of the factor @xmath266 truncated at @xmath122 order in @xmath5 and evaluated at @xmath140 converges to 0 as @xmath127 . the prediction from prescription i should therefore approach the prediction from prescription iii as @xmath8 increases , which suggests that it will converge to the correct answer from below . because the @xmath255 prediction from prescription i is greater than the correct result by about a factor of 2.3 , there must be a crossover in @xmath8 where the prediction goes from values greater than the correct result to smaller values . the high accuracy of the 4@xmath252 order result suggests that this crossover occurs at @xmath204 . thus we expect higher order predictions from prescription i to first overshoot the correct value and then approach it from below . before the lattice monte carlo calculation of ref . @xcite , there were many previous attempts to calculate the coefficient @xmath16 in the expression ( [ c - def ] ) for the leading order shift in @xmath30 for bose - einstein condensation . for most of these methods @xcite , there is no known systematic way of improving the calculation to make it more accurate . such a method is best regarded as just a model of the critical @xmath1 field theory whose accuracy can only be determined by comparison with a more reliable calculational method . a systematically improvable method has two great advantages over such a model . first , the accuracy of the prediction can be improved with additional effort . second , one can make a reliable estimate of the error in the prediction . one example of a systematically improvable method is the @xmath19 expansion @xcite , which has been used to calculate @xmath26 to leading order @xcite and to next - to - leading order @xcite . comparison with the lattice monte carlo results show that these first two approximations have errors of 77% and 30% , respectively . our demonstration that the lde converges in the large @xmath4 limit suggests that the lde may also be a systematically improvable method for @xmath2 . although the convergence rate of the optimized lde appears to be exponential in the large-@xmath4 limit , it is still rather slow . one must calculate to about @xmath265 order in @xmath5 to achieve 10% accuracy . for the case @xmath2 relevant to bose - einstein condensation , we have calculated @xmath88 to @xmath32 order in the lde . it may be feasible to extend the calculation to @xmath267 order , but it would be very difficult to extend it to much higher order . the slow convergence in the large-@xmath4 limit suggests that even if the lde also converges for @xmath2 , a strict expansion in @xmath5 is not useful for quantitative calculations . it may however be possible to use order - dependent mappings to change the expansion in @xmath5 into a more rapidly converging expansion @xcite . in conclusion , we have shown that the lde for the quantity @xmath88 that determines the leading order shift in @xmath30 converges in the large @xmath4 limit if we use an appropriate prescription . if the pms criterion is used to optimize the convergence , the errors seem to decrease exponentially in the order of the lde . this suggests that the lde may also be a systematically improvable approximation scheme in the case @xmath2 relevant to bose - einstein condensation . with the use of order - dependent mappings , it may be possible to to develop the lde into a general and powerful tool for quantitative calculations in superrenormalizable field theories , even at a critical point . this research was supported by doe grant no . de - fg02 - 91-er4069 . in this appendix , we give the expressions for the diagrams that contribute to @xmath91 through order @xmath240 . the diagrams for @xmath91 are obtained by inserting propagator corrections into the diagram in fig . [ fig : delta ] , with the propagator corrections subtracted at @xmath228 momentum . in addition , any 2-loop self - energy subdiagram like that in fig . [ fig:2loop ] must also be subtracted at @xmath228 momentum . we first introduce a compact notation for the massive propagator and for the functions of the momentum that correspond to subdiagrams in which two points are connected by 2 or 3 lines : @xmath268 where @xmath269 denotes the 3-dimensional integral over the momentum @xmath270 @xmath271 there is a simple analytic expression for the function @xmath272 : @xmath273 the function @xmath274 is ultraviolet divergent , but the divergence can be eliminated by a subtraction at @xmath228 momentum . the subtracted function can be evaluated analytically : @xmath275 .\ ] ] we proceed to give the expression for the contributions to @xmath88 using the prescription ii given in ( [ d - ii ] ) . the contribution from the 2-loop propagator correction in fig . [ fig:2loop ] is @xmath276 . \label{a8}\end{aligned}\ ] ] the contribution from the sum of the 3-loop propagator corrections in fig . [ fig:3loop ] is @xmath277 d_{2}^{2}\left ( q\right ) . \label{a10}\end{aligned}\ ] ] the contributions from the 4-loop propagator corrections in fig . [ fig:4loop ] are @xmath278 where the momentum integrals are @xmath279 d_{2}^{3}\left ( q\right ) , \label{a16 } \\ i_{4b } & = & \int_{p}\frac{1}{\left ( p^{2}\right ) ^{2}}\int_{qr}d_{1}\left ( q\right ) d_{1}\left ( q - r\right ) d_{2}\left ( r\right ) \nonumber \\ & & \hspace{3 cm } \times \left [ d_{1}\left ( r - p\right ) d_{2}\left ( p - q\right ) -d_{1}\left ( r\right ) d_{2}\left ( q\right ) \right ] , \label{a17 } \\ i_{4c } & = & \int_{p}\frac{1}{\left ( p^{2}\right ) ^{2}}\int_{qrs}\left [ d_{1}\left ( p+q\right ) -d_{1}\left ( q\right ) \right ] d_{1}\left ( r\right ) d_{1}\left ( q+r\right ) \nonumber \\ & & \hspace{3 cm } \times d_{1}\left ( s\right ) d_{1}\left ( q+s\right ) d_{2}\left ( r - s\right ) , \label{a18 } \\ i_{4d } & = & \int_{p}\frac{1}{\left ( p^{2}\right ) ^{2}}\int_{q}\left [ d_{2}\left ( p+q\right ) -d_{2}\left ( q\right ) \right ] d_{1}^{2}\left ( q\right ) \left [ d_{3}\left ( q\right ) -d_{3}\left ( 0\right ) \right ] , \label{a19 } \\ i_{4e } & = & \int_{p}\frac{1}{\left ( p^{2}\right ) ^{3}}\left [ d_{3}\left ( p\right ) -d_{3}\left ( 0\right ) \right ] ^{2}. \label{a20}\end{aligned}\ ] ] if one uses the prescription iii for @xmath88 given in ( [ d - ii ] ) , the corresponding expressions for the diagrams are obtained by replacing each factor of @xmath280 in the integrand by @xmath281 . we denote the integral corresponding to @xmath232 by @xmath233 . if one uses the prescription i of ref . @xcite , there is also a diagram that is @xmath282 order in @xmath5 : @xmath283 . \label{j0}\end{aligned}\ ] ] the @xmath32 in the class of diagrams that survive in the large-@xmath4 limit is shown in fig . [ fig : largen ] . the expression for the first three diagrams in the series are given in ( [ a7 ] ) , ( [ a9 ] ) and ( [ a11 ] ) . the general expression for the @xmath284-loop diagram is @xmath285 ^{n-1}\left [ d_{1}\left ( p+q\right ) -d_{1}\left ( q\right ) \right ] . \label{a22}\end{aligned}\ ] ] if we use prescription iii , the corresponding integral is @xmath286^{n-1 } \left [ d_{1}(p+q ) -d_{1}(q ) \right ] . \label{a23}\ ] ] a closed - form expression for the @xmath287 factor associated with this class of diagrams is @xmath288 where we have chosen the overall normalization such that @xmath289 for @xmath290 and @xmath291 this expression reproduces the @xmath287 factor in ( [ a7 ] ) , ( [ a9 ] ) and ( [ a11 ] ) . note that for @xmath292 the large-@xmath4 limit of the coefficient is @xmath293 , \qquad n\geq 3.\ ] ] the corrections are suppressed by powers of @xmath294 if we use the simpler expression @xmath295 the error is of order @xmath19 for @xmath296 and of order @xmath297 for @xmath8 of order @xmath4 . in this appendix , we evaluate the momentum integrals that appear in the expressions for the diagrams in appendix a. we first consider the integrals @xmath232 that arise if we use prescription ii for @xmath88 given in ( [ d - ii ] ) . the 3-loop and 4-loop integrals ( [ a8 ] ) and ( [ a10 ] ) can be evaluated analytically : @xmath298 \frac{1}{m^{2}}.\end{aligned}\ ] ] the remaining integrals must be evaluated numerically . three of the 5-loop integrals can be reduced to integrals over a single momentum . the integral @xmath299 in ( [ a20 ] ) is already in such a form . the integrals @xmath300 in ( a16 ) and @xmath301 in ( [ a19 ] ) can be reduced to this form by first integrating over @xmath302 @xmath303 & = & -\frac{m}{4\pi } \frac{1}{\left ( q^{2}+m^{2}\right ) ^{2 } } , \\ \int_{p}\frac{1}{\left ( p^{2}\right ) ^{2}}\left [ d_{2}\left ( p+q\right ) -d_{2}\left ( q\right ) \right ] & = & -\frac{1}{32\pi ^{2}}\frac{1}{q^{2}+4m^{2}}.\end{aligned}\ ] ] the resulting expression for the integrals are @xmath304 \frac{1}{q^{2}+4m^{2}}.\end{aligned}\ ] ] reducing these integrals to 1-dimensional integrals and evaluating then numerically , we obtain @xmath305 the remaining two 5-loop integrals can be expressed as 3-loop integrals with the mercedes - benz topology . the integral @xmath244 in ( [ a17 ] ) already has this form . the integral @xmath245 in ( [ a18 ] ) can be put in this form by integrating over @xmath182 : @xmath306 these integrals can be reduced to 7-dimensional integrals that are then evaluated numerically using the monte carlo method to obtain @xmath307 the integrals @xmath233 required if @xmath88 is calculated using the prescription iii differ from the integrals @xmath190 calculated above only in the substitution @xmath308 . the 3-loop integral can be evaluated analytically : @xmath309 the 4-loop and 5-loop integrals must be evaluated numerically . four of them can be reduced to integrals over a single momentum . the integrals @xmath310 and @xmath311 already have this form . the integral @xmath312 and @xmath313 can be put into this form by first integrating over @xmath182 : @xmath314 & = & -\frac{3}{8\pi } \frac{m}{\left ( q^{2}+m^{2}\right ) \left ( q^{2}+4m^{2}\right ) } , \\ \int_{p}d_{1}^{2}\left ( p\right ) \left [ d_{2}\left ( p+q\right ) -d_{2}\left ( q\right ) \right ] & = & \frac{1}{32\pi ^{2}}\frac{1}{mq } \left [ \arctan \frac{q}{3m}-\arctan \frac{q}{2m}\right ] .\end{aligned}\ ] ] reducing the final momentum integrals to a 1-dimensional integral and then evaluating it numerically , we obtain @xmath315 the remaining two 5-loop integrals can be expressed as 3-loop integrals with the mercedes - benz topology . the integral @xmath244 in ( [ a17])already has this form . the integral @xmath245 in ( [ a18 ] ) can be put in this form by integrating over @xmath316 these integrals can be reduced to 7-dimensional integrals that are then evaluated using the monte carlo method to obtain @xmath317 if we use the prescription i of ref . @xcite , we must also evaluate the integral @xmath318 defined in ( [ j0 ] ) : @xmath319 the integrals @xmath190 that survive in the large-@xmath4 limit can all be reduced to integrals over a single momentum . after interchanging orders of integration in ( [ a22 ] ) and integrating over @xmath182 , the remaining integral reduces to @xmath320 ^{n-1 } , \label{b24}\ ] ] where the function @xmath195 is defined in ( [ a - y ] ) . applying the same strategy to ( [ a23 ] ) , we obtain @xmath321 ^{n-1}. \label{b25}\ ] ] p. arnold and g. d. moore , phys . lett . * 87 * , 120401 ( 2001 ) . p. arnold , g. d. moore and b. tomasik , phys . a * 65 * , 013606 ( 2002 ) . stoof , phys . a * 45 * , 8398 ( 1992 ) ; m. bijlsma and h.t.c . stoof , phys . a * 54 * , 5085 ( 1998 ) . a. duncan and h. f. jones , phys . d * 47 * , 2560 ( 1993 ) ; c. arvanitis , h. f. jones and c. s. parker , phys . rev . d * 52 * , 3704 ( 1995 ) . r. guida , k. konishi and h. suzuki , ann . phys . * 241 * , 152 ( 1995 ) ; ann . phys . * 249 * , 109 ( 1996 ) . w. janke and h. kleinert , quant - ph/9502019 ; h. kleinert and w. janke , phys . * a206 * , 283 ( 1995 ) . b. bellet , p. garcia and a. neveu , int . j. mod . phys . a * 11 * , 5587 ( 1996 ) .

we have carried out a blind search in the general direction of the galactic anticenter for absorption of the cosmic microwave background ( cmb ) radiation near 4.83 ghz by molecular clouds containing gaseous ortho - formaldehyde ( hco ) . the observations were done using the 25-m radio telescope at onsala in sweden , and covered strips in galactic latitude at several longitudes in the region . spectra were obtained in these strips with a grid spacing corresponding to the telescope resolution of . we have detected hco cmb absorption at of the survey pointings . this detection rate is likely to increase with further improvements in sensitivity , and may become comparable to the detection rate expected from a blind co survey with a corresponding sensitivity limit . we have mapped some of these detections in more detail and compared the hco absorption to existing maps of co(1 - 0 ) emission in the same regions . there appears to be a rough correlation between the velocity - integrated line strength of the co(1 - 0 ) emission and that of the hco absorption . however , the scatter in this correlation is significantly larger than the measurement errors , indicating differences of detail at and below the linear resolution of our observations ( pc ) . although these two tracers are expected to have similar excitation requirements on the microscopic level characteristic of warm , tk , dense , condensations in molecular clouds , the co(1 - 0 ) line is expected to be optically thick , whereas the hco line is not . this latter difference is likely to be responsible for a significant part of the scatter in the correlation we have found .

astro-ph0607616

following on the discovery of the anomalous absorption of the cosmic microwave background ( cmb ) at 4.83 ghz by gas - phase ortho - formaldehyde ( h@xmath0co ) molecules in a few nearby galactic dark nebula @xcite , several surveys were carried out in hopes of establishing the general galactic distribution of distant dusty molecular clouds containing gaseous h@xmath0co . @xcite used the nrao 140-foot telescope ( beam fwhm @xmath9 , system temperature t@xmath10 k ) to survey 30 positions spread out along the galactic plane near @xmath11 in the range @xmath12 and chosen to be free of radio continuum emission . with the exception of the one position near to the galactic center ( which did indeed appear to have continuum emission in the general area ) , no detections of cmb absorption could be registered , with a typical peak limits ( 5@xmath13 ) of 0.07 k in a velocity channel of width 1.6 km s@xmath14 . gordon & roberts concluded either that the excitation temperature of the general galactic distribution of h@xmath0comust be close to the brightness temperature of the cmb ( now known to be 2.73 k ) , or that the dust clouds harboring the h@xmath0co must be much smaller than the telescope beam . this disappointing result was corroborated with additional observations by @xcite using the 25-m radio telescope of the onsala space observatory ( oso ) in sweden ( fwhm @xmath15 , t@xmath16 k ) in order to map three @xmath17 fields in the galactic plane at @xmath18 , and @xmath19 . spectra were obtained in each of these fields on a grid with separation @xmath3 . no emission or absorption was found , with a typical peak limit of 0.15 k in a 0.62 km s@xmath14 channel . the first large - scale survey for h@xmath0co along the galactic plane was carried out by @xcite using the jodrell bank mark ii radio telescope ( fwhm @xmath20 , t@xmath21 k ) . observations were made at @xmath11 every @xmath22 of galactic longitude in the range @xmath23 and every @xmath24 in the range @xmath25 . these observations were successful in recording h@xmath0coabsorption in the inner galaxy , and some approximate information on the spatial distribution was obtained . however , the signal dropped to undetectable levels beyond @xmath26 , and no observations were attempted in the outer galaxy . in fact , at no position did the absorption - line profile depth exceed the observed continuum temperature . one can therefore safely conclude that what was being measured was not the anomalous cmb absorption ( which was still apparently too weak ) , but rather absorption of the galactic background radio radiation , which is strongest in the inner galaxy . although gaseous h@xmath0co may be nearly absent in parts of the ism because it is dissociated or frozen out on grains , these early surveys provided an indication that the physical conditions under which detectable anomalous h@xmath0co absorption occurs may not be common in the galaxy . observers turned their attention to other , more easily detected molecules , notably co , and we now have extensive surveys of the co(1 - 0 ) line over large sections of the galactic plane @xcite . these surveys , along with many detailed studies of specific molecular clouds in a wide range of molecular tracers , have all contributed to a much more complete ( and much more complicated ) view of physical conditions in the cool molecular ism . an explanation for the anomalous cmb absorption by h@xmath0co was first suggested by @xcite using a classical calculation for collisional excitation . subsequent work , especially by evans and his collaborators @xcite , confirmed this result using quantum mechanical calculations and observations . the collisional pumping mechanism is more effective at high collision rates , so in general the absorption is strongest at higher densities and temperatures . however , the calculations by @xcite showed that the mechanism would still be effective at rather low temperatures , below about 10k . more precise quantum mechanical calculations reported by @xcite suggested a smaller effect at very low kinetic temperatures , but both methods involved approximations . this leaves open the possibility that high - density , cold clumps of molecular gas in the ism may be detectable in h@xmath0co absorption , and has provided part of the motivation for our observing program . our approach is to carry out long integrations at a set of blindly - selected positions in the general direction of the galactic anticenter . for these observations we have again used the 25-m oso radio telescope at onsala , the same telescope used by gordon & hglund more than 33 years ago in their failed attempt to detect the general h@xmath0co cmb absorption . our present success is due entirely to the availability of more sensitive receivers and to generous allocations of observing time on this telescope . our survey has two main features : first , it is a `` blind '' survey ; we purposely avoided using maps of any other ism tracer to construct the observing program . second , we chose to observe in the galactic anticenter region , i.e. in the general direction of the outer galaxy . the galactic nonthermal background at 6 cm is exceedingly faint in this direction , so we can be fairly confident that any absorption we might detect is indeed anomalous cmb absorption . we will return to this point later . also , the velocity gradient owing to galactic rotation is small in the anticenter direction , so we might hope for some degree of `` bunching '' of absorption features , enhancing the probability of detection . the observations were obtained during two sessions , the first in september - october 2004 , and the second in may 2005 , with the 25 m onsala radio telescope . at 6 cm , the angular resolution of the telescope is fwhm @xmath27 , and the pointing precision is better than @xmath28 . the local oscillator was operated in frequency - switching mode with a `` throw '' of 0.8 mhz , and both ( circular ) polarizations of the incoming signal were recorded simultaneously in two independent units of a digital autocorrelator spectrometer . each of these units provided 800 channels spaced by 4 khz . at the frequency of the 1@xmath29 - 1@xmath30 line of ortho - h@xmath0co ( 4829.660 mhz ) , this setup provided a total bandwidth of 3.2 mhz = 199 km s@xmath14 and a ( twice - hanning - smoothed ) velocity resolution of 16 khz = 0.99 km s@xmath14 . daily observations of cas a were made in order to check the overall performance of the receiver . the system temperature during our sessions varied from 33 to 36 k. the off - line data reduction was done with the class / gildas software system @xcite , and involved only the subtraction of ( flat ) baselines from individual integrations and the averaging of all spectra taken at the same pointing position . the total integration time at each pointing position was @xmath31 hours , yielding a final typical rms noise level of 0.0035 k ( t@xmath32 ) and a 5@xmath13 detection limit of 0.0175 k in each @xmath33 km s@xmath14 channel . for our blind search , integrations were made every @xmath3 in 11 strips perpendicular to the galactic plane from @xmath34 to @xmath35 at intervals of @xmath22 in galactic longitude from @xmath36 to @xmath37 . the spectrometer was centered at a slightly different systemic velocity for each strip ( cf . table [ table : velocity ] ) , following the expected run of radial velocity according to the `` standard '' model of rotation in the outer galaxy using hi data from @xcite . cccccccccccc longitude ( deg . ) & 170 & 172 & 174 & 176 & 178 & 180 & 182 & 184 & 186 & 188 & 190 + velocity ( km s@xmath14 ) & -12 & -9.7 & -8.5 & -7.3 & -5.4 & -3.1 & -1.0 & 1.4 & 3.1 & 4.3 & 6.2 the final ( averaged ) spectra are shown as a mosaic in figure [ fig : blindsearch ] . there is clear evidence for absorption in 10 12 positions , and hints of absorption in several more . it is evident that , with our increased level of sensitivity , we have successfully detected h@xmath0co absorption at roughly 10% of the total of 143 positions observed . it is also interesting to note that we have not recorded any h@xmath0co emission ; this may indicate that very dense gas with @xmath38 ( see [ sec : discussion ] ) is rare , with a low area filling factor , and/or the gas is much colder than about 10 k. for reference , figure [ fig : blindsearch ] also shows the disposition of our survey positions with respect to the co emission obtained from the survey by @xcite . we will return to this comparison later , but we wish to emphasize here that our choice of h@xmath0cosurvey positions took no prior account of the distribution of co emission . nevertheless , it is clear from this figure that the h@xmath0co absorption is seen most strongly in regions of strong co emission . figure [ fig : blindsearch ] also shows regions of faint co emission without corresponding h@xmath0co absorption ; for instance several survey points at @xmath39 are located in a region of faint co emission . the absence of h@xmath0co absorption at such positions is likely to be a reflection of the sensitivity limit of our observations . the presence of two regions of relatively strong absorption can be identified in figure [ fig : blindsearch ] , the first at @xmath40 , and the second at @xmath41 . we have mapped these two regions in more detail in order to examine the distribution of h@xmath0co absorption and to permit a point - by - point comparison with the existing co(1 - 0 ) emission surveys in this region of the galaxy . the receiver settings for these two maps were identical to those used for the blind survey , and the total integration times per pointing position were similarly long . figure [ fig : firstmosaic]a ( left panel ) shows profiles at 63 of 66 positions on a grid centered ( position 0,0 ) at @xmath42 , @xmath11 , with an interval of @xmath3 . similarly , figure [ fig : secondmosaic]a ( left panel ) shows profiles at 101 of 121 positions on a grid centered ( position 0,0 ) at @xmath43 , @xmath11 , also with an interval of @xmath3 . how can we be sure that the absorption we have recorded , both in our blind search and in our two detailed maps , is actually absorption of the cmb ? we have earlier argued that it is not likely to be absorption of the galactic nonthermal background , since this background is very faint in the outer galaxy at 6 cm , so we can confidently rule out the type of absorption recorded in the inner galaxy by @xcite . but what about discrete continuum sources in the background , perhaps distant regions or radio galaxies ? in order to examine this possibility , we have retrieved radio continuum survey data at 21-cm from @xcite covering the two regions we have mapped in detail . figure [ fig:21 cm ] shows the 21-cm continuum in these two regions observed with the effelsberg 100-m telescope at a resolution of @xmath44 ( contours ) overlaid on our h@xmath0co absorption data ( grey scale , resolution @xmath27 ) . from the general lack of correspondence of the radio continuum peaks with the h@xmath0co absorption maxima we can safely conclude that it is indeed the cmb absorption we have recorded in h@xmath0co . furthermore , the continuum sources are either behind the h@xmath0co absorbers , or the area filling factor of the absorbing clouds is small and the line of sight to the sources just misses the clouds . our blind search has succeeded in recording anomalous h@xmath0coabsorption at approximately 10% of the survey positions . this is a significant improvement over the earlier blind searches of @xcite and @xcite , which failed to record any such absorption . since we have no reason to believe that the regions chosen for the early searches are in any way peculiar , we conclude that our success is likely due to the large improvement in sensitivity of our observations . our rms noise level is typically 0.0035 k ( t@xmath32 ) for a 5@xmath13 detection limit of 0.0175 k. when compared at the same velocity resolution , this is an improvement of @xmath45 over the observations of @xcite and a factor @xmath46 over those of @xcite . this result strongly suggests that the detection rate is simply sensitivity limited , and that it would increase with further increases in sensitivity . future searches would benefit from using larger telescope apertures with more collecting area ; beam dilution would also be reduced on the more distant dust clouds . we have retrieved the co(1 - 0 ) emission line data for our two mapped fields from the survey by @xcite , smoothed that data slightly from its original @xmath47 resolution to the @xmath3 resolution of our h@xmath0co observations , and extracted profiles at the same positions as on our maps near @xmath42 , @xmath11 and @xmath43 , @xmath11 . the results are shown in the right panels of figure [ fig : firstmosaic ] and figure [ fig : secondmosaic ] . a cursory inspection of these mosaics shows that there is an overall general correlation between the two tracers , in particular , at every position where we have detected h@xmath0co , co(1 - 0 ) emission is also detected . the converse is not always true ; there are many positions where co(1 - 0 ) is easily detected which show no corresponding h@xmath0co above the noise . although this may simply be a consequence of a lower s / n for the h@xmath0co observations , a more detailed look at these profiles reveals that s / n is not the whole story . in particular , the h@xmath0co absorption profiles are not merely scaled versions of the co(1 - 0 ) emission profiles . consider the profiles on the mosaic of figure [ fig : secondmosaic ] at @xmath43 , in particular those at ( -20,+10 ) , ( -10,+10 ) , and ( 0,+10 ) . in h@xmath0co this sequence of 3 profiles shows a uniform decrease in the depth of the absorption from about -0.05 to -0.04 to -0.03 k , but in co the emission profiles peak at about 6 , 7 , and 6 k respectively . another way of showing the general correlation between co and h@xmath0co is depicted in figures [ fig : firstcomparison ] and [ fig : secondcomparison ] , which show contour maps of the two components in a latitude - velocity plot integrated over a small range in longitude . figure [ fig : firstcomparison ] shows the two velocity components in this region , a weaker component at @xmath48 km s@xmath14 , and a stronger component at @xmath49 km s@xmath14 . figure [ fig : secondcomparison ] shows the data for the region centered at @xmath43 in the same latitude - velocity presentation . the principal component here is observed at @xmath50 km s@xmath14 . these figures confirm the general similarity of the co emission and the h@xmath0co absorption . the h@xmath0co cmb absorption clearly generally traces the same molecular gas seen in the more ubiquitous ( and easier to observe ) co molecule . a general correspondence of h@xmath0co and @xmath51co(1 - 0 ) was also noted by @xcite in their extensive mapping study of h@xmath0co and oh in the orion region . those authors found a broad general agreement in that well - known star - forming region , but concluded that the detailed agreement was poor . they found a better correspondence with @xmath52co(1 - 0 ) , and concluded that the reason for this was that the optically - thick @xmath51co(1 - 0 ) line was primarily tracing gas temperature , whereas the @xmath52co(1 - 0 ) emission and the h@xmath0co absorption are both optically - thin lines that will trace primarily the density . in general we agree with this interpretation , although as we will discuss in more detail later the dependence of the h@xmath0co absorption on kinetic temperature is also likely to be playing a role . in order to study the relation between h@xmath0co absorption and co(1 - 0 ) emission in more detail , we have computed the moments of the profiles shown in figures [ fig : firstmosaic ] and [ fig : secondmosaic ] . the results are listed in tables [ table : mom182 ] and [ table : mom190 ] ( _ see the electronic version of this paper _ ) . if two components are visible at a given pointing , the moments for each have been computed separately ; in many cases a second component was too weak to be identified in h@xmath0co . components that exceed the @xmath53 limit are indicated in column 6 of these tables and their values plotted in the correlation diagram of figure [ fig : intensityratio ] . since the data in the two fields did not seem to show any different trends , we have included the profile intensities from both fields in figure [ fig : intensityratio ] . we have fitted the data in figure [ fig : intensityratio ] by least squares to a straight line ; the result is : @xmath54 yielding a value of @xmath55 and @xmath56 . while the fit is not bad , in several cases the data points lie significantly off the best - fit line . we will return to this point in the discussion below . in order to have some idea of the distances to our detections , we have searched the catalog of star - forming regions in the galaxy @xcite for objects at known distances which may be associated with them . table [ table : iras ] lists several iras sources which are candidates , and their associated distances . these distances were obtained from the references listed in the table , and were determined from optical ( regions ) and radio spectroscopy ( co(1 - 0 ) in molecular clouds ) and using a model rotation curve of the outer galaxy . ccccc iras 05490 + 2658 & 182.4 , + 0.3 & s242 & 2.0 - 2.1 & 1 , 2 + iras 05431 + 2629 & 182.1 , -1.1 & & 0.1 * & 3 + iras 06067 + 2138 & 189.1 , + 1.1 & & 0.9 * & 3 + iras 06051 + 2041 & 189.7 , + 0.3 & & 2.3 * & 3 + iras 06055 + 2039 & 189.8 , + 0.3 & gem ob1 , s252 & 1.5 - 2.9 * & 3 , 4 , 5 + iras 06063 + 2040 & 189.9 , + 0.5 & gem ob1 , afgl5183 & 1.5 - 2.8 * & 3 , 4 , 6 + iras 06068 + 2030 & 190.1 , + 0.5 & s252 , afgl5184 & 1.5 - 2.7 * & 1 , 3 , 6 , 7 + iras 06061 + 2028 & 190.0 , + 0.3 & & 2.3 * & 3 + iras 06079 + 2007 & 190.5 , + 0.6 & & 2.6 * & 3 + in figures [ fig : iras180 ] and [ fig : iras190 ] we show contour plots of the h@xmath0co and co profile fluxes integrated over velocity for our two detailed fields . the iras sources from table [ table : iras ] are indicated with the black triangles . we conclude that the correspondence is only marginal , but the results suggest that the regions we have found are likely to be located at distances of 1.5 - 3 kpc , probably in the perseus arm of the galaxy ( 2 3 kpc from the sun ) . in that case our @xmath3 beam corresponds to a linear resolution of 4 - 9 pc . the general correlation between the integrated line intensities of the co(1 - 0 ) emission and the h@xmath0co absorption suggests that the excitation characteristics of these two lines are similar . in fact , both of these lines trace warm , dense molecular gas , as various calculations have shown in many historical papers . for instance , @xcite show examples of emergent brightness calculations for co(1 - 0 ) in the galactic gmc lynds 1204 in the context of one popular model ( the large - velocity - gradient model ) . as their figure 1 shows , the observed brightness of this ( usually optically - thick ) line is essentially proportional to the kinetic temperature of the gas as long as the density exceeds the critical density of @xmath57 . it follows that any sensitivity - limited co(1 - 0 ) survey will therefore preferentially record the warm , dense regions of the ism . the details of h@xmath0co 6-cm line formation have been studied for example in several papers by evans and co - workers ( @xcite ; @xcite ) . the 6 cm absorption line arises through collisions to higher rotational levels followed by radiative decay to the ground level , which becomes overpopulated owing to small differences in collision cross sections . the line is therefore generally stronger if the collision rate is higher , so this too favors warm , dense clouds . however , the absorption line is quenched at very high collision rates , when the level populations approach a boltzmann distribution , and the 6 cm line then goes into emission . in a future paper on h@xmath0co mapping of the large galactic dust cloud lynds 1204 @xcite we will present model computations of h@xmath0co absorption in more detail . for the moment we note that the 6 cm absorption is strongest in the density range of @xmath58 . in our model the absorption is deepest at @xmath59 , first appearing at @xmath60 k , and saturating at @xmath61 k in the sense that the absorption hardly increases for higher temperatures . this dependence on kinetic temperature is therefore somewhat different than is the case for the co(1 - 0 ) line ; this is mostly a consequence of the fact that the h@xmath0coline remains optically thin , so it more closely reflects the specific excitation conditions . one of the premises for doing the blind search for h@xmath62co towards the galactic anticenter was the possibility of finding molecular gas in regions where co emission had not previously been detected . this could potentially reveal a component of the molecular gas characterized by an excitation temperature lower than that corresponding to the energy level of the first rotational transition of co ( or any other commonly observed molecule ) . such an excitationally cold molecular gas component would only be observable in absorption , either against a continuum background source or , when using the anomalous absorption of h@xmath62co , against the cmb . the fact that we do not see h@xmath62co absorption outside the regions where co emission is observed , suggests that either such cold molecular gas is not present , or , as we will argue , the anomalous h@xmath62co absorption is not sensitive to excitationally cold molecular gas . as shown by townes & cheung ( 1969 ) and evans et al . ( 1975 ) , the anomalous h@xmath62co absorption is due to collisional pumping and most effective at densities between @xmath63 @xmath8 . our observations of the 6-cm h@xmath62co line therefore seem to exclude the existence of dense molecular gas outside the regions probed by co emission . however , the effectiveness of the collisional pumping at low temperatures has not been decisively determined ( cf . evans 1975 ; garrison et al . 1975 ) , and we can not exclude the possibility of very cold molecular gas outside the co emission regions , even if it meets the density requirement for inversion . in rodriguez et al . ( 2006 ) we found that the effectiveness of the collisional pumping leading to the anomalous excitation of the 6-cm h@xmath62co transition is indeed temperature dependent , requiring a relatively high temperature . the combined requirement of a high temperature and density in order to render the anomalous h@xmath62co absorption line observable means that whenever present , the h@xmath62co absorption and co emission lines will be spatially co - existing . hence , the 6-cm h@xmath62co absorption seen against the cmb is not a viable tracer of cold molecular gas . this is discussed in more detail in rodriguez et al . * we have successfully detected anomalous cmb absorption by h@xmath0co - bearing dust clouds at @xmath4 of our blind survey positions in the direction of the outer galaxy . no emission profiles were found . * our success is likely due to the large improvement in sensitivity of our observations over that of earlier surveys . this strongly indicates that the detection statistics will improve further if even higher - sensitivity searches are carried out . since the absorption signal strength from more distant clouds will suffer from increasing beam dilution , future searches ought to be done with larger radio telescopes . * h@xmath62co absorption and co emission lines are spatially co - existing . all observed h@xmath0co absorption features were associated with know co emission . * we have found a rough correspondence between the h@xmath0co and the @xmath51co(1 - 0 ) line fluxes in detailed maps of two regions in our survey area . we have argued that both lines generally trace warm , dense gas in the ism , with the situation for co(1 - 0 ) being somewhat simpler owing to the fact that this line is usually optically thick . we thank prof . roy booth , director ( ret . ) of the radio observatory at onsala , for generous allocations of telescope time and for his warm hospitality during our several visits to the observatory . we are also grateful to the observatory technical and administrative staff for their capable assistance with our observing program . t. dame kindly provided a digital copy of relevant parts of the cfa co(1 - 0 ) survey of the galaxy . we acknowledge the financial support of dgapa , unam and conacyt , in mxico , and the _ director s discretionary research fund _ at the space telescope science institute . rrrrrc 30 , -70 & & & 6.0 @xmath64 1.7 & -9.6 @xmath64 2.8 + 20 , -70 & @xmath65 -9.3 & & 7.2 @xmath64 1.7 & -10.6 @xmath64 2.7 + 10 , -70 & @xmath65 -14.0 & & 6.4 @xmath64 0.8 & -11.9 @xmath64 1.8 + & & & 6.7 @xmath64 1.3 & -4.5 @xmath64 1.0 + 0 , -70 & @xmath65 -14.0 & & 4.5 @xmath64 0.8 & -11.9 @xmath64 2.6 + & & & @xmath65 3.9 & + -10 , -70 & & & @xmath66 5.7 & + -20 , -70 & & & @xmath65 6.0 & + 30 , -60 & @xmath65 -5.7 & & @xmath67 & + 20 , -60 & @xmath65 -6.9 & & 10.0 @xmath64 1.6 & -9.3 @xmath64 1.6 + 10 , -60 & -6.7 @xmath64 1.9 & -5.7 @xmath64 3.2 & 7.3 @xmath64 1.3 & -4.2 @xmath64 0.8 & y + & -4.5 @xmath64 1.2 & -10.3 @xmath64 3.3 & 12.2 @xmath64 0.9 & -10.7 @xmath64 0.9 & y + 0 , -60 & @xmath65 -3.0 & & @xmath65 3.0 & + & & & 6.7 @xmath64 0.9 & -10.7 @xmath64 1.8 + -10 , -60 & @xmath65 -6.3 & & @xmath65 6.0 & + -20 , -60 & @xmath65 -6.9 & & @xmath65 6.0 & + 30 , -50 & @xmath65 -6.6 & & 6.9 @xmath64 1.7 & -9.1 @xmath64 2.3 + 20 , -50 & @xmath65 -6.6 & & 10.0 @xmath64 1.7 & -9.4 @xmath64 1.7 + 10 , -50 & -7.8 @xmath64 2.1 & -3.2 @xmath64 2.4 & 15.0 @xmath64 1.0 & -10.2 @xmath64 0.8 & y + & & & @xmath65 3.9 & + & & & @xmath65 3.3 & + 0 , -50 & -5.0 @xmath64 1.0 & -9.7 @xmath64 2.3 & 11.3 @xmath64 1.6 & -8.9 @xmath64 1.3 & y + -10 , -50 & @xmath65 -6.0 & & 5.2 @xmath64 1.2 & -10.6 @xmath64 2.7 + & & & @xmath65 4.5 & + -20 , -50 & @xmath65 -6.0 & & @xmath65 6.9 & + 30 , -40 & @xmath65 -6.3 & & @xmath65 6.0 & + 20 , -40 & @xmath65 -7.2 & & 8.0 @xmath64 1.8 & -8.6 @xmath64 2.1 + 10 , -40 & @xmath65 -4.2 & & 17.1 @xmath64 1.2 & -10.5 @xmath64 0.8 + & & & @xmath65 3.0 & + & & & @xmath65 2.5 & + 0 , -40 & @xmath65 -3.9 & & 13.6 @xmath64 1.2 & -10.2 @xmath64 1.1 + & & & @xmath65 3.0 & + & & & @xmath65 2.5 & + -10 , -40 & @xmath65 -4.5 & & 7.6 @xmath64 1.5 & -10.5 @xmath64 2.3 + & & & @xmath65 5.1 & + -20 , -40 & @xmath65 -7.5 & & @xmath65 6.3 & + 30 , -30 & @xmath65 -8.4 & & 8.3 @xmath64 1.8 & -8.3 @xmath64 1.9 + 20 , -30 & @xmath65 -6.3 & & 12.6 @xmath64 2.0 & -8.7 @xmath64 1.4 + 10 , -30 & -8.7 @xmath64 2.5 & -8.8 @xmath64 2.6 & 18.8 @xmath64 1.0 & -10.6 @xmath64 0.7 & y + & & & @xmath65 3.9 & + 0 , -30 & -8.6 @xmath64 1.3 & -9.7 @xmath64 1.3 & 14.3 @xmath64 1.1 & -10.3 @xmath64 0.9 & y + & & & @xmath65 4.2 & + -10 , -30 & @xmath65 -9.0 & & 5.8 @xmath64 0.8 & -10.1 @xmath64 1.7 + & & & @xmath65 3.6 & + -20 , -30 & -7.3 @xmath64 2.3 & 0.7 @xmath64 2.7 & @xmath65 5.7 & + 30 , -20 & @xmath65 -7.5 & & @xmath65 6.3 & + 20 , -20 & -10.3 @xmath64 1.9 & -9.4 @xmath64 1.9 & 11.8 @xmath64 1.8 & -8.6 @xmath64 1.4 & y + 10 , -20 & -15.8 @xmath64 1.8 & -9.7 @xmath64 0.9 & 19.1 @xmath64 1.0 & -9.8 @xmath64 0.6 & y + & & & @xmath65 3.3 & + 0 , -20 & -9.1 @xmath64 1.3 & -9.7 @xmath64 1.2 & 13.4 @xmath64 1.1 & -10.4 @xmath64 1.0 & y + & & & 4.6 @xmath64 1.3 & 3.6 @xmath64 1.2 + -10 , -20 & @xmath65 -7.5 & & 4.6 @xmath64 1.2 & -10.1 @xmath64 3.1 + & & & 5.2 @xmath64 1.5 & 3.3 @xmath64 1.1 + -20 , -20 & @xmath65 -7.5 & & @xmath65 3.6 & + & & & @xmath65 4.5 & + 30 , -10 & @xmath65 -7.5 & & @xmath65 5.7 & + 20 , -10 & @xmath65 -4.8 & & @xmath65 6.0 & + 10 , -10 & -5.7 @xmath64 1.3 & -9.6 @xmath64 0.8 & 10.0 @xmath64 1.0 & -9.9 @xmath64 1.1 & y + & & & 4.0 @xmath64 1.1 & 3.8 @xmath64 1.2 + 0 , -10 & -11.3 @xmath64 1.5 & -9.7 @xmath64 0.8 & 9.4 @xmath64 1.0 & -10.2 @xmath64 1.2 & y + & & & 6.3 @xmath64 1.2 & 2.8 @xmath64 1.3 + -10 , -10 & @xmath65 -6.0 & & @xmath65 3.6 & + & & & @xmath65 4.8 & + -20 , -10 & @xmath65 -6.9 & & @xmath65 4.5 & + & & & @xmath65 3.1 & + 30 , 0 & @xmath65 -6.0 & & @xmath65 3.6 & + & & & @xmath65 3.0 & + 20 , 0 & @xmath65 -6.6 & & @xmath65 3.3 & + & & & @xmath65 2.7 & + 10 , 0 & @xmath65 -6.3 & & 9.0 @xmath64 1.1 & -10.4 @xmath64 1.4 + & & & @xmath65 4.2 & + 0 , 0 & @xmath65 -4.8 & & 8.1 @xmath64 1.0 & -10.4 @xmath64 1.5 + & & & @xmath65 3.9 & + -10 , 0 & @xmath65 -5.7 & & @xmath65 3.9 & + & & & @xmath65 3.8 & + -20 , 0 & @xmath65 -7.2 & & @xmath65 3.6 & + & & & @xmath65 2.4 & + 30 , 10 & @xmath65 -7.5 & & 7.0 @xmath64 1.2 & 0.8 @xmath64 1.3 + & & & @xmath65 3.0 & + 20 , 10 & @xmath65 -5.7 & & 7.0 @xmath64 1.4 & 1.4 @xmath64 1.3 + & & & @xmath65 5.1 & + 10 , 10 & -8.3 @xmath64 2.3 & 2.2 @xmath64 0.8 & 7.3 @xmath64 1.3 & 2.0 @xmath64 1.3 & y + & & & 8.5 @xmath64 1.0 & -10.5 @xmath64 1.5 + 0 , 10 & @xmath65 -4.5 & & 7.2 @xmath64 1.1 & 1.8 @xmath64 1.3 + & & & 8.5 @xmath64 1.0 & -9.7 @xmath64 1.3 + -10 , 10 & @xmath65 -6.6 & & 6.7 @xmath64 1.2 & 2.6 @xmath64 1.3 + & & & 6.3 @xmath64 0.9 & -9.9 @xmath64 1.7 + -20 , 10 & @xmath65 -8.1 & & @xmath65 4.2 & + & & & @xmath65 3.0 & + 30 , 20 & @xmath65 -8.2 & & @xmath65 3.9 & + & & & 3.7 @xmath64 1.0 & -9.7 @xmath64 3.2 + 20 , 20 & @xmath65 -6.6 & & 6.9 @xmath64 1.4 & 1.2 @xmath64 1.3 + & & & @xmath65 3.0 & + 10 , 20 & @xmath65 -6.0 & & 7.4 @xmath64 1.2 & 1.9 @xmath64 1.3 + & & & 4.3 @xmath64 1.2 & -9.1 @xmath64 2.8 + 0 , 20 & -7.3 @xmath64 1.6 & 1.5 @xmath64 0.5 & 10.3 @xmath64 1.2 & 2.3 @xmath64 1.3 & y + & & & 6.7 @xmath64 1.2 & -8.7 @xmath64 1.7 + -10 , 20 & -7.1 @xmath64 1.0 & 2.1 @xmath64 0.4 & 10.1 @xmath64 1.2 & 2.5 @xmath64 1.3 & y + & @xmath65 -4.5 & & 7.7 @xmath64 1.2 & -8.8 @xmath64 1.6 + -20 , 20 & @xmath65 -5.7 & & @xmath65 4.5 & + & & & 3.6 @xmath64 1.1 & -8.9 @xmath64 3.2 + 30 , 30 & @xmath65 -6.9 & & @xmath65 3.9 & + & & & 3.7 @xmath64 1.0 & -9.4 @xmath64 3.0 + 20 , 30 & @xmath65 -9.0 & & @xmath65 3.0 & + & & & @xmath65 3.6 & + 10 , 30 & @xmath65 -9.9 & & @xmath65 4.5 & + & & & @xmath65 3.3 & + 0 , 30 & @xmath65 -3.9 & & 4.5 @xmath64 1.0 & -8.6 @xmath64 2.1 + & & & @xmath65 3.7 & + -10 , 30 & @xmath65 -7.2 & & @xmath65 3.9 & + & & & 5.4 @xmath64 1.0 & -8.7 @xmath64 1.9 + -20 , 30 & @xmath65 -5.7 & & @xmath65 3.3 & + & & & @xmath65 3.0 & + rrrrrc 50 -30 & & & 5.8 @xmath64 0.9 & 6.6 @xmath645.6 + 40 -30 & @xmath65 -6.0 & & 6.2 @xmath64 0.7 & 8.5 @xmath64 6.2 + & & & 6.3 @xmath64 0.5 & 0.2 @xmath64 6.0 + 30 -30 & @xmath65 -8.1 & & 7.1 @xmath64 0.9 & 9.0 @xmath64 6.0 + & & & @xmath65 3.3 & + 20 -30 & @xmath65 -5.4 & & 7.1 @xmath64 0.7 & 11.6 @xmath64 5.0 + & & & 9.1 @xmath64 0.7 & 5.4 @xmath64 3.4 + & & & @xmath65 1.5 & + 10 -30 & @xmath65 -5.1 & & 7.4 @xmath64 0.9 & 6.7 @xmath64 5.2 + 0 -30 & @xmath65 -3.6 & & 5.8 @xmath64 0.8 & 5.1 @xmath64 4.6 + -10 -30 & @xmath65 -5.7 & & 6.0 @xmath64 1.0 & 6.3 @xmath64 6.3 + -20 -30 & @xmath65 -5.4 & & @xmath65 2.3 & + & & & 9.5 @xmath64 0.5 & 4.8 @xmath64 2.2 + & & & @xmath65 2.3 & + -30 -30 & @xmath65 -5.1 & & 7.7 @xmath64 1.1 & 7.4 @xmath64 6.3 + -40 -30 & @xmath65 -6.9 & & @xmath65 2.9 & + -50 -30 & & & @xmath65 3.0 & + 50 -20 & & & @xmath65 2.4 & + & & & @xmath65 3.0 & + 40 -20 & @xmath65 -5.7 & & 6.0 @xmath64 0.8 & 7.0 @xmath64 6.3 + & & & 5.3 @xmath64 0.5 & -0.5 @xmath64 5.0 + 30 -20 & @xmath65 -5.7 & & 10.1 @xmath64 0.7 & 9.4 @xmath64 3.9 + & & & @xmath65 3.9 & + 20 -20 & @xmath65 -5.4 & & 8.6 @xmath64 0.7 & 11.2 @xmath64 4.0 + & & & 7.9 @xmath64 0.7 & 5.4 @xmath64 3.7 + & & & @xmath65 1.5 & + 10 -20 & @xmath65 -5.7 & & 12.1 @xmath64 1.0 & 9.1 @xmath64 4.2 + 0 -20 & @xmath65 -3.3 & & 10.2 @xmath64 1.0 & 6.7 @xmath64 3.7 + -10 -20 & @xmath65 -6.0 & & 5.9 @xmath64 0.9 & 5.6 @xmath64 5.2 + -20 -20 & @xmath65 -5.4 & & 9.8 @xmath64 1.0 & 6.7 @xmath64 4.0 + -30 -20 & @xmath65 -7.2 & & 10.3 @xmath64 0.8 & 7.1 @xmath64 3.1 + -40 -20 & @xmath65 -4.8 & & 7.2 @xmath64 1.0 & 9.8 @xmath64 6.0 + & & & @xmath65 3.2 & + -50 -20 & & & @xmath65 4.0 & + 50 -10 & & & 10.0 @xmath64 0.9 & 9.3 @xmath64 5.2 + 40 -10 & @xmath65 -5.4 & & @xmath65 3.3 & + 30 -10 & @xmath65 -5.7 & & 9.5 @xmath64 0.8 & 7.9 @xmath64 4.1 + 20 -10 & @xmath65 -5.1 & & 16.5 @xmath64 0.9 & 6.5 @xmath64 2.3 + 10 -10 & @xmath65 -7.8 & & 20.3 @xmath64 1.0 & 8.7 @xmath64 2.7 + 0 -10 & @xmath65 -3.0 & & 19.2 @xmath64 0.9 & 8.1 @xmath64 2.4 + -10 -10 & @xmath65 -6.6 & & 10.9 @xmath64 1.4 & 7.0 @xmath64 5.4 + -20 -10 & -10.7 @xmath64 2.9 & 6.7 @xmath64 6.7 & 16.9 @xmath64 1.0 & 7.7 @xmath64 2.8 & y + -30 -10 & @xmath65 -5.4 & & 10.5 @xmath64 0.7 & 8.4 @xmath64 3.8 + & & & @xmath65 3.3 & + -40 -10 & @xmath65 -6.9 & & 11.2 @xmath64 0.9 & 9.2 @xmath64 4.4 + & & & @xmath65 4.2 & + -50 -10 & & & 6.4 @xmath64 1.0 & 8.7 @xmath64 7.6 + 50 0 & & & 8.7 @xmath64 1.1 & 7.8 @xmath64 5.9 + 40 0 & @xmath65 -6.0 & & 6.1 @xmath64 0.8 & 6.6 @xmath64 5.1 + 30 0 & @xmath65 -5.4 & & 5.3 @xmath64 1.1 & 7.0 @xmath64 8.9 + 20 0 & @xmath65 -5.1 & & 9.9 @xmath64 1.2 & 6.8 @xmath64 4.8 + 10 0 & @xmath65 -6.6 & & 17.8 @xmath64 0.8 & 8.1 @xmath64 2.3 + 0 0 & -8.2 @xmath64 1.3 & 7.9 @xmath64 4.8 & 20.9 @xmath64 1.0 & 8.2 @xmath64 2.4 & y + -10 0 & -5.8 @xmath64 1.6 & 7.8 @xmath64 8.8 & 21.2 @xmath64 0.9 & 7.6 @xmath64 2.0 & y + -20 0 & -14.8 @xmath64 2.4 & 8.5 @xmath64 5.4 & 28.2 @xmath64 1.0 & 7.6 @xmath64 1.8 & y + -30 0 & @xmath65 -5.4 & & 9.1 @xmath64 0.7 & 1.2 @xmath64 3.1 + & & & 15.4 @xmath64 0.9 & 8.7 @xmath64 3.1 + -40 0 & @xmath65 -6.6 & & 9.1 @xmath64 0.6 & 1.2 @xmath64 1.7 + & & & 12.0 @xmath64 0.8 & 8.9 @xmath64 3.9 + -50 0 & & & @xmath65 4.5 & + & & & @xmath65 2.6 & + 50 10 & & & 5.4 @xmath64 1.1 & 6.2 @xmath64 7.6 + 40 10 & @xmath65 -6.9 & & @xmath65 3.6 & + 30 10 & @xmath65 -6.0 & & 7.7 @xmath64 1.1 & 5.5 @xmath64 4.8 + 20 10 & @xmath65 -4.8 & & 9.7 @xmath64 1.1 & 6.2 @xmath64 4.3 + 10 10 & -14.8 @xmath64 2.3 & 6.3 @xmath64 4.7 & 25.7 @xmath64 0.9 & 7.7 @xmath64 1.7 & y + 0 10 & -11.0 @xmath64 1.7 & 7.0 @xmath64 5.0 & 31.3 @xmath64 1.4 & 7.9 @xmath64 2.1 & y + -10 10 & -16.5 @xmath64 2.2 & 8.3 @xmath64 5.2 & 40.5 @xmath64 0.8 & 7.5 @xmath64 1.2 & y + -20 10 & -22.2 @xmath64 1.9 & 7.9 @xmath64 3.2 & 38.9 @xmath64 0.8 & 7.4 @xmath64 1.1 & y + -30 10 & @xmath65 -6.6 & & 18.2 @xmath64 0.6 & 8.5 @xmath64 2.0 + & & & 9.6 @xmath64 0.5 & 0.7 @xmath64 1.5 + -40 10 & @xmath65 -5.7 & & 14.1 @xmath64 0.6 & 7.8 @xmath64 2.5 + & & & 9.1 @xmath64 0.5 & 1.4 @xmath64 1.7 + -50 10 & & & 5.7 @xmath64 1.1 & 4.1 @xmath64 5.3 + 50 20 & & & 15.4 @xmath64 1.0 & 6.3 @xmath64 2.5 + 40 20 & @xmath65 -6.0 & & 9.9 @xmath64 1.0 & 6.1 @xmath64 3.4 + 30 20 & @xmath65 -5.1 & & 7.7 @xmath64 0.9 & 5.7 @xmath64 3.8 + 20 20 & @xmath65 -6.9 & & 7.0 @xmath64 0.9 & 7.0 @xmath64 5.4 + 10 20 & @xmath65 -6.6 & & 15.0 @xmath64 0.8 & 7.6 @xmath64 2.6 + 0 20 & -17.3 @xmath64 1.7 & 8.2 @xmath64 3.8 & 29.4 @xmath64 0.9 & 7.6 @xmath64 1.5 & y + -10 20 & -29.4 @xmath64 2.0 & 8.3 @xmath64 2.7 & 50.2 @xmath64 0.6 & 7.9 @xmath64 0.9 & y + -20 20 & -19.8 @xmath64 2.0 & 8.0 @xmath64 3.8 & 19.7 @xmath64 0.5 & 8.3 @xmath64 1.4 & y + -30 20 & @xmath65 -7.5 & & 17.7 @xmath64 0.7 & 8.7 @xmath64 2.0 + -40 20 & @xmath65 -6.9 & & 10.5 @xmath64 0.6 & 9.6 @xmath64 3.5 + & & & @xmath65 4.8 & + -50 20 & & & 5.2 @xmath64 0.9 & 4.2 @xmath64 4.9 + 50 30 & @xmath65 -11.4 & & 14.6 @xmath64 1.2 & 7.1 @xmath64 3.7 + 40 30 & @xmath65 -5.7 & & 17.6 @xmath64 1.3 & 6.2 @xmath64 2.3 + 30 30 & -9.3 @xmath64 1.8 & 7.2 @xmath64 6.4 & 11.4 @xmath64 1.0 & 6.7 @xmath64 3.5 & y + 20 30 & @xmath65 -4.2 & & 7.0 @xmath64 0.8 & 6.5 @xmath64 4.2 + 10 30 & @xmath65 -5.7 & & 24.6 @xmath64 1.1 & 8.0 @xmath64 2.3 + 0 30 & -10.5 @xmath64 1.3 & 6.4 @xmath64 6.3 & 22.9 @xmath64 0.9 & 8.1 @xmath64 1.9 & y + -10 30 & -9.6 @xmath64 2.2 & 4.7 @xmath64 5.4 & 30.5 @xmath64 0.7 & 7.2 @xmath64 1.1 & y + -20 30 & @xmath65 -5.4 & & 12.9 @xmath64 0.6 & 7.3 @xmath64 1.9 + -30 30 & @xmath65 -4.5 & & @xmath65 4.9 & + & & & 6.8 @xmath64 0.5 & 9.4 @xmath64 4.1 + -40 30 & @xmath65 -6.6 & & 6.9 @xmath64 0.5 & 2.8 @xmath64 4.8 + & & & 8.1 @xmath64 0.5 & 9.9 @xmath64 4.0 + -50 30 & & & 9.4 @xmath64 0.9 & 4.5 @xmath64 2.8 + 50 40 & @xmath65 -12.0 & & 18.9 @xmath64 0.8 & 6.7 @xmath64 1.9 + 40 40 & @xmath65 -6.3 & & 22.4 @xmath64 1.0 & 7.3 @xmath64 2.0 + 30 40 & @xmath65 -6.6 & & 13.5 @xmath64 0.9 & 6.7 @xmath64 2.8 + 20 40 & @xmath65 -6.6 & & 8.1 @xmath64 1.0 & 6.7 @xmath64 4.8 + 10 40 & @xmath65 -6.3 & & 28.4 @xmath64 1.0 & 7.6 @xmath64 1.7 + 0 40 & -12.8 @xmath64 1.4 & 7.2 @xmath64 3.6 & 22.9 @xmath64 1.0 & 7.8 @xmath64 2.0 & y + -10 40 & -11.0 @xmath64 2.2 & 7.4 @xmath64 5.7 & 22.3 @xmath64 0.8 & 6.9 @xmath64 1.7 & y + -20 40 & @xmath65 -7.5 & & 12.5 @xmath64 0.8 & 6.4 @xmath64 2.4 + -30 40 & @xmath65 -7.8 & & 15.5 @xmath64 0.8 & 5.7 @xmath64 1.7 + -40 40 & @xmath65 -6.0 & & 10.2 @xmath64 0.4 & 3.8 @xmath64 1.4 + & & & @xmath65 3.8 & + -50 40 & & & 21.8 @xmath64 1.0 & 4.0 @xmath64 1.4 + 50 50 & @xmath65 -8.4 & & 6.0 @xmath64 1.1 & 5.6 @xmath64 6.0 + 40 50 & @xmath65 -8.1 & & 14.7 @xmath64 1.2 & 6.0 @xmath64 2.8 + 30 50 & @xmath65 -6.6 & & 16.9 @xmath64 0.9 & 6.6 @xmath64 2.3 + 20 50 & @xmath65 -7.5 & & 12.9 @xmath64 1.1 & 7.2 @xmath64 3.8 + 10 50 & @xmath65 -7.8 & & 7.4 @xmath64 1.1 & 8.1 @xmath64 6.9 + 0 50 & @xmath65 -3.6 & & @xmath65 4.0 & + -10 50 & @xmath65 -6.0 & & @xmath65 4.5 & + -20 50 & @xmath65 -5.7 & & 11.1 @xmath64 0.8 & 5.4 @xmath64 2.4 + -30 50 & @xmath65 -6.6 & & 15.2 @xmath64 1.2 & 4.4 @xmath64 2.2 + -40 50 & -13.6 @xmath64 1.4 & 6.7 @xmath64 4.4 & 17.2 @xmath64 1.0 & 4.6 @xmath64 1.8 & y + -50 50 & -13.1 @xmath64 2.4 & 7.0 @xmath64 4.3 & 31.2 @xmath64 0.8 & 6.0 @xmath64 1.1 & y + 50 60 & & & 5.7 @xmath64 1.0 & 4.8 @xmath64 4.9 + 40 60 & @xmath65 -4.2 & & 8.3 @xmath64 1.2 & 4.4 @xmath64 3.8 + 30 60 & @xmath65 -5.7 & & 14.5 @xmath64 0.9 & 5.9 @xmath64 2.2 + 20 60 & -11.9 @xmath64 1.8 & 7.9 @xmath64 5.1 & 8.7 @xmath64 1.0 & 7.9 @xmath64 3.8 & y + & & & @xmath65 2.2 & + 10 60 & @xmath65 -6.9 & & @xmath65 1.5 & + & & & @xmath65 4.2 & + & & & @xmath65 2.5 & + 0 60 & @xmath65 -4.2 & & @xmath65 3.6 & + -10 60 & @xmath65 -5.4 & & @xmath65 4.5 & + -20 60 & @xmath65 -8.4 & & 7.3 @xmath64 0.9 & 4.3 @xmath64 3.2 + -30 60 & @xmath65 -6.0 & & 7.1 @xmath64 0.7 & 5.2 @xmath64 3.3 + -40 60 & @xmath65 -7.8 & & 11.7 @xmath64 1.0 & 5.0 @xmath64 2.6 + -50 60 & -23.2 @xmath64 2.7 & 5.8 @xmath64 3.3 & 37.6 @xmath64 0.9 & 6.2 @xmath64 1.0 & y + 50 70 & & & @xmath65 4.6 & + 40 70 & & & @xmath65 3.0 & + 30 70 & & & 8.3 @xmath64 1.1 & 7.2 @xmath64 5.7 + 20 70 & & & 7.5 @xmath64 0.9 & 9.8 @xmath64 5.0 + & & & @xmath65 2.4 & + 10 70 & & & @xmath65 5.1 & + 0 70 & @xmath65 -6.0 & & @xmath65 3.9 & + -10 70 & @xmath65 -7.8 & & @xmath65 4.6 & + -20 70 & @xmath65 -6.9 & & @xmath65 4.6 & + -30 70 & @xmath65 -7.5 & & @xmath65 2.6 & + & & & @xmath65 4.8 & + -40 70 & @xmath65 -6.6 & & 7.4 @xmath64 0.7 & 2.1 @xmath64 5.7 + & & & 7.1 @xmath64 0.7 & 8.9 @xmath64 5.8 + -50 70 & -18.5 @xmath64 2.7 & 4.6 @xmath64 3.5 & 34.7 @xmath64 0.9 & 4.6 @xmath64 1.0 & y + dummytext avedisova , v.s . , 2002 , , 79 , 216 blitz , l. , fich , m. , & stark , a.a . 1982 , , 49 , 183 carpenter , j.m . , snell , & r.l . , schloerb , f.p . 1995 , , 445 , 246 cohen , r.j . , matthews , n. , few , r.w . , & booth , r.s . 1983 , , 203 , 1123 combes , f. 1991 , , 29 , 195 dame , t.m . , ungerechts , h. , cohen , r.s . , de geus , e.j . , grenier , i.a . , may , j. , murphy , d.c . , nyman , l.a . , & thaddeus , p. 1987 , , 322 , 706 dame , t.m . , hartmann , d. , thaddeus , p. 2001 , , 547 , 792 evans , n.j . ii , , 201 , 112 evans , n.j . ii , zuckerman , b. , morris , g. , & sato , t. 1975 , , 196 , 433 few , r.w . 1979 , , 187 , 161 garrison , b.j . , lester , w.a . jr . , miller , w.h . , & green , s. 1975 , , 200 , l175 gordon , m.a . , & roberts , m.s . 1971 , , 170 , 277 gordon , m.a . , & hglund , b. 1973 , , 182 , 41 guilloteau s. , & forveille t. 1989 , _ grenoble image and line data analysis system ( gildas ) _ , iram , http://www.iram.fr/iramfr/gildas hartmann , d. , burton , w.b . 1997 , atlas of galactic neutral hydrogen helfer , t.t . , & blitz , l. 1997 , , 478 , 233 humphreys , r.m . 1978 , , 38 , 309 , 116 , 1899 moffat , a.f.j . , jackson , p.d . , & fitzgerald , m.p . 1979 , , 38 , 197 mundy , l.g . , evans , n.j . , snell , r.l . , & goldsmith , p.f . 1987 , , 318 , 392 palmer , p. , zuckerman , b. , buhl , d. , & snyder , l.e . 1969 , , 156 , l147 reich , p. , reich , w. , & furst , e. , 1997 , a&as 126 , 413 rodriguez , m.i . et . 2006 , in preparation snell , r.l . , huang , y .- l . , dickman , r.l . , & claussen , m.j . 1988 , , 325 , 853 snell , r.l . , dickman , r.l . , huang , y .- l . 1990 , , 352 , 139 townes , c.h . , & cheung , a.c . 1969 , , 157 , l103 wouterloot , j.g.a . , & brand , j. 1989 , , 80 , 149 young , k.e . , lee , j .- e . , evans , n.j . , goldsmith , p.f . , & doty , s.d . 2004 , , 614 , 252

we observed the bright phase of the 2003 outburst of the galactic black hole candidate h 1743 in x - rays simultaneously with _ chandra _ and _ rxte _ on four occasions . the _ chandra_/hetgs spectra reveal narrow , variable ( he - like ) fe xxv and ( h - like ) fe xxvi resonance absorption lines . in the first observation , the fe xxvi line has a fwhm of km / s and a blue - shift of km / s , suggesting that the highly ionized medium is an outflow . moreover , the fe xxv line is observed to vary significantly on a timescale of a few hundred seconds in the first observation , which corresponds to the keplerian orbital period at approximately ( where ) . our models for the absorption geometry suggest that a combination of geometric effects and changing ionizing flux are required to account for the large changes in line flux observed between observations , and that the absorption likely occurs at a radius less than for a black hole . viable models for the absorption geometry include cyclic absorption due to an accretion disk structure , absorption in a clumpy outflowing disk wind , or possibly a combination of these two . if the wind in h 1743 has unity filling factor , the highest implied mass outflow rate is 20% of the eddington mass accretion rate . this wind may be a hot precursor to the seyfert - like , outflowing `` warm absorber '' geometries recently found in the galactic black holes gx 339 - 4 and xte j1650 . we discuss these findings in the context of ionized fe absorption lines found in the spectra of other galactic sources , and connections to warm absorbers , winds , and jets in other accreting systems . [ firstpage ]

astro-ph0406272

the galactic x - ray transient h 1743@xmath0322 was classified as a black hole candidate by white & marshall ( 1984 ) based on its x - ray spectral characteristics . the source was discovered during its 19771978 outburst with _ heao-1 _ and _ ariel v _ ( doxsey et al . 1977 , kaluzienski & holt 1977 ) . the next convincing detection of h 1743@xmath0322 was made on 2003 march 21 with _ integral _ ( revnivtsev et al . 2003 ) and later with _ rxte _ ( markwardt & swank 2003a ) , earning the source the additional designations igr j17464@xmath03213 and xte j1746@xmath0322 . markwardt & swank ( 2003b ) pointed out that although the original _ _ detection reported two possible locations , it was very likely that the potentially new transient was actually h 1743@xmath0322 . following its 2003 re - activation in x - rays , the source was extensively monitored in optical , ir , and radio bands ( see , e.g. , rupen , mioduszewski , & dhawan 2003 ; steeghs et al . 2003 and ref . therein ) . two discoveries made during the 2003 outburst of h 1743@xmath0322 have provided additional evidence that this source harbors a black hole primary . first , homan et al . ( 2003 ) reported the detection of quasi - periodic oscillations ( qpos ) at 240 hz in the _ rxte _ x - ray flux . this frequency is typical of black hole candidates , and is tantalizingly close to the keplerian orbital frequency expected at the innermost stable circular orbit ( isco ) around a schwarzschild black hole ( @xmath6 ) . moreover , there is evidence for qpos at 160 hz , and therefore in the 2:3 ratio which has recently emerged in a few galactic black hole systems ( gro j1655@xmath040 : strohmayer 2001a , grs 1915@xmath7105 : strohmayer 2001b , xte j1550@xmath0564 : miller et al . 2001 , and possibly xte j1650@xmath0500 : homan et al . . the apparent 2:3 pairing may be evidence of a parametric epicyclic resonance in black hole accretion disks ( see , e.g. , abramowicz & kluzniak 2003 ) . however , there is evidence that black hole qpo pairs may drift out of a 2:3 ratio ( miller et al.2001 , remillard et al . 2002 ) , and the 2:3 ratio described by the resonance model for so - called khz qpos in neutron star systems is at odds with data ( miller 2003 ) . second , radio observations have revealed a jet in h 1743@xmath0322 , moving at @xmath8 ( rupen , mioduszewski , & dhawan 2004 ) . on the basis of this relativistic jet , h 1743@xmath0322 may be termed a `` microquasar '' . indeed , the x - ray spectroscopic results presented in this paper will further establish close similarities between h 1743@xmath0322 and microquasars like gro j1655@xmath040 and grs 1915@xmath7105 , in addition to their radio jet and x - ray timing similarities . although highly relativistic jets are not a unique black hole primary signature ( fender et al . 2004 ) , such jets are are more commonly observed in black hole systems than in neutron star systems . we observed h 1743@xmath0322 simultaneously with _ chandra _ and _ rxte _ on four occasions during the bright phase of its 2003 outburst , as part of our larger multi - wavelength black hole observing program . herein , we report the results of an analysis of the x - ray spectra resulting from these observations . most importantly , the _ chandra_/hetgs spectra reveal variable he - like fe xxv and h - like fe xxvi resonance absorption lines , which may be due to absorption in an outflow . variable , seyfert - like `` warm - absorber '' outflows have recently been detected in the x - ray spectra of the galactic black holes xte j1650@xmath0500 and gx 339@xmath04 ( miller et al.2004a , miller et al . we examine the absorption in h 1743@xmath0322 within the context of these recent results . the results of an analysis of the x - ray timing properties of the _ rxte _ observations will be reported in a companion paper ( homan et al.2004 ) . the results of radio observations made in partial support of this program will be reported in a separate paper by rupen et al . ( 2004 , in prep . ) , and results from optical and ir observations will be reported in steeghs et al . ( 2004 , in prep . ) we observed h 1743@xmath0322 with _ chandra _ and _ rxte _ on four occasions during the bright phase of its 2003 outburst , as part of ongoing and complementary target of opportunity programs to study galactic black holes in outburst in x - rays . the _ chandra _ observations were each approximately 50 ksec in duration . _ rxte _ made observations simultaneous with each of these _ chandra _ observations , with individual exposure times ranging between 0.6 ksec and 25 ksec . the observation dates , starting times , and net exposure times are given in table 1 . at the time of writing , pcu-2 is the best - calibrated pcu in the _ rxte_/pca , based on simple power - law fits to the crab spectrum . in order to best search for subtle features like broad fe k@xmath9 lines , we restricted our pca spectral reduction and analysis to pcu-2 . spectra from the hexte - a cluster were included for spectral reduction and analysis to constrain the high energy continuum . _ rxte _ data reduction tools in the heasoft version 5.2 suite were used to screen the event and spectral files . spectra from pcu-2 were extracted from data taken in `` standard 2 '' mode , providing full coverage of the 2.0 - 60.0 kev pca bandpass in 129 channels every 16 seconds . data from all of the xe gas layers in pcu-2 was combined . background spectra were made using the tool `` pcabackest '' using the latest `` bright source '' background model . redistribution matrix files ( rmfs ) and ancillary response files ( arfs ) were made and combined into a single file using the tool `` pcarsp . '' the background spectra were subtracted from the total spectrum using the tool `` mathpha . '' it is well - known that fits to pca spectra of the crab with a simple power - law model reveal residuals as large as 1% , but adding 1% systematic errors to pca data is often an over - estimate . using the tool `` grppha '' , we added 0.6% systematic errors to the spectra from pcu-2 . the lowest energy bins in the pca are poorly calibrated , and the calibration above 25 kev is uncertain ; we therefore restricted our pcu-2 spectral analysis to the 325 kev range . the hexte - a spectra were made from standard `` archive '' mode data , which has a nominal time resolution of 32 s and covers the 10250 kev range with 61 channels . all spectra were background - subtracted and deadtime corrected using the standard procedures . the calibration of the energy channels below 20 kev is poor , and we therefore restricted our analysis of the hexte spectra to energies above 20 kev . as some of the spectra are soft and therefore dominated by background above 100 kev , we fixed this energy as an upper bound for consistency . however , analysis of the spectra up to 250 kev was undertaken for those spectra with sufficient signal . the _ rxte _ spectra were analyzed using xspec version 11.2.0 ( arnaud & dorman 2000 ) . all errors on fits to the _ rxte _ spectra are 90% confidence errors . in each _ chandra _ observation , the high energy transmission grating spectrometer ( hetgs ) was used to disperse the incident flux onto the acis - s ccd array . the acis - s array was read - out in `` continuous - clocking '' mode ( nominal frametime : 2.8 msec ) to prevent photon pile - up . information is collapsed into one dimension to achieve the fast read - out in this mode ; thus , our _ chandra _ observations were purely spectroscopic as imaging information was not preserved . a 100-column `` gray filter '' window was used around the aimpoint on the acis - s3 chip , which acted to read - out only only one in every ten zeroth - order events . this step was taken to prevent telemetry saturation and frame drop - out on the s3 chip . we shifted the location of the aimpoint from the nominal s3 aimpoint . a y - coordinate offset of 0.33 arcmin was used to shift the wavelength at which ccd chip gaps occur in order to ensure a simple response function in the fe k line region , and a sim - z offset of @xmath07.5 mm was used to push the aimpoint toward the ccd read - out to limit charge transfer inefficiencies and to limit wear and tear on the prime imaging area of the s3 chip . spectra and lightcurves were extracted from the _ chandra _ event lists using the ciao version 2.3 reduction and analysis suite . the `` evt1 '' file was filtered to accept only the standard event grades , to accept only events from the nominal good time intervals , and to exclude bad pixels . the `` destreak '' tool was run to remove the effects of `` streaking '' in the ccds . although imaging information is not preserved in the mode used for these observations , the ciao hetgs spectral extraction routines are robust for locating the observation aimpoint , defining a wavelength grid , and extracting spectra . spectra were extracted using the default settings for the tools `` tg_resolve_events '' and `` tgextract . '' we used the canned redistribution matrix files ( rmfs ) to make ancillary response files ( arfs ) via the `` fullgarf '' tool . the first - order heg spectra and arfs were added to make a single spectrum and single arf using the tool `` add_grating_spectra '' ; the same addition was done with the first - order meg spectra . in the analysis detailed below , we only claim spectral features which are readily detected in the individual dispersed spectra ( prior to addition ) . chandra_/hetgs spectra were fit with isis version 1.1.3 ( houck & denicola 2000 ) . all errors on the absorption line parameters measured from the hetgs spectra are @xmath10 errors . in this paper , we have restricted our analysis to the first - order heg spectra . moreover , we fit only phenomenological power - law continua and simple gaussian line functions in a range spanning less than 1 . the primary reasons for these restrictions are ( 1 ) the low - energy portions of the meg and heg spectra do not reveal any evidence for narrow emission or absorption lines , and ( 2 ) only the heg spectra have sufficient sensitivity through the fe k line range to allow meaningful analysis and constraints . for the benefit of scientists examining these spectra in the future , we note some additional reasons behind the above restrictions , since they are related to nuances of the telescope and the observing mode employed . h 1743@xmath0322 had a flux roughly equivalent to or greater than 1 crab in the _ rxte_/asm ( 1.512.0 kev ) during the first _ observations , _ despite _ the high column density along the line of sight to this source . when fit with the spectral shapes that describe the _ rxte _ spectra well , the _ chandra_/hetgs spectra show positive residuals above approximately 8 kev , and also at energies below approximately 2 kev . these effects are inconsistent with photon pile - up , and inconsistent with the effects expected due to contaminant build - up on the acis chips . rather , the positive residuals above @xmath118 kev are likely due to incident photons scattering off the instrument support structure and falling into the high energy portion of the dispersed spectra ( which is closest to the zeroth order ) . the nature of the positive residuals below @xmath112 kev is less certain . if the standard ciao tools over - correct for charge transfer inefficiency at our offset , such a correction might partially account for the positive residuals . while these problems do not affect our ability to search for narrow emission and absorption lines through the full hetgs bandpass , we were unable to fit the full hetgs continuum with meaningful models , and we do not report such fits in this work . we note , however , that in the 2.57.5 kev range , the _ chandra_/hetgs spectra broadly agree with the _ rxte _ spectra . the positive residuals above @xmath118 kev made it impossible to measure the depth of the ( he - like ) fe xxv and ( h - like ) fe xxvi k edges at 8.8 kev and 9.3 kev ( respectively ) . balancing the optical depth of these k - shell absorption edges with the optical depth of the fe xxv and fe xxvi absorption lines we detected ( see below ) would have allowed tighter constraints . nevertheless , a number of important constraints can be made using only the variable fe xxv and fe xxvi absorption lines detected in these spectra , and we report measurements of the line parameters , a model for the absorption , and implications for the accretion inflow and outflow in h 1743@xmath0322 in this work . we made fits to each of the time - averaged _ rxte _ spectra listed in table 1 with simple , phenomenological spectral models . initial fits revealed that an equivalent neutral hydrogen column density of @xmath12 was easily within the error bars of each individual measurement . therefore , we fixed this value in all subsequent fits . this column is higher than that predicted from radio measurements ( @xmath13 , dickey & lockman 1990 ) ; however , the radio estimates are limited by the coarse spatial resolution of the survey observations . our spectral model consisted of the multicolor disk blackbody ( mcd ; mitsuda et al . 1984 ) and power - law components . as the name implies , the mcd model is included to account for soft , thermal , blackbody - like emission thought to arise from an optically - thick but geometrically - thin accretion disk . the power - law component is likely primarily the result of compton - upscattering of disk photons by a central , hot , electron - dominated `` corona '' . this region may not be distinct from the base of a jet ; if it is distinct , processes like synchrotron self - compton emission from the base of a jet may also contribute to the hard power - law component . the results of spectral fits in the 3100 kev band with this absorbed mcd plus power - law are reported in table 2 , and an example spectral fit is shown in figure 1 . the spectral parameters obtained through our fits are typical of galactic black hole candidates in bright phases . in general , statistically acceptable fits are obtained ( @xmath14 , where @xmath15 is the number of degrees of freedom in the fit ) with this simple model . ( the fits would be statistically improved if the xe edge at approximately 4.8 kev was better described in the detector response . ) the disk temperature varies between @xmath16 kev in our observations , indicating that the disk is likely important throughout the bright phase of the outburst and close to the innermost stable circular orbit ( isco ) . the broad - band spectral properties indicate that h 1743@xmath0322 was likely in an extended `` very high '' or `` steep power - law '' state in observations 1a , 1b , 1c , 2a , 2b , 2c . observation 3 occurred in a phase intermediate between the `` high / soft '' an `` very high '' ( `` steep power law '' ) state , but was likely more similar to the `` high / soft '' ( or , `` thermal - dominant '' ) state . observations 4a , 4b , and 4c likely occurred in the `` high / soft '' ( or , `` thermal - dominant '' ) state . the states observed in h 1743@xmath0322 differ in subtle ways from canonical black hole states , however , and the reader is referred to homan et al . 2004 ( in prep . ) for a more complete analysis and discussion of the states observed . the highest flux inferred in the 0.510.0 kev band is @xmath17 , which corresponds to a luminosity of @xmath18 ( see table 2 ) . at present , neither the mass of h 1743@xmath0322 nor the distance to the source are known . given its galactic center position and rather high column density , however , it is reasonable to use a galactic center distance of 8.5 kpc as an estimate of the distance to the source . at such a distance , the source emission was ( just ) sub - eddington for a @xmath19 black hole ( and more comfortably sub - eddington for a @xmath5 black hole ) . since homan et al . ( 2003a ) detected qpos at 240 hz and 160 hz in h 1743@xmath0322 , which may be indicative of black hole spin as the qpos are in a 2:3 ratio ( abramowicz & kluzniak 2003 ) , it is worth addressing whether or not our spectral fits support the inference of a disk extending very close to a potentially spinning black hole . the mcd model allows a measure of the inner disk radius through its normalization : @xmath20^{2 } \times cos(\theta)$ ] ( where @xmath21 is the inclination of the disk to the line of sight ) . the smallest disk normalization is measured in observation 2b . via the normalization , @xmath22 km for @xmath23 kpc and @xmath24 , and @xmath25 km for @xmath23 kpc and @xmath26 . the isco around a non - spinning black hole is @xmath27 ( where @xmath28 ; @xmath29 ) . for black hole masses greater than @xmath5 , @xmath30 km may imply spin and support any such implications from the qpos previously detected . however , this simple analysis has a few important problems and the inferred radii should be regarded cautiously . the mcd model does not include an inner boundary condition , and inclination effects and radiative processes in the disk can be important and have not been accounted for in this exercise ( see , e.g. , shimura & takahara 1995 , and merloni , fabian , & ross 2000 ) . moreover , disk continuum spectroscopy is particularly sensitive to the absorption model and to the model for the hard component . finally , qpo frequencies may not be simple keplerian frequencies . the broad - band _ rxte _ spectra are remarkable for what they _ do not _ show : first , although we have only fitted the data up to 100 kev ( since the softer spectra have little flux above this energy ) , there is no evidence for a break in the power - law component in any of the spectra . breaks may indicate the electron temperature of the putative corona . the absence of such an observed break in these spectra may indicate that the coronal electron temperature is generally very high ( @xmath31 kev , or greater ) . observation 2b provides an illustrative case . this observed spectrum is reasonably hard and has a high s / n , with h 1743@xmath0322 clearly detected out to 250 kev without a break . replacing the power - law component in our spectral model with the comptt comptonization model ( in which the coronal electron temperature is a variable , titarchuk 1994 ) , @xmath32 kev is obtained . the model underestimates the hard flux above 100 kev and is statistically poor ( @xmath33 ) , but indicates that the coronal temperature may be very hot indeed . second , the spectra are free of any features expected to arise from disk reflection ( see fig . 1 ) . hard x - ray irradiation of the ( relatively ) cool accretion disk by the corona is expected to produce a characteristic set of features : an fe k@xmath9 emission line , a corresponding fe k edge , and a compton - backscattering hump peaking between 2030 kev . disk reflection features are an extremely common , if not ubiquitous feature of broad - band galactic black hole x - ray spectra . yet , there is no evidence for any of these features in any of the broad - band _ rxte _ spectra we obtained . in table 2 , the 95% confidence upper limits on the strength of narrow and broad fe k@xmath9 emission lines are given ( fwhm@xmath34 kev and fwhm@xmath35 kev , as per the doppler broadening expected at the inner disk around a schwarzschild black hole ) . in most cases , the upper limits are stringent ( @xmath36 ev ) . for a disk at the isco , reflection from a neutral disk is expected to produce an emission line with an equivalent width of @xmath37 ev ( george & fabian 1991 ) . especially in bright , high accretion rate phases , the disks around galactic black holes are likely to be ionized , and ionized disk reflection models predict line equivalent widths that are a few times higher than the neutral case ( see , e.g. , ross , fabian , & young 1999 ) . we made broad - band reflection fits to observation 2b to further address the nature of the inner accretion flow geometry in h 1743@xmath0322 . we replaced the simple power - law model with the `` constant density ionized disk model '' ( cdid ross , fabian , & young 1999 ; ballantyne , ross , & fabian 2001 ) . the cdid model is particularly well - suited to highly ionized reflection , and includes fe k@xmath9 line emission self - consistently . we `` smeared '' the cdid model by convolving it with the laor relativistic emission line function ( laor 1991 ) . the disk emissivity ( @xmath38 ; @xmath39 is expected for a standard disk ) , the inner disk radius @xmath40 ( in units of @xmath4 ) , and inclination parameters in the smearing function variables in the spectral model . in all cases where the reflection fraction @xmath41 was allowed to vary , values along these lines were measured : log@xmath42 , @xmath43 , @xmath44 , @xmath45 , and @xmath46 ( 95% confidence ) . such fits were formally acceptable ( @xmath47 ) though clearly not statistically required over the phenomenological mcd plus power - law model . in fits with @xmath48 fixed , even for @xmath49 ( extreme smearing ) the fit was not acceptable @xmath50 . fits with higher values of @xmath51 fixed were statistically much worse . the _ rxte _ variability analysis conducted in this work follows the methods and procedures detailed in homan et al . ( 2004b ) . the reader is directed to that work for a full discussion of the timing analysis presented here . an example lightcurve showing strong qpo - like variability on the timescale of @xmath52 s is shown in figure 2 . a power density spectrum of _ rxte _ observation 1a is shown in figure 3 ; the variability seen in the lightcurve corresponds to the broad @xmath53 hz peak in the power density spectrum . it is clear that much of the power is concentrated in this feature . the term `` qpo '' is usually reserverd for features with @xmath54 when fit with a lorentzian ( where @xmath55 ) . as @xmath56 for the @xmath57 hz feature , it may not be termed a qpo . however , it is important to note that , like a qpo , its frational rms _ increases _ with increasing energy ( homan et al.2004b ; see figures 2 and 5 ) . we began by analyzing the combined first - order heg and meg time - averaged spectra from each observation in narrow 2 slices , through the 1.225 range . the continuum in each slice was fitted with a simple phenomenological power - law model . where required , absorption edge(s ) due to neutral ( atomic ) photoelectric absorption in the ism were included and scaled to @xmath58 ( see table 2 ) . line features were fit with simple gaussian models . line identifications are based on the calculations and line lists tabulated by verner , verner , & ferland ( 1996 ) . column density estimates were derived using the relation : @xmath59 + where @xmath60 is the column density of a given species , @xmath61 is the oscillator strength , @xmath62 is the equivalent width of the line in cm units , and @xmath63 is the wavelength of the line in cm units ( spitzer 1978 ) . in using this relation , we have assumed that the lines are optically thin and that we are on the linear part of the curve of growth . in observations 1 , 3 , and 4 , we clearly detect absorption lines which we identify as he - like fe xxv @xmath64 ( @xmath65 ) and h - like fe xxvi @xmath66 ( @xmath67 ) resonance absorption lines . the line wavelengths , velocity shifts , line breadths , fluxes , equivalent widths , columns , and equivalent neutral hydrogen columns are listed in table 3 . the spectra are shown in figure 4 . there is no clear evidence for any narrow emission or absorption lines in the rest of the spectrum . there are some line - like features in the region of the neutral si edge which might be he - like si xiii and h - like si xiv ; however , these are more likely due to uncertainties in the instrument response in this region ( the spectrum peaks in this region , and the ccds contain si ) . we have measured significant variability in the line equivalent widths and other parameters between observations , which demands changing absorption and clearly ties the lines to a region close to ( or within ) h 1743@xmath0322 . in observation 2 , when the source flux was high and twice as spectrally hard as in the other observations , the fe xxv and fe xxvi lines are absent . in observation 1 , the lines are significantly blue - shifted , suggesting that the absorbing matter was flowing into our line of sight . the fe xxvi line in observation 3 is also slightly blue - shifted . in all other cases , the lines are consistent with no net shift . in observations 1 and 4 , the fe xxvi lines are resolved , and have velocity widths ( fwhm ) of @xmath68 km / s and @xmath69 km / s , respectively . the rest of the lines are not resolved and likely have smaller velocity widths . we made lightcurves of the dispersed spectrum from each observation in the 0.510.0 kev band . the lightcurves are presented in figure 5 . strong variability is present in all observations on the timescale of @xmath52 s , but particularly in the first and second observations . this variability is confirmed to also be present in the lightcurves of the simultaneous _ rxte _ lightcurves of h 1743@xmath0322 ( see fig . 2 and fig . a 1012% `` dip '' feature may be present in the third _ chandra _ observation . a preliminary examination of other _ rxte _ observations in the public archive revealed much stronger dipping activity typical of dipping black hole binaries viewed at high inclinations ( see , e.g. , kuulkers et al . 1998 ) . to explore whether the absorption lines vary within the dip , we made spectra of observation 3 from before the dip and within the dip ( see figure 5 ) , and fit the spectrum in the manner noted above . again , the only lines apparent are the fe xxv and fe xxvi absorption lines found in the time - averaged spectrum . the results of fits to the spectrum prior to the dip and within the dip are given in table 4 , and the spectra are shown in figure 6 . both absorption lines are stronger within the dip . while the statistical significance of the variability in the fe xxvi line is marginal , the variability in the strength of the fe xxv line is clearly significant . indeed , in observation 3 , the fe xxv line is not clearly detected in the pre - dip spectrum . next , we investigated whether the absorption lines in observation 1 , and 4 vary on the @xmath52 s flaring variability timescale . we calculated the mean count rate for observations 1 , 2 , and 4 ( see fig . 5 ) , produced lists of time intervals for count rates above and below the mean rate , and extracted spectra from the time intervals above and below the mean rates . those spectra were also fitted in the manner described above ; the spectra are shown in figure 6 . as was the case with the time - averaged spectrum of observation 2 , the count - rate selected spectra of observation 2 show no absorption lines . in observation 4 , the strength of the lines does not vary significantly between the high rate and low rate spectra . in observation 1 , however , the absorption lines are marginally stronger in the low rate spectra than in the high rate spectra . this raises the interesting possibility that the absorption may vary on a timescale of @xmath52 s. to establish this more clearly , we created a list of time intervals in which the source count rate was more than 5 counts / s above the mean rate , and more than 5 counts / s below the mean rate , and extracted spectra from these time intervals . the resultant spectra are shown in figure 7 . the fe xxv absorption line is clearly stronger in the low count rate spectrum than in the high count rate spectrum , and the variation is statistically significant ( see table 4 ) . to interpret the iron absorption lines measured in the time - averaged spectra in more detail , we computed line profiles from a spherical wind with a photoionization code . we used the atomic physics packages from the x - ray illuminated accretion disk models of raymond ( 1993 ) . in this case , since the gas is optically thin , it is only necessary to consider a single slab . we used the spectral parameters taken from table 5 for each of the 4 observations and specify a density , slab thickness and distance from the central object , to model a 1-d , optically - thin wind . the code iterates to find a self - consistent temperature and ionization state , then computes equivalent widths assuming a doppler profile with widths ranging from 100 to 2000 @xmath70 . it assumes ionization equilibrium , which is a good approximation for the densities and dynamical times estimated below in all cases . from the output of the code we find the parameters that match the observed fe xxv and fe xxvi absorption line equivalent widths . the fe xxv and fe xxvi lines can not originate in exactly the same gas , because the line widths and line shifts differ . a plausible physical picture might consist of denser , slower clumps containing more fe xxv embedded in a less dense gas where fe xxvi dominates . however , a modest density contrast of order 2 ( approximately ) is sufficient to shift the balance between fe xxv and fe xxvi , so an average density that produces both lines is meaningful , and we do not have enough measured parameters to justify the additional free parameters of a more complex model ( the line widths are uncertain ) . we therefore choose densities to match the observed equivalent widths and doppler velocity widths intermediate between those of fe xxv and fe xxvi . there is a maximum radius at which a slab can plausibly be causing the observed absorption . the fe xxvi / xxv line flux ratio gives an ionization parameter ( @xmath71 ) , and for any given distance from the central source @xmath72 , a specific density @xmath73 is required . a specific slab thickness , which can not exceed @xmath72 , is required to produce the measured line column densities , so that @xmath74 . because the density that gives the ionization parameter scales as @xmath75 , the slab thickness scales as @xmath75 ( therefore @xmath76 ) , and there is a maximum radius at which a slab of thickness @xmath72 can reproduce the lines observed . we call this parameter @xmath77 . the lines can be formed at smaller @xmath72 , and if the lines are formed in a wind with @xmath75 density profile , the density at @xmath77 is correspondingly smaller . table 5 shows the derived parameters . we can also compute the mass loss rate in a wind of any velocity , because @xmath78 is constant . the velocity shifts in table 3 are generally uncertain , so we present mass loss rates for a reference wind speed of 300 @xmath79 . the table shows that the lack of absorption lines in the second observation can not entirely be attributed to a higher ionizing flux . although the ionizing flux in this observation is about double that of the first observation , an order of magnitude lower column density of absorbing material is needed to match the upper limits to the equivalent widths . an interesting result in table 5 is the small value of @xmath77 for the 4th observation , which results primarily from the small ionizing flux . the lines do not show a significant doppler shift , but they arise in a white dwarf - size region near the central source . in the first and third observations , the absorbing regions have a size comparable to a few solar radii , which is similar to the separation of the components in several binary black hole systems . the second interesting aspect of the table is the indication that the observed absorption lines are formed in a wind with a mass loss rate of order @xmath80 comparable to the inferred mass accretion rate . if we assume a distance of 8.5 kpc , take the highest inferred 0.510.0 kev luminosity of @xmath81 and assume an efficiency of 10% in the equation @xmath82 , we obtain @xmath83 . taking the highest measured 0.510.0 kev luminosity to be a lower limit on the true eddington luminosity , the highest inferred mass loss in the wind ( @xmath84 ) corresponds to an outflow rate that is 22% of the eddington mass accretion rate ( for a unity filling factor ) . note , however , that the fraction depends crucially on the filling factor , the energy range used , and the distance assumed . an outflow of this kind may be analogous to the accretion disk winds from cataclysmic variables in their high states and to t tauri stars in high states ( fu ori stars ) , which have wind mass loss rates of order 10% of their accretion rates . however , observation 4 shows a much smaller mass loss rate along with its lower luminosity . the high equivalent neutral hydrogen column density along the line of sight is consistent with a central galactic distance to h 1743@xmath0322 . for distances greater than @xmath853 kpc , the highest 0.510.0 kev unabsorbed flux observed ( see table 2 ) exceeds the eddington limit for a neutron star . moreover , the parameters obtained from broad - band spectral fits to h 1743@xmath0322 with the phenomenological mcd plus power - law model are typical for galactic black hole systems in bright phases ( for a review , see mcclintock & remillard 2003 , see also homan et al . 2001 ) . these facts and the presence of 240 hz qpos in h 1743@xmath0322 make it a near certainty that h 1743@xmath0322 harbors a black hole primary . however , the apparent absence of reflection features in the spectrum , even in phases with 1 ) a strong hard component , 2 ) a small inferred inner disk radius , and 3 ) high - frequency qpos , is at odds with the majority of galactic black holes . it is likely that the low reflection fraction obtained is due to a combination of astrophysical effects ( ionization and disk inclination ) and the limitations of our reflection modeling . zdziarski et al.(2003 ) have shown that when the disk is fully ionized , the reflected spectrum is nearly featureless , and might therefore be confused with a zero reflection scenario . high disk ionization tends to smooth - out reflection features ; the dips we have detected in the lightcurve of _ chandra _ observation 3 and in additional public _ rxte _ observations of h 1743@xmath0322 point to a system viewed at high inclination . at high inclinations , doppler shifts become very important , and tend to smear - out reflection features . the cdid reflection model we fit to _ rxte _ observation 2b indicates that the disk in h 1743@xmath0322 is likely highly ionized and seen at a high inclination . while these constraints are likely robust , it must be acknowledged that even the cdid model has limitations . it is implemented into xspec as a library of angle averaged spectra . finally , we note that if any reflection spectrum also passes through the hard x - ray emitting corona , which is likely if a corona hugs a disc viewed at high inclination , then the reflection component will be comptonized and much less detectable ( petrucci et al 2001 ) . it is possible , then , that the total reflection model the cdid model smeared with a line function ( for which inclination is a fit parameter ) does not address high inclinations correctly . were the combination of ionization and inclination effects more accurately described by our model , it is likely that a high reflection fraction would have been within errors . we have clearly detected variable , narrow fe xxv and fe xxvi resonance absorption lines in the spectra of h 1743@xmath0322 . in the recent past , similar absorption lines have been detected in the microquasars gro j1655@xmath040 ( ueda et al . 1998 ) , and grs 1915@xmath7105 ( kotani et al.2000 ) with _ asca _ , and in grs 1915@xmath7105 with _ ( lee et al . 2002 ) . unfortunately , each of these prior spectra suffered heavily from photon pile - up . the extraordinary sensitivity of the h 1743@xmath0322 spectra has allowed us to improve upon previous implications based on fe xxv / xxvi absorption lines in microquasars in two important ways : first , we have measured blue - shifts in the lines detected in observation 1 , providing direct evidence that the highly ionized absorbing geometry may be an outflow ( previously , outflows were only asserted ) . second , we have resolved the lines in two cases , and in other cases the lines are likely narrower than the instrumental resolution . the fwhm of the lines is vital for constraining the nature of the absorption . viable descriptions of the absorption geometry in h 1743@xmath0322 must be able to account for a number of our key findings , including : ( 1 ) the blue - shifts observed in the first _ observation , ( 2 ) the large line flux changes observed between observations ( in particular , the absence of absorption lines in the second _ chandra _ observation ) , and the stronger absorption seen in the dip observed during the third _ chandra _ observation , and ( 3 ) the absence of fe emission lines in the spectra , ( 4 ) the variability of the absorption lines in observation 1 on a timescale of a few hundred seconds , and ( 5 ) the absorption must occur at a radius smaller than the maximum radius at which our photoionization models suggest the lines may be produced . two geometries satisfy or partially satisfy most of these requirements : cyclic absorption in a disk structure resulting from the high inclination , and absorption in an ( potentially clumpy ) outflowing disk wind . the absence of fe emission lines in the _ chandra _ and _ rxte _ spectra suggests that the absorbing geometry is not spherical ( else , emission line photons would have scattered into our line of sight ) . indeed , modeling of the absorbers in gro j1655@xmath040 ( ueda et al.2004 ) and grs 1915@xmath7105 ( kotani et al . 2000 ) also suggested a cylindrical , perhaps pancake - like absorbing geometry . absorption in a disk structure perhaps a disk atmosphere is consistent with this constraint . however , this picture relies on relatively large covering and filling factors . the lack of emission lines in the spectra is also consistent with the absorption occurring in relatively dense clumps in a disk wind , with large covering factors but small filling factors . a clumpy wind is particularly appealing when one considers that at high mass accetion rates ( such as in the first _ chandra _ observation ) , our models for the absorber indicate that the wind would already be optically - thick at approximately @xmath86 . yet , we clearly see continuum spectral components consistent with viewing the innermost flow without significant obscuration . moreover , if qpos are tied to keplerian frequencies , the qpos detected ( homan et al . 2004 , in prep . ) indicate that emission from the innermost flow is either seen directly or scattered into our line of sight . a significant fraction of the absorption needs to occur in clumps with a high covering factor but small filling factor to be consistent with the observed continuum spectral properties and variability properties . the absence of absorption lines due to a combination of geometric effects and ionizing flux ( as implied by our photoionization models ) in the second _ chandra _ observation is easily explained by the clumpy outflow model . simply , a clump may not have intersected our line of sight to the central accretion observation within this observation . it is more difficult to explain the absence of absorption lines if the absorption occurs via cyclic absorption in a disk structure . the disk structure would have to be dampened or disappear in the second _ chandra _ observation . however , the ionizing flux is higher in this observation , and it has been shown that irradiation can induce warps in accretion disks ( see , e.g. , maloney , begelman , & pringle 1996 ) . to explain the enhanced absorption line fluxes measured within the dip in the third _ chandra _ observation , it is easy to envision a cool clump absorbing fe xxv more strongly than the hotter surrounding wind . alternatively , it is also possible that the dip observed is due to an unrelated geometric effect , and that the enhanced absorption line flux seen within the dip occurs because the wind along our line of sight is temporarily shielded and allowed to cool . if the absorption lines are instead due to absorption in a disk structure , a second warp or disk structure may need to be invoked to explain the dip itself . the variability of the fe xxv absorption line flux on timescales of a few hundred seconds in the first _ chandra _ observation can be explained by both models . indeed , this short timescale line variability is one of the most compelling arguments in favor of cyclic absorption in a disk structure . the period of the line variability broadly agrees with the keplerian orbital period at @xmath3 for a @xmath5 black hole . if h 1743@xmath0322 harbors a black hole with @xmath87 , then @xmath3 corresponds to @xmath88 cm , which is less than @xmath77 in observations 1 and 3 ( see table 5 ) . however , the observed variability can not be termed a qpo , and it is therefore less easily tied to a cyclic disk structure than if it qualified as a true qpo . if the absorption occurs in a disk structure , it would suggest that the variability itself is due to changes in the absorption in our line of sight . this is at odds with the fact that similar variability is observed in the absence of absorption lines in the second _ chandra _ observation . if the variability were due to absorption , the variability should be more pronounced at low energies ; in fact , the fractional rms of the peak at @xmath57 hz increases with higher energy ( homan et al . 2004b ; see figures 2 and 5 ) . although it is less exciting , absorption in a clumpy outflowing wind can likely account for the rapid line flux variability at least as well as cyclic absorption in a disk structure . simply , the gas ionization and recombination timescale is much faster than the variability timescale observed . if the variability is not due to a disk structure , but rather just due to fluctuations in the intensity of the central engine , an absorbing clump may be able to rapidly respond to the changing flux . finally , the blue - shifts observed in the first _ observation are more easily explained through an outflowing disk wind , than by absorption in the disk . the modest blue - shifts of the lines are a small fraction of the speed of light , and similar to the blue - shifted lines observed in gx 339@xmath04 and xte j1650@xmath0500 ( miller et al . 2004a ) , which are consistent with a disk wind . if the absorption is instead due to a cyclic disk structure , it is rather hard to explain the blue - shifts without significant fine - tuning . if the structure were a warp , its motion should be tangential to our line of sight , not into our line of sight . instead , the absorption might need to occur in a disk wind flowing outward along the disk , and further postulate that the wind density is modulated by a cyclic disk structure underlying the atmosphere . it has been shown that accretion disks may drive outflows with a significant component along the disk ( see , e.g. , murray & chiang 1997 , proga 2003 ) . on balance , absorption in an outflowing , clumpy wind appears to be the more compelling than absorption in a disk structure . both possibilities require a degree of fine - tuning to satisfy the observational constraints . a combination of these simple models may also be a viable description of the absorption geometry . the highly ionized outflow revealed in these _ chandra _ observations of h 1743@xmath0322 is very likely connected to the variable , outflowing , seyfert - like warm absorbers recently discovered in gx 339@xmath04 and xte j1650@xmath0500 ( miller et al . 2004a , miller et al.2004b ) . in those systems , the outflowing absorbers are consistent with disk winds or shell ejections . the absorbers in gx 339@xmath04 and xte j1650@xmath0500 are blue - shifted , and so clearly outflowing . in gx 339@xmath04 , he - like ne ix , mg xi , and o vii absorption lines were detected simultaneously with more strongly blue - shifted ne ii and ne iii lines , suggesting a `` clumpy '' outflow with regions of differing temperature and density . the fe xxv and fe xxvi lines in h 1743@xmath0322 also need to come from distinct regions , though the contrast needed is much less than for the wind observed in gx 339@xmath04 . the outflow observed in gx 339@xmath04 was likely detected at a lower fraction of the eddingtion mass accretion rate in that source ; its lower ionization can therefore be attributed to a lower central source flux . moreover , the outflow in gx 339@xmath04 implied an order of magnitude lower mass outflow rate than than the highest rate implied in h 1743@xmath0322 ( perhaps as high as 22% of the eddington inflow rate ) . the outflow in h 1743@xmath0322 may be clumpy , however , and have a low filling factor ; in that case , the outflow rate in h 1743@xmath0322 could be lower by a factor of @xmath89 , or more . it is plausible that the outflow in h 1743@xmath0322 is a hot , high @xmath90 precursor to the cooler , lower @xmath90 outflows observed in gx 339@xmath04 and xte j1650@xmath0500 ( miller et al . 2004a , 2004b ) . the mass outflow rate in the fourth _ chandra _ observation of h 1743@xmath0322 is far less than that in the first observation ( see table 5 ) , further indicating an evolution along the lines we suggest here . the outflows in these black hole systems are likely disk winds , and resemble the warm absorbers revealed in some seyfert galaxies ( see , e.g. , reynolds 1997 , elvis 2000 , morales & fabian 2002 ) . such absorption regions could be tied to rather equatorial disk outflows ( murray & chiang 1997 , proga 2003 ) , rather than to a highly collimated flow perpendicular to the disk . focused studies of galactic black hole winds in the future may enable us to learn about the evolution seyfert warm absorbers ; their evolution is difficult to study directly given the inherently longer timescales . it is worth addressing the possibility that a putative disk structure and the outflow are not merely coincident , but potentially causally related . for instance , a family of models has been discussed wherein jets or a wind can be launched vertically from a disk through a coupling between a spiral density wave in the disk and a rossby vortex ( tagger & pellat 1999 ; varniere , rodriguez , & tagger 2002 ; rodriguez , varniere , tagger , & durouchoux 2002 ) . it is possible , then , that the @xmath52s variations seen in h 1743@xmath0322 are due to a density structure which acts to drive jets close to the black hole , and that the same wave modulates a disk wind far from the black hole . it is not yet clear if such a mechanism could be at work in driving both jets and a disk wind , but if so , then the appearance of ionized fe absorption lines in strong jet sources like gro j1655@xmath040 , grs 1915@xmath7105 , and now h 1743@xmath0322 would be explained naturally . we note that using reduction methods exactly like those detailed for h 1743@xmath0322 , we have detected fe xxv and fe xxvi lines in a 60 ksec _ chandra_/hetgs spectrum of cygnus x-1 obtained on 3 march 2004 , when the source was in a canonical low / hard state ( miller et al.2004c , in prep . ) . cygnus x-1 is known to produce a compact , steady radio jet in this state ( stirling et al . highly ionized fe absorption lines have also been detected in the neutron star system gx 13@xmath71 ( sidoli et al . interestingly , gx 13@xmath71 is also a radio emitter in which a delay between its x - ray spectral hardness and radio brightness has been observed which is very similar to that seen in grs 1915@xmath7105 ( homan et al . 2004a ) , strongly suggesting a compact jet is also at work in this system . however , simultaneous radio observations of h 1743@xmath0322 with the vla / vlba reveal that the radio source was quite strong during the first and second _ chandra _ observations nearly 10 mjy at ghz frequencies ) but only 2 mjy during the third observation , and less than 1 mjy during the fourth observation ( rupen et al . 2004 , in prep . ) . thus , strong radio activity ( indicative of a jet ) is seen during the second _ chandra _ observation ( where absorption lines are not observed ) , and absorption lines are clearly observed in the third and fourth _ chandra _ observations when the radio flux was low . at minimum , this suggests that if jets and ionized absorption are causally related , they are not related in a simple way . both may merely be the effect of a strongly active corona , for instance . moreover , the detection of highly ionized fe absorption lines in high inclination neutron star binaries may signify that it is unlikely that ionized absorption lines are tied to jet production through a mechanism like aei . similar lines have also been seen in the persistent emission of the ( edge - on ) dipping neutron star binaries x 1254@xmath0690 ( boirin & parmar 2003 ) , x 1624@xmath0490 ( parmar et al.2002 ) , mxb 1658@xmath029 ( sidoli et al . 2001 ) and 4u 1916@xmath0053 ( boirin et al . 2004 ) . in contrast to the absorption lines in the black hole systems h 1743@xmath0322 , gx 339@xmath04 , and xte j1650@xmath0500 , the absorption lines in these edge - on neutron star binaries are not significantly blue - shifted . although the appearance of lines in both black hole and neutron star systems suggests a related physical origin , the lack of blue - shifts in the lines seen in neutron stars may hint at a different physical picture . without significant blue - shifts , it is difficult to clearly tie the lines to an outflow . the lines may not be due to absorption in an outflow , but might be due to absorption in ambient gas within the system . the absorption lines in mxb 1658@xmath029 , for instance , are not observed to vary with orbital phase or even during dipping events ( sidoli et al . 2001 ) , which suggests that the absorbing gas is not as close to the compact object as the absorbing gas in black hole systems . + we wish to thank cxc director harvey tananbaum and the cxc staff , and jean swank and the _ rxte _ staff , for making these too observations . we thank john houck for assistance with isis . jmm acknowledges helpful discussions with mike nowak , david huenemoerder , aneta siemiginowska , rob fender , herman marshall , and patrick wojdowski . jmm gratefully acknowledges support from the nsf through its astronomy and astrophysics postdoctoral fellowship program . whgl and jh gratefully acknowledge support from nasa . ds acknowledges support from the sao clay fellowship . this research has made use of the data and resources obtained through the heasarc on - line service , provided by nasa - gsfc . houck , j. c. , & denicola , l. a. , 2000 , astronomical data analysis software and systems ix , in asp conference proceedings , vol . 216 , eds . n. manset , c. veillet , & d. crabtree ; isis is available at http://space.mit.edu/cxc/isis lll instrument & start time & net exposure ( ksec ) + cxo / acis@xmath7hetgs & 2003 - 05 - 01t21:41:03 & 48.8 + rxte / pca 1a & 2003 - 05 - 01t17:00:32 & 25.2 + rxte / pca 1b & 2003 - 05 - 02t00:02:24 & 0.8 + rxte / pca 1c & 2003 - 05 - 02t00:54:08 & 3.3 + cxo / acis@xmath7hetgs & 2003 - 05 - 28t04:09:21 & 45.5 + rxte / pca 2a & 2003 - 05 - 28t05:28:48 & 1.5 + rxte / pca 2b & 2003 - 05 - 28t06:44:16 & 16.1 + rxte / pca 2c & 2003 - 05 - 28t14:28:32 & 5.6 + cxo / acis@xmath7hetgs & 2003 - 06 - 23t15:56:10 & 50.0 + rxte / pca 3 & 2003 - 06 - 23t17:05:20 & 13.3 + cxo / acis@xmath7hetgs & 2003 - 07 - 30t15:57:58 & 50.2 + rxte / pca 4a & 2003 - 07 - 30t19:48:48 & 0.6 + rxte / pca 4b & 2003 - 07 - 30t21:22:40 & 0.6 + rxte / pca 4c & 2003 - 07:31t00:32:48 & 0.6 + lllllllll observation & 1a & 1b & 1c & 2a & 2b & 2c & 3 & 4 + @xmath91 & @xmath92 & @xmath92 & @xmath92 & @xmath92 & @xmath92 & @xmath92 & @xmath92 & @xmath92 + @xmath93 ( kev ) & @xmath94 & @xmath94 & @xmath95 & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 + @xmath100 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & @xmath106 & @xmath107 & @xmath108 + @xmath109 & @xmath110 & @xmath111 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath116 & @xmath117 + @xmath118 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & @xmath125 & @xmath126 + @xmath127 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & @xmath134 & @xmath135 + @xmath136 ( 3 - 100 ) & @xmath137 & @xmath137 & @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142 + @xmath143 ( 3 - 100 ) @xmath144 & @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 + @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 + @xmath136 ( 0.5 - 10 ) & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 & @xmath168 & @xmath169 + @xmath143 ( 0.5 - 10 ) @xmath144 & @xmath170 & @xmath171 & @xmath172 & @xmath156 & @xmath173 & @xmath174 & @xmath175 & @xmath161 + fe k@xmath9 ( fwhm@xmath1850 kev ) ( ev ) & @xmath186 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath190 & @xmath191 + fe k@xmath9 ( fwhm@xmath1852 kev ) ( ev ) & @xmath191 & @xmath188 & @xmath188 & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath189 + lllllllllllll obs . & & theor . & meas . & shift & & flux & w & n@xmath196 & n@xmath197 + & & ( ) & ( ) & ( km / s ) & ( @xmath198 ) & ( km / s ) & ( @xmath198 ph/@xmath199/s ) & ( m ) & ( @xmath200 ) & ( @xmath201 ) + 1 & & 1.850 & 1.848(1 ) & @xmath202 & @xmath203 & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 + 1 & & 1.780 & 1.776(1 ) & @xmath209 & @xmath210 & @xmath1 & @xmath211 & @xmath212 & @xmath213 & @xmath214 + 3 & & 1.850 & 1.851(2 ) & @xmath221 & @xmath222 & @xmath223 & @xmath224 & @xmath225 & @xmath226 & @xmath227 + 3 & & 1.780 & 1.778(1 ) & @xmath228 & @xmath229 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & @xmath234 + 4 & & 1.850 & 1.850(1 ) & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 + 4 & & 1.780 & 1.781(1 ) & @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & @xmath119 + lllllllllllll obs . & & theor . & meas . & shift & & flux & w & n@xmath196 & n@xmath197 + & & ( ) & ( ) & ( km / s ) & ( @xmath198 ) & ( km / s ) & ( @xmath198 ph/@xmath199/s ) & ( m ) & ( @xmath200 ) & ( @xmath201 ) + 1 ( @xmath248mean ) & & 1.850 & & & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath220 & @xmath217 + 1 ( @xmath253mean ) & & 1.850 & 1.848(1 ) & @xmath202 & @xmath249 & @xmath250 & @xmath254 & @xmath148 & @xmath255 & @xmath256 + 1 ( @xmath248mean ) & & 1.780 & 1.776(1 ) & @xmath209 & @xmath257 & @xmath258 & @xmath134 & @xmath259 & @xmath260 & @xmath151 + 1 ( @xmath253mean ) & & 1.780 & 1.776(1 ) & @xmath209 & @xmath261 & @xmath262 & @xmath128 & @xmath156 & @xmath172 & @xmath263 + 1 ( mean@xmath75 c / s ) & & 1.850 & & & @xmath249 & @xmath250 & @xmath220 & @xmath264 & @xmath265 & @xmath218 + 1 ( mean@xmath05 c / s ) & & 1.850 & 1.849(1 ) & @xmath266 & @xmath249 & @xmath250 & @xmath234 & @xmath133 & @xmath267 & @xmath268 + 1 ( mean@xmath75 c / s ) & & 1.780 & 1.775(1 ) & @xmath269 & @xmath257 & @xmath258 & @xmath128 & @xmath270 & @xmath271 & @xmath145 + 1 ( mean@xmath05 c / s ) & & 1.780 & 1.777(1 ) & @xmath272 & @xmath273 & @xmath274 & @xmath206 & @xmath156 & @xmath172 & @xmath263 + 3 ( norm . ) & & 1.850 & & & @xmath249 & @xmath250 & @xmath218 & @xmath275 & @xmath251 & @xmath217 + 3 ( dip ) & & 1.850 & 1.850(1 ) & @xmath235 & @xmath249 & @xmath250 & @xmath276 & @xmath277 & @xmath206 & @xmath278 + 3 ( norm . ) & & 1.780 & 1.777(1 ) & @xmath272 & @xmath249 & @xmath250 & @xmath279 & @xmath280 & @xmath281 & @xmath226 + 3 ( dip ) & & 1.780 & 1.778(1 ) & @xmath228 & @xmath249 & @xmath250 & @xmath254 & @xmath282 & @xmath283 & @xmath145 + c r r r r r r r r r observation & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 + 1 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & @xmath293 + 2 & @xmath294 & & @xmath295 & @xmath296 & + 3 & @xmath297 & @xmath298 & @xmath299 & @xmath300 & @xmath301 + 4 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & @xmath306 +

we present the results of extensive multi - frequency monitoring of the radio galaxy 3c 120 between 2002 and 2007 at x - ray ( 2 - 10 kev ) , optical ( r and v band ) , and radio ( 14.5 and 37 ghz ) wave bands , as well as imaging with the very long baseline array ( vlba ) at 43 ghz . over the 5 yr of observation , significant dips in the x - ray light curve are followed by ejections of bright superluminal knots in the vlba images . consistent with this , the x - ray flux and 37 ghz flux are anti - correlated with x - ray leading the radio variations . furthermore , the total radiative output of a radio flare is related to the equivalent width of the corresponding x - ray dip . this implies that , in this radio galaxy , the radiative state of accretion disk plus corona system , where the x - rays are produced , has a direct effect on the events in the jet , where the radio emission originates . the x - ray power spectral density of 3c 120 shows a break , with steeper slope at shorter timescale and the break timescale is commensurate with the mass of the central black hole based on observations of seyfert galaxies and black hole x - ray binaries . these findings provide support for the paradigm that black hole x - ray binaries and both radio - loud and radio - quiet active galactic nuclei are fundamentally similar systems , with characteristic time and size scales linearly proportional to the mass of the central black hole . the x - ray and optical variations are strongly correlated in 3c 120 , which implies that the optical emission in this object arises from the same general region as the x - rays , i.e. , in the accretion disk - corona system . we numerically model multi - wavelength light curves of 3c 120 from such a system with the optical - uv emission produced in the disk and the x - rays generated by scattering of thermal photons by hot electrons in the corona . from the comparison of the temporal properties of the model light curves to that of the observed variability , we constrain the physical size of the corona and the distances of the emitting regions from the central bh . in addition , we discus physical scenarios for the disk - jet connection that are consistent with our observations .

0909.2051

stellar - mass black hole x - ray binaries ( bhxrbs ) and active galactic nuclei ( agns ) are both powered by accretion onto a black hole ( bh ) . in many cases these systems emit radiation over several decades of frequency and possess relativistic jets @xcite . the above similarities in the basic generation of energy and observational properties have led to the paradigm that these two systems are fundamentally similar , with characteristic time and size scales linearly proportional to the mass of the central bh ( @xmath010 m@xmath1 for bhxrbs and @xmath2 to @xmath3 m@xmath1 for agns ) . although this paradigm has given rise to the expectation that we might test models of agns with observations of bhxrbs , such an approach is unjustified until detailed , possibly quantitative connections between bhxrb systems and agns become well - established . the comparison of bhxrbs and agns is complicated by the fact that a single agn usually does not show the entire range of properties that we wish to compare . for example , seyfert galaxies are the agns that most resemble bhxrbs , but their radio jets tend to be weak and non - relativistic ( e.g. * ? ? ? * ) . on the other hand , in radio - loud agns with strong , highly variable nonthermal radiation ( blazars ) , the doppler beamed emission from the jet at most wavelengths masks the thermal emission from the accretion disk and its nearby regions . one well - established property of bhxrbs is the connection between accretion state and events in the jet . in these objects , transitions to high - soft x - ray states are associated with the emergence of very bright features that proceed to propagate down the radio jet @xcite . similarly , an accretion disk - jet connection was established in agns by @xcite , who reported a relationship between x - ray and radio events in the radio galaxy 3c 120 ( redshift of 0.033 ) . during @xmath0 3 yr of monitoring of this object , four dips in the x - ray flux , accompanied by spectral hardening , preceded the appearance of bright superluminal knots in the radio jet . this fanaroff - riley class i radio galaxy has a prominent radio jet that displays strong variability in flux and jet structure @xcite . the jet lies at an angle @xmath4 to our line of sight , significantly wider than is typically the case for blazars @xcite . at optical and x - ray frequencies , 3c 120 possesses properties similar to seyfert galaxies and bhxrbs , e.g. , a prominent iron emission line at a rest energy of 6.4 kev @xcite . hence , most of the x - rays are produced in the immediate environs of the accretion disk : the corona , a hot wind , or the base of the jet . since the superluminal knots are disturbances propagating down the jet , a link between decreases in x - ray production and the emergence of new superluminal components demonstrates the existence of a strong disk - jet connection . in this paper we present the results of extensive multi - frequency monitoring of 3c 120 between 2002 and 2007 at x - ray energies ( 2.4 - 10 kev ) , optical @xmath5 and @xmath6 bands , and radio frequencies 37 and 14.5 ghz , as well as imaging with the very long baseline array ( vlba ) at 43 ghz . we use these data to obtain quantitative support for the disk - jet connection claimed by @xcite . times of ejections " are defined as the extrapolated time of coincidence of a moving knot with the position of the ( presumed quasi - stationary ) core in the vlba images . @xcite have found that ejection times are generally associated with an increase in flux at 37 ghz . therefore , the 37 ghz flux should be anti - correlated with the x - ray flux if decreases in x - ray production are linked to ejections . here , we confirm the existence of this anti - correlation and test for a relation between the amplitude of the x - ray dips and that of the associated 37 ghz flares . another similarity between bhxrbs and seyfert galaxies lies in their x - ray power spectral densities ( psds ) . the psd corresponds to the power in the variability of flux on different timescales . the x - ray psds of seyfert galaxies can be fit by piece - wise power laws with a break @xcite . this is generally similar to that of bhxrbs although in many cases the latter have more than one break and sharp peaks due to quasi - periodic oscillations . we use our data set to verify that the x - ray psd of 3c 120 also has a break , as inferred by @xcite , who calculated the psd from somewhat less extensive observations that partially overlap with ours . the bulk of the optical continuum from non - blazar agns is thought to emanate from the accretion disk @xcite . in order to test this , we cross - correlate the optical flux variations of 3c 120 with those at x - ray energies . however , the coefficients of raw correlations may be artificially low due to the uneven sampling present in the data . in light of this , we use simulated light curves , based on the underlying psd , to estimate the significance of the derived coefficients of both the x - ray / optical correlation and x - ray/37 ghz anti - correlation found in the 3c 120 data . in 2 we present the observations and data reduction procedures , while in 3 we describe the power spectral analysis and its results . in 4 we cross - correlate the x - ray and 37 ghz light curves and quantitatively investigate the relation between the x - ray dips , radio flares and ejections , as well as discuss the x - ray / optical correlation . in 5 we discuss and interpret the results , focusing on the implications regarding the accretion disk - jet connection and the source of the optical emission in 3c 120 . 6 presents conclusions , including comparison of 3c 120 with bhxrbs table [ data ] summarizes the intervals of monitoring at different frequencies for each of the three wave bands in our data set . we term the entire light curve `` monitor data '' ; shorter segments of more intense monitoring are described below . the x - ray light curves are based on observations of 3c 120 with the _ rossi _ x - ray timing explorer ( rxte ) from 2002 march to 2007 may . we observed 3c 120 with the rxte pca instrument with typical exposure times of 1 - 2 ks . for each exposure , we used routines from the x - ray data analysis software ftools and the program xspec to calculate and subtract an x - ray background model from the data and to fit the x - ray spectrum from 2.4 to 10 kev as a power law with low - energy photoelectric absorption by intervening gas in our galaxy . for the latter , we used a hydrogen column density of @xmath7 atoms @xmath8 @xcite . the sampling of the x - ray flux varied . normally , observations were made 2 - 3 times per week except during 8-week intervals each year when the radio galaxy is too close to the sun s celestial position to observe safely . in order to sample shorter - term variations , between 2006 november and 2007 january we obtained , on average , four measurements per day for almost two months . we refer to these observations as the `` medium '' data . xmm - newton observed 3c 120 quasi - continuously for about 130 ks on 2002 december 13 and 14 , during which all instruments were operating normally . the data were processed with the latest software ( sas version 5.3.3 ) . light curves were extracted in two energy bands , 0.3 - 10 kev and 4 - 10 kev , and were background - subtracted and binned to 100 s time intervals . the 4 - 10 kev light curve , re - binned to an interval of 0.01 day , were used as the longlook " data . we use the 4 - 10 kev data since the energy range is similar to that of our rxte data ( 2.4 - 10 kev ) . figure [ xdata ] presents these three data sets . we also monitored 3c 120 in the optical @xmath5 and @xmath6 bands over a portion of the time span of the x - ray observations . the majority of the measurements in r band are from the 2 m liverpool telescope ( lt ) at la palma , canary islands , spain , supplemented by observations at the 1.8 m perkins telescope of lowell observatory , flagstaff , arizona . the v - band photometry was obtained with the 0.4-m telescope of the university of nebraska . on each night a large number of one - minute images ( @xmath9 ) were taken and measured separately . details of the observing and reduction procedure are as described in @xcite . comparison star magnitudes were calibrated using @xcite . to minimize the effects of variations in the image quality , fluxes were measured through an aperture of 8 arcseconds radius . the errors given for each night are the uncertainties in the means . the miller observatory observations were taken with a 0.4-m telescope in nebraska and reduced in similar manner as were the university of nebraska observations . observations at the shanghai astronomical observatory were obtained with the 1.56-m telescope at sheshan station with standard johnson - cousins v , r , and i filters , and all the magnitudes were scaled to the v passband . the reductions were as for the nebraska observations . early observations ( 2002 to 2004 ) carried out at the perkins telescope were in v - band . for these measurements , we used stars d , e , and g from @xcite to calculate the v - magnitudes . we use three comparison stars in the field of 3c 120 to calculate the r - magnitude . we determined the r - magnitudes of these three stars based on @xmath020 frames obtained within 2 yr . we use the flux - magnitude calibration of @xcite and correct for galactic extinction for both r and v bands . since the optical sampling in r or v band individually is not as frequent as the x - ray sampling , we construct a better - sampled optical light curve by combining these two bands . we find that the r and v band light curves have 38 data points that were measured within 0.5 day of each other . these data are shown in figure [ vr_correction ] . the equation of the best - fit line is @xmath10 , where @xmath11 and @xmath12 are the fluxes in v and r band , respectively . we convert the v band fluxes into r band using this equation . we also add some data from lowell observatory while constructing the combined light curve , which we present in figure [ opvr ] . we have compiled a 37 ghz light curve with data from the 13.7 m telescope at metshovi radio observatory , finland . the flux density scale is set by observations of dr 21 . sources 3c 84 and 3c 274 are used as secondary calibrators . a detailed description on the data reduction and analysis is given in @xcite . we also monitored 3c 120 at 14.5 ghz with the 26 m antenna of the university of michigan radio astronomy observatory . details of the calibration and analysis techniques are described in @xcite . at both frequencies the flux scale was set by observations of cassiopeia a @xcite . starting in 2001 may , we observed 3c 120 with the very long baseline array ( vlba ) at 43 ghz at roughly monthly intervals , with some gaps of 2 - 4 months . the sequence of images from these data provide a dynamic view of the jet at an angular resolution @xmath00.1 milliarcseconds ( mas ) in the direction of the jet , corresponding to 0.064 pc for hubble constant @xmath13 = 70 km s@xmath14 mpc@xmath14 . we processed the data in the same manner as described in @xcite . we modeled the brightness distribution at each epoch with multiple circular gaussian components using the task modelfit of the software package difmap @xcite . this represents the jet emission downstream of the core by a sequence of knots ( also referred to as `` components '' ) , each characterized by its flux density , fwhm diameter , and position relative to the core . the apparent speeds of the moving components with well - determined motions are all within , @xmath15 . the ejection time @xmath16 is the extrapolated time of coincidence of a moving knot with the position of the ( presumed stationary ) core in the vlba images . table [ ejecflare ] lists the ejection times and figure [ vlba1 ] and [ vlba2 ] display the vlba images . [ xopradspec ] presents the x - ray , optical and radio light curves . in the top panel of the figure , the arrows represent the times of superluminal ejections , while the line segments perpendicular to the arrows show the uncertainties in the values of @xmath16 . figure [ xop_commondips ] displays the x - ray and optical light curves in the interval between 2006 october and 2007 april . it shows the best - sampled pair of dips in the x - ray and optical fluxes . we use a variant of the power spectrum response method ( psresp * ? ? ? * ) to determine the intrinsic psd of the x - ray light curve . our realization of psresp is described in @xcite . psresp gives both the best - fit psd model and a `` success fraction '' @xmath17 ( fraction of simulated light curves that successfully represent the observed light curve ) that indicates the goodness of fit of the model . at first we fit a simple power - law model to the x - ray psd , but found that the value of @xmath17 was unacceptably low . this implies that a simple power - law is not the best model for this psd . then we fit a bending power - law model ( broken power law with a smooth break ) to the x - ray psd , @xmath18^{-1 } p(\nu)=a\nu^{\alpha_l}[1+(\frac{\nu}{\nu_{b}})^{(\alpha_{l}-\alpha_{h})}]^{-1}.\ ] ] here , @xmath19 is a normalization constant , @xmath20 is the break frequency , and @xmath21 and @xmath22 are the slopes of the power - laws above and below the break frequency , respectively @xcite . during the fitting , we varied @xmath23 from @xmath24 to @xmath25 hz in steps of @xmath26 , @xmath27 from @xmath28 to @xmath29 in steps of 0.1 , and @xmath30 from @xmath31 to @xmath28 in steps of 0.1 . these ranges include the values of @xmath32 found in the light curves of bhxrbs in the high - soft state , for which @xmath33 and @xmath27 between @xmath34 and @xmath35 ( e.g. * ? ? ? this procedure yields a much higher success fraction than the simple power - law model . based on the model with the highest success fraction , we obtain a best fit with the parameters @xmath36 , @xmath37 , and log@xmath38 hz . the success fraction for this fit is high , @xmath39 . the intrinsic scatter of power is very high for slopes below @xmath34 due to red noise leak . as a result , trial psd models with a high frequency slope of any value between @xmath34 and @xmath35 give very similar success fractions . this causes the larger uncertainty in the value of the high frequency slope relative to that at lower frequencies . figure [ psd ] presents this best - fit model and the corresponding psd . as seen in the figure , the high frequency part of the psd is dominated by poisson noise . that is because this part of the psd is generated from the longlook light curve , and fluxes in the longlook light curve have larger uncertainties owing to shorter exposure times than those in the other light curves . the figure shows that when the estimated poisson noise is added to the best - fit model psd , it matches the observed psd quite well . the nature of the x - ray psds and the values of break frequencies found previously in seyfert galaxies are consistent with those of bhxrbs in high - soft state @xcite . this is consistent with accretion rates in all these objects which are above 10% , more than the typical value of 2% indicating the transition from low to high state in galactic bhxrbs @xcite . in 3c 120 , the accretion rate @xmath030% which implies that it is also a high state system . the psd break frequency in bhxrbs and seyfert galaxies scales inversely with the mass of the black hole while being proportional to the accretion rate @xcite . using the best - fit values and uncertainties in the relation between break timescale , bh mass , and accretion rate obtained by @xcite , we estimate the expected value of the break frequency in the x - ray psd of 3c 120 to be @xmath40 hz . hence , our derived break frequency lies within the expected range . here we adopt a bh mass of @xmath41 m@xmath1 from emission - line reverberation mapping @xcite and a bolometric luminosity of the big blue bump of @xmath42 ergs s@xmath14 @xcite . we note that the high and low frequency slopes , as well as the break frequency calculated by @xcite ( @xmath43 , @xmath44 , @xmath45 hz ) , are consistent with the best - fit values in this paper within the uncertainties . the difference in the central values probably results from our use of a different longlook light curve , which is from the epic instrument onboard xmm - newton . the sensitivity of this detector is higher than the pca of rxte , allowing our light curve to have a much better sampling rate . this is useful for extending the lower end of the frequency range of the psd , leading to a more accurate measurement of the high frequency slope and break frequency . we employ the discrete cross - correlation function ( dccf * ? ? ? * ) method to search for connections between variations at pairs of wave bands . we determine the significance of the correlations by performing the analysis on @xmath46 simulated light curves generated randomly with a psd matching that of the real data . details of this procedure are described in @xcite . in the present case , we correlate simulated x - ray light curves with the observed optical / radio light curves to calculate the significance . the 37 ghz light curve has an average sampling frequency of about once per week . we bin the x - ray and the 37 ghz light curves in 7-day intervals before performing the cross - correlation so that the light curves being compared have similar sampling frequency . as determined by the dccf ( top panel of figure [ xradcor ] ) , the x - ray flux variations are anti - correlated with those at 37 ghz in 3c 120 . the highest amplitude of the x - ray versus 37 ghz dccf is at a value of @xmath47 , which corresponds to a 90% significance level . the time lag of the peak indicates that the x - ray lead the radio variations by @xmath48 days . we have used the fr - rss technique proposed by @xcite to calculate the mean value and uncertainty of the cross - correlation time lag . this procedure gives quantitative support to the trend that is apparent by inspection of the light curve , i.e. , x - ray dips are followed by appearance of new superluminal knots and hence enhancement in the 37 ghz flux . the bottom panel of figure [ xradcor ] shows the x - ray/37 ghz dccf but without the data after 2006 april , in order to exclude the two deepest x - ray dips and highest amplitude flare at 37 ghz toward the end of our monitoring program . the x - ray/37 ghz anti - correlation remains , although the minimum of the dccf is smaller , @xmath49 , with a significance level of 72% . the x - ray variations lead the radio by @xmath50 days . thus , the x - ray/37 ghz anti - correlation is not just the consequence of a singular event . the longer time delay of the major event in late 2006 was caused by the longer time between the start of the outburst and the peak relative to other radio flares , especially the one at 2003.7 . this is naturally explained as a consequence of higher optical depth of stronger radio outbursts . in order to determine the physical link between the accretion disk and jet , we check whether the amplitudes of the x - ray dips and associated radio flares are related . to test this , we calculate the equivalent width of each x - ray dip and the area under the curve of each 37 ghz flare to measure the total energy involved in the events . we approximate that the radio light curve is a superposition of a constant baseline of 1.5 jy from the more extended jet and long - term flares . the baseline is chosen as approximately the lowest level of the 37 ghz light curve . we then follow @xcite by decomposing the baseline - subtracted light curve into individual flares , each with exponential rise and decay . we use four parameters to describe each flare : the rise and decay timescales , and the height and epoch of the peak . before the decomposition , we smooth the light curve using a gaussian function with a 10-day fwhm smoothing time . the details of the decomposition procedure are described in @xcite . the x - ray light curve has a long - term trend , i.e. , the baseline is not constant . we define the baseline x - ray flux as a cubic - spline fit of the annual mean plus one standard deviation . although this is an arbitrary definition , this baseline reproduces reasonably well the mean flux level in between obvious dips and flares . there is a long - term trend in the spectral index variations as well . the x - ray spectral index @xmath51 , defined by @xmath52 , where @xmath53 is the x - ray flux density and @xmath54 is the frequency , varied between @xmath55 and @xmath56 , with an average value of @xmath57 and a standard deviation of 0.10 over the 5.2 yr of observation . in addition to the short term fluctuations , there is a long - term trend of increasing values of @xmath51 during this interval . we calculate a baseline in the same manner as for the x - ray flux variations to highlight the change in spectral index during the dips ( see figure [ baseline1sig ] ) . before we perform the above analysis to calculate the baseline , we smooth the x - ray light curve using a gaussian function with a 10-day fwhm smoothing time . we proceed by subtracting the baseline and then modeling the dips as inverted exponential flares . figure [ xradmodelfit_1sig ] shows the model fits to both the x - ray and 37 ghz light curves . table [ ejecflare ] lists the parameters of the x - ray dips and the corresponding 37 ghz flares along with the times of superluminal ejections . since the correlation functions with and without the data in 2007 ( during the large amplitude 37 ghz flare ) show that the x - ray dips lead the 37 ghz flares by @xmath48 and @xmath50 days , respectively , and we have identified 15 significant dips during the 5.2 yr of monitoring , we assume that a radio flare that peaks between 20 and 180 days after an x - ray minimum is physically related to it . using the results of this analysis , we plot the equivalent width of x - ray dips versus the energy output of the corresponding 37 ghz flares in figure [ dipejecarea ] . the 37 ghz flare at 2003.72 has a wide decaying wing and there are two x - ray dips ( at 2003.58 and 2003.82 ) that may be related to this flare as well as another smaller 37 ghz flare ( at 2003.92 ) on top of that wing . similarly , there are two x - ray dips at 2005.12 and 2005.39 , and there are four corresponding 37 ghz flares very close to each other in time at 2005.23 , 2005.36 , 2005.57 , 2005.80 . in the above cases , we plotted the total energy output of the group of 37 ghz flares against the total equivalent width of the corresponding group of x - ray dips . the energy output of the flares and the dips were corrected by adding the residuals shown in figure [ xradmodelfit_1sig ] . the plot in figure [ dipejecarea ] reveals a positive correlation between the flare and dip strengths . the level of change in the accretion disk and/or corona is therefore closely related to the amount of excess energy injected into the jet . it can be seen from table [ ejecflare ] that 14 out of 15 significant x - ray dips are followed by a superluminal ejection . the time delay between the start of the x - ray dips and the time of ejection " of the corresponding superluminal knot varies between @xmath58 yr to @xmath59 yr , with a mean value of @xmath60 yr . we plot the times of ejection of new knots along with the corresponding times of start of the x - ray dips in figure [ dip_ejec_times ] . the uncertainties in the start time are proportional to the widths of the dips and those in the ejection times are as given in table [ ejecflare ] . a straight line fits the data extremely well with small scatter , which indicates that there is a clear association between x - ray dips and superluminal ejections . the best - fit line through the points has a slope of 1 and y - intercept of @xmath61 , which is consistent with the mean delay of @xmath60 yr given above . this strongly supports the proposition @xcite that in 3c 120 , a decrease in the x - ray production is linked with increased speed in the jet flow , causing a shock front to subsequently move downstream . generally , there is a close correspondence of superluminal ejections with 37 ghz flares , but sometimes the decrease in flux farther out in the jet offsets much of the flux increase from the appearance of a new knot . in such cases , the corresponding increase of flux was not large enough to be detected in our decomposition of the smoothed light curve . this causes the minimum in the x - ray dccf to be less significant than it would if old knots were to completely fade before new ones appear . despite this complication , the x - ray/37 ghz cross - correlation ( 4.1 ) verifies with an objective statistical method that radio events in the jet are indeed associated with x - ray dips . the apparent speeds of the moving components with well - determined motions are all similar , @xmath15 . therefore , a knot moves a distance of @xmath62 pc in @xmath63 yr , projected on the plane of the sky . since the angle of the jet axis of 3c 120 to the line of sight @xmath020@xmath64 , the actual distance traveled by the knot @xmath65 pc hence , we derive a distance @xmath65 pc from the corona ( where the x - rays are produced ) to the vlba 43 ghz core region . this confirms that the core is offset from the position of the bh . this is one of the few cases where we are able to probe the upstream region of the core using the time variable emission at a combination of wave bands . we bin the x - ray and the combined optical light curves in 2-day intervals before performing the cross - correlation , so that the light curves being compared have similar sampling frequency . as determined by the dccf ( figure [ xop ] ) , we find that the x - ray variations are very strongly correlated with those at optical wavelengths in 3c 120 . the peak x - ray versus optical dccf is @xmath66 , which corresponds to a 99% significance level . the position of the peak of the correlation function indicates the relative time delay between the variations at the two wavelengths . in this case , the peak is very wide , so that the value of the relative time delay can not be easily estimated from the dccf plot . we used the fr - rss technique proposed by @xcite to calculate the mean value and the uncertainty of the cross - correlation time lag . this method indicates that the x - ray variations lead the optical by @xmath67 days . the highly significant correlation and short time delay between the x - ray and optical variations indicates that emission at these wave bands is at least partially co - spatial . the asymmetry in the correlation function at time delays above @xmath0100 days indicate that the x - ray variations lead those in the optical at these longer timescales . to characterize the variation of the x - ray / optical time lag over the years , we divide both light curves into two intervals , 2004 july to 2005 may and 2005 june to 2007 may , and repeat the dccf analysis on each segment . the result ( figure [ xop_tw ] ) indicates that in the first segment , the x - ray variations lead those in the optical by @xmath025 days while in the second segment the correlation function is similar to what we obtained for the entire time interval , with similar non - zero time delay ( figure [ xop ] ) . this variation of the time lag over the years may be the cause of the observed wide peak in the correlation function . between 2006 november and 2007 january , the x - ray light curve ( medium " light curve described in 2 ) was sampled 4 times per day and the combined optical light curve has a sampling rate of twice per day , on average . we bin these light curves in 0.5 day intervals . we cross - correlate the binned light curves in order to compare the correlation function with that of the long - term light curves . the correlation function , shown in figure [ xop_small ] , has a similarly significant correlation coefficient and time delay . the similar values of correlation coefficient and time delay using very well - sampled light curves illustrates the robustness of the correlation result . the x - ray to optical time delay calculated by @xcite has the same sign as determined in this paper , although the magnitude is significantly different ( @xmath68 days ) . we note that ( 1 ) this work includes more data points in the r band light curve , including the period @xmath69 , and 2 ) we construct a combined light curve using the r band data as well as scaled v band data . we use this more completely sampled light curve for the correlation analysis . this , in addition to the longer time lag from 2004.5 to 2005.5 , can account for the difference between our result and that obtained by @xcite . the interpolated cross - correlation function ( iccf ) is another method to calculate the cross - correlation of unevenly sampled discrete data @xcite . we have repeated all the above correlation analysis using the iccf and have found that the dccf and iccf results are consistent with each other for the data presented in this work . we have cross - correlated the hard ( 4 - 10 kev ) and soft ( 0.2 - 4 kev ) longlook x - ray light curves , finding an excellent correlation with nearly zero time lag ( figure [ softhard ] ) . this indicates that in 3c 120 , variations in soft and hard x - rays follow each other very closely . the physical cause of the connection between events in the central engine and the jet of bhxrbs and agns is currently a matter of considerable speculation . if the jet is magnetically launched from the accretion disk @xcite , then there must be a link between the magnetic state at the base of the jet and the accretion state in the inner disk . one scenario , proposed for bhxrbs by @xcite and @xcite , involves a change in the magnetic field configuration in the inner disk from a turbulent condition in the high - soft state ( when the x - ray flux is relatively higher and softer ) to mainly poloidal in the low - hard x - ray state . the turbulence is needed for viscous heating , which in the bhxrb case leads directly to bright x - ray emission with a soft spectrum . in an agn it causes strong ultraviolet emission , which is compton scattered in the corona to a hard x - ray spectrum . if the field switches from chaotic to mainly poloidal , which @xcite suggest can occur by random episodes of near - alignment of the field in the relatively small number of turbulent cells in the inner disk , then the radiation in the inner disk will be quenched at the same time as energy flow into the jet is promoted . the transition back to the turbulent , radiative inner disk of the high - soft state would need to involve a surge of energy injected into the jet in order to send a shock wave @xcite down the jet . perhaps global magnetic reconnection could cause this , but no detailed mhd model has been published to date . alternatively , it is possible that the corona , " where the x - ray emission seen in agn supposedly arises from compton up - scattering of softer accretion - disk photons , might be the base of the jet @xcite . if this is the case , then the x - ray flux will be related to the number of electrons residing there and available for scattering to create x - rays . the mass loading of the jet should also affect the asymptotic lorentz factor of the flow downstream if the jet is magnetically driven ( e.g. * ? ? ? * ) . the same decrease in electron number that causes a drop in scattered x - ray emission near the disk would lead to a time - delayed increase in the speed of the jet downstream . the flatter - spectrum nonthermal x - ray emission from the downstream jet would then play a larger relative role in the x - ray emission , causing the observed hardening of the spectrum during the dips . the increase in flow speed of the jet could form a shock wave , seen as a superluminal radio knot . it is difficult to speculate why the mass loading should change , since we do not understand the processes by which material from the disk and/or ergosphere are injected into the base of the jet . however , observations of the microquasar grs 1915 + 105 suggest that outflow of matter from the disk switches from mainly a wind to the jets @xcite . in an agn with a magnetically driven jet , it may be the case that lower mass injection into the jet actually enhances the jet emission owing to an increase in the flow speed driving a shock wave down the jet , as described above . the strong correlation between the x - ray and optical variations in 3c 120 implies that the emission at both wave bands arises from the same general region . since the x - rays originate in the corona , the optical emission is probably thermal emission from the accretion disk @xcite . we can reject the alternative hypothesis that the main component of optical emission is synchrotron radiation from the jet . in that case , the emission should be significantly polarized , contrary to observations showing the optical linear polarization to be @xmath70 @xcite . furthermore , the anti - correlation of the x - ray and 37 ghz emission ( the latter of which is produced in the jet ) implies that any optical synchrotron emission from the jet should also anti - correlate with the x - ray flux , contrary to the strong observed correlation . although the x - rays are predominantly produced by inverse compton ( ic ) scattering of the thermal optical / uv seed photons from the accretion disk by hot , but non - relativistic , electrons in the corona , the optical / uv emission and the x - rays are tied together by another mechanism : some of the optical / uv radiation is produced by heating of the accretion disk by x - rays produced in the above process ( feedback " mechanism ) . the amplitude of the short timescale ( @xmath0days ) variations in the x - ray light curve is larger than that in the optical . this can happen if a significant fraction of the optical emission is due to heating of the disk by x - rays , since this reprocessing may smooth out the short timescale variations . feedback can make the time delay more difficult to define by producing a fraction of the optical / uv photons with different temporal properties from that of the direct emission . there is a weak correlation between the x - ray spectral index and the x - ray flux , indicating that the spectrum becomes harder during a decrease in the production of x - rays ( fig [ baseline1sig ] ) . this kind of pivoting " has been seen before @xcite . the x - rays may be produced mainly by up - scattering of uv photons and not optical photons , consistent with the correlation found by @xcite between the x - ray and uv variations in 3c 120 . this could occur if the flux of optical photons reaching the corona is much smaller than that of uv photons . such a scenario is likely if the corona is small such that the region where the uv photons are generated is much closer to the corona than the region where the majority of the optical photons are produced . any disturbance propagating outwards in the accretion disk will cause a change in the uv flux ( and a resultant nearly immediate change in the x - ray emission ) followed by a similar change in the optical flux . the sign of the time delay will switch if the disturbance propagates inwards . the observed mean time delay between x - ray and optical variations may be due to such propagation time delays . if we adopt typical parameters such as @xmath0 0.1 for accretion efficiency , l / l@xmath71 @xmath0 0.3 @xcite , and l@xmath72 2@xmath73 @xmath74 ergs / sec @xcite and neglect general relativistic effects , the region in the accretion disk where the emission peaks in the extreme uv / soft x - ray range ( @xmath75 nm ) should be very close to the innermost stable orbit ( @xmath0 5 @xmath76 , where @xmath76 is the gravitational radius of the black hole ) . the region where the emission peaks in the optical ( @xmath77 nm ) lies at @xmath0 1000@xmath76 . for 3c 120 , with a bh mass of @xmath0 5@xmath78 m@xmath1 , 1000 @xmath76 is equivalent to @xmath04 light days . we consider a model that includes both time delays and coupling of emission from the corona and disk . we have performed a theoretical calculation to produce multi - wavelength light curves from a disk - corona system . we assume that ( 1 ) the temperature of the disk changes with its radial distance from the center according to t@xmath79 @xcite , ( 2 ) each annulus radiates as a perfect black - body , ( 3 ) the x - rays are produced in a spherical region ( corona ) close to the center of the disk by inverse - compton scattering of the disk photons that reach the corona , and ( 4 ) lower energy radiation ( uv - optical ) is produced both by blackbody radiation in the disk and reheating of the disk by the x - rays from the corona . for simplicity , we assume that the energy distribution of electrons is uniform throughout the volume of the corona . the fraction of the uv - optical photons that reach the corona are up - scattered to x - rays and a part of the x - rays re - heat the disk to produce additional thermal photons . we treat the x - ray production in the corona as a reflection that increases the energy of the radiation . the computer code that we use for the model also follows the travel time of photons from different radii of the disk to the corona and _ vice versa_. the temporal nature of the emission from the corona and the part of the disk close to it are very similar owing to the small light travel time . figure [ theory0 ] shows the variation of the total uv and optical intensity emitted by an annulus at radius @xmath80 of thickness @xmath81 , i.e. , @xmath82 versus @xmath80 . this shows that the annulus that produces the largest amount of uv radiation ( region uv " ) lies @xmath05 @xmath76 from the center , and for optical emission ( region op " ) it is at @xmath075 @xmath76 . this distance is much smaller than what would be expected if most of the optical radiation were to come from radii near where the planck curve @xmath83 peaks at r band . these latter radii lie @xmath04 light - days ( 1000 @xmath76 ) from the center , while regions close to the center are at higher temperature , and therefore emit at a higher blackbody intensity at all wavelengths . although @xmath83 increases monotonically with decreasing radii , @xmath82 peaks at a radius @xmath075 @xmath76 at r band . we use a computer code to model this disk - corona system and then introduce a disturbance in the temperature of the accretion disk that propagates at a speed @xmath00.1@xmath84 from the center toward the outside or _ vice versa_. due to the disturbance , the temperature at a given annulus varies with time according to a gaussian profile . this causes a flare in the emission of the entire system at all wavelengths , although the flare starts and peaks at different times at different wavelengths . we produce these light curves at x - ray , uv , and optical bands , including delays from internal light - travel time . the nature of the flare and relation between flares at multiple wave bands depend on the speed and direction of propagation of the disturbance since emission from different annuli of the disk have different wavelength dependence . the top panel of figure [ theory1 ] shows the x - ray and optical light curves from the above calculation when the disturbance propagates toward smaller radii . as a result , the emission from the outer disk flares before that closer to the bh . therefore , the optical variation leads that in the x - rays . in this simulation , the time delay between the peaks of the x - ray and optical light curves @xmath04 days . in 3c 120 , the time delay is centered on 0.5 day with an uncertainty of @xmath85 days , hence the direction of the time delay can not be specified with certainty . the correlation function in figure [ xop_small ] also shows a relatively broad peak centered on zero , with a similar uncertainty in the time delay . the cross - correlation function for longterm light curves sometimes show a broad peak due to changes in the direction and magnitude of the time delay over the timespan of the light curves . if the broad peak in these correlation functions is due to the above - mentioned reason , it may indicate a dichotomy in the speed of propagation of the disturbances and their directions , i.e. , the uncertainty corresponds to an actual range of time delays . this also constrains the size of the corona . for example , if the corona were spread such that the regions op and uv were at the same distance from the corona , then these regions would contribute equally to the flux of seed photons that are up - scattered to x - rays . in that case , the x - ray flares would be much broader than observed and no optical / x - ray time delay would be present . this is contrary to the observation of relatively sharp x - ray flares . in addition , although overall we find a short mean time delay of @xmath86 day ( i.e. , if the uncertainty does not correspond to an actual range of time delays but rather is consistent with zero ) , in some of the individual flares the time delay is longer ( @xmath025 days ; see top panel of figure [ xop_tw ] ) . also , if optical and uv photons were up - scattered to soft and hard x - rays , respectively , a significant soft - hard " time delay would be expected . but from figure [ softhard ] , it is clear that the soft - hard time lag can not be more than 0.25 day , which is essentially unobservable with the longterm light curves . from the above discussion , we conclude that the corona is situated close to the center of the accretion disk , and its size is such that the flux of photons reaching the corona from region op is negligible with respect to that from region uv . based on the solid angle subtended , the coronal radius should be less than 40@xmath76 for the flux from region op to be less than 10% of that from region uv . in figure [ theory1 ] , the two panels show results for different levels of feedback ( the fraction of the x - rays produced in the corona that reflect toward and re - heat the disk ) . the top panel has feedback fraction equal to zero , i.e. , none of the x - rays re - heat the disk . we can see from the figure that , as feedback is introduced , the resulting time delay becomes smaller and less precise . figure [ theory2 ] shows the light curves when the disturbance is propagating outwards from the bh . in this case , as expected , the x - ray variations lead those in the optical ( by similar amount as in figure [ theory1 ] ) for the same propagation speed . the panels show , as above , that including feedback makes the time delay shorter . hence , feedback from the corona may also contribute to the range of time delays that we see in the data . in another possible scenario , the x - ray variability is caused by intrinsic changes in the hot electrons in the corona and the uv - optical changes are due to feedback , i.e. , there is no intrinsic variability in the accretion disk . in this case , the x - ray / optical time delay will solely be due to light travel time from the corona to the accretion disk . this will produce time lags of a small fraction of a day @xcite , which is too small to observe with the sampling of the data used in this work . the observed essentially zero time lag between the long - term x - ray and optical light curves ( @xmath67 days ) and smaller variability amplitude in the short - term optical light curve than that in the x - rays are consistent with the above scenario . in fact the steepening of the x - ray spectrum and decrease in the mean x - ray flux level after 2006.0 ( see fig . [ baseline1sig ] ) suggest a long - term steepening of the energy distribution of electrons in the corona . but the longer x - ray / optical time delay over a significant portion of the dataset ( figure [ xop_tw ] ) and comparable x - ray / optical b - band variability amplitude in the long - term light curves @xcite with b - band variations sometimes leading , indicate that propagation of disturbances in the accretion disk must produce at least part of the x - ray / optical variability in this radio galaxy . figure [ cartoon ] shows a sketch of the accretion disk - corona system as derived in this paper . the black filled circle is the position of the bh , the temperature of the accretion disk is shown by gray scale with lighter color meaning higher temperature , and the larger circular area filled with dots is the corona . region uv " and region op " are shown as thick solid lines on the accretion disk . the radius of the corona and the distance of the relevant emission regions from the bh are shown in units of @xmath76 . from the above model and the correlation results , we conclude that most , if not all , of the optical emission in 3c 120 is produced in the accretion disk . the x - rays are produced by scattering of ( mostly ) uv radiation in the corona . optical / uv emission due to re - heating of the disk by the x - rays is also a possible ongoing mechanism that may cause the spread of the time delay . the presence of sharp x - ray flares and the x - ray / optical time delay of weeks in some individual flares indicate that the corona must lie sufficiently close to the bh that the flux of seed photons reaching the corona is dominated by uv light . the light curve during the dip at x - ray and optical wave bands between 2006 october and 2007 april , displayed in figure [ xop_commondips ] , is very well - sampled . it is evident that the optical flux starts to decrease after @xmath0mjd @xmath87 , @xmath88 days earlier than the decrease starts in the x - rays . the minimum in optical occurs @xmath040 days earlier than that in the x - rays as well . this indicates that the decrease was caused by a disturbance propagating from the outer radii of the accretion disk toward the bh . if this disturbance is a thermal fluctuation propagating inward , then it should have an effective speed @xmath89 @xmath90 to cause a time delay of @xmath040 days . this is one order of magnitude higher than the sound speed for a gas pressure dominated disk of temperature @xmath91 @xmath92 k. therefore , the above disturbance can not be transmitted by sound waves @xcite unless the relevant regions of the disk are dominated by radiation pressure , which can not be ruled out in 3c 120 given its high accretion rate , nearly the eddington value ( 0.3 @xmath93 ) . on the other hand , this scenario is consistent with the model proposed by @xcite , in which large scale alignment of poloidal magnetic field in the inner accretion disk from random fluctuations causes the decrease in the x - ray flux . such alignment occurs at a timescale @xmath94 , where @xmath95 is the radius to thickness ratio of the disk , @xmath96 is a constant @xmath010 @xcite , and @xmath97 is the disk dynamical timescale at radius @xmath5 . this alignment timescale , for a @xmath98 bh , is few seconds , which translates to @xmath050 days for a @xmath99 bh in 3c 120 . model light curves of @xcite also contain short timescale , small amplitude fluctuations on top of the big flares and dips caused by small - scale alignment of the poloidal magnetic field , similar to that observed in the light curves of 3c 120 presented in this paper . this paper presents well - sampled , 5-yr - long light curves of 3c 120 between 2002 and 2007 at x - ray , optical , and radio wavebands , as well as monthly images obtained with the vlba at 43 ghz . we have calculated the broad - band x - ray power spectral density ( psd ) of 3c 120 . the x - ray and radio light curves were decomposed into individual dips and flares , respectively , and the properties of the contemporaneous dips and flares were compared . that comparison , as well as , the x - ray / radio correlation analysis were used to find a connection between events in the jet and the accretion disk . we have also cross - correlated the x - ray and optical light curves and have determined the significance of the correlations with simulated light curves based on the psd . using a numerical model , we have produced x - ray and optical light curves from the accretion disk - corona system . from the comparison of the temporal properties of the simulated light curves to that of the observed variability , we infer the physical size of the corona and the distances of the emitting regions from the central bh . our main conclusions are as follows : + ( 1 ) the x - ray psd of 3c 120 is best fit by a bending power - law model where there is a smooth change in the slope above a break frequency . the best fit value of break frequency for 3c 120 is @xmath100 hz , which agrees very well with the relation between break timescale , bh mass and accretion rate obtained by @xcite spanning a range of bh mass from 10 @xmath101 to @xmath102 @xmath101 . this indicates that the accretion process in 3c 120 is similar to that of the bhxrbs . + ( 2 ) the x - ray and 37 ghz variations are anti - correlated with the x - ray leading the radio by @xmath0120 days . the anti - correlation remains even if the large x - ray dips and radio flares after 2006 are excluded . + ( 3 ) almost all x - ray dips are followed by the ejection of a new knot in the vlba images . this and the anti - correlation mentioned in ( 2 ) imply that decrease in the x - ray production is linked with increased speed in the jet flow , causing a shock front to form and move downstream . this property of 3c 120 is also similar to the galactic black hole systems where transitions to high - soft x - ray states are associated with the emergence of very bright features that proceed to propagate down the radio jet . + ( 4 ) we derive a distance @xmath00.5 pc from the corona ( where the x - rays are produced ) to the vlba 43 ghz core region using the average time delay between the start of the x - ray dips and the time of ejection " of the corresponding superluminal knots . + ( 5 ) the x - ray and optical variations in 3c 120 are very strongly correlated ( see also @xcite ) . this correlation , absence of significant optical polarization , and anti - correlation of x - ray and 37 ghz variations ( the latter of which are produced in the jet ) , together imply that the optical emission is blackbody radiation from the accretion disk . + ( 6 ) the x - rays are produced by scattering of ( mostly ) uv radiation in the corona . comparison of simulated light curves from a disk - corona system and the observed variation imply that the corona must lie sufficiently close to the bh that the flux of seed photons reaching the corona is dominated by uv light . we thank the anonymous referee for many useful comments which helped to improve the quality of the manuscript . we would also like to thank p. uttley for many valuable discussions . the research at boston university was funded in part by the national science foundation ( nsf ) through grant ast-0406865 and by nasa through several rxte guest investigator program grants , most recently nng05gm33 g , nng05gm64 g , and most recently astrophysical data analysis program grant nnx08aj64 g . the metshovi team acknowledges the support from the academy of finland . the university of michigan radio astronomy observatory was supported by funds from the nsf , nasa , and the university of michigan . the vlba is an instrument of the national radio astronomy observatory , a facility of the national science foundation operated under cooperative agreement by associated universities , inc . the liverpool telescope is operated on the island of la palma by liverpool john moores university in the spanish observatorio del roque de los muchachos of the instituto de astrofisica de canarias , with financial support from the uk science and technology facilities council . mg thanks the us national science foundation for support through grants ast 03 - 07912 and ast 08 - 03883 . hp , jt , and zs thank the laboratory for the optical astronomy of the chinese academy of sciences for partial support and the national natural science foundation of china for support through grants 10633010 and 10625314 . aller , h. d. , aller , m. f. , latimer , g. e. , & hodge , p. e. 1985 , , 59 , 513 angione , r. j. 1971 , , 76 , 412 baars , j. w. m. , genzel , r. , pauliny - toth , i. i. k. , & witzel , a. 1977 , , 61 , 99 blandford , r. d. , & payne , d. g. 1982 , , 199 , 883 chatterjee , r. , et al . 2008 , , 689 , 79 doroshenko , v. t. , sergeev , s. g. , merkulova , n. i. , sergeeva , e. a. , golubinsky , y. v. , & pronik , v. i. 2006 , astronomical society of the pacific conference series , 360 , 69 doroshenko , v. t. , sergeev , s. g. , efimov , y. s. , klimanov , s. a. , nazarov , s. v. 2009 , astronomy letters , 2009 , vol . 35 , no . 6 , 361 ( original russian text published in pisma v astronomicheski zhurnal , 2009 , vol . 6,403 edelson , r. a. , & krolik , j. h. 1988 , , 333 , 646 edelson , r. a. , & nandra k. 1999 , , 514 , 682 elvis , m. , lockman , f. j. , & wilkes , b. j. 1989 , , 97 , 777 fender , r. p. , belloni , t. m. & gallo , e. 2004 , , 355 , 1105 fender , r. p. , homan , j. & belloni , t. m. 2009 , , 396 , 1370 gaskell , c. m. , & peterson , b. m. 1987 , , 65 , 1 gomez , j. l. , marscher , a. p. , alberdi , a. , jorstad , s. g. & agudo , i. 2001 , , 561 , l161 grandi , p. et al . 1997 , , 487 , 636 greiner , j. , cuby , j. g. , & mccaughrean , m. j. , 2001 , nature , 414 , 522 jorstad , s. g. , marscher , a. p. , lister , m. l. , stirling , a. m. , cawthorne , t. v. , gmez , j .- l . , & gear , w. k. 2004 , , 127 , 3115 jorstad , s. g. , et al . 2005 , , 130 , 1418 jorstad , s. g. , et al . 2007 , , 134 , 799 kazanas , d. & nayakshin , s. 2001 , , 550 , 655 king , a. r. , pringle , j. e. , west , r. g. , & livio , m. 2004 , , 348 , 111 klimek , e. s. , gaskell , c. m. , & hedrick , c. h. 2004 , , 609 , 69 krolik , j. h. , horne , k. , kallman , t. r. , malkan , m. a. , edelson , r. a. , & kriss , g. a. 1991 , , 371 , 541 livio , m. , pringle , j. e. , & king , a. r. , 2003 , , 593 184 maccarone , t. j. 2003 , , 409 , 697 malkan , m. a. 1983 , , 268 , 582 maraschi , l. , chiapetti , l. , falomo , r. , garilli , b. , malkan , m. , tagliaferri , g. , tanzi , e. g. , & treves , a. 1991 , , 368 , 138 markoff , s. , nowak , m. a. , & wilms , j. 2005 , , 635 , 1203 markowitz , a. , et al . 2003 , , 593 , 96 marscher et al . 2002 , nature , 417 , 625 , 2002 marshall , k. , ryle , w. t. , miller , h. r. , marscher , a. p. , jorstad , s. g. , chicka , b. , & mchardy , i. m. 2009 , , 696 , 601 mchardy , i. m. , koerding , e. , knigge , c. , uttley , p. , & fender , r. p. 2006 , nature , 444 , 730 mchardy i. m. , papadakis i. e. , uttley p. , page m. j. , & mason k. o. 2004 , , 348 , 783 mchardy , i. m. , gunn , k. f. , uttley , p. & goad , m. r. 2005 , , 359 , 1469 meier , d. l. , koide , s. , & uchida , y. , 2000 , science , 291 , 84 mead , a. r. g. , ballard , k. r. , brand , p. w. j. l. , hough , j. h. , brindle , c. , & bailey , j. a. 1990 , , 83 , 183 miller - jones , j. c. a. , mccormick , d. g. , fender , r. p. , spencer , r. e. , muxlow , t. w. b. , & pooley , g. g. 2005 , , 363 , 867 mirabel , i. f. and rodrguez , l. f. , 1999 , , 37 , 409 neilsen , j. & lee , j. c. , 2009 , nature , 458 , 481 nowak , m. a. , et al . 1999 , , 510 , 874 ogle , p. m. , davis , s. w. , antonucci , r. r. j. , colbert , j. w. , malkan , m. a. , page , m. j. , sasseen , t. p. , & tornikoski , m. 2005 , , 618 , 139 peterson , b. m. , wanders , i. , horne , k. , collier , s. , alexander , t. , kaspi , s. , & maoz , d. 1998 , pasp , 110 , 660 peterson , b. m. , et al . 2004 , , 613 , 682 pounds , k. a. , edelson , r. a. , markowitz , a. , & vaughan , s. 2001 , , 550 , l15 remillard , r. a. , & mcclintock , j. e. 2006 , , 44 , 49 sasseen , t. p. , & tornikoski , m. 2005 , , 618 , 139 sambruna , r. m. , eracleous , m. , & mushotzky , r. f. 1999 , , 526 , 60s shakura , n. f. & sunyaev , r. a. 1973 , , 24 , 337 shepherd , m. c. 1997 , in asp conf . 125 , astronomical data analysis software and systems vi , ed . g. hunt & h. e. payne ( san francisco : asp ) , 77 savolainen , t. , wiik , k. , valtaoja , e. , jorstad , s. g. , & marscher , a. p. 2002 , , 394 , 851 stone , j. m. , hawley , j. f. , gammie , c. f. , & balbus , s. a. 1996 , , 463 , 656 tersranta , h. et al . 1998 , a&as , 132 , 305 tout , c. a. , & pringle , j. e. 1992 , , 259 , 604 ulvestad , j. s. , et al . 1999 , apjl , 517 , l81 uttley , p. , mchardy , i. m. , & papadakis , i. e. 2002 , , 332 , 231 valtaoja e. , et al . 1999 , , 120 , 95 vlahakis , n. & knigl , a. 2004 , , 605 , 656 walker , r. c. , benson , j. m. , unwin , s. c. , lystrup , m. b. , hunter , t. r. , pilbratt , g. , & hardee , p. e. 2001 , , 556 , 756 woo , j. h. , & urry , c. m. 2002 , , 579 , 530 zdziarski , a. & grandi , p. 2001 , , 551 , 186 ccccccc + & data set & start & end & t ( days ) & @xmath103 t ( days ) & n@xmath104 + & longlook & 2002 december 13 & 2002 december 14 & 1.5 & 0.01 & 150 + x - ray & medium & 2006 november & 2007 january & 60.0 & 0.25 & 240 + & monitor & 2002 march & 2007 may & 1910.0 & 15.0 & 1050 + @xmath105 & 2002.15 & @xmath105&@xmath105 & 2002.23 @xmath106 0.03&02a&2002.39 & 195 & 95 + 2002.19 & 2002.30 & 657 & 120 . & 2002.65 @xmath106 0.04 & 02b & 2002.58 & 404 & 75 + 2002.75 & 2002.76 & 12 & 5 . & & & & & + 2003.02 & 2003.12 & 29 & 22.5 & 2003.35 @xmath106 0.15 & 03a & 2003.35 & 17 & 17 + 2003.32 & 2003.58 & 172 & 52.5 & 2003.67 @xmath106 0.02 & 03b & 2003.72 & 441 & 85 + 2003.66 & 2003.82 & 184 & 40 . & 2003.81 @xmath106 0.03 & 03c & 2003.92 & 24 & 20 + 2003.95 & 2003.98 & 44 & 15 . & 2003.98 @xmath106 0.03 & 03d & & & + 2004.12 & 2004.17 & 20 & 15 . & 2004.16 @xmath106 0.05 & 04a & & & + 2004.21 & 2004.37 & 364 & 65 . & 2004.37 @xmath106 0.03 & 04b & 2004.38 & 126 & 52 + & & & & & & 2004.49 & 60 & 25 + 2004.62 & 2004.66 & 30 & 12.5 & 2004.82 @xmath106 0.05 & 04c & 2005.05 & 61 & 27 + 2005.09 & 2005.12 & 33 & 17.5 & 2005.14 @xmath106 0.03 & 05a & 2005.23 & 21 & 15 + 2005.19 & 2005.39 & 351 & 65 . & 2005.34 @xmath106 0.02 & 05b & 2005.36 & 10 & 10 + & & & & & & 2005.57 & 92 & 57 + & & & & & & 2005.80 & 127 & 57 + 2005.94 & 2006.04 & 287 & 70 . & 2006.00 @xmath106 0.03 & 06a & 2006.43 & 126 & 105 + 2006.38 & 2006.44 & 382 & 72.5 & 2006.72 @xmath106 0.05 & 06b & 2006.88 & 598 & 62 + 2006.96 & 2007.09 & 193 & 62.5 & 2007.05 @xmath106 0.02 & 07a & 2007.43 & 132 & 72 + & & & & @xmath105 & & 2007.87 & 122 & 60 +

we present an extended version of an invited talk given on the international conference turbulent mixing and beyond " . the dynamical and statistical description of stably stratified turbulent boundary layers with the important example of the stable atmospheric boundary layer in mind is addressed . traditional approaches to this problem , based on the profiles of mean quantities , velocity second - order correlations , and dimensional estimates of the turbulent thermal flux run into a well known difficulty , predicting the suppression of turbulence at a small critical value of the richardson number , in contradiction with observations . phenomenological attempts to overcome this problem suffer from various theoretical inconsistencies . here we present an approach taking into full account all the second - order statistics , which allows us to respect the conservation of total mechanical energy . the analysis culminates in an analytic solution of the profiles of all mean quantities and all second - order correlations removing the unphysical predictions of previous theories . we propose that the approach taken here is sufficient to describe the lower parts of the atmospheric boundary layer , as long as the richardson number does not exceed an order of unity . for much higher richardson numbers the physics may change qualitatively , requiring careful consideration of the potential kelvin - helmoholtz waves and their interaction with the vortical turbulence . * nomenclature * + + [ cols="<,^,<,^,<,^ , < " , ] + +

0812.4017

the lower levels of the atmosphere are usually strongly influenced by the earth s surface . known as the atmospheric boundary layer , this is the part of the atmosphere where the surface influences the temperature , moisture , and velocity of the air above through the turbulent transfer of mass . the stability of the atmospheric boundary layer depends on the profiles of the density and the temperature as a function of the height above the ground . during normal summer days the land mass warms up and the temperature is higher at lower elevations . if it were not for the decrease in density of the air as a function of the height , such a situation of heating from below would have been always highly unstable . in fact , the boundary layer is considered stable as long as the temperature decreases at the dry adiabatic lapse rate ( @xmath0 per kilometer ) throughout most of the boundary layer . with such a rate of cooling one balances out the decrease in density . with a higher degree of cooling one refers to the atmospheric boundary layer as unstably stratified , whereas with a lower degree of cooling the situation is stably stratified . stably stratified boundary layer occurs typically during clear , calm nights . in extreme cases turbulence tends to cease , and radiational cooling from the surface results in a temperature that increases with height above the surface . the tendency of the atmosphere to be turbulent does not depend only on the rate of cooling but also on the mean shear in the vertical direction . the commonly used parameter to describe the tendency of the atmosphere to be turbulent is the gradient " richardson number ( richardson , 1920 ) , defined as @xmath1 ^ 2 } \,,\ ] ] where @xmath2 is the stream - wise direction , @xmath3 is the height above the ground , @xmath4 is the mean potential temperature profile , ( which differs from the mean temperature profile @xmath5 by accounting for the adiabatic cooling of the air during its expansion : @xmath6 ) , @xmath7 is the buoyancy parameter in which @xmath8 is the adiabatic thermal expansion coefficient ( for an ideal gas @xmath9 ) , and @xmath10 is the gravitational acceleration . the mean shear @xmath11 is defined in terms of the mean velocity @xmath12 , which in the simplest case of flat geometry depends only on the vertical coordinate @xmath3 . the parameter @xmath13 represents the ratio of the generation or suppression of turbulence by buoyant production of energy to the mechanical generation of energy by wind shear . this paper is an extended presentation of an invited talk given on international conference turbulent mixing and beyond " devoted , in particular , to the problems of fluid dynamics , turbulence , geophysics and statistics , that are long - standing challenging tasks . here we consider the description of stably stratified turbulent boundary layers ( tbl ) , taking as an example the case of stable thermal stratification . since the 50 s of twentieth century , traditional models of stratified tbl generalize models of unstratified tbl , based on the budget equations for the kinetic energy and mechanical momentum ; see reviews of umlauf and burchard ( 2005 ) , weng and taylor ( 2003 ) . the main difficulty is that the budget equations are not closed ; they involve turbulent fluxes of mechanical moments @xmath14 ( known as the reynolds stress " tensor ) and a thermal flux @xmath15 ( for the case of thermal stratification ) : @xmath16 where @xmath17 and @xmath18 stand for the turbulent fluctuating velocity and the potential temperature with zero mean . the nature of the averaging procedure behind the symbol @xmath19 will be specified below . earlier estimates of the fluxes ( [ def3 ] ) are based on the concept of the down - gradient turbulent transport , in which , similarly to the case of molecular transport , a flux is taken proportional to the gradient of transported property times a corresponding ( turbulent ) transport coefficient : [ dt ] @xmath20 here the turbulent - eddy viscosity @xmath21 and turbulent thermal conductivity @xmath22 are estimated by dimensional reasoning via the vertical turbulent velocity @xmath23 and a scale @xmath24 ( which in the simplest case is determined by the elevation @xmath3 ) . the dimensionless coefficients @xmath25 and @xmath26 are assumed to be of the order of unity . this approach meets serious difficulties ( zeman , 1981 ) , in particular , it predicts full suppression of turbulence when the stratification exceeds a critical level , for which @xmath27 . on the other hand , in observations of the atmospheric turbulent boundary layer turbulence exists for much larger values than @xmath28 : experimentally above @xmath29 and even more ( see galperin et al . ( 2007 ) and references therein ) . in models for weather predictions this problem is fixed " by introducing fit functions @xmath30 and @xmath31 instead of the constant @xmath32 and @xmath26 in the model parametrization ( [ dt ] ) . this technical solution " is not based on any physical derivation and just masks the shortcomings of the model . to really solve the problem one has to understand its physical origin , even though from a purely formal viewpoint it is indeed possible that a dimensionless coefficient like @xmath26 can be any function of @xmath13 . to expose the physical reason for the failure of the down - gradient approach , recall that in a stratified flow , in the presence of gravity , the turbulent kinetic energy is _ not _ an integral of motion . only the total mechanical energy , the sum of the kinetic and the potential energy , is conserved in the inviscid limit . as it was shown already by richardson , the difficult point is that an important contribution to the potential energy comes not just from the mean density profile , but from the density fluctuations . clearly , any reasonable model of the turbulent boundary layer must obey the conservation laws . the physical requirement of conserving the total mechanical energy calls for an explicit consideration not only of the mean profiles , but also of _ all _ the relevant second - order , one - point , simultaneous correlation functions of _ all _ the fluctuating fields together with some closure procedure for the appearing third order moments . first of all , in order to account for the important effect of stratification on the anisotropy , we must write explicit equations for the entire reynolds stress tensor , @xmath33 . next , in the case of the temperature stratified turbulent boundary layers we follow tradition [ see , e.g. zeeman ( 1981 ) , hunt et al . ( 1988 ) , schumann and gerz ( 1995 ) , hanazaki and hunt ( 2004 ) , keller and van atta ( 2000 ) , stretch et al . ( 2001 ) , elperin et al . ( 2002 ) , cheng et al . ( 2002 ) luyten et al . ( 2002 ) , and rehmann and hwang ( 2005 ) ] and account for the turbulent potential energy which is proportional to the variance of the potential temperature deviation , @xmath34 . and last but not least , we have to consider explicitly equations for the vertical fluxes , @xmath35 and @xmath36 , which include the down gradient terms proportional to the velocity and temperature gradients , and counter - gradient terms , proportional to @xmath37 ( in the equation for @xmath35 ) and to @xmath38 ( in the equation for @xmath36 ) . unfortunately , the resulting second order closure seems to be inconsistent with the variety of boundary - layer data , and many authors took the liberty to introduce additional fitting parameters and sometimes fitting functions to achieve a better agreement with the data ( see reviews of umlauf and burchard ( 2005 ) , weng and taylor ( 2003 ) , zeeman , ( 1981 ) , melor and yamada ( 1974 ) , and references therein ) . moreover , in the second order closures the problem of critical richardson number seems to persists ( cheng et al . , 2002 ; canuto , 2002 ) . notice that in spite of obvious inconsistency of the first - order schemes , most of the practically used turbulent models are based on the concept of the down - gradient transport . one of the reasons is that in the second - order schemes instead of two down - gradient equations ( [ dt ] ) one needs to take into account eight nonlinear coupled additional equations i.e. four equations for the reynolds stresses , three equations for the heat fluxes and equation for the temperature variance . as the result , the second - order schemes have seemed to be rather cumbersome for comprehensive analytical treatment and have allowed to find only some relationships between correlation functions ( see , e.g. , cheng et al . , 2002 ) . unfortunately , the numerical solutions to the complete set of the second - order schemes equations which involve too many fitting parameters are much less informative in clarification of physical picture of the phenomenon than desired analytical ones . in this paper we suggest a relatively simple second - order closure model of turbulent boundary layer with stable temperature stratification that , from one hand , accounts for main relevant physics in the stratified tbl and , from the other hand , is simple enough to allow complete analytical treatment including the problem of critical ri@xmath39 . to reach this goal we approximate the third order correlations via the first- and second - order ones , accounting only for the most physically important terms . we will try to expose the approximations in a clear and logical way , providing the physical justification as we go along . resulting second - order model consist of nine coupled equations for the mean velocity and temperature gradients , four components of the reynolds stresses , two components of the temperature fluxes and the temperature variance . thanks to the achieved simplicity of the model we found an approximate analytical solution of these equations , expressing all nine correlations as functions of only one governing parameter , @xmath40 , where @xmath41 is the outer scale of turbulence ( depending on the elevation @xmath3 and also known as the dissipation scale " ) and @xmath42 is the obukhov length . we would like also to stress , that in our approach @xmath40 is an external parameter of the problem . for small elevations @xmath43 , it is well accepted that @xmath41 is proportional to @xmath3 , while the @xmath41 dependence is still under debate for @xmath3 comparable or exceeding @xmath42 . for @xmath44 the assignment and discussion of the actual dependence of the outer scale of turbulence , @xmath41 , which is manifested in the nature is out of the scope of this paper , and is remained for future work . at time being , we can analyze consequences of our approach for the following versions of @xmath41 dependence at @xmath45 : + function @xmath41 is saturated at some level of the order of @xmath42 . for concreteness we can take @xmath46 @xmath41 is again proportional to @xmath3 for elevations much larger than @xmath42 : @xmath47 but with the proportionality constant @xmath48 . if so , we can also study the case @xmath49 even though such a condition may not be realizable in nature . in that case our analysis of the limit @xmath49 has only a methodological character : it allows to derive an approximate analytic solution for all the objects of interest as functions of @xmath40 that is also valid for the outer scale of turbulence not exceeding a value of the order of @xmath42 . it should be noticed that traditional turbulent closures ( including ours ) can not be applied for strongly stratified flows with @xmath50 ( may be even at @xmath51 ) . the main reason is that these closures are roughly justified for developed _ vortical _ turbulence , in which the eddy - turnover time is of the order of its life time ; in other words , there are no well defined quasi - particles " or waves . this is not the case for stable stratification with @xmath50 , in which the brunt - visl frequency @xmath52 is larger then the eddy - turnover frequency @xmath53 . it means that for @xmath54 there are weakly decaying kelvin - helmoholtz internal gravity waves ( with characteristic frequency @xmath55 and decay time above @xmath56 ) , propagating on large distances , essentially effecting on tbl , as pointed out by zilitinkevich , ( 2002 ) . we concentrate in our paper on self - consistent description of the lower part of the atmospheric tbl , in which turbulence has vortical character and consequently , large values of @xmath13 do not appear . we relate large values of @xmath13 in the upper part of tbl with contributions of the internal gravity waves to the energy and the energy flux in tbl , to the momentum flux , and to the production of ( vortical ) turbulent energy . due to their instability in a shear flow , the waves can break and create turbulent kinetic energy . all these effects are beyond the framework of our paper . their description in the upper potential - wave " tbl and intermediate region with the combined vortical - potential " turbulent velocity field is in our nearest agenda . to make the paper more transparent for wide audience , not necessarily experts in atmospheric tbl , we attempt to present the material in a self - contained manner , and organized it as follows . in sect . [ s : balance]a we use the oberbeck - boussinesq approximation and apply the standard reynolds decomposition ( into mean values and turbulent zero - mean fluctuations of the velocity and temperature fields ) to derive equations for the mean values and balance equations for all relevant second - order correlation functions . in sect . [ s : balance]b we demonstrate that the resulting balance equations exactly preserve ( in the non - dissipative limit ) the total mechanical energy of the system , that consists of the kinetic energy of the mean flow , kinetic energy and potential energy of the turbulent subsystem . in sect . [ s : closure ] we describe the proposed closure procedure that results in a model of stably stratified tbl , that accounts explicitly for all relevant second - order correlations . the third order correlations which appear in the theory are modeled in terms of second - order correlations in sects . [ s : closure]a and b. further simplifications are presented in sects . [ s : closure]c and d for stationary turbulent flows in a plane geometry outside the viscous and buffer layers . in sect . [ s : closure]e we suggest a generalization of the standard wall - normalization " to obtain the model equations in a dimensionless form with only one governing parameter , @xmath40 . section [ ss : strat41 ] contains approximate analytical solution of the model . it is shown that the analytical solution deviates from the numerical counterpart in less than a few percent in the entire interval @xmath57 . the last sect . [ s : res ] is devoted to a detailed description of our results : profiles of the mean velocity and potential temperature ( sect . [ s : res]a ) , profiles of the turbulent kinetic and temperature " energies , profiles of the anisotropy of partial kinetic energies ( sect . [ s : res]b ) , profiles of the turbulent transport parameters @xmath58 and @xmath59 , profiles of the gradient- and flux - richardson numbers @xmath13 and @xmath60 , and the dependence of the turbulent prandtl number pr@xmath61 vs. @xmath62 and @xmath13 , sect . [ s : res]c . in conclusion sect . [ s : res]d , we consider the validity of the down - gradient transport concept ( [ dt ] ) and explain why it is violated in the upper part of tbl . the problem of critical @xmath13 is also discussed . the aim of this section is to consider the simplified dynamics of a stably temperature - stratified turbulent boundary layer , aiming finally at an explicit description of the height dependence of important quantities like the mean velocity , mean temperature , turbulent kinetic and potential energies , etc . in general one expects very different profiles from those known in standard ( unstratified ) wall - bounded turbulence . we want to focus on these differences and propose that they occur already relatively close to the ground allowing us to neglect ( to the leading order ) the dependence of the density on height and the coriolis force . we thus begin by simplifying the hydrodynamic equations which are used in this section . first we briefly overview the derivation of the governing equations in the boussinesq approximation . the system of hydrodynamic equations describing a fluid in which the temperature is not uniform consists of the navier - stokes equations for the fluid velocity , @xmath63 , a continuity equation for the space and time dependent ( total ) density of the fluid , @xmath64 , and of the heat balance equation for the ( total ) entropy per unit mass , @xmath65 , landau and lifshitz , 1987 . these equations are considered with boundary conditions that maintain the solution far from the equilibrium state , where @xmath66 . these boundary conditions are @xmath67 at zero elevation , @xmath68 at a high elevation of a few kilometers . this reflects the existence of a wind at high elevation , but we do not attempt to model the physical origin of this wind in any detail . the only important condition with regards to this wind is that it maintains a momentum flux towards the ground that is prescribed as a function of the elevation . similarly , we assume that a stable temperature stratification is maintained such that the heat flux towards the ground is prescribed as well . we neglect the viscous entropy production term assuming that the temperature gradients are large enough such that the thermal entropy production term dominates . for simplicity of the presentation we restrict ourselves by relatively small elevations and disregard the coriolis force ( for more details , see wyngaard , 1992 ) . on the other hand we assume that the temperature and density gradients in the entire turbulent boundary layer are sufficiently small to allow employment of local thermodynamic equilibrium . in other words , we assume the validity of the equation of state . as a basic reference state " ( brs ) denoted hereafter by a subscript @xmath69 " we use the isentropic model of the atmosphere , where the entropy is considered space homogeneous . now assuming smallness of deviations of the density and pressure from their brs values and exploiting the equation of state , one obtains a simplified equation , which is already very close to the standard navier - stokes equation in the boussinesq approximation . introducing ( generalized ) potential temperature , one results in the well - known system of hydrodynamic equations in the boussinesq approximation . close to the ground , where one can neglect the dependence of the density on height , the system reads : @xmath70here @xmath71 is the convection time derivative , @xmath72 deviation of pressure from brs , @xmath73 is the density in brs , @xmath74 is the buoyancy parameter ( @xmath75 , @xmath76 , @xmath10 is the gravity acceleration and @xmath77 is the thermal expansion coefficient , which is equal to @xmath78 , reciprocal molecular temperature , for an ideal gas ) , @xmath79 is the deviation of the potential temperature from brs value , @xmath80 kinematic viscosity and @xmath81 is the kinematic thermal conductivity . to develop equations for the mean quantities and correlation functions one applies the reynolds decomposition : @xmath82 . here the average @xmath83 stands for an averaging over a horizontal plane at a constant elevation . this leaves the average quantities with a @xmath84 dependence only . substituting in eqs . ( [ goveq ] ) one gets equations of motion for the mean velocity and mean temperature profiles @xmath85 here @xmath86 is the mean convection derivative . the total ( molecular and turbulent ) momentum and thermal fluxes are @xmath87 where @xmath88 is the reynolds stress tensor describing the turbulent momentum flux , and @xmath89 is the turbulent thermal flux . in order to derive equations for these correlation functions , one considers the equations of motion for the fluctuating velocity and temperature : [ fluct]@xmath90 the whole set of the second order correlation functions includes the reynolds stress , @xmath14 , the turbulent thermal flux , @xmath15 , and the temperature energy " @xmath91 , which is denoted and named by analogy with the turbulent kinetic energy density ( per unit mass and unit volume ) , @xmath92 . using ( [ fluct ] ) one gets the following balance equations " : [ corr]@xmath93 here we denoted the dissipations of the reynolds - stress , heat - flux and the temperature energy by @xmath94 the last term on the lhs of each of eqs . ( [ corr ] ) describes _ spatial _ flux of the corresponding quantity . in models of wall bounded unstratified turbulence it is known that these terms are very small almost everywhere . we do not have sufficient experience with the stratified counterpart to be able to assert that the same is true here . nevertheless , for simplicity we are going to neglect these terms . it is possible to show that the accounting for these terms does not influence much the results . note that keeping these terms turns the model into a set of differential equations which are very cumbersome to analyze . this is a serious uncontrolled step in our development , so we cross our fingers and proceed with caution . since these terms are neglected we do not provide here the explicit expressions for @xmath95 , @xmath96 , and @xmath97 . the first term on the rhs of the balance eq . ( [ corra ] ) for the reynolds stresses is the energy production tensor " @xmath98 , describing the production of the turbulent kinetic energy from the kinetic energy of the mean flow , proportional to the gradient of the mean velocity : [ defs1 ] @xmath99 the second term on the rhs of eq . ( [ corra ] ) , @xmath100 , will be referred hereafter to as the energy conversion tensor " . it describes the conversion of the turbulent kinetic energy into potential energy . this term is proportional to the buoyancy parameter @xmath101 and the turbulent thermal flux @xmath15 : @xmath102the next term in the rhs of eq . ( [ corra ] ) is known as the pressure - rate - of - strain tensor " : @xmath103 in incompressible turbulence its trace vanishes , therefore @xmath104 does not contribute to the balance of the kinetic energy . as we will show in sec . [ sss : prs ] , this tensor can be presented as the sum of three contributions ( zeman , 1981 ) , @xmath105 in which @xmath106 is responsible for the nonlinear process of isotropization of turbulence and is traditionally called the return - to - isotropy " , @xmath107 is similar to the energy production tensor ( [ defs1a ] ) and is called isotropization of production " . a new term , appearing in the stratified flow , @xmath108 , is similar to the energy conversion tensor ( [ defs1b ] ) and will be refereed to as the isotropization of conversion " . consider the balance of the turbulent thermal flux @xmath15 , eq . ( [ corrb ] ) . the first term in the rhs , @xmath109 , describes the source of @xmath15 and , by analogy with the energy - production tensor , @xmath110 , is called thermal - flux production vector " . like @xmath110 , eq . ( [ defs1b ] ) , it has the contribution , @xmath111 , proportional to the mean velocity gradient : @xmath112 and two additional contributions , related to the temperature gradient and to the temperature energy " , @xmath113 , and the buoyancy parameter . one sees , that in contrary to the oversimplified assumption ( [ dth ] ) , the thermal flux in such a turbulent media can not be considered as proportional to the temperature gradient . it has also a contribution proportional to the velocity gradient and even to the square of the temperature fluctuations . moreover , the rhs of the flux - balance eq . ( [ corrb ] ) has an additional term , the pressure - temperature - gradient vector " which , similarly to the pressure - rate - of - strain tensor ( [ rof ] ) , can be divided into three parts ( zeman , 1981 ) : @xmath114as we will show in sec . [ sss : prs ] the first contribution , @xmath115 is responsible for the nonlinear flux of @xmath15 in the space of scales toward smaller scales , similarly to the correlation @xmath116 , which is responsible for the flux of kinetic energy @xmath117 toward smaller scales . the correlation @xmath118 may be understood as the nonlinear contribution to the dissipation of the thermal flux . correspondingly we will call it renormalization of the thermal - flux dissipation " and will supply it with a superscript @xmath119 " . the next two terms in the decomposition ( [ defs1f ] ) are @xmath120 and @xmath121 . they describe the renormalization of the thermal - flux production terms @xmath122 and @xmath123 , accordingly . the total mechanical energy of temperature stratified turbulent flows consists of three parts with densities ( per unit mass ) : @xmath124 , where @xmath125 is the density of kinetic energy of the mean flow , @xmath126 is the density of turbulent kinetic energy and @xmath127 is the density of potential energy , associated with turbulent density fluctuation @xmath128 , caused by the ( potential ) temperature fluctuations @xmath18 , and @xmath129 . the balance eq . for @xmath130 follows from eq . ( [ mean ] ) : [ bale]@xmath131 + \tau_{ij}\nabla_{\!\ ! j } \ , u_i\,,\end{aligned}\ ] ] with the help of identity : @xmath132 and definition ( [ def - tau ] ) . the terms on the lhs of this eq . , proportional to @xmath80 and @xmath133 respectively , describe the dissipation and the spatial flux of @xmath134 . the term [ source @xmath135 on the rhs of eq . ( [ baletm ] ) describes the external source of energy , originating from the boundary conditions described above , and @xmath136 describes the kinetic energy out - flux from the mean flow to turbulent subsystem . the balance eq . for the turbulent kinetic energy follows directly from eq . ( [ corra ] ) : @xmath137/2=- \tau_{ij } \nabla_{\!\ ! j } \ , u_i + \beta f_z\ . \]]on the lhs of eq . ( [ balett ] ) one sees the dissipation and spatial flux terms . the first term on the rhs originates from the energy production , @xmath138 , defined by eq . ( [ defs1a ] ) . this term has an opposite sign to the last term on the rhs of eq . ( [ baletm ] ) and describes the production of the turbulent kinetic energy from the kinetic energy of the mean flow . the last term on the rhs of eq . ( [ balett ] ) originates from the energy conversion tensor @xmath139 , eq . ( [ defs1b ] ) , and describes the conversion of the turbulent kinetic energy into potential one . according to the last of eqs . ( [ corr ] ) , one gets the balance equation for the potential energy @xmath140 ; multiplying eq . ( [ corrc ] ) for @xmath113 by @xmath141:@xmath142/s_\theta = -\beta f_z\ .\ ] ] the rhs of this eq . [ coinciding up to a sign with the last term on the rhs of eq . ( [ balett ] ) ] is the source of potential energy ( from the kinetic one ) . in the sum of the three balance equations , the conversion terms ( of the kinetic energy from the mean to turbulent flows and of the turbulent kinetic energy to the potential one ) cancel and one gets the total mechanical energy balance : @xmath143 + { { \bm{\nabla}}}\ , [ \mbox{flux } e]= [ \mbox{source } e_{{{\mathcal{k}}}}]\ .\ ] ] this equation exactly respects the conservation of total mechanical energy in the dissipation - less limit , irrespective of the closure approximations . this is because the energy production and conversion terms are exact and do not require any closures , while the pressure - rate - of - strain tensor , that requires some closure , does not contribute to the total energy balance . in this section we describe the proposed closure procedure that results in a model of stably stratified tbl . in developing this model we strongly rely on the analogous well developed modeling of standard ( unstratified ) tbl . the final justification of this approach can be done only in comparison to data from experiments and simulations . we will do below what we can to use the existing data , but we propose at this point that much more experimental and simulational work is necessary to solidify all the steps taken in this section . the correlation functions @xmath144 and @xmath145 , defined by eqs . ( [ defs1c ] ) and ( [ defs1f ] ) , include fluctuating part of the pressure @xmath146 . the poisson s equation for @xmath146 follows from eq . ( [ fluct ] ) : @xmath147 $ ] . the solution of this equation includes a harmonic part , @xmath148 , which is responsible for sound propagation and does not contribute to turbulent dynamics at small mach numbers . thus this contribution can be neglected . the inhomogeneous solution includes three parts @xmath149 $ ] , where @xmath150 and the inverse laplace operator @xmath151 is defined as usual in terms of an integral over the green s function . correspondingly the correlations @xmath144 and @xmath145 consist of three terms , eqs . ( [ rof ] ) and ( [ defs1f ] ) , in which @xmath152 all of those terms originating from @xmath153 are the most problematic because they introduce coupling to triple correlation functions : @xmath154 and @xmath155 . thus they require closure procedures whose justification can be only tested a - posteriori against the data . having in mind to simplify the model in most possible manner , we adopt for the diagonal part of the return - to - isotropy tensor , the simplest rota form ( rotta , 1951 ) [ rof]@xmath156 in which @xmath157 is the relaxation frequency of diagonal components of the reynolds - stress tensor toward its isotropic form , @xmath158 . the parametrization of @xmath157 will be discussed later . the tensor @xmath106 is traceless , therefore the frequency @xmath157 must be the same for all the diagonal components of @xmath159 . on the other hand there are no reasons to assume that off - diagonal terms have the same relaxation frequency . therefore , following lvov et al . ( 2006a ) we assume that @xmath160with , generally speaking , @xmath161 . moreover , on the intuitive level , we can expect that off - diagonal terms should decay faster then the diagonal ones , i.e. @xmath162 . indeed , our analysis of dns results shows that @xmath163 ( lvov et al . , 2006b ) . the term @xmath164 also describes return - to - isotropy due to nonlinear turbulence self interactions ( zeman , 1981 ) , and may be modeled as : @xmath165 this equation dictates the vectorial structure of @xmath166 , which will be confirmed below . the rest can be understood as the definition of the @xmath167 as the relaxation frequency of the thermal flux . its parametrization is the subject of further discussion in sec . [ ss : closure ] . the traceless isotropization - of - production " tensor , @xmath107 , has a very similar structure to the production tensor , @xmath98 , eq . ( [ defs1a ] ) , and thus is traditionally modeled in terms of @xmath98 ( pope , 2001):@xmath168the accepted value of the numerical constant @xmath169 ( pope , 2001 ) . the traceless isotropization - of - conversion " tensor , @xmath108 does not exist in unstratified tbl . its structure is very similar to the conversion tensor , @xmath170 , eq . ( [ defs1b ] ) . therefore it is reasonable to model it in the same way in terms of @xmath170 ( zeman , 1981 ) : @xmath171with some new constant @xmath172 . the renormalization of production terms @xmath173 and @xmath174 are very similar to the corresponding thermal flux production terms , @xmath175 and @xmath176 , defined by eqs . ( [ defofa ] ) . therefore , in the spirit of eqs . ( [ ip ] ) and ( [ ic ] ) , they are modeled as follows:@xmath177 using this and ( [ pr1 ] ) one finds the sign of @xmath178 : @xmath179to estimate @xmath178 we assume that on the gradient scales the temperature fluctuations are roughly isotropic , and therefore we can estimate @xmath180 . introducing this estimate and integrating by parts leads to @xmath181 . far away from the wall and for large reynolds numbers the dissipation tensors are dominated by the viscous scale motions , at which turbulence can be considered as isotropic . therefore , the vector @xmath182 should vanish , while the tensor @xmath183 , eq . ( [ diss ] ) , should be diagonal : [ distau1 ] @xmath184where the numerical prefactor @xmath185 is chosen such that @xmath186 becomes the relaxation frequency of the turbulent kinetic energy . under stationary conditions the rate of turbulent kinetic energy dissipation is equal to the energy flux through scales , that can be estimated as @xmath187 , where @xmath188 is the outer scale of turbulence . therefore , the natural estimate of @xmath186 involves the triple - velocity correlator , @xmath189 , exactly in the same manner , as the return - to - isotropy frequencies , @xmath190 and @xmath191 in eqs . ( [ ria ] ) and ( [ rib ] ) . similarly , @xmath192 in the plane geometry , the equations simplify further . the mean velocity is oriented in the ( streamwise ) @xmath193 direction and all mean values depend on the vertical ( wall - normal ) coordinate @xmath3 only : @xmath194 , @xmath195 , @xmath196 , @xmath197 , @xmath198 . therefore @xmath199 , and in the stationary case , when @xmath200 , the mean convective derivative vanishes : @xmath201 . moreover due to the @xmath202 symmetry of the problem the following correlations vanish : @xmath203 . the only non - zero components of the mean velocity and temperature gradients are : @xmath204 having in mind eqs . of sec . [ ss : balance ] and integrating eqs . ( [ mean ] ) for @xmath205 and @xmath206 over @xmath3 , one gets equations for the total ( turbulent and molecular ) mechanical - momentum flux , @xmath207 , and thermal flux , @xmath208 , toward the wall [ tbl - sim ] @xmath209 the total flux of the @xmath2-component of the mechanical moment in @xmath3-direction is @xmath210 . generally speaking , @xmath211 depends on @xmath3 . for example , for the pressure driven planar channel flow ( of the half - wight @xmath212 ) @xmath213 . relatively close to the ground , where @xmath214 , the @xmath3 dependence of @xmath215 can be neglected . in the absence of the mean horizontal pressure drop and spatial distributed heat sources @xmath216 and @xmath208 are @xmath3-independent , and thus equal to their values at zero elevation , as indicated in eqs . ( [ tbl - sim ] ) after @xmath217"-sign . notice , that in our case of stable stratification both vertical fluxes , the @xmath2-component of the mechanical momentum , @xmath218 , and the thermal flux , @xmath219 , are directed toward the ground , i.e. negative . for the sake of convenience , we introduce in eqs . ( [ tbl - sim ] ) notations for their ( positive ) zero level absolute value : @xmath220 and @xmath221 . recall that in the plain geometry @xmath222 . nevertheless one can write an equation for @xmath223 : @xmath224 which describes a turbulent correction ( @xmath225 ) to the hydrostatic equilibrium . actually , this equation determines the profile of @xmath226 , that does not appear in the system of balance equations ( [ tbl - sim ] ) . consider first the balance eqs . ( [ corra ] ) for the diagonal components of the reynolds - stress tensor in algebraic model ( which arises when we neglect the spatial fluxes ) : @xmath227where @xmath228 . the lhs of these equations includes the dissipation and return - to - isotropy terms . on the rhs we have the kinetic energy production and isotropization of production terms ( both proportional to @xmath229 ) together with the conversion and isotropization of conversion terms , that are proportional to the vertical thermal flux @xmath36 . the horizontal component of the thermal flux , @xmath37 , does not appear in these equations . system ( [ taup ] ) allows to find anisotropy of the turbulent - velocity fluctuations and to get the balance eqs . for the turbulent kinetic energy with the energy production and conversion terms on the rhs : [ tau1 ] @xmath230 e{_{\scriptscriptstyle\rm{k } } } \\ \nonumber & & -(3 -2c{_{\scriptscriptstyle\rm{ip } } } + c{_{\scriptscriptstyle\rm{ic}}})\ , \beta f_z / \gamma{_{\scriptscriptstyle\rm{ri } } } \big\ } \ , , \\ 3\tau_{yy } & = & 2 \big\ { [ ( c{_{\scriptscriptstyle\rm{ip}}}-1)\gamma_{uu}/\gamma{_{\scriptscriptstyle\rm{ri } } } + 1 ] e{_{\scriptscriptstyle\rm{k } } } \\ \nonumber & & -(c{_{\scriptscriptstyle\rm{ip } } } + c{_{\scriptscriptstyle\rm{ic } } } ) \beta f_z/\gamma{_{\scriptscriptstyle\rm{ri } } } \big\ } \ , , \\ 3\tau_{zz } & = & 2 \big\ { [ \left ( c{_{\scriptscriptstyle\rm{ip}}}-1 \right ) \gamma_{uu}/\gamma{_{\scriptscriptstyle\rm{ri } } } + 1 ] e{_{\scriptscriptstyle\rm{k } } } \\ \nonumber & & - ( c{_{\scriptscriptstyle\rm{ip } } } -2c{_{\scriptscriptstyle\rm{ic } } } -3 ) \ , \beta f_z/\gamma{_{\scriptscriptstyle\rm{ri}}}\big\ } \ , , \\ \label{bale } \gamma_{uu}e{_{\scriptscriptstyle\rm{k } } } & = & -\tau_{xz}s_{_u}+ \beta f_z \,,\ ] ] equation ( [ bale ] ) includes the only non - vanishing tangential ( off - diagonal ) reynolds stress @xmath35 and has to be accompanied with an equation for this object : @xmath231 this equation manifests that the tangential reynolds stress @xmath35 , that determines the energy production [ according to eq . ( [ bale ] ) ] , influences , in its turn , on the value of the streamwise thermal flux @xmath37 , which therefore effects on the turbulent kinetic energy production . as we mentioned , in the plain geometry @xmath232 . equations ( [ corrb ] ) for the @xmath37 and @xmath36 in this case take the form : @xmath233 in which the rhs describes the thermal - flux production , corrected by the isotropization of production terms . the last eq . ( [ corrc ] ) for @xmath234 , represents the balance between the dissipation ( lhs ) and production ( rhs ) : @xmath235 at this point we follow a tradition in modeling of all the nonlinear inverse time - scales by dimensional estimates ( kolmogorov , 1941 ) : @xmath236remember that @xmath237 is the outer scale of turbulence " . this scale equals to @xmath3 for @xmath238 , where @xmath42 is the obukhov length ( definition is found below ) . detailed analysis of experimental , dns and les data ( see lvov et al . , 2006 , and references therein ) shows that for unstratified flows , @xmath239 , the anisotropic boundary layers exhibits values of the reynolds stress tensor that can be well approximated by the values @xmath240 , @xmath241 . in our approach this dictates the choice @xmath242 . also we can expect that @xmath243 is almost not affected by buoyancy . this gives simply @xmath244 . if so , eqs . ( [ tau1 ] ) with the parametrization ( [ freqs ] ) can be identically rewritten as follows : [ simple ] @xmath245 for completeness we also repeated here the parametrization ( [ freqs ] ) of @xmath186 . finally we present the version of the balance eqs . for the thermal flux ( [ sim - f - x ] ) , ( [ sim - f - z ] ) , and for the temperature energy " , ( [ sim - et ] ) , after all the simplified assumptions : @xmath246 the analysis of the balance eqs . ( [ simple ] ) is drastically simplified if they are presented in a dimensionless form . traditionally , the conventional wall units " are introduced via the wall friction velocity @xmath247 , and the viscous length - scale @xmath248 . a wall unit for the temperature @xmath249 is defined via the thermal flux at the wall and friction velocity . subsequently , @xmath250 , @xmath251 , @xmath252 , @xmath253 , @xmath254 , @xmath255 , etc . then the governing eqs . ( [ goveq ] ) take the form:@xmath256 these eqs . include two dimensionless parameters : the conventional prandtl number pr@xmath257 , and @xmath258 the obukhov length @xmath42 measured in wall units : @xmath259 , @xmath260 . we used here the modern definition of the obukhov length , which differs from the old one by the absence of the von - krmn constant @xmath261 in its denominator ( monin and obukhov , 1954 ) . outside of the viscous sub - layer , where the kinematic viscosity and kinematic thermal conductivity can be ignored , @xmath258 is the only dimensionless parameter in the problem , which separates the region of weak stratification , @xmath262 , and the region of strong stratification , where @xmath263 . given the generalized wall normalization we introduce objects with a superscript @xmath264 " in the usual manner : @xmath265 in the turbulent region , governed by @xmath258 only , eqs . ( [ tbl - sim ] ) simplify to @xmath266 , @xmath267 . outside of the viscous region , where eqs . ( [ simple ] ) were derived , the problem has only one characteristic length , i.e. the obukhov scale @xmath269 . correspondingly , one expects that the only dimensionless parameter that governs the turbulent statistics in this region should be the ratio of the outer scale of turbulence , @xmath188 , to the obukhov length - scale @xmath42 , which we denote as @xmath270 . indeed , introducing @xmath268-objects " : @xmath271 and using eqs . ( [ plus - m ] ) one rewrites the balance eqs . ( [ simple ] ) as follows : [ mm]@xmath272 these equations are the main result of current sec . [ s : closure ] . it may be considered as minimal model " for stably stratified tbl , that respects the conservation of energy , describes anisotropy of turbulence and all relevant fluxes explicitly and , nevertheless is still simple enough to allow comprehensive analytical analysis , that results in an approximate analytical solution ( with reasonable accuracy ) for the mean velocity and temperature gradients @xmath273 and @xmath274 , and all second - order ( cross)-correlation functions . as expected , the only parameter appearing in the minimal model ( [ mm ] ) is @xmath275 . the outer scale of turbulence , @xmath188 , does not appear by itself , only via the definition of @xmath275 ( [ ddagb ] ) . therefore our goal now is to solve eqs . ( [ mm ] ) in order to find five functions of only one argument @xmath276 : @xmath277 , @xmath278 , @xmath279 , @xmath280 and @xmath281 . after that we can specify the dependence @xmath282 and then reconstruct the @xmath283-dependence of these five objects . this subsection is devoted to an analytical and numerical analysis of the minimal - model ( [ mm ] ) . an example of numerical solution of eqs . ( [ mm ] ) ( with some reasonable choice of the phenomenological parameters ) is shown in fig . [ f : s ] . nevertheless , it would be much more instructive to have approximate analytical solutions for all correlations that will describe their @xmath275-dependence with reasonable accuracy . the detailed cumbersome procedure of finding these solutions is skipped here , but a brief overview is as follows . the eqs . ( [ mm ] ) can be reformulated as a polynomial equation of ninth order for the only unknown @xmath284 . an analysis of its structure helps to formulate an effective interpolation formula ( [ inter ] ) , discussed below . hence , we found the solutions of eqs . ( [ mm ] ) at neutral stratification , @xmath285 , corrected up to the linear order in @xmath275 . its comparison with the existing dns data resulted in an estimate for the constants @xmath286 , and @xmath287 . then , we considered the region @xmath288 . even though such a condition may not be realizable in nature , from a methodological point of view , as we will see below , it enables to obtain the desired analytical approximation . the @xmath288 asymptotic solution with corrections , linear in the small parameter @xmath289 , were found . now we are armed to suggest an interpolation formula [ inter ] @xmath290 that coincides with the exact solutions for @xmath285 and for @xmath291 , including the leading corrections to both asymptotics , linear in @xmath275 , and @xmath292 . moreover , in the region @xmath293 , eq . ( [ intera ] ) accounts for the structure of the exact polynomial . as a result , the interpolation formula ( [ intera ] ) is close to the numerical solution with deviations smaller than 3% in the entire region @xmath294 , see upper middle panel on fig . [ f : s ] . together with eq . ( [ 9-eqc ] ) it produces a solution for @xmath295 , that can be written as @xmath296 where @xmath297 and @xmath261 is the von - krmn constant . this formula gives the same accuracy @xmath298 , see upper left panel in fig . [ f : s ] . we demonstrate below that the proposed interpolation formulae describe the @xmath275-dependence of the correlations with a very reasonable accuracy , about @xmath299 , for any @xmath300 , see black dashed lines in figs . [ f : s ] . unfortunately , a direct substitution of the interpolation formula ( [ inter ] ) into the exact relation for @xmath301 obtained from the system ( [ mm ] ) works well only for small @xmath302 , in spite of the fact that the interpolation formula is rather accurate in the whole region . we need therefore to derive an independent interpolation formula for @xmath301 . using expansions for small @xmath303 and large @xmath304 we suggest @xmath305 in which @xmath306 and @xmath307 satisfies @xmath308with @xmath309 equation ( [ inter1a ] ) is constructed such that the leading and sub - leading asymptotics for small and large @xmath310 coincide with the first two terms in the exact expansions at `` almost '' neural stratification and extremely strong stratification . as a result , ( [ inter1a ] ) approximates the exact solution with errors smaller then 5% for @xmath311 and @xmath312 and with errors smaller than 10% for any @xmath310 , see lower left panel in figure [ f : s ] . substituting the approximate eqs . ( [ inter ] ) into the exact relations ( [ mm ] ) , one gets approximate solutions @xmath313 and @xmath314 with errors smaller than 10% , see rightmost panels in figure [ f : s ] . + in principle , integrating the mean shear @xmath315 and the mean temperature gradient @xmath316 , one can find the mean velocity and temperature profiles . unfortunately , to do so we need to know @xmath317 and @xmath316 as functions of the elevation @xmath3 , while in our approach they are found as functions of @xmath62 . remember , that the external parameter @xmath188 is the outer scale of turbulence that depends on the elevation @xmath3 . the importance of an accounting for the proper physically motivated dependance of @xmath188 on @xmath3 for an example of channels and pipes has been recently shown by lvov et al . ( 2008 ) . for the problem at hands , we can safely take @xmath318 if @xmath319 , however when @xmath320 the function @xmath41 is not found theoretically although it was discussed phenomenologically with support of observational , experimental and numerical data . it is traditionally believed that for @xmath321 the scale @xmath188 saturates at some level of order @xmath42 [ see , e.g. eq . ( [ sless ] ) ] . the resulting plots of @xmath322 are shown on figure [ f : means ] , left panel . even taking @xmath323 one gets a very similar velocity profile , see figure [ f : means ] , right panel . with @xmath323 we found an analytical expression for the mean - velocity profile using the interpolation eq . ( [ interb ] ) for @xmath315 : @xmath324+\frac{z } { l_1 } \ . \ ] ] here @xmath325 is the roughness length . the resulting mean velocity profiles have logarithmic asymptotic for @xmath326 and a linear behavior for @xmath320 in agreement with meteorological observations . usually the observations are parameterized by a so - called log - linear approximation ( monin and obukhov , 1954 ) : [ ll - apr ] @xmath327 which is plotted in figure [ f : means ] by dotted lines . one sees some deviation in the region of intermediate @xmath3 . the reason is that the real profile [ see , e.g. eq . ( [ vel ] ) ] has a logarithmic term that saturates for @xmath45 , while in the approximation ( [ ll - aprv ] ) this term continues to grow . to fix this one can use eq . ( [ vel ] ) ( with @xmath328 for simplicity ) , or even its simplified version @xmath329 this approximation is plotted as a dashed line on figure [ f : means ] for comparison . one sees that the approximation ( [ ll - aprv2 ] ) works much better than the traditional one . thus we suggest eq . ( [ ll - aprv2 ] ) for parameterizing meteorological observations . the temperature profiles in our approach look similar to the velocity ones : they have logarithmic asymptotic for @xmath330 and linear behavior for @xmath331 . correspondingly , they can be fitted by a log - linear approximation , like ( [ ll - aprv ] ) , or even better , by an improved version of it , like eq . ( [ ll - aprv2 ] ) . clearly , the values of constants will be different : @xmath332 , @xmath333 , etc . the computed profiles of the turbulent kinetic and temperature energies , horizontal thermal flux profile and the anisotropy profiles are shown on figure [ f : s ] in the middle and right panels . the anisotropy profiles , lower middle panel , saturate at @xmath334 , therefore they are not sensitive to the @xmath3-dependence of @xmath41 ; even quantitatively one can think of these profiles as if they were plotted as a function of @xmath335 . another issue is the profiles of @xmath336 ( upper middle panel ) and of @xmath337 and @xmath314 ( rightmost panels ) , that are @xmath338 for @xmath339 ( if realizable ) . with the interpolation formula ( [ sless ] ) the profiles of the second order correlations have to saturate at levels corresponding to @xmath340 . this sensitivity to the @xmath3-dependence of @xmath41 makes a comparison of the prediction with experimental data very desirable . in our notations the turbulent viscosity and thermal conductivity , turbulent prandtl number , the gradient- and flux - richardson numbers are [ num ] @xmath341 with eqs . ( [ nu ] ) and ( [ chi ] ) we introduce also two dimensionless functions @xmath342 and @xmath343 that are taken as @xmath344-independent constants in the down - gradient transport approximation ( [ dt ] ) described in the introduction . we will show , however , that these functions have a strong dependence on @xmath344 , going to zero in the limit @xmath345 as @xmath346 . therefore this approximation is not valid for large @xmath344 even qualitatively . + as we mentioned in the introduction , the concept of the down - gradient transport assumes that the momentum and thermal fluxes are proportional to the mean velocity and temperature gradients , see eqs . ( [ dt ] ) : @xmath347 where @xmath58 and @xmath59 are effective turbulent viscosity and thermal conductivity , that can be estimated by dimensional reasoning . equations ( [ dt ] ) , giving this estimates , include additional physical arguments that vertical transport parameters should be estimated via vertical turbulent velocity , @xmath23 , and characteristic vertical scale of turbulence , @xmath24 . the relations between the scales @xmath348 in different @xmath349-directions in anisotropic turbulence can be found in the approximation of time - isotropy , according to which @xmath350 here @xmath53 is a characteristic isotropic frequency of turbulence , that for concreteness can be taken as the kinetic energy relaxation frequency @xmath351 . the approximation ( [ ti ] ) is supported by experimental data , according to which in anisotropic turbulence the ratios @xmath352 ( @xmath353 ) are larger then the ratios @xmath354 that are close to unity . with this approximations @xmath58 and @xmath59 can be estimated as follows:@xmath355 where , according to the approximation of down - gradient transport , the dimensionless parameters @xmath25 and @xmath26 are taken as constants , independent of the level of stratification . in order to check how the approximation ( [ appr ] ) , ( [ est1 ] ) works in the stratified tbl for both fluxes , one can consider eqs . ( [ appr ] ) as _ definitions _ of @xmath58 and @xmath59 and eqs . ( [ est1 ] ) as _ definitions _ of @xmath25 and @xmath25 . this gives [ defc ] @xmath356 recall , that in this paper the down - gradient approximation is not used at all . instead , we are using exact balance equations for all relevant second order correlations , including @xmath35 and @xmath37 . substituting our results in the rhs of the definitions ( [ defc ] ) we can find , how @xmath25 and @xmath26 depend on @xmath357 that determines the level of stratification in our approach . the resulting plots of the ratios @xmath358 and @xmath359 are shown in the leftmost panel in figure [ f : res ] . one sees that the @xmath360 and @xmath361 can be considered approximately as constants only for @xmath362 . for larger @xmath363 both @xmath360 and @xmath361 rapidly decrease , more or less in the same manner , diminishing by an order of magnitude already for @xmath364 . for larger @xmath62 one can use the asymptotic solution according to which @xmath365 this means that both functions vanish as @xmath366 : @xmath367 where numerical prefactors account for the accepted values of the dimensionless fit parameters . the physical reason for the strong dependence of @xmath368 and @xmath26 on stratification is as follows : in the rhs of eq . ( [ txz ] ) for the momentum flux and eq . ( [ sim - f - z ] ) for the vertical heat flux there are two terms . the first ones , proportional to @xmath369 and velocity ( or temperature ) gradients correspond to the approximation ( [ appr ] ) , giving ( in our notations ) @xmath370const and @xmath371const , in agreement with the down - gradient transport concept . however , there are second contributions to the vertical momentum flux @xmath372 and to the vertical heat flux , that is proportional to @xmath373 . in our approach both contributions are negative , giving rise to the _ counter - gradient fluxes_. what follows from our approach , is that these counter - gradient fluxes cancel ( to the leading order ) the down - gradient contributions in the limit @xmath345 . as a result , in this limit the effective turbulent diffusion and thermal conductivity vanish , making the down - gradient approximation for them ( with constant @xmath25 and @xmath26 ) irrelevant even qualitatively for @xmath374 . in our picture of stable temperature - stratified tbl , the turbulence exists at any elevations , where one can neglect the coriolis force . moreover , the turbulent kinetic and temperature energies increase as @xmath375 for @xmath376 , see figure [ f : s ] . at the same time , the mean velocity and potential temperature change the @xmath377-dependence from logarithmic lo linear , see figure [ f : means ] and ( modified ) log - linear interpolation formula ( [ ll - aprv2 ] ) . correspondingly , the shear of the mean velocity and _ the mean temperature gradient _ saturate at some elevation ( and at some @xmath62 ) , and @xmath13 _ saturates as well_. this predictions agree with large eddy simulation by zilitinkevich and esau ( 2006 ) , where @xmath13 can be considered as saturating around 0.4 for @xmath378 . notice that the turbulent closures of kind used above can not be applied for strongly stratified flows with @xmath50 ( may be even at @xmath51 ) . there are two reasons for that . the first one was mentioned in the introduction . namely , for @xmath50 the brunt - visl frequency @xmath379 , @xmath380 , is larger then the eddy - turnover frequency and therefore there are weakly decaying kelvin - helmoholtz internal gravity waves which , generally speaking , have to be accounted for in the momentum and energy balance equations . the second reason , that makes the results very sensitive to the contribution of internal waves follows from the fact that vortical turbulent fluxes vanish ( at fixed velocity and temperature gradients ) . therefore even relatively small contributions of different nature to the momentum and thermal fluxes may be important . the final conclusion is that the tbl modeling at large level of stratification requires an accounting for turbulence of the internal waves together with the vortical turbulence . definitely , new observations , laboratory and numerical experiments with control of internal wave activity are very likely . vl kindly acknowledge the possibility to give an invited lecture on the problems , discussed in this paper , at the international conference `` turbulent mixing and beyond '' , which was held august 2007 at the abdus salam international center for theoretical physics , trieste , italy . vl and or also acknowledge the support of the transnational access programme at risc - linz , funded by the european commission framework 6 programme for integrated infrastructures initiatives under the project science ( contract no . let us look more carefully at the approximation ( [ freqs ] ) , which is @xmath381the dimensional reasoning that leads to this approximation is questionable for problems having a dimensionless parameter @xmath344 . generally speaking , all constants " @xmath382 and @xmath383 in eq . ( [ app - appr ] ) can be any functions of @xmath344 . presently we just hope that a possible @xmath344 dependence of these functions is relatively weak and does not affect the qualitative picture of the phenomenon . moreover , even the assumption ( [ distau1a ] ) that the dissipation of the thermal flux @xmath384 is proportional to the thermal flux and the assumption ( [ distau1b ] ) that the dissipation of @xmath113 , @xmath385 are also questionable . formally speaking , one can not guarantee that the triple cross - correlator @xmath386 that estimates @xmath387 , can be ( roughly speaking ) decomposed like @xmath388 , i.e really proportional to @xmath389 as it stated in eq . ( [ distau1a ] ) . theoretically , one can not exclude the decomposition @xmath390 , i.e. a contribution to @xmath391 . similarly , the dissipation @xmath392 in the balance ( [ corrc ] ) of @xmath113 , that is determined by the correlator ( [ distau1b ] ) , is @xmath393 , as it follows from the decomposition @xmath394 and is stated in eq . ( [ distau1b ] ) . this correlator allows , for example , the decomposition @xmath395 , i.e. contribution to @xmath396 . this discussion demonstrates , that the situation with the dissipation rates is not so simple , as one may think and thus requires careful theoretical analysis that is in our agenda for future work . our preliminary analysis of this problem shows that all fitting constants are indeed functions of @xmath344 . fortunately , they vary within finite limits in the entire interval @xmath397 . therefore we propose that the approximations used in this paper preserve the qualitative picture of the phenomenon . once again , the traditional down - gradient approximation does not work even qualitatively because corresponding constants " @xmath25 and @xmath26 vanish in the limit @xmath345 . hunt , j. c. r. , stretch , d. d. , and britter , r. e. , 1988 : length scales in stably stratified turbulent flows and their use in turbulence models . in : _ i.m.a . conference on `` stably stratified flow and dense gas dispersion '' _ ( j. s. puttock , ed . ) , clarendon press , 285 - 322 . luyten , p. j. , carniel , s. , and umgiesser , g. , 2002 : validation of turbulence closure parameterisations for stably stratified flows using the provess turbulence measurements in the north sea , _ j. sea research _ , * 47 * , 239 - 267 . stretch , d. d. , rottman , j. w. , nomura , k. k. , and venayagamoorthy , s. k. , 2001 : transient mixing events in stably stratified turbulence , in : _ 14th australasian fluid mechanics conference _ , adelaide , australia , 10 - 14 december 2001 . zilitinkevich s.s . , 2002 : third - order transport due to internal waves and non - local turbulence in the stably stratified surface layer , _ quarterly journal of the royal meteorological society _ , * 128 * , 913 - 925 . zilitinkevich , s.s . , elperin , t. , kleeorin , n. , and rogachevskii , i. , 2007 : energy- and flux - budget ( efb ) turbulence closure model for stably stratified flows . part i : steady - state , homogeneous regimes , _ boundary - layer meteorology _ * 125 * , 167 - 191 . zilitinkevich , s.s . , and essau , i. : similarity theory and calculation of turbulent fluxes at the surface for the stably stratified atmospheric boundary layer , _ boundary - layer meteorology _ * 125 * , 193 - 205 ( 2007 ) .

we have observed a sample of 10 white dwarf candidates in the rich open cluster ngc 2323 ( m50 ) with the keck low - resolution imaging spectrometer . the spectroscopy shows eight to be da white dwarfs , with six of these having high s / n appropriate for our analysis . two of these white dwarfs are consistent with singly evolved cluster membership , and both are high mass.07 m , and give equivalent progenitor masses of 4.69 m . to supplement these new high - mass white dwarfs and analyze the initial - final mass relation ( ifmr ) , we have also looked at 30 white dwarfs from publicly available data that are mostly all high - mass (.9 m ) . these original published data exhibited significant scatter , and to test if this scatter is true or simply the result of systematics , we have uniformly analyzed the white dwarf spectra and have adopted thorough photometric techniques to derive uniform cluster parameters for their parent clusters . the resulting ifmr scatter is significantly reduced , arguing that mass - loss rates are not stochastic in nature and that within the ranges of metallicity and mass analyzed in this work mass loss is not highly sensitive to variations in metallicity . lastly , when adopting cluster ages based on y isochrones , the slope of the high - mass ifmr remains steep and consistent with that found from intermediate - mass white dwarfs , giving a linear ifmr from progenitor masses between 3 to 6.5 m . in contrast , when adopting the slightly younger cluster ages based on parsec isochrones , the high - mass ifmr has a moderate turnover near an initial mass of 4 m .

1601.03053

performing stellar archeology on white dwarfs , by far the most common stellar remnant , provides valuable information for not only understanding stellar evolution and mass loss but also galactic evolution . one of the fundamental relations in the analysis of white dwarfs is the initial - final mass relation ( hereafter ifmr ) , where the masses of white dwarfs are compared directly to the zero - age main sequence mass of their progenitors . this semi - empirical relation is critical to our understanding of integrated mass - loss over the lifetime of a star and how it changes with stellar mass . the ifmr has a variety of additional applications including predicting type ia supernovae rates ( pritchet et al . 2008 ; greggio 2010 ) and overall stellar feedback in galaxy models ( agertz & kravtsov 2014 ) , interpreting the white dwarf luminosity function ( cataln et al . 2008 ) , and providing a technique for measuring the age of the galactic halo ( kalirai 2013 ) . analysis of the ifmr began with weidemann ( 1977 ) , where it was shown that the models of the time greatly underestimated the observed stellar mass loss . subsequent work on the ifmr by a number of groups ( see weidemann 2000 for review ) resulted in a broad but sparsely populated relation that showed a clear trend with higher - mass main sequence stars producing increasingly more massive white dwarfs . in the past 15 years the amount of ifmr data has greatly increased ( e.g. , claver et al . 2001 ; dobbie et al . 2004 , 2006a ; williams et al . 2004 ; kalirai et al . 2005 ; liebert et al . 2005 ; williams & bolte 2007 ; kalirai et al . 2007 ; kalirai et al . 2008 ; rubin et al . 2008 ; kalirai et al . 2009 ; williams et al . 2009 ; dobbie & baxter 2010 ; dobbie et al . 2012 ) . these newer data retain the general trend of the previous ifmr work , but the scatter in the data remains significant . the source of this scatter may be attributable to several factors including the possible stochastic nature of mass loss , effects from variation in metallicity or environment , or systematic differences between the studies . one important systematic is the challenge in defining the ages of the clusters these white dwarfs belong to , which creates uncertainty in the derived lifetimes of their progenitor stars . cummings et al . ( 2015 ; hereafter paper i ) began to analyze the important intermediate - mass ifmr ( progenitor masses of 3 - 4 m@xmath1 ) from the rich ngc 2099 ( m37 ) . this work strengthened the observational evidence that in this mass range the ifmr is steep , where the final white dwarf mass increases more rapidly with increasing progenitor mass . comparison to the rich population of comparable mass white dwarfs in both the hyades and praesepe from kalirai et al . ( 2014 ) showed strongly consistent ifmrs . this consistency also suggests that across this mass range the slightly metal - rich progenitor stars from the hyades and praesepe ( [ fe / h]@xmath00.15 ) have no significant increase in mass - loss rates compared to those in the solar metallicity ngc 2099 . expanding beyond paper i , we now look at the challenging higher - mass region ( m@xmath4 of 46.5 m@xmath1 ) of the ifmr by focusing on white dwarfs in younger clusters . while younger clusters do not provide a broad mass range of white dwarfs , they provide several important advantages . first , the highest - mass white dwarfs are the most compact and lowest luminosity , and because they form first they remain bright only in the youngest clusters ( @xmath5 200 myr ) . second , the cooling rates are far more rapid in young and hot white dwarfs and , as a result , errors in temperature lead to far smaller errors in both cooling age and luminosity . third , high - mass white dwarfs ( @xmath20.9 m@xmath1 ) may be prone to be ejected from their parent population clusters , either due to dynamical interactions or potential velocity kicks due to asymmetric mass loss during their formation ( fellhauer et al . 2003 ; tremblay et al . therefore , the probability of finding high - mass white dwarfs still within their cluster population may decrease with age . these three reasons are why younger clusters provide far more advantages in analyzing high - mass white dwarfs with the best precision . in this paper we begin our analysis with the rich , young , and nearby cluster ngc 2323 ( m50 ) . based on population analysis , kalirai et al . ( 2003 ) find that it is approximately three times as rich as the pleiades , making it an excellent environment to search for the rare remnants of higher - mass stars . additionally , to expand our sample we self - consistently reanalyze publicly available data on all published high - mass white dwarfs from clusters ( pleiades [ m45 ] , ngc 2516 , ngc 2287 , ngc 3532 , and ngc 2168 [ m35 ] ) and sirius b. to further limit systematics , we also self - consistently analyze the cluster parameters for all of the parent cluster populations based on published high - quality ubv photometry . to look at the broader picture , we connect these high - mass data to the moderate - mass data from paper i and further analyze the broader characteristics of the ifmr . the structure of this paper is as follows , in section 2 we discuss the spectroscopic white dwarf observations of ngc 2323 , the publicly available data we have used , and our reduction and analysis techniques . in section 3 we discuss the ubv photometry based cluster parameters for our six open clusters being analyzed . in section 4 we discuss the cluster membership of our white dwarf candidates in ngc 2323 . in section 5 we discuss the high - mass ifmr and compare to the intermediate - mass ( 34 m@xmath1 progenitors ) ifmr from paper i. in section 6 we summarize our study . based on the deep bv photometric observations of ngc 2323 ( kalirai et al . 2003 ) with the canada - france - hawaii telescope and the cfh12k mosaic camera , a sample of white dwarf candidates in ngc 2323 were spectroscopically observed at keck i using the low resolution imaging spectrometer ( lris ; oke et al . 1995 ) . in total , 10 of these candidates had sufficient signal to properly analyze their characteristics . the 400/3400 grism was used with 1 " slits giving a spectral resolution of @xmath06 , which provides us the wavelength coverage of @xmath03000 to 5750 and a series of 5 balmer lines ( h@xmath6 , h@xmath7 , h@xmath8 , h@xmath9 , and h8 ) . these observations were performed on 2008 december 23 and 24 , on 2011 december 27 and 28 , and on 2015 february 19 . for spectral flux calibration , flux standards were observed each night . three independent lris masks were used in 2008 to observe white dwarfs candidates from ngc 2323 . mask 1 was observed for 40 minutes , mask 2 was observed for 40 minutes , and mask 3 was observed for 2 hours and 40 minutes . in 2011 , individual longslit observations were performed for five different white dwarf candidates in ngc 2323 ranging from 20 minutes to 1 hour and 50 minutes . lastly , in 2015 an additional 70 minutes was acquired on wd11 , which is a white dwarf candidate of interest that only had 20 minutes of observation in 2011 . we have reduced and flux calibrated our lris observations using the idl based xidl pipeline . of our total observed sample of 10 white dwarfs candidates , eight are da white dwarfs and wd23 and wd38 have no clear spectral features . to provide additional high - mass white dwarfs for comparison we have taken from the vlt archive the observations of seven white dwarf members of ngc 3532 , three white dwarf members of ngc 2287 , and four white dwarf members of ngc 2516 ( based on observations made with eso telescopes under program ids : 079.d-0490(a ) ; 080.d-0654(a ) ; 084.d-1097(a ) ; pi : dobbie ) . these observations were performed with fors1 and fors2 using the 600b grism ( appenzeller et al . 1998 ) giving comparable spectral resolution to our lris observations of @xmath06 . analysis of the parameters and membership for the white dwarfs of ngc 2287 , ngc 2516 , and ngc 3532 were originally published in dobbie et al . ( 2009 ; 2012 ) . additionally , from the keck archive we have taken the observations of 11 white dwarf members of ngc 6128 ( program i d : u49l-2002b ; u60l-2004a ; u15l-2004b ; u18l-2005b ; pi : bolte ) . similar to our ngc 2323 observations , these observations were performed using lris with a majority of them using the 400/3400 grism and 1 " slits , giving the same characteristics to our data . analysis of the parameters and membership for the white dwarfs of ngc 6128 were original published in williams et al . ( 2009 ) . we have performed our own reductions and analyses of these data from both the vlt and keck , but we do not redetermine their membership status in this paper . the vlt data were reduced using the standard iraf techniques for reduction of longslit data , while the keck data were reduced using the same xidl pipeline used to analyze our ngc 2323 data . as a test for our vlt data reduction , we were provided with the published spectrum of white dwarf j0646 - 203 ( ngc 2287 - 4 ) ( p.d . dobbie ; private communication 2014 ) , and we found that there are no meaningful differences in our reduced spectra of the same data . hence , there are no systematics caused by our spectral reduction techniques , and the systematic differences between our parameters and those presented in dobbie et al . ( 2012 ) are due to differences in our applied white dwarf models and fitting techniques . l c c c c c c c c c c * - ngc 2323 white dwarf initial and final parameters * + id&t@xmath10&log g&m@xmath11 & m@xmath12&t@xmath13 & y@xmath3 m@xmath14 & parsec m@xmath14 & m@xmath15 & m@xmath16&s / n + & ( k ) & & ( m@xmath1 ) & & ( myr ) & ( m@xmath1)&(m@xmath1)&(m@xmath1)&(m@xmath1 ) & + + ngc 2323-wd10 & 52800@xmath171350 & 8.68@xmath170.09 & 1.068@xmath170.045 & 10.36@xmath170.19 & 1.6@xmath18 & 4.69@xmath19 & 5.07@xmath20 & 4.98 & 4.45 & 85 + ngc 2323-wd11 & 54100@xmath171000 & 8.69@xmath170.07 & 1.075@xmath170.032 & 10.36@xmath170.13 & 1.3@xmath21 & 4.69@xmath19 & 5.07@xmath22 & 4.98 & 4.45 & 130 + + ngc 2323-wd21 & 18200@xmath17850 & 8.26@xmath170.15 & 0.779@xmath170.096 & 11.33@xmath170.25 & 170@xmath23 & & & & & 28 + ngc 2323-wd7 & 16800@xmath17250 & 7.90@xmath170.05 & 0.559@xmath170.024 & 10.92@xmath170.07 & 112@xmath24 & & & & & 122 + ngc 2323-wd17 & 19800@xmath17300 & 8.12@xmath170.05 & 0.694@xmath170.028 & 10.97@xmath170.07 & 96@xmath24 & & & & & 111 + ngc 2323-wd12 & 17100@xmath17400 & 7.88@xmath170.07 & 0.550@xmath170.037 & 10.87@xmath170.11 & 101@xmath25 & & & & & 60 + + ngc 2323-wd22 & 24400@xmath171550 & 8.08@xmath170.22 & 0.681@xmath170.128 & 10.53@xmath170.36 & 33@xmath26 & & & & & 13 + ngc 2323-wd30 & 13400@xmath17900 & 8.04@xmath170.18 & 0.629@xmath170.106 & 11.52@xmath170.28 & 290@xmath27 & & & & & 23 + for the high - mass lb 1497 , from the pleiades , and sirius b we have taken the t@xmath10 and log g parameters from gianninas et al . we also have taken these parameters from gianninas et al . ( 2011 ) for the supermassive gd50 and pg 0136 + 251 white dwarfs , where dobbie et al . ( 2006b ) used three - dimensional space velocities to argue that gd50 s progenitor was related to and coeval with the pleiades cluster . similarly , but based on only proper motions , dobbie et al . ( 2006b ) argued that pg 0136 + 251 s progenitor is likely consistent with coeval formation with the pleiades . gianninas et al . ( 2011 ) use white dwarf atmospheric models and fitting techniques equivalent to ours , and we adopt their t@xmath10 and log g in our analysis of these four white dwarfs . for our spectroscopic analysis we adopted the same analysis techniques as those described in paper i. in brief , we used the recent white dwarf spectroscopic models of tremblay et al . ( 2011 ) with the stark profiles of tremblay & bergeron ( 2009 ) , and the automated fitting techniques described by bergeron et al . ( 1992 ) to fit our balmer line spectra and derive t@xmath10 and log g. however , for our derived parameters ( mass , luminosity , and cooling age ) we expand upon the methods of paper i because our current sample has a far broader mass range . for deriving the parameters for white dwarfs of mass less than 1.10 m@xmath1 we applied our t@xmath10 and log g to the cooling models for a carbon / oxygen ( co ) core composition with a thick hydrogen layer by fontaine et al . for the highest - mass white dwarfs ( @xmath28 1.1 m@xmath1 ) we derived the parameters based on the oxygen / neon ( one ) core models of althaus et al . ( 2005 ; 2007 ) . of the cooling rates for the several white dwarf models that we have discussed , in addition to the basti co cooling models from salaris et al . ( 2010 ) . in the lower panel , the hottest temperatures are shown and the systematics are always @xmath01.5 myr or less in magnitude , which is not concerning . at cooler temperatures , all models other than the carbon core models of wood ( 1995 ) are consistent . this is consistent with the systematic effects of differing core composition , which grow as white dwarfs cool further . ] a discussion of our adopted models is warranted . first , we do not adopt the more recent white dwarf atmospheric models of tremblay et al . ( 2013 ) because those focus only on 3d modeling of convective atmospheres ( @xmath2914,000 k ) , and our current analysis looks hot fully radiative white dwarfs . for our cooling models , while the fontaine et al . ( 2001 ) models are widely used , we should acknowledge two limitations they have . first , they assume a 50/50 carbon and oxygen core composition , which based on full stellar evolutionary models is not accurate ( e.g. , romero et al . the effect of this on the calculated cooling ages can be important , but for the relatively young white dwarfs we are analyzing this effect remains small . second , these cooling models do not begin at the tip of the asymptotic giant branch ( hereafter agb ) and instead begin at a t@xmath10 of @xmath060,000 k. in comparison , the widely used wood ( 1995 ) co cooling models do begin at the tip of the agb and they do not adopt a simple 50/50 carbon and oxygen core composition . unfortunately , their co core models have an upper mass limit of 1.0 m@xmath1 , which limits their application in our analysis but they provide an important test for systematics . in figure 1 we compare the cooling ages from the fontaine et al . ( 2001 ) co core models at 1.0 m@xmath1 to both the co and pure - carbon core models of wood ( 1995 ) and the basti co models from salaris et al . the basti co cooling models adopted c / o ratio profiles based on the basti scaled solar stellar evolution models . the lower panel of figure 1 shows that at high t@xmath30@xmath2845,000 k all of the differences between these models never result in cooling age differences of more than 1.5 myr . the upper panel of figure 1 similarly shows that at cooler temperatures 30,000@xmath5t@xmath10@xmath540,000 k the co models , irrespective of their adopted c / o ratios or starting points , all give strong agreement in cooling ages , but the systematic effects introduced from pure carbon core models ( @xmath010 myr ) are now clearly seen . for ultramassive white dwarfs , the mass at which they transition to one white dwarfs remains uncertain , and it also likely depends on metallicity ( doherty et al . here we have adopted a somewhat conservative 1.10 m@xmath1 , but we note that the models of garcia - berro et al . ( 1997 ) argue that it may be as low as 1.05 m@xmath1 . reassuringly , in this mass range the co and one cooling ages at constant mass are consistent for such young white dwarfs , but for these white dwarfs the dependence of the mass - radius relationship on core - composition is very important . for example , applying the gravities of white dwarfs in the mass range of 1.05 to 1.10 m@xmath1 ( based on co - core models ) to the one models derives masses @xmath00.05 m@xmath1 lower and places them all below 1.05 m@xmath1 . therefore , it is appropriate to adopt 1.10 m@xmath1 as the transition mass . in table 1 we present our white dwarf parameters for the eight da white dwarfs from ngc 2323 . for clarity in table 1 we have distinguished between members and nonmembers , where membership is based on our comparisons of model based and observed photometry in addition to comparisons of cooling ages to the cluster age ( see our detailed discussion of membership in section 4 ) . additionally , we have considered s / n ( per resolution element ) and the resulting errors given in table 1 . the wd22 and wd30 spectra have low s / n and mass errors greater than 0.1 m@xmath1 . therefore , their parameters are presented for reference but have been cut from our final analysis . this is because their membership determinations are unreliable , but their low masses do suggest they are field white dwarfs . with ifmr analysis , the parameters of the star clusters are as critical as the white dwarfs themselves . this is particularly true for the highest - mass white dwarfs , where the derived masses of their progenitors change rapidly with evolutionary time . uniform photometric data sets are not available for these six clusters . but in paper i we showed that for ngc 2099 , the adopted isochrones and fitting techniques had as large , if not larger , of an effect on its cluster parameters as did any systematics between the cluster s different photometry sets . this is best illustrated by the systematic effects on derived cluster ages , where the yi et al . ( 2001 ; hereafter y@xmath3 ) and the ventura et al . ( 1998 ) isochrones both gave our final adopted age of 520 myr for ngc 2099 . the parsec version 1.2s isochrones ( bressan et al . 2012 ) derived a comparable age of 540 myr , and lastly the girardi et al . ( 2000 ) and bertelli et al . ( 1994 ) isochrones both gave significantly younger ages of 445 myr . between these isochrone sets there is nearly a 100 myr range of derived ages when using identical photometry , and further systematics can be introduced based on how the isochrones are fit to the data . in this paper we have redetermined as uniformly as possible the reddenings , distance moduli , and ages for these six clusters using available high - quality ubv photometry that covers up to the full turnoff . the y@xmath3 isochrones provide our final adopted cluster ages , but to broaden our results we also determine ages with the parsec isochrones . these two isochrones give only slightly different cluster ages , but in these younger clusters the masses of the progenitor stars have a far more significant dependence on evolutionary time , and even a 20 myr systematic has a significant effect on our results . in our cluster photometric analysis we first make use of color - color diagrams ( b - v vs u - b ) , which provide direct photometric information on the cluster reddening . the photometric metallicity can also be derived but it is quite sensitive to systematics in u magnitude , a concern considering the typically more complex standardization process for u magnitudes and the varying sources of our photometry therefore , in the case of the pleiades , ngc 2168 , and ngc 2516 we consider more detailed spectroscopic metallicities , but for ngc 2287 , ngc 2323 , and ngc 3532 we will simply adopt solar metallicity . however , we note that these adopted metallicities do show strong consistency with the observed photometry . our color - color analysis adopts two techniques , the first is semi - empirical and based on the hyades fiducial and the second is based directly on the y@xmath3 isochrones , which reach to higher masses than those available in the hyades fiducial . for both methods the reddening relation adopted is that of cardelli et al . ( 1989 ) and the metallicity correction is based on that of the y@xmath3 isochrones . the methods using the hyades fiducial have been developed in deliyannis et al . ( in prep ) , where the fiducial was derived from single - star cluster members ( see perryman et al . 1998 ) . the hyades ubv photometry of johnson & knuckles ( 1955 ) was adopted with a cluster [ fe / h ] of + 0.15 and e(b - v ) of 0 . the pleiades provides a good example for our color - color analysis techniques . in the left panel of figure 2 we have plotted the photoelectric ubv photometry from johnson & mitchell ( 1958 ) with several reddening curves based on the hyades fiducial , and we have applied a metallicity of [ fe / h]=0.01 ( z=0.0185 ) to match the reddening insensitive region where all of the reddening curves intersect near b - v of 0.6 . fitting by eye the blueward color - color trend we find that a reddening curve of e(b - v)=0.03@xmath170.02 matches the pleiades ubv photometry the best . both this reddening and metallicity are consistent with the typically derived values and spectroscopic analyses of the pleiades ( e.g. , [ fe / h]=0.01@xmath170.02 schuler et al . 2010 ; [ fe / h]=0.03@xmath170.02@xmath170.05 soderblom et al . 2009 ) we also note that the hyades fiducial ends at b - v@xmath00.1 , where the older hyades turnoff occurs . in the right panel of figure 2 we fit the higher - mass stars bluer than b - v=0.0 with three different 135 myr y@xmath3 isochrones of differing metallicity . all three metallicities fit a reddening of e(b - v)=0.03 in these bluest stars . this demonstrates that these higher - mass stars create a nearly linear trend that is insensitive to variations in metallicity . we note that the position of this blue linear trend is also insensitive to cluster age , where as we look at older clusters the trend only shortens in length and does not shift its position . therefore , this linear trend s position provides a reliable reddening measurement independent of all other cluster parameters . in the upper left panel of figure 3 we compare directly our hyades fiducial fit and our y@xmath3 fit for the pleiades . there are systematic differences that are noted in the b - v range of 0.1 - 0.3 , where it appears that the y@xmath3 isochrones are too blue in u - b relative to both the data and the hyades fiducial . however , when adopting a y@xmath3 isochrone with an age consistent to the hyades ( 650 myr ; not shown ) the isochrone is nearly identical to that of the hyades fiducial , so at these younger ages the isochrones appear to overestimate the u flux at this intermediate color range . at all other colors this does not seem to be a concern because , reassuringly , both the hyades fiducial and the y@xmath3 isochrones find the same reddening in their respective regions , and they also agree in the region sensitive to metallicity ( b - v@xmath00.6 ) and redder . in regard to the parsec isochrones , they are not independently considered in our color - color analysis because their u - b colors do not fit the observations . a key advantage to color - color analysis is that it can be used to clean young cluster turnoffs . dating younger clusters is prone to several difficulties , including that turnoffs are relatively sparse even in the richest clusters and that these higher - mass turnoff stars have high binarity fraction ( kouwenhoven et al . 2007 ) . additionally , many higher - mass stars fall into the peculiar groups of blue stragglers or be stars . for example , mermilliod 1982a , 1982b , and ahumada & lapasset ( 2007 ) found that several of the brightest stars in the clusters being analyzed in this paper are blue stragglers that are far too blue or too bright , if not both , for their age . lu et al . ( 2011 ) modeled the formation of blue stragglers on the short time scales necessary in these young clusters and found that they can rapidly be created through binary mass transfer . identifying these peculiar stars can greatly improve the fit of the turnoff ages of these younger clusters . mermilliod 1982b do find that when plotted in color - color space be stars deviate from the approximately linear trend of the normal " high - mass stars . additionally , several of our clusters may have variable reddening , and fitting this high - mass linear trend identifies the richest group of cluster stars with consistent reddening . in the pleiades shown in the upper - left panel of figure 3 , we mark in red the high - mass stars consistent with the trend , and several stars deviate from this trend that are likely peculiar . however , we do acknowledge that the brightest star in the pleiades , alcyone , does not deviate from this trend . alcyone is a multiple system and has several peculiar characteristics like spectral emission and rapid rotation ( hoffleit & jaschek 1991 ) and is commonly referred to as both a blue straggler and be dwarf . this suggests that while this color - color method does identify many problematic stars , it does not remove all peculiar stars . in the upper right panel of figure 3 we have similarly plotted the ngc 2323 ubv photometry from claria et al . we fit the bluest stars and see that independent of an assumed metallicity , we find a large reddening of e(b - v)=0.23@xmath170.06 at ( b - v)@xmath31=0 , where we deem this reddening large enough to account for the color dependence of reddening ( see fernie 1963 and our discussion in paper i ) . when we have adopted a color dependent reddening we will define the reddening at ( b - v)@xmath31=0 . again , our selected final turnoff stars are shown in red . l c c c c c c * - open cluster parameters * + cluster & e(b - v)@xmath32 & ( m - m)@xmath31 & [ fe / h ] & y@xmath3 age ( myr ) & parsec age ( myr ) & photometric sources + pleiades & 0.03@xmath170.02 & 5.67@xmath170.10 & + 0.01 & 135@xmath1715 & 110@xmath1715 & 1 + ngc 2323 & 0.23@xmath170.06 & 10.0@xmath170.15 & 0.00 & 140@xmath1720 & 115@xmath1720 & 2,3 + ngc 2516 & 0.10@xmath170.03 & 8.20@xmath170.12 & + 0.065 & 170@xmath1720 & 150@xmath1720 & 4,5 + ngc 2168 & 0.25@xmath170.04 & 9.66@xmath170.10 & -0.143 & 190@xmath1720 & 170@xmath1720 & 6 + ngc 2287 & 0.035@xmath170.025 & 9.52@xmath170.12 & 0.00 & 220@xmath1730 & 205@xmath1730 & 7,8 + ngc 3532 & 0.04@xmath170.025 & 8.46@xmath170.14 & 0.00 & 320@xmath1720 & 325@xmath1720 & 9 + l c c c c c c c c * - membership data * + id&@xmath33&@xmath8&m@xmath12 & v & ( b - v)@xmath31 & b - v & t@xmath13 + & ( j2000 ) & ( j2000 ) & ( model)&(obs.)&(model)&(obs.)&(myr ) + + ngc 2323-wd10 & 7:02:41.02 & -8:26:12.8 & 10.36@xmath170.19 & 20.62@xmath170.009 & -0.303@xmath170.002 & -0.020@xmath170.060 & 2.9@xmath34 + ngc 2323-wd11 & 7:03:22.14 & -8:15:58.7 & 10.36@xmath170.13 & 20.67@xmath170.008 & -0.305@xmath170.002 & 0.050@xmath170.025 & 2.6@xmath35 + + ngc 2323-wd21 & 7:02:08.56 & -8:25:48.0 & 11.33@xmath170.25 & 21.91@xmath170.026 & -0.013@xmath170.022 & 0.149@xmath170.158 & 170@xmath23 + ngc 2323-wd7 & 7:02:47.87 & -8:35:56.4 & 10.92@xmath170.07 & 19.50@xmath170.004 & -0.005@xmath170.007 & 0.078@xmath170.004 & 112@xmath24 + ngc 2323-wd17 & 7:03:12.62 & -8:30:53.6 & 10.97@xmath170.07 & 20.96@xmath170.010 & -0.057@xmath170.007 & 0.107@xmath170.015 & 96@xmath24 + ngc 2323-wd12 & 7:03:17.79 & -8:19:59.7 & 10.87@xmath170.11 & 20.52@xmath170.007 & -0.014@xmath170.010 & 0.116@xmath170.079 & 101@xmath25 + + ngc 2323-wd22 & 7:02:33.93 & -8:31:04.3 & 10.53@xmath170.36 & 21.89@xmath170.024 & -0.135@xmath170.025 & 0.180@xmath170.013 & 33@xmath26 + ngc 2323-wd30 & 7:03:22.56 & -8:29:25.7 & 11.52@xmath170.28 & 22.04@xmath170.028 & 0.111@xmath170.026 & 0.485@xmath170.029 & 290@xmath27 + ngc 2323-wd38 & 7:03:31.87 & -8:28:25.0 & & 22.93@xmath170.062 & & 0.645@xmath170.058 & + ngc 2323-wd23 & 7:03:39.85 & -8:28:16.7 & & 21.42@xmath170.016 & & 0.375@xmath170.044 & + our four additional clusters are also shown in figure 3 . in ngc 2516 , we have used two ubv photometric studies , dachs ( 1970 ) for the brightest stars and sung et al . ( 2002 ) for the fainter stars shown in blue . for the moderately large reddening we fit a e(b - v ) of 0.10@xmath170.03 at ( b - v)@xmath31=0 and a [ fe / h]=0.065 , and as with the pleiades this is consistent with the typically adopted parameters and our spectroscopic analysis ( cummings 2011 ) . for ngc 2287 we have used the ubv photometry of ianna et al . ( 1987 ) and fit a e(b - v ) of 0.035@xmath170.025 with an assumed solar metallicity . for ngc 3532 we have used the ubv photometry of fernandez & salgado ( 1980 ) and fit a e(b - v ) of 0.04@xmath170.025 with an assumed solar metallicity . lastly , for ngc 2168 we have used the ubv photometry of sung & bessell ( 1999 ) . we derive a e(b - v ) of 0.25@xmath170.04 at ( b - v)@xmath31=0 for this cluster and adopt a metal - poor [ fe / h ] of -0.143 ( steinhauer & deliyannis 2004 ) , which provides a reliable fit to the bluest stars and follows the u - b range well but it appears that there is a systematic shift in b - v where the stars becoming increasingly too blue at redder colors . otherwise , changes in either adopted metallicity or reddening can not fit the full b - v color range . in figure 4 we display our by eye turnoff age fits with y@xmath3 isochrones in green when adopting from figure 3 the reddenings , metallicities , and the cleaned turnoff stars shown in red . in magenta we similarly display the parsec isochrone fits , where we have adopted the same reddenings , distance moduli , and metallicities ( z ) . in table 2 the derived cluster parameters and the photometric sources are listed . we note that in the youngest clusters the parsec isochrones systematically derive ages 25 myr younger , while in the 320 myr ngc 3532 they derive ages 5 myr older , and lastly from paper i the parsec isochrones derive an age 20 myr older in the 520 myr old ngc 2099 . therefore , these two isochrones not only have changing systematics at differing ages , but they are in opposite directions in young versus older clusters . the possible causes of the systematics between these two isochrones include the differences in their adopted opacities , equations of state , and solar compositions . while the y@xmath3 isochrones do not consider evolution past the tip of the red giant branch ( rgb ) , in these typically young clusters we can not reliably fit the giants because their populations are very sparse or they have no giants at all . therefore , we have chosen the y@xmath3 isochrones for our final parameters because they more successfully fit both the color - color data and the the main sequence features . these cluster ages provide several advantages over adopting literature values . 1 ) the ages are based on a uniform system of isochrones , while literature values adopt wide ranging models that have systematic differences that become more pronounced at younger ages . 2 ) the fitting techniques applied are by eye but consistent , while fitting techniques for literature values can greatly vary . 3 ) the difficulty of peculiar turnoff stars are addressed in a systematic way , while their consideration can have important differences in the literature values , if they are considered at all . we will not comment on the absolute accuracy of the various isochrone model ages , but in this study uniformity and precision is the goal . we must also reiterate two remaining limitations with our cluster parameters . first , all of our photometries are primarily from differing groups , which still may leave important systematics remaining in our parameter analyses . additionally , uniformly measured spectroscopic metallicities are also needed to address the metallicity sensitivity in the turnoff isochrone fits . lastly , for sirius b there is no parent cluster that we can self - consistently analyze for the total age , but the age of the sirius system is well studied and here we adopt solar metallicity and the age of 237.5@xmath1712.5 myr determined in liebert et al . , h@xmath7 , h@xmath8 , h@xmath9 , and h8 fits are shown from bottom to top . ] ( solid line ) , 1.0 , 0.8 , and 0.6 m@xmath1 ( dotted lines ) , and 0.4 m@xmath1 ( dashed line ) . the mean distance modulus and reddening derived from the white dwarf members are applied to these cooling models . ] cluster membership determination is key in analyzing the formation history of these white dwarfs and applying them to the ifmr . in the case of ngc 2323 , we have the advantage that because it is such a young cluster , any hot and young high - mass white dwarfs observed in its field would already have reliable membership . for example , in the sloan digital sky survey sample of field white dwarfs , only @xmath02.6% of da white dwarfs have a mass greater than 1 m@xmath1 and less than 10% of these are young with a t@xmath10@xmath2820,000 k ( e.g. , kleinman et al . 2013 ; kepler et al . 2016 ) . to further confirm cluster membership in ngc 2323 , however , we have also performed magnitude and color analysis . in table 3 we list the observed photometry for these white dwarfs and also list their cooling age and model based m@xmath12 and ( b - v)@xmath31 . for our membership analysis we directly compare the model based and observed magnitudes and colors in figure 5 to derive an effective distance modulus and reddening for each white dwarf . we define our 1@xmath36 color and magnitude errors by adding in quadrature the respective model fitting errors , the observational errors , and in the case of magnitude the distance modulus errors or in the case of color the reddening errors . we select white dwarfs as consistent with membership if their effective distance modulus and reddening are within 2@xmath36 of the cluster parameters derived in section 3 . of our six high signal white dwarfs , only wd10 , wd11 , and wd21 have both magnitudes and colors consistent with membership . however , wd21 has a cooling age longer than the age of the cluster . therefore , we do not consider it a member but a field white dwarf at comparable distance to ngc 2323 . in figure 6 we show the balmer line fits of wd10 and wd11 , the two cluster members . they both are high mass at @xmath01.07 m@xmath1 and have very high temperature ( @xmath2850,000 k ) with accordingly extremely short cooling times of @xmath03 myr . their membership further suggests that all of our other observed white dwarfs , which all have both lower masses and longer cooling times are not consistent with ngc 2323 membership . in figure 7 we display the cmd of all observed candidates relative to several white dwarf cooling models taken from http://www.astro.umontreal.ca/.17exbergeron/coolingmodels/ ( holberg & bergeron 2006 , kowalski & saumon 2006 , tremblay et al . 2011 , and bergeron et al . l c c c c c c c c * - reanalyzed white dwarf initial and final parameters * + id&t@xmath10&log g&m@xmath11 & m@xmath12 & t@xmath13 & y@xmath3 m@xmath14 & parsec m@xmath14 & s / n + & ( k ) & & ( m@xmath1 ) & & ( myr ) & ( m@xmath1 ) & ( m@xmath1 ) & + ngc 2287 - 2 & 25900@xmath17350 & 8.45@xmath170.05 & 0.909@xmath170.028 & 11.01@xmath170.09 & 76@xmath37 & 4.61@xmath38 & 4.81@xmath39 & 150 + ngc 2287 - 4 & 26500@xmath17350 & 8.71@xmath170.05 & 1.065@xmath170.027 & 11.44@xmath170.10 & 127@xmath40 & 5.50@xmath41 & 5.93@xmath42 & 130 + ngc 2287 - 5 & 25600@xmath17350 & 8.44@xmath170.04 & 0.901@xmath170.028 & 11.02@xmath170.08 & 77@xmath37 & 4.63@xmath38 & 4.83@xmath39 & 170 + ngc 2516 - 1 & 30100@xmath17350 & 8.47@xmath170.04 & 0.925@xmath170.027 & 10.74@xmath170.08 & 42@xmath43 & 4.85@xmath44 & 5.19@xmath45 & 160 + ngc 2516 - 2 & 35500@xmath17550 & 8.55@xmath170.07 & 0.981@xmath170.040 & 10.58@xmath170.13 & 24@xmath46 & 4.61@xmath47 & 4.88@xmath48 & 90 + ngc 2516 - 3 & 29100@xmath17350 & 8.46@xmath170.04 & 0.918@xmath170.027 & 10.78@xmath170.08 & 48@xmath46 & 4.94@xmath48 & 5.31@xmath49 & 170 + ngc 2516 - 5 & 32200@xmath17400 & 8.54@xmath170.05 & 0.970@xmath170.027 & 10.73@xmath170.09 & 38@xmath50 & 4.80@xmath51 & 5.12@xmath38 & 170 + ngc 3532 - 1 & 23100@xmath17300 & 8.52@xmath170.04 & 0.950@xmath170.026 & 11.33@xmath170.08 & 131@xmath52 & 4.17@xmath44 & 4.13@xmath44 & 240 + ngc 3532 - 5 & 27700@xmath17350 & 8.28@xmath170.05 & 0.804@xmath170.028 & 10.57@xmath170.08 & 31@xmath50 & 3.57@xmath53 & 3.55@xmath53 & 220 + ngc 3532 - 9 & 31900@xmath17400 & 8.18@xmath170.04 & 0.752@xmath170.026 & 10.13@xmath170.07 & 9.3@xmath54 & 3.48@xmath19 & 3.46@xmath19 & 210 + ngc 3532 - 10 & 26300@xmath17350 & 8.34@xmath170.04 & 0.838@xmath170.027 & 10.78@xmath170.08 & 51@xmath55 & 3.66@xmath56 & 3.64@xmath56 & 200 + ngc 3532-j1106 - 590 & 21100@xmath17350 & 8.48@xmath170.05 & 0.922@xmath170.031 & 11.43@xmath170.09 & 163@xmath57 & 4.46@xmath58 & 4.41@xmath58 & 110 + ngc 3532-j1106 - 584 & 20200@xmath17300 & 8.52@xmath170.05 & 0.945@xmath170.029 & 11.58@xmath170.09 & 197@xmath59 & 4.91@xmath60 & 4.83@xmath60 & 120 + ngc 3532-j1107 - 584 & 20700@xmath17300 & 8.59@xmath170.05 & 0.990@xmath170.028 & 11.66@xmath170.09 & 211@xmath61 & 5.16@xmath62 & 5.07@xmath62 & 180 + sirius b & 26000@xmath17400 & 8.57@xmath170.04 & 0.982@xmath170.024 & 11.21@xmath170.08 & 99@xmath63 & 4.69@xmath64 & 4.69@xmath64 & - + pleiades - lb 1497 & 32700@xmath17500 & 8.67@xmath170.05 & 1.046@xmath170.028 & 10.94@xmath170.09 & 54@xmath65 & 5.86@xmath66 & 6.85@xmath67 & 170 + pleiades - gd50 & 42700@xmath17800 & 9.20@xmath170.07 & 1.246@xmath170.021 & 11.58@xmath170.15 & 70@xmath40 & 6.41@xmath68 & 8.05@xmath69 & - + pleiades - pg0136 + 251 & 41400@xmath17800 & 9.03@xmath170.07 & 1.186@xmath170.027 & 11.28@xmath170.15 & 48@xmath24 & 5.64@xmath70 & 6.52@xmath71 & - + ngc 2168-lawds1 & 33500@xmath17450 & 8.44@xmath170.06 & 0.911@xmath170.039 & 10.47@xmath170.11 & 19@xmath50 & 4.27@xmath72 & 4.48@xmath73 & 95 + ngc 2168-lawds2 & 33400@xmath17600 & 8.49@xmath170.10 & 0.940@xmath170.061 & 10.57@xmath170.18 & 25@xmath74 & 4.33@xmath75 & 4.55@xmath76 & 70 + ngc 2168-lawds5 & 52700@xmath17900 & 8.21@xmath170.06 & 0.801@xmath170.031 & 9.49@xmath170.10 & 1.0@xmath77 & 4.10@xmath19 & 4.28@xmath19 & 210 + ngc 2168-lawds6 & 57300@xmath171000 & 8.05@xmath170.06 & 0.731@xmath170.029 & 9.13@xmath170.11 & 0.5@xmath77 & 4.10@xmath19 & 4.28@xmath19 & 260 + ngc 2168-lawds11 & 19900@xmath17350 & 8.35@xmath170.05 & 0.834@xmath170.035 & 11.31@xmath170.09 & 149@xmath57 & 7.93@xmath78 & 11.60@xmath79 & 100 + ngc 2168-lawds12 & 34200@xmath17500 & 8.60@xmath170.06 & 1.009@xmath170.037 & 10.73@xmath170.12 & 36@xmath65 & 4.44@xmath44 & 4.69@xmath38 & 100 + ngc 2168-lawds14 & 30500@xmath17450 & 8.57@xmath170.06 & 0.988@xmath170.038 & 10.89@xmath170.12 & 54@xmath63 & 4.67@xmath80 & 4.98@xmath58 & 100 + ngc 2168-lawds15 & 30100@xmath17400 & 8.61@xmath170.06 & 1.009@xmath170.032 & 10.98@xmath170.11 & 64@xmath81 & 4.80@xmath82 & 5.16@xmath83 & 130 + ngc 2168-lawds22 & 53000@xmath171000 & 8.22@xmath170.06 & 0.807@xmath170.035 & 9.50@xmath170.11 & 1.0@xmath77 & 4.10@xmath19 & 4.28@xmath19 & 250 + ngc 2168-lawds27 & 30700@xmath17400 & 8.72@xmath170.06 & 1.071@xmath170.031 & 11.16@xmath170.11 & 78@xmath24 & 5.06@xmath84 & 5.49@xmath85 & 120 + ngc 2168-lawds29 & 33500@xmath17450 & 8.56@xmath170.06 & 0.984@xmath170.034 & 10.70@xmath170.11 & 34@xmath55 & 4.42@xmath86 & 4.67@xmath87 & 120 + ngc 2168-lawds30 & 29700@xmath17500 & 8.39@xmath170.08 & 0.878@xmath170.048 & 10.63@xmath170.13 & 33@xmath88 & 4.41@xmath48 & 4.65@xmath39 & 60 + with the white dwarf cluster members , a simple comparison of their cooling age to the total cluster age gives the evolutionary time to the tip of the asymptotic giant branch ( agb ) for their progenitor . application of this time to evolutionary models gives the white dwarf s progenitor mass . in paper i we adopted the models of hurley et al . ( 2000 ) for the progenitor masses in ngc 2099 ( 3 - 4 m@xmath1 ) . at these masses , the difference between predicted progenitor mass by the hurley et al . ( 2000 ) models and the parsec isochrones are less than 1% . we note that the parsec models do not include the thermally pulsing - agb phase , but this phase is very short and does not meaningfully affect the resulting progenitor masses . for the y@xmath3 isochrones we can not infer progenitor masses directly because these isochrones do not evolve beyond the rgb . but we note that while the y@xmath3 isochrones predict slower evolution to the turnoff in these younger clusters , they predict more rapid evolution through the rgb . this results in the total evolutionary time scales to the tip of the rgb being comparable in both model isochrones for all but the highest masses . for example , as we reach evolutionary times of 100 myr or shorter ( progenitors of @xmath25.3 m@xmath1 ) , the systematic differences become significant between all three models . for our current analysis , with both our y@xmath3 and parsec ages , we use progenitor masses derived from the parsec isochrones because it will provide the strongest consistency with our cluster age fits . for ngc 2323-wd10 and wd11 , with their cooling times and our y@xmath3 isochrones based age for ngc 2323 of 140 myr , the corresponding progenitor masses for both white dwarfs are 4.69 m@xmath1 . ( see table 1 , where we also give the progenitor masses based on the cluster age errors [ 140@xmath1720 myr ] and the parsec based age of 115 myr . ) it is quite remarkable that we have two independently formed high - mass white dwarfs from the same cluster that are so consistent in both initial and final mass . across their lifetimes they both lost 77.2% of their total mass . this argues for the consistency of single - star mass loss at higher masses . to look at these data in context , in table 4 we present the initial - final mass data for 30 mid to very high - mass ( 0.73 - 1.25 m@xmath1 ) white dwarfs that have been self - consistently analyzed from publicly available data . in the left panel of figure 8 we plot the initial and final masses of the two analyzed white dwarfs from ngc 2323 and the 30 white dwarfs we have reanalyzed from the literature , adopting the y@xmath3 ages . in the right panel of figure 8 we plot these same data , but with application of the parsec based ages . we also compare these high - mass white dwarf data to the sample of 31 intermediate - mass white dwarfs taken from ngc 2099 , the hyades , and praesepe ( paper i ) . for the higher - mass white dwarfs , while there is some dispersion in the trend there is only one clear outlier that is from ngc 2168 . lawds11 is an extreme outlier with a far longer cooling time than the other members of ngc 2168 , giving it a massive progenitor . for clusters in the rich galactic plane , contamination from common field white dwarfs is expected and likely explains this white dwarf . proper motions may be necessary to further constrain its membership , but in our current analysis and ifmr fits we do not consider lawds11 . in our initial analysis from paper i , we demonstrated that the intermediate - mass white dwarfs ( 0.70.9 m@xmath1 ) create a steep ifmr slope . with the continued adoption of y@xmath3 ages in this paper , which systematically derive older ages in young clusters , we find there is no meaningful change in slope at higher masses and the data appears to be defined well with a weighted linear relation : @xmath89 the linear nature of the ifmr across such a broad range ( 36.5 m@xmath4 ) is of interest . for example , this may suggest we can extrapolate to derive the progenitor of a chandrasekhar mass limit white dwarf to be @xmath07.75 m@xmath1 , but the still limited data at the highest masses and the remaining uncertainties in the evolutionary models suggests this is unreliable . furthermore , in theoretical models a moderate turnover in the slope of the ifmr is predicted near initial mass of 4 m@xmath1 ( e.g. , marigo & girardi 2007 and meng et al . this predicted turnover is the result of the second dredge - up , which only occurs in stars of @xmath04 m@xmath1 and higher . this diminishes their core mass and hence their final white dwarf mass . a comparable turnover could still be lost in our data s remaining scatter , causing further problems with a linear extrapolation . in figure 9 we more closely analyze the residuals of this linear data fit . to illustrate the importance of our adopted isochrones , the right panel of figure 8 shows a clear turnover in the ifmr near an initial mass of 4 m@xmath1 when adopting parsec isochrone based ages , consistent with theoretical predictions . we note that while our comparison ifmr data from paper i adopts y@xmath3 isochrone ages , this is not the cause of the strong overturn . this is because in these older clusters ages from paper i the systematics shift direction and the parsec isochrones give cluster ages 20 myr older than the y@xmath3 isochrones . for the intermediate - mass comparison ifmr this would increase its slope by @xmath05% , and this would further increase the magnitude of the turnover . we fit the data based on the parsec ages for our younger clusters with a two - piece linear function with a kink at initial mass of 4 m@xmath1 : @xmath90 this is a strong turnover , one that is stronger than predicted in marigo & girardi ( 2007 ) or meng et al . ( 2008 ) . taking a moment to focus specifically on ngc 3532 may help to clarify these systematic differences . this is both because it is the only cluster with progenitors that cover both above and below initial mass of 4 m@xmath1 , and it is the cluster least effected by systematic differences between the y@xmath3 and parsec isochrones . the ngc 3532 data alone is limited but consistent with an overturn , which suggests the y@xmath3 may be overestimating the ages of the youngest clusters , but the massive progenitors of the pleiades white dwarfs in the right panel of figure 8 conversely suggests that the parsec isochrones may be underestimating the ages of the older clusters . further work is needed on isochrones and evolutionary times at these higher masses , but for now we continue to adopt our y@xmath3 isochrone results as final . in figure 10 we compare to multiple empirical ifmrs from weidemann ( 2000 ) , dobbie et al . ( 2006a ) , cataln et al . ( 2008 ) , casewell et al . ( 2009 ) , kalirai et al . ( 2009 ) , and salaris et al . we find that our relation based on y@xmath3 ages gives that the ifmr is steeper at higher masses than all of these relations besides those of dobbie et al . ( 2006a ) and cataln et al.(2008 ) , which have comparable slopes . our relation is strongly consistent with that of cataln et al . ( 2008 ) , but the zeropoint of dobbie et al . ( 2006a ) is @xmath00.05 m@xmath1 lower . therefore , in comparison to most previous relations , ours predicts that white dwarf mass increases more rapidly with increasing progenitor mass , and that overall mass - loss rates are lower . this finding is primarily the result of our ages being systematically older due in part to our analysis cleaning the turnoffs of peculiar stars . additionally , the y@xmath3 isochrones give systematically older ages in comparison to the commonly adopted girardi et al . ( 2000 ) isochrones . for example , a consistent turnoff fit to the pleiades with the isochrones of girardi et al . ( 2000 ) gives a 100 myr age , 35 myr younger than the y@xmath3 isochrones . the most remarkable difference to previous ifmrs is that the large scatter has significantly decreased by @xmath050% . this not only illustrates the importance of doing uniform analysis of the white dwarf spectra , but of the consistent examination of the cluster parameters with detailed consideration of peculiar stars in the turnoffs . our ifmr s relatively smaller scatter also begins to show that while variations in mass - loss rates may still occur during a progenitor s lifetime , in singly evolved stars it is likely quite minor and the previously observed scatter is not indicative of stochastic mass loss . metallicity dependence of the ifmr is another important parameter to consider , but it will not be quantitatively analysed here with such incomplete metallicity information . for our three clusters pleiades , ngc 2168 , and ngc 2516 with spectroscopic metallicity determinations , however , we note that the high - mass white dwarfs from the metal - poor ngc 2168 primarily fall above the mean of the ifmr while the white dwarfs from the slightly metal - rich ngc 2516 fall below the mean . direct comparison of the ngc 2323 white dwarfs to the rest of our sample shows that they deviate more from the relation than a majority of the others . individually , their errors in white dwarf mass are large enough to make these deviations not appear significant , but both white dwarfs being so consistent suggests that this deviation may be telling us something . for example , in comparison to the progenitors of the three comparable mass white dwarfs in other clusters ( lb 1497 , ngc 2168-lawds27 , and ngc 2287 - 4 ) , their three progenitors are all higher mass with lb 1497 giving a @xmath01.2 m@xmath1 larger progenitor . is this indicative of mass - loss variations between these different clusters or remaining systematics ? for example , a possible explanation for this is that our assumed solar metallicity for ngc 2323 is not appropriate . a more metal - rich isochrone would give a younger age and help to decrease the observed differences in progenitor masses . or this deviation may simply be the result of potential systematic differences between our adopted photometries , and in general this may be a cause for much of the remaining scatter in our total ifmr . in any case , this new ngc 2323 white dwarf analysis further emphasizes the importance of this nearby , rich , and young open cluster , and of the need for spectroscopic metallicity and precise ubv photometry spanning from its turnoff to its faint white dwarfs . we have spectroscopically observed a sample of 10 white dwarf candidates in the young and rich ngc 2323 . while we only found that two of these white dwarfs are consistent with membership in ngc 2323 , they are both newly discovered high - mass white dwarfs at @xmath01.07 m@xmath1 . to supplement these new white dwarfs we have reanalyzed all published , publicly available high - mass white dwarfs from the pleiades , ngc 2516 , ngc 2168 , ngc 2287 , and ngc 3532 star clusters and sirius b. at these higher masses , the inferred progenitor masses are increasingly dependent on the adopted star cluster ages . therefore , we uniformly analyzed available high - quality ubv photometry for these clusters . due to the difficulties of analyzing younger cluster ages , we have analyzed both color - color diagrams and color - magnitude diagrams . the color - color diagrams provide a self - consistent manner to not only determined a cluster reddening but to identify peculiar stars in each cluster s turnoff . with the cleaned turnoffs for each cluster , we can more precisely determine a cluster age . in general , we find cluster ages moderately older than those adopted in previous ifmr analyses , and we attribute this both to cluster peculiar stars affecting previous age determinations and the systematically older ages derived from the y@xmath3 and parsec isochrones . application of our derived white dwarf cooling ages and cluster turnoff ages gives us the progenitor masses for each white dwarf . the two massive white dwarfs from ngc 2323 at 1.07 m@xmath1 both find remarkably consistent progenitor masses of 4.69 m@xmath1 . with the addition of the publicly available high - mass white dwarfs , the derived ifmr is quite clean , and we find that with y@xmath3 ages the ifmr can be well defined by a linear relation from progenitor masses of 3 to 6.5 m@xmath1 . in contrast , adoption of the only moderately younger cluster ages from the parsec isochrones gives a moderate turnover in ifmr slope near initial mass of 4 m@xmath1 . in our current analysis we continue to adopt the y@xmath3 results , but this draws attention to the work that is needed on evolutionary timescales at these higher masses . our ifmr also finds a significantly decreased scatter in comparison to recent ifmr work . this shows that the typically observed scatter in the ifmr is likely not indicative of real variations in mass loss for singly evolved stars , but was the result of remaining systematics . the ifmr relation is now reassuringly consistent across multiple star clusters . metallicity , however , still may play an important role , in particular at these higher masses . to further improve the ifmr , not only are newly discovered high - mass white dwarfs needed , but uniform ubv photometry and spectroscopic metallicity analyses are also needed . this will further limit systematics that may still remain and give the foundation for detailed quantitative analysis of the metallicity dependence of the ifmr . this project was supported by the national science foundation ( nsf ) through grant ast-1211719 . this work was also supported by a nasa keck pi data award , administered by the nasa exoplanet science institute . data presented herein were obtained at the w. m. keck observatory from telescope time allocated to the national aeronautics and space administration through the agency s scientific partnership with the california institute of technology and the university of california . the observatory was made possible by the generous financial support of the w. m. keck foundation . p - e.t . was supported during this project by the alexander von humboldt foundation and by nasa through hubble fellowship grant hf-51329.01 , awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , incorporated , under nasa contract nas5 - 26555 . this research has made use of the keck observatory archive ( koa ) , which is operated by the w. m. keck observatory and the nasa exoplanet science institute ( nexsci ) , under contract with the national aeronautics and space administration . lastly , we would like to thank c. deliyannis for his helpful discussions of open cluster color - color analysis .

strong lensing is a powerful probe of the distribution of matter in the cores of clusters of galaxies . recent studies suggest that the cold dark matter model predicts cores that are denser than those observed in galaxies , groups and clusters . one possible resolution of the discrepancy is that the dark matter has strong interactions ( sidm ) , which leads to lower central densities . a generalized form of the navarro , frenk and white profile ( zhao profile ) may be used to describe these halos . in this paper we examine gravitational lensing statistics for this class of model . the optical depth to multiple imaging is a very sensitive function of the profile parameters in the range of interest for sidm halos around clusters of galaxies . less concentrated profiles , which result from larger self - interaction cross - sections , can produce many fewer lensed pairs . furthermore , profiles that result in a small optical depth exhibit reduced typical splittings , but produce multiple images that are more highly magnified . however the resulting increased magnification bias does not alter our conclusions . we find that lensing statistics based on profile parameters obtained from fits out to the virial radius are dependent on the minimization scheme adopted , and may be seriously in error . however , profile fits weighted towards the core region have parameter degeneracies that are approximately equivalent to those for strong lensing cross - sections . lensing statistics provide a powerful test for sidm . more realistic and observationally oriented calculations remain to be done , however larger self - interaction cross - sections may well be ruled out by the very existence of strong lenses on galaxy cluster scales . the inclusion of centrally dominant cluster galaxies should boost the cross - section to multiple imaging . however our preliminary calculations suggest that the additional multiple imaging rate is small with respect to the differences in multiple imaging rate for different halo profiles . in future statistical studies , it will be important to properly account for the scatter among halo profiles since the optical depth to multiple imaging is dominated by the most concentrated members of a cluster population .

astro-ph0007354

the cold dark matter plus cosmological constant model ( @xmath0cdm ) appears to be remarkably successful in describing cosmology on scales much larger than a mpc ( e.g. bahcall , ostriker , perlmutter & steinhardt 1999 ) . however recent observations on smaller scales find @xmath0cdm wanting : @xmath0cdm predicts dwarf galaxy halos ( e.g. moore et al . 1999 ) that are denser and more concentrated than observed ( dalcanton & bernstein 2000 ; firmani et al . the persistence of bars in galaxies like our own , suggests that the dark matter is not centrally concentrated ( sellwood 2000 ) . in addition , the intensively studied cl0024 + 1654 is nearly spherical ( colley , tyson & turner 1996 ; tyson , kochanski & dellantonio 1998 ) , at odds with @xmath0cdm predictions of triaxial systems . spergel & steinhardt ( 2000 ) have recently suggested that these and other discrepancies ( see for example dav@xmath1 , spergel , steinhardt & wandelt 2000 ; wandelt et al . 2000 and references therein ) could be resolved if cold dark matter were weakly self interacting with a large scattering cross - section ( sidm ) . several authors ( burkert 2000 ; yoshida , springel , white & tormen 2000 ; dav@xmath1 et al . 2000 ) have since undertaken n - body simulations of halos with self interactions in the regime suggested by spergel & steinhardt ( 2000 ) . these works find that sidm halos produce flatter , smoother and more spherical cores than their @xmath0cdm counterparts . we are interested in the statistics of multiple imaging due to a generalized nfw profile ( zhao 1996 ) which may be used to describe both cdm and sidm halos . we concentrate our efforts on lensing by clusters of galaxies , since single galaxies contain a significant mass in baryons , distributed to produce a flat rotation curve . this makes sidm in galaxies difficult to probe using gravitational lensing . conversely , clusters of galaxies have a much larger mass to light ratio , and baryons which are less centrally concentrated ( particularly in the absence of a central dominant galaxy ) . we assume spherically symmetric profiles and concentrate on the statistics of the optical depth to multiple imaging . the total magnification , and the maximum image splitting are also considered . in secs . [ darkmatter ] and [ lensing ] we discuss the properties of dark matter halos and introduce the formalism for computing the lensing statistics of the zhao profile . in secs . [ results ] , [ degeneracy ] and [ implications ] we discuss the statistics obtained , contrast the parameter degeneracies present in the lens statistics and in profile fitting , and present some implications for sidm cluster halos . in sec . [ halodist ] we discuss the important implications for lens statistics of the distribution of halo profile - parameters . we also discuss two potentially important caveats which turn out not to effect our conclusions : we discuss magnification bias ( sec . [ amp_dist_bias ] ) , and give a preliminary account of the effect of massive central galaxies on the multiple imaging rate ( sec . [ galaxies ] ) . throughout the paper we have assumed a standard @xmath2 , @xmath3 and @xmath4 filled beam cosmology . navarro , frenk & white ( 1997 ) claimed the nfw profile ( with an inner density @xmath5 varying with radius @xmath6 as @xmath7 ) to be universal . more recently moore et al . ( 1999 ) and ghingna et al . ( 2000 ) have claimed a different universal profile , having @xmath8 in the innermost regions . however jing & suto ( 2000 ) find a systematic correlation of inner profile slope with mass . moreover , they find some scatter in the slope at a fixed mass . on the other hand , from their convergence study , klypin , kravtsov , bullock & primack ( 2000 ) find that while there is some scatter , in general halos are equally well fit in the region where convergence is achieved by either an nfw or a moore profile , independent of the halo mass . they interpret the correlation found by jing & suto ( 2000 ) as a trend in halo concentration rather than in asymptotic slope . dav@xmath1 , spergel , steinhardt & wandelt ( 2000 ) present a comprehensive suite of simulations of both cdm and sidm dwarf galaxy halos . they find typical profiles for cdm halos in the inner most regions of @xmath9 , with some scatter . however they found sidm halos to be less cuspy , having central profiles ranging from @xmath10 to @xmath11 . the uncertainty as to the optimal halo profile as well as the variation introduced by sidm motivate an exploration of lensing by profiles that differ from the analytically treatable nfw form . following the introduction of the navarro - frenk - white profile ( nfw ) ( navarro , frenk & white 1995,1996,1997 ) , bartelmann ( 1996 ) was the first to consider its lensing implications , showing that the profile is capable of producing radial arcs like those observed in several clusters ( e.g. mellier , fort & kneib 1993 ; smail et al . there have been many subsequent efforts contrasting resultant lensing predictions with observation . bartelmann et al . ( 1997 ) found the optical depth for large arcs for lensing by nfw profiles and found significant dependence on cosmology , though no models could reproduce the observed number . wright & brainerd ( 2000 ) and asano ( 2000 ) have contrasted weak lensing mass measurements and 2-image system image positions flux ratios based on the nfw with those for a singular isothermal sphere ( sis ) . maoz , rix , gal - yam & gould ( 1997 ) carried out a survey for wide separation lensed quasars , and compared their results with a thorough statistical study of the number of expected cluster lensed quasars . modeling clusters using the nfw profile , they found that their null result for observed lensed quasars was consistent with the lensing statistics . molikawa , hattori , kneib & yamashita ( 1999 ) investigated giant arc statistics and found that the nfw profile can reproduce the number of observed lensed arcs . however their study finds some discrepancy between virial and x - ray temperature for some clusters . broadhurst , huang , frye & ellis ( 2000 ) have found that an nfw profile ( singular core ) can explain strong lensing in cl0024 + 1654 , contrary to earlier claims of a flat core ( tyson et al . however shapiro & iliev ( 2000 ) find that the velocity dispersion implied by the nfw fit is a factor of two larger than that observed . the shallower cores resulting from sidm simulations may resolve this discrepancy . on the other hand , smith et al . ( 2000 ) find that radial arcs imply a core in a383 which has a steeper logarithmic slope than an nfw profile . williams , navarro & bartelmann ( 1999 ) find that arc properties for the most massive clusters are consistent with nfw profiles derived from n - body simulations , but that clusters with velocity dispersions of @xmath12 require the presence of a massive central galaxy to reconcile the models with observation . fox , & pen ( 2000 ) have considered lensing by galaxy groups modeled as nfw profiles in the hubble deep field . they find that the null result for lenses is consistent with lensing probability . in addition , they compare the probability for multiple imaging due to an nfw profile with the probability for an sis , and find that the probability is significantly reduced . recent complimentary studies have also considered lensing statistics for zhao profiles . molikawa & hattori ( 2000 ) find that the ratio of radial to tangential arcs is a sensitive function of the inner slope of the density profile . li & ostriker ( 2000 ) consider a semi - analytic treatment of lensing by the zhao profile . they find that lensing probability is sensitive to the profile considered , as well as cosmology . in addition , they find a probability for large splittings which is inconsistent with the observed distribution . the nfw halo density profile can be generalized to describe a profile @xmath5 as a function of radius @xmath6 with an arbitrary power - law shaped central cusp @xmath13 and outer regions that fall off as @xmath14 ( zhao 1996 ) , @xmath15 navarro , frenk and white ( 1996 ) defined the _ concentration parameter _ @xmath16 where @xmath17 is the radius enclosing a mass @xmath18 with an over - density of 200 above the critical density . characteristic over - density _ @xmath19 in eqn . [ prof ] is @xmath20 where @xmath21 therefore , given a cluster mass @xmath18 , the zhao profile has two free parameters @xmath22 and @xmath23 . @xmath23 is quoted for a profile at red - shift zero . our lensing calculations assume a non - evolving cluster - lens population . the concentration parameter @xmath24 and characteristic over - density @xmath25 of a halo placed at red - shift @xmath26 are therefore the solution of @xmath27 and @xmath28 these transformations leave @xmath29 and @xmath30 independent of @xmath26 . from the study of the evolution of a large sample of @xmath0cdm halos , bullock et al . ( 2000 ) find @xmath31 , consistent with eqn . [ cnfwz ] . note however , that the inclusion of a finite self - interaction cross - section should result in a different profile evolution . the zhao profile is spherically symmetric , while real clusters , and n - body simulations tend to be elliptical or tri - axial . blandford & kochanek ( 1987 ) and kochanek & blandford ( 1987 ) find that the the introduction of ellipticities smaller than @xmath32 into nearly singular profiles has little effect on the cross - section and image magnification . furthermore , the images remain on opposite sides of the potential so that the variation in image separations is similar to the axis ratio of the potential . however profiles containing a finite core and ellipticity show a significant increase in cross - section and qualitative differences in image positions . we make the thin screen approximation for gravitational lensing by extended bodies . the surface mass density @xmath33 at a radius @xmath34 may be written @xmath35 where @xmath36 . the mass @xmath37 enclosed within the cylinder of radius @xmath34 and the bend angle @xmath38 of a ray with impact parameter @xmath34 are @xmath39 and @xmath40 if @xmath41 , eqns . [ surf ] , [ mass ] and [ bend ] reduce to the expressions computed by bartelmann ( 1996 ) . the lens equation describes the location @xmath34 of an image in the lens plane given a source position @xmath42 ( both measured from the observer source line of sight ) . the angular diameter distances from the observer to a lens at red - shift @xmath43 , from the observer to a source at @xmath44 and from the lens to the source are denoted by @xmath45 , @xmath46 and @xmath47 . with these , the lens equation may be written ( e.g. schneider , ehlers & falco 1992 ) @xmath48 where @xmath49 and @xmath50 are the angular positions of the image and source respectively . the point source magnification @xmath51 of an image at @xmath52 is @xmath53 the zhao profile has a critical radius @xmath54 separating regions of the source plane that produce 1 or 3 images . the corresponding critical impact parameter @xmath55 is the solution of @xmath56 and the critical angle for multiple imaging is @xmath57 for later comparison , the bend angle for a sis is @xmath58 in this case @xmath59 so that @xmath60 for the computation of lens statistics for an unevolving population of lenses with constant co - moving number density @xmath61 we find ( turner , ostriker & gott 1984 ) that for an @xmath62 universe the differential and total cross - section may be written @xmath63 and @xmath64 turner , ostriker & gott ( 1984 ) defined a dimensionless quantity @xmath65 that measures the efficiency of cosmologically distributed siss for producing multiple images . @xmath66 we use @xmath65 to normalize @xmath67 . more realistic calculations of @xmath68 for the nfw ( @xmath41 ) profile have previously been made within the contexts of wide separation quasars ( maoz , rix , gal - yam & gould 1997 ) , giant arc statistics ( bartelmann et al . 1997 ; molikawa , hattori , kneib & yamashita 1999 ) and distant supernovae ( porciani & madau 2000 ) . however in this work we are interested in relative values of @xmath67 produced by different profiles , and so have neglected quantitatively important effects such as formation history . one image is present for all source positions . we denote this image to be at @xmath69 . note that @xmath70 for all @xmath22 . for @xmath71 there are two additional images with positions @xmath72 and @xmath73 . we label the corresponding point source image magnifications as @xmath74 , @xmath75 and @xmath76 respectively . if @xmath77 then @xmath78 , @xmath79 , @xmath80 and @xmath81 . in addition @xmath82 . for @xmath71 , we define the average image separation of a multipli - imaged source at @xmath44 due to a deflector at @xmath43 to be @xmath83 similarly the average magnification of a multipli - imaged source at @xmath44 due to a deflector at @xmath43 is @xmath84 it follows that the average image splittings @xmath85 and total magnifications @xmath86 for a source at @xmath44 are @xmath87 and @xmath88 in this section we present the differential and total cross - sections ( eqns . [ eqn_difod ] and [ eqn_od ] ) , distributions of image separations , and distributions of magnifications ( eqns . [ sep ] and [ mag ] ) for zhao profiles having different parameters @xmath22 and @xmath23 . [ profsurf ] shows @xmath5 ( top ) and @xmath33 ( bottom ) for three different zhao profiles . these are labeled profile @xmath89 ( solid lines ) , @xmath90 ( dashed lines ) and @xmath91 ( dotted lines ) , have parameters of @xmath92 , @xmath93 , and @xmath94 respectively , and are increasingly steep cusps in their central regions in addition to being increasingly centrally concentrated . the model clusters have an @xmath18 equal to that of a sis ( @xmath95 ) with a velocity dispersion corresponding to a large cluster @xmath96 : @xmath97 the solid light lines show @xmath98 and @xmath99 of the sis for comparison . the horizontal dashed lines show the critical surface mass density @xmath100 for a source at @xmath101 and lenses at @xmath102 , @xmath103 and @xmath104 . comparison of @xmath105 with @xmath33 indicates a significant variation of the lensing probability between profiles having the same @xmath18 but different profile parameters . profiles @xmath89 and @xmath90 reach the critical density over a much larger area than profile @xmath91 , which is only just supercritical at @xmath106 , and sub - critical at @xmath107 . the variation of cross - section with the profile parameters is quantified by fig . [ od ] which shows @xmath68 for zhao profiles @xmath108 . the cross - section is given in units of @xmath65 ( eqn . [ f ] ) . for comparison , the cross - section for a sis ( @xmath109 , light line , turner , ostriker & gott 1984 ) is also shown in fig . examples of the effect of different cosmologies on @xmath109 ( and on @xmath110 in fig . [ difod ] ) can be found in turner et al . ( 1984 ) and turner ( 1990 ) . there is a striking variation in @xmath67 over these three profiles . profiles @xmath90 and @xmath91 produce @xmath67s which differ by an order of magnitude , while there are 4 orders of magnitude between @xmath67 for profiles @xmath89 and @xmath91 . more concentrated profiles are much more efficient for producing multiple images due to their higher central densities . for larger source red - shifts the sis has a similar cross - section to profile @xmath90 , while at smaller red - shifts the cross - section is similar to that of profile @xmath89 , indicating that different profiles are efficient lenses at different red - shifts . in fig . [ difod ] we have plotted @xmath110 for profiles @xmath89 ( solid lines ) , @xmath90 ( dashed lines ) and @xmath91 ( dotted lines ) with @xmath111 ( upper panel ) and @xmath112 ( lower panel ) . the light lines show the corresponding distributions for a sis . less concentrated profiles have distributions that are more sharply peaked , with modes at lower red - shift ( c.f . kochanek & blandford 1987 ) . the lower cross - section to multiple imaging is therefore due to a combination two effects . firstly , the halo reaches the critical density to lensing only over a small region in the center due to the low concentration or central slope . in addition , there is a decreased redshift range in which the central surface density exceeds the critical value anywhere . note that the assumption in eqns . [ eqn_difod ] and [ eqn_od ] of an unevolving population of sidm halos is aided in the case of low - concentration profiles since these preferentially lens at low red - shift . to further investigate the variation of cross - section with @xmath22 and @xmath23 we have plotted contours of @xmath113 with @xmath101 , over a parameter space of @xmath22 and @xmath23 ( fig . [ contod ] ) . two additional dark lines are shown . the upper dark line is a contour of unity and marks the parameters that produce an cross - section equal to that of a sis . the lower dark line marks the boundary of the region of parameter space where no multiple imaging is produced ( on this and subsequent plots this line is the approximated by the @xmath114 contour ) . the contours describe sets of parameters @xmath22 and @xmath23 that are equally efficient for producing multiple images . profiles @xmath89 , @xmath90 and @xmath91 lie on a line in @xmath115 space that is approximately orthogonal to the contours of @xmath68 . however , figure [ contod ] demonstrates that dramatic variability in @xmath68 may be obtained by variation of either @xmath22 or @xmath23 . we have calculated @xmath113 , @xmath116 and @xmath86 over a small grid of @xmath22 and @xmath23 ( see tab . [ sepmagtab ] ) . we find that profiles which produce a small cross - section to multiple imaging also produce smaller image splittings , although only by a factor of a few compared with more highly concentrated profiles . however , the image magnification is significantly larger for the less concentrated profiles . distributions of image separations ( @xmath117 ) for the zhao profile are presented in fig . [ sep_dist ] for the values of @xmath22 and @xmath23 in tab . [ sepmagtab ] . for comparison , the light line shows the distribution of splitting angles for a sis ( @xmath118 ) . there is a trend for the distributions with lower @xmath116 to also have a narrower range of @xmath119 . in contrast to the sis distribution , the zhao profile distributions are asymmetric about the mode , having a longer large splitting tail . [ amp_dist ] shows magnification distributions @xmath120 for the values of @xmath22 and @xmath23 in tab . [ sepmagtab ] . the distribution for an sis , @xmath121 is shown for comparison ( light line ) . from fig . [ amp_dist ] we see that the high mean magnifications shown in tab . [ sepmagtab ] result from a higher minimum magnification of multiple images . lenses with low concentration and small @xmath22 result in a low @xmath68 . very close alignment is therefore required to produce multiple imaging , and the close alignment results in high magnification . more concentrated lenses do not require such high alignment for multiple imaging . this results in a larger cross - section and reduced mean magnification . the basic lensing properties of the zhao profile may therefore be summarized as follows . profiles that have a small concentration and flat central profile produce significantly less multiple imaging than a profile with either a high concentration or steep central profile ( or both ) . however , these few multiple images are significantly magnified . in any lens survey , the increased @xmath122 will partially offset the decreased @xmath68 in determining the lens frequency . this magnification bias is discussed in sec . [ amp_dist_bias ] . for clusters of galaxies , the reduced splittings for low density profiles are @xmath123 and should not pose a significant selection bias for well defined surveys . a property of the zhao profile ( eqn . [ prof ] ) is the ability to describe a high central density either by a steep central cusp , or by a large concentration parameter ( so that the turnover separating the core regions from the outer halo occurs closer to the center ) . this degeneracy is present between other profiles exhibiting a power - law core and a characteristic radius . klypin , kravtsov , bullock & primack ( 2000 ) have investigated the relative suitability of moore et al . ( 1999 ) and nfw ( @xmath41 ) profiles as descriptions of cdm halos . for galaxy size halos , they find differences which are small at radii above @xmath125 . larger differences are found for lower concentration cluster halos as the profiles central slope begins to approach its asymptotic value . moreover , using a very general 4-parameter nfw profile , klypin et al . ( 2000 ) obtained equally good fits to an n - body halo for several different sets of parameters . our interest lies in the relationship between the @xmath124 degeneracy found in in the contexts of multiple imaging statistics and profile fits to simulated halos . our approach is to select a collection of profiles @xmath126 having @xmath127 but different concentrations @xmath128 . keeping @xmath18 constant ( for comparison with our strong lensing calculations ) , we fit a new profile @xmath129 with @xmath23 as a function of @xmath22 . fits are obtained by minimizing each of three different quantities which emphasize different parts of the halo : @xmath130 @xmath131 ( klypin , kravtsov , bullock & primack 2000 ) , and @xmath132 minimization is performed over the range @xmath133 . some examples of the fitted profiles and their fractional residuals are presented in fig . [ fitprofa ] ( light lines ) . the dark lines show the profile being fitted . in each case we quote the parameters @xmath134 , @xmath128 and @xmath22 , @xmath23 , as well as the average absolute relative difference over the fitting volume @xmath135 the central density dominates minimization of @xmath136 ( top panels ) so that the fitted profile has the correct central density at the cost of error over large regions around @xmath137 . alternatively if we set @xmath23 , and adjust @xmath22 , the fitted profile has the correct central slope and density , but is systematically high or low over the outer regions . the initial and fitted halos have average absolute relative differences of a few to @xmath138 . as noted by klypin , kravtsov , bullock & primack ( 2000 ) , minimization of @xmath139 ( central panels ) often does a much better job of reproducing the density at small @xmath6 . if @xmath22 is the fitted parameter rather than @xmath23 , a similar result is obtained , the central slope adjusts so as to give the correct density at small @xmath6 . minimization of @xmath139 produces halo fits with similar average absolute relative differences to @xmath136 . finally , minimizing @xmath140 ( lower panels ) obtains good fits beyond @xmath141 at the expense of an incorrect central density . the mean relative error is substantially reduced , with average absolute relative differences of a fraction of a percent . the dot - dashed lines in the upper , central and lower plots of fig . [ fitprofb ] are depictions of the @xmath115 degeneracies corresponding to the minimization quantities @xmath136 , @xmath139 and @xmath140 . since @xmath136 produces good fits to the core region , which is known to be most important for strong lensing ( e.g. turner , ostriker & gott 1984 ) , the lines of degeneracy are nearly parallel to those describing the cross - section to multiple imaging over much of the parameter space . however for combinations of low or high concentration and low inner slope , the approximately degenerate profiles do not provide a accurate description of the lensing statistics . minimization of @xmath139 results in profile degeneracies similar to those for strong lensing unless the profile has large @xmath23 and small @xmath22 . finally , minimization of @xmath140 leads to degeneracies that differ significantly from those for strong lensing . fits that are weighted towards the central regions , and which result in highly concentrated profiles may therefore be converted to the analytically treatable nfw ( @xmath41 ) profile while approximately preserving the lensing statistics . this is particularly true for fits based on minimization of @xmath139 . on the other hand , lensing calculations based on profiles which best describe the density over the largest volume ( out to @xmath142 ) may severely under or over - estimate the strong lensing statistics . for example , zhao halos having the same @xmath18 , but different parameters ( @xmath143 ) = ( 1.5,6.0 ) and ( 1.0,10.0 ) are degenerate to fitting by the minimization of @xmath140 , but differ in @xmath67 by an order of magnitude . thus fig . [ fitprofb ] may be considered a quantification ( for zhao profiles ) of the statement that lensing is sensitive to the size of the halo core . statistics generated from surveys for gravitational lenses are subject to several classes of bias . kockanek ( 1991 ) has given a detailed discussion of selection biases in optical , imaging , galaxy mass gravitational lens surveys . two primary sources of bias against finding a lens are selection against small separations due to the finite angular resolution of the survey , and the obscuration or differential reddening of one or both images due to dust in the lens galaxy . when considering lensing by clusters , the splittings are nearly always larger than the survey resolution . in addition , the lensed images are viewed through the halo rather than the outskirts of a galaxy . these biases should not therefore operate on any significant level in the analysis of cluster lensing statistics . the magnification of a lensed image has the potential to boost the observed flux of a source which is intrinsically faint above the survey detection limit . lensed images therefore sample fainter members of the source population . this is magnification bias ( gott & gunn 1974 ; turner 1980 ) , and it a significant factor in any lens survey . as mentioned in sec . [ results ] , lenses with low concentration and a shallow cusp result in a low cross - section . very close alignment is therefore required to produce multiple imaging , and this results in a high minimum magnification . more concentrated lenses do not require such high alignment to produce multiple imaging , thus the cross section is larger , and the minimum magnification not so high . in the latter case , although the mean is lower , high magnifications are still present where the alignment is high . as a result , the mean magnification is not necessarily a fair indication of the magnification bias . magnification bias operates differently in surveys for galaxy and cluster lenses . the typical separation of a galaxy lens is smaller than most optical survey resolution limits , and the total magnification must be used to compute the magnification bias . however , the total magnification is not the important quantity for cluster lensing surveys since the separate images are generally resolved ( see also chiba & yoshii 1997 ) . there are different approaches to making a wide separation gravitational lens survey . firstly , candidate lenses may be selected from galaxy or cluster catalogues having one image of a source in good alignment with a potential lens . follow up imaging can then be used to look for a second , fainter image . this is the approach taken by systematic surveys to date ( e.g. maoz et al . 1997 ; phillips , browne & wilkinson 2000 ) . here the magnification bias which results in the inclusion of the brighter image into the initial survey is important ( though the overall bias for the survey must also consider the detectability of the faint image in follow up observations ) . a second method looks for pairs of lensed images in existing data . in this instance , the magnification bias which also includes the fainter image ( the second brightest for three image lenses , i.e. image 3 ) , and thus both images into the survey is the relevant quantity . to compute magnification bias , a number magnitude relation for the sources is required . we use the function described by kochanek ( 1996 ) , based on data from boyle , shanks & peterson ( 1988 ) and hartwick & schade ( 1990 ) . he uses a broken - power - law form ( turner 1990 ) . to simplify our calculations and to keep the conclusions as general as possible , we assume that the survey depth is deeper than the break b - magnitude at @xmath101 of @xmath144 . the magnitude - number relation may then be written as @xmath145 where @xmath146 is a normalizing constant and @xmath147 is the logarithmic slope . for a survey with a limit @xmath148 , the magnification bias is defined as ( fukugita & turner 1991 ) @xmath149 in application to a real lens survey , the magnification bias may be written more conveniently as @xmath150 where the limits @xmath151 and @xmath152 are the range of the bright image magnification @xmath153 such that at least one other image is visible . note that only the magnification distribution for the bright image is directly relevant . in this paper we adopt the simpler and non - specific case of magnification bias for bright and faint images based on eqn . [ biaseqn ] . [ amp_dists_fig ] shows the magnification distributions for the bright ( left hand panel ) and faint ( right hand panel ) images for zhao profiles . the light lines show the distributions for a sis : @xmath154 and @xmath155 for comparison . the image magnification distributions for the zhao profiles exhibit similar behavior to the sis , and there is a high probability that the two images will have different magnifications . however unlike the sis , the zhao profiles have finite @xmath156 as @xmath157 . the faint image therefore has a finite minimum magnification . the magnification bias described by eqn . [ biaseqn ] describes the additional number of lensed pairs . if the quantity of interest is the density of lensed images on the sky , then an additional factor of @xmath158 must be inserted under the integral to account for the smaller region of source plane sampled per unit area of lens plane . inserting eqn . [ numbercount ] into eqn . [ biaseqn ] we find @xmath159 for magnification bias due to bright images , and @xmath160 for magnification bias due to faint images . note that the magnification bias is equal to the mean magnification for @xmath161 . since the high magnification tail of any magnification distribution has @xmath162 the integral in eqns . [ bbright ] and [ bfaint ] is finite for @xmath163 , and in this range no low - magnitude cutoff need be assumed . using eqns . [ psis_faint ] and [ psis_bright ] the magnification biases for the faint and bright images of the sis are @xmath164 and @xmath165 fig . [ bright_amp_bias ] shows the absolute magnification bias ( left hand panel ) , and the magnification bias relative to the sis ( right hand panel ) based on the bright image as a function of @xmath147 . plots are shown for the values of @xmath22 and @xmath23 in tab . [ sepmagtab ] . the light line in the left hand panel shows the magnification bias curve for the sis . fig . [ faint_amp_bias ] shows the corresponding results for the faint image . values of absolute and relative ( to the sis ) magnification bias based on the faint and bright image are presented in tabs . [ abbright ] , [ abbrightsis ] , [ abfaint ] and [ abfaintsis ] for @xmath166 . for quasars kockanek ( 1996 ) finds @xmath167 . the absolute magnification biases for the bright images range up to a few , a few tens and a few hundred for @xmath168 and @xmath169 respectively . however , as discussed at the top of this section , the quantity of interest is the relative magnification bias . the values in tab . [ abbrightsis ] demonstrate the reduced amplitude of the relative ( to the sis ) bias , which is similar for @xmath170 , but reduced by a factors of @xmath171 and @xmath172 for @xmath161 and @xmath169 . similar behavior is seen in magnification bias of faint images . the absolute magnification biases for the faint images produced by zhao profiles are slightly reduced from those computed for the bright image . however the magnification biases for the faint images relative to the sis are increased by a factor of two over the corresponding bright image case . this apparent contradiction arises because the faint image produced by a sis may be arbitrarily faint . the magnification bias results in additional sources being available for lensing . since the relative magnification bias is greater than 1 where the ratio of cross - sections is less than one , the magnification bias counteracts the dramatic dependence of @xmath67 on @xmath22 and @xmath23 seen in fig . [ contod ] . the relevant quantity is therefore the optical depth to multiple imaging @xmath173 or @xmath174 . [ abbrighttao ] and [ abfainttao ] show values of the optical depth relative to the sis for biases computed using bright and faint images . values are shown assuming @xmath170 , @xmath175 and @xmath169 . the magnification bias reduces the difference between cross - sections for different profiles , though the variation with @xmath22 and @xmath23 is still very dramatic . however , profiles with no cross - section to multiple imaging are not subject to magnification bias . magnification bias therefore increases the dependence of the cross - section to multiple imaging on @xmath22 and @xmath23 near the multiple - imaging / no - multiple - imaging boundary . miralda - escude ( 2000 ) argued that strong lensing was a powerful test of sidm models . assuming a cross - section independent of velocity , he pointed out that if sidm were responsible for the cores in dwarf galaxies , then scaling arguments lead to clusters cores that are larger than observed . while miralda - escude used a scaling law that is not a good fit to the sidm simulations ( dav@xmath1 , spergel , steinhardt & wandelt 2000 ) , his basic argument , that lensing is a powerful test of the model remains valid . furthermore , miralda - escude ( 2000 ) points out that the mis - alignment of the images in the cluster lens ms2137 - 23 can not be reconciled with a spherical lens implied by sidm halos . recently , yoshida , springel , white & tormen ( 2000 ) have produced simulations of a cluster with @xmath176 . they model the cluster using collision - less cold dark matter ( labeled s1 ) , as well as weakly self interacting cold dark matter having collision cross - sections of @xmath177 , 1.0 and @xmath178 ( labeled s1w - a , s1w - b , and s1w - c ) . halo density profiles are presented at @xmath179 for all 4 realizations . in addition , they present halo profiles for the simulation s1w - c at a range of different epochs . during the evolution of the halo , the central densities are most extreme when the cosmological expansion factor @xmath180 ( normalized to its present value ) takes the values @xmath181 and @xmath182 . the 6 halo profiles are reproduced in fig . [ yoshfit ] ( dots ) . to estimate the lensing rate due to these different profiles we have fitted each with the zhao profile ( eqn . [ prof ] ) . the fitting was achieved via minimization of @xmath183 ( solid lines ) and @xmath139 ( dashed lines ) . the constraint that @xmath184 was imposed and the halos s1w - c ( @xmath180=0.78 ) and s1w - c ( @xmath180=1.72 ) were fitted assuming an @xmath142 ( and hence @xmath18 ) corresponding to that at @xmath180=1 . yoshida et al . ( 2000 ) note that the halo approximately doubled its @xmath18 between @xmath185 and @xmath182 . if half this gain was made before ( after ) @xmath186 , then @xmath23 for the halo s1w - c ( @xmath180=0.78 ) ( s1w - c ( @xmath180=1.72 ) ) should be revised downwards ( upwards ) by @xmath187 . both the fits and fractional residuals are plotted in fig . [ yoshfit ] . the zhao profile provides an adequate fit to halos s1 and s1w - a . however for the larger interaction cross - sections , the core density is overestimated by the fitted profile . the corresponding lensing rate may therefore also be overestimated in these cases . the estimates for the profile parameters @xmath23 and @xmath22 of the yoshida , springel , white & tormen ( 2000 ) halos are plotted over contours of @xmath113 in fig . [ yosh_points ] . each parameter set is labeled and the upper and lower plots show points corresponding to the fits obtained by minimization of @xmath183 and @xmath139 respectively . from fig . [ yosh_points ] we find that @xmath67 falls off very rapidly for small @xmath22 and @xmath23 , the region of interest for sidm halos . the halos s1w - b and s1w - c at the current epoch do not produce any multiple images . the halo s1w - a ( at the current epoch ) lenses at a significantly reduced rate with respect to the cdm halo s1 . the variation in the density profile at different epochs is due to successive mergers ( yoshida et al . the spread in parameters obtained from profile fitting is representative of the range of lensing strengths expected . a population of s1w - c ( @xmath182 ) halos would yield a finite lensing rate , comparable with that of s1w - a halos , while at the other epochs considered ( @xmath181 and @xmath188 ) the halos would produce no multiple images . for comparison , we have also included in fig . [ yosh_points ] , points corresponding to parameters found by two complimentary studies of cdm halos . jing & suto ( 2000 ) have estimated of the spread in the distribution of concentration parameter for cdm halos by fitting the zhao form to 4 cluster mass halos ( @xmath189 ) using both @xmath41 and @xmath190 . their results are plotted in the lower panel of fig . [ yosh_points ] ( labeled 1 - 4 as per jing & suto ( 2000 ) ) . this range is in agreement with the 68% range for @xmath23 ( extrapolated to @xmath191 ) found by bullock et al . ( 2000 ) ( plotted in the upper panel of fig . [ yosh_points ] ) . the halo s1 has a @xmath23 which is consistent with the range of values found by these authors . it is important to note that the range of parameters describe halos that produce lensing cross - sections differing by several orders of magnitude . the lensing statistics will therefore be dominated by the more concentrated halos , indicating that the medians of profile parameter distributions may not provide a reasonable estimation of the lensing rate for a particular dark matter model . we return to this point in the following section . more detailed modeling including evolution , non - spherical mass distributions , and a range of formation red - shifts , as well as observational details including magnification bias etc . ( e.g. turner , ostriker & gott 1984 ; kochanek 1991 ) appropriate for surveys such as that being undertaken by the sdss collaboration ( e.g. york et al . 2000 ) , is required before quantitative constraints on sidm models can be derived . in addition , since lensing probes the central regions of the cluster , the contribution to the potential of a centrally dominant galaxy may be significant and must also be considered . we offer a preliminary discussion of the contribution of central galaxies in sec . [ galaxies ] , however , we conclude that the sensitivity of @xmath67 to @xmath22 and @xmath23 indicates that cluster lensing statistics will provide a powerful probe of the self - interaction cross - section . while lensing constrains the dark matter properties for dark matter collisions with velocities of @xmath192 , observations of dwarfs constrain dark matter properties with @xmath193 . if the dark matter particle is a composite particle , analogous to a nucleon with a low energy resonance , then its cross - section scales as @xmath194 for fixed cross - sections on the dwarf scale , a cross - section that is inversely proportional to the characteristic velocity will produce significantly smaller , less spherical cores in clusters and large cores in dwarfs ( yoshida , springel , white & tormen 2000 ) , consistent with observations of cluster cores like cl0024 + 1654 and cl0016 + 16 ( avila - reese , firmani , donghia & hernandez 2000 ; firmani et al . 2000 ) . alternatively , if the dark matter can self annihilate at late times ( kaplinghat , knox & turner 2000 ) then @xmath195 , this leads to low density tri - axial cores and differing lensing predictions . recent studies of n - body simulations have described the distribution of halo concentrations ( bullock et al . 2000 ; jing & suto 2000 ; dav@xmath1 , spergel , steinhardt & wandelt 2000 ) . moreover , the simulations of yoshida , springel , white & tormen ( 2000 ) demonstrate large variation in the concentration of a sidm halo during its evolution . an important quantity which has hitherto been neglected in parametric cluster lens analyses is the distribution of halo profiles . the upper panels of fig . [ od_spread ] show cross - sections @xmath113 ( thick light lines ) and optical depths @xmath196 ( thick dark lines ) to multiple imaging with respect to an sis as a function of @xmath23 for @xmath197 ( left hand panel ) and @xmath127 ( right hand panel ) . we have assumed @xmath170 , appropriate for the faint end of the quasar luminosity function . the light curves are sections through fig . [ contod ] . the corresponding curves for @xmath196 show the decreased slope of @xmath196 with @xmath23 for large @xmath23 , and the increased slope near the no - multiple - imaging cutoff discussed at the end of sec . [ amp_dist_bias ] . the upper panels also show hypothetical @xmath23 distributions having means of @xmath198 ( thin solid lines ) and @xmath199 ( thin dashed lines ) . in each case one delta function and one gaussian distribution ( half - width of @xmath200 ) are plotted . the lower panels show the dependence on @xmath201 of convolutions of @xmath113 ( light lines ) and @xmath196 ( dark lines ) with gaussian distributions of @xmath23 . the solid and dashed lines correspond to convolutions with @xmath23 distributions having @xmath198 and @xmath202 respectively . the convolutions in the lower left and right panels assumed @xmath197 and @xmath127 . the optical depth shows significant dependence on the range of @xmath23 . for example , the simulations of bullock et al . ( 2000 ) yield @xmath41 , @xmath203 and @xmath204 for cdm halos . the spread of @xmath204 results in a 2 fold increase in optical depth to multiple imaging . much more dramatic examples are obtained where @xmath205 lies close to , or below the no - multiple - imaging boundary , and the significant portion of optical depth is produced by outliers in the concentration population . this is the region of parameter space important for sidm halos . we use the s1w - b halo of yoshida , springel , white & tormen ( 2000 ) as an example . using the @xmath124 degeneracy lines in fig . [ fitprofb ] , we find that this halo may be approximated by a zhao profile with @xmath127 and @xmath206 . these parameters place the s1w - b halo in the no - multiple imaging regime . however , if halos have concentrations distributed about @xmath207 , the optical depth is finite . taking the example of @xmath208 , we find that 2 orders of magnitude separate the optical depth due to distributions of halos with @xmath209 and @xmath200 , and that 3 orders of magnitude separate the optical depth between distributions having @xmath209 and @xmath210 . note that the s1w - c halo of yoshida et al . ( 2000 ) has a concentration that differs by up to @xmath211 from the present day value during its evolution . these results demonstrate that knowledge of the distribution of halo concentrations is an important consideration for cluster lensing studies . the statistics of multiple imaging due to clusters of galaxies are sensitive to the inner most regions of the mass distribution , where the contribution of a central galaxy may be important . one expects that this is particularly true where the halo is less concentrated or or has a shallower cusp , as is the case for sidm halos . the presence of an isothermal central galaxy renders the cross - section to multiple imaging finite for all halos . williams , navarro & bartelmann ( 1999 ) have investigated the effect of including a massive central galaxy . they find that arc properties of the most massive clusters are consistent with predictions of the nfw profile , but that observed separations are too large in the lower mass cluster lenses . they demonstrate that a centrally dominant galaxy with a dark halo can resolve the discrepancy . however there are several caveats . as noted by williams et al . ( 1999 ) , their cluster sample over represents the expected fraction of massive clusters ( based on the algorithm of press & schechter 1974 ) , possibly indicating a selection bias towards large splittings and highly elongated arcs . furthermore , the higher resolution of simulations of moore et al . ( 1998 ) show a steep core @xmath9 , indicating that the concentrations found by nfw may be a factor of @xmath212 too low . this reduces the required mass of the centrally dominant galaxy . more importantly , bullock et al . ( 2000 ) find significant scatter in concentration for a halo of given mass . high concentration halos are more likely to lens , produce larger image splittings , and so are more likely to be observed as cluster lenses . in the previous section this was shown to be a significant effect and might give the appearance that the profiles found from n - body simulations can not account for observations of cluster lenses . on the other hand , bullock et al . ( 2000 ) also find a concentration that decreases faster with redshift than originally thought ( navarro , frenk & white 1997 ) . this will exasperate the problem pointed out by williams et al . ( 1999 ) for cluster lenses above a redshift of @xmath213 . on the other hand , a cluster need not have a single centrally dominant galaxy ( e.g. cl0024 + 1654 ) . flores , maller & primack ( 2000 ) have analyzed the inclusion of cluster galaxies on the statistics of giant arc properties . they find that observationally based constraints suggest that there are not enough massive galaxies in a cluster to significantly alter arc statistics . in a complementary study , menghetti et al . ( 2000 ) artificially added galaxies to n - body simulations . when the properties of short arcs are excluded , their analysis suggests that the contribution of cluster galaxies is negligible . a proper account of the cluster density profile in the presence of a centrally dominant galaxy requires a full treatment of galaxy formation at the center of the cluster . the studies mentioned artificially add intrinsically or extrinsically truncated parametric galactic halos to the cluster halo . williams , navarro & bartelmann ( 1999 ) ( following navarro , frenk & white ( 1997 ) ) take the additional step of modifying the cluster halo adiabatically in response to the formed central galaxy . profiles with large and small @xmath22 describe @xmath214 and @xmath215 halos respectively . the response of the dark matter to the formation of a central galaxy can not be commonly approximated in these two cases . in cdm halos , the central galaxy may form adiabatically ( see blumenthal , faber , flores & primack 1984 ) , resulting in additional contraction of the cluster dark matter halo . in this scenario , it is possible that the central galaxy may retain its own halo if it is not tidally striped during formation . note however that n - body simulations of cluster halos are expected to properly describe the halo of the central galaxy , since the cusp in the cluster density profile is formed from the infall of smaller halos . to reconcile observations with cdm nfw halo models , williams et al . ( 1999 ) find that an additional central cluster galaxy of mass @xmath216is required . with a stellar velocity dispersion of @xmath217 , this implies that the central galaxy must retain its halo during formation . to find an approximate contribution to the multiple imaging cross - section we have taken the most simple approach and superimposed the dark matter halo profile of brainerd , blandford & smail ( 1996 ) , @xmath218 onto the center of the zhao profile . here @xmath219 is the central velocity dispersion and @xmath220 is the characteristic outer scale on the center of the zhao profile . the profile has a total mass @xmath221 we have calculated cross - sections to multiple imaging for zhao profiles having @xmath222 in the presence of an additional central galaxy of mass @xmath223 with a central stellar velocity dispersion of @xmath217 . the resulting cross - section calculations are applicable to zhao profiles with large @xmath22 and @xmath23 appropriate for cdm halos . the left hand panels of tabs . [ cdtab ] and [ cdtabrel ] show values of the cross - section to multiple imaging @xmath224 , and the fractional increase in cross - section @xmath225 over a small grid of @xmath22 and @xmath23 . in the self interacting scenario , the response of the halo to the formation of the central galaxy should be larger . the contraction heats up the dark matter , after which interactions transfer the heat outwards , resulting in cooling and additional halo contraction . an isothermal approximation is therefore appropriate for the response of the sidm halo to the formation of a central galaxy . consider an initial dark matter density profile resulting from an isothermal distribution function : @xmath226 where @xmath227 is boltzmann s constant and @xmath228 the temperature . the ratio of the dark matter density before and after the addition of baryons is @xmath229 thus the density increase of the cluster halo with @xmath230 and central galaxy velocity dispersion @xmath231 , the density increase is of order 20% . this density increase is negligible with respect to the orders of magnitude which separate the densities of different profiles in the inner most regions ( see fig . [ profsurf ] ) . if the cluster has a sidm halo , the central galaxy can not have retained its own halo during formation . the additional galactic mass should therefore consist only of stars , and is thus more confined . we have calculated cross - sections to multiple imaging for zhao profiles having @xmath222 in the presence of an additional central of mass @xmath232 ( appropriate for the mass in stars of a typical central galaxy ) with a central stellar velocity dispersion of @xmath217 . the calculation is applicable to all the zhao profiles assuming that the central dark - matter distribution is described solely by the zhao profile . the right hand panels of tabs . [ cdtab ] and [ cdtabrel ] show the resulting values of the cross - section to multiple imaging @xmath224 , and the fractional increase in cross - section @xmath225 where a central galaxy is included over a small grid of @xmath22 and @xmath23 . @xmath225 is of order 1 for all but the shallowest halo profiles . the addition of the extra mass increases the cross - section of all profiles with respect to the sis . we have not re - normalized the mass of the zhao - profile - plus - galaxy as we are interested in the relative rather than the absolute cross - sections of different zhao - profile - plus - galaxies . the addition of the central isothermal galaxy sets a lower limit on the cross - section , and reduces the dependence of multiple imaging on the cluster profile parameters , though the differences are still orders of magnitude . the values in tab . [ cdtab ] may be compared directly with the magnification bias ( tabs . [ abbright ] and [ abfaint ] ) . these show that the contribution of the central galaxy to the optical depth is not as important as that of magnification bias . note that while magnification bias and the addition of a central galaxy both preferentially increase the cross - section of less concentrated clusters , the addition of a central galaxy reduces the magnification bias due to the increased cross - section . thus , while a self - consistent calculation is yet to be done , we conclude that centrally dominant galaxies will not prevent lens statistics from being used as a probe of sidm halos . we have calculated the differential and total optical depths to multiple imaging , the average image splitting and the total magnification for a constant co - moving number density of unevolving generalized nfw ( zhao ) profile cluster mass gravitational lenses . we find that the number of expected strongly lensed quasars is a very sensitive function of the profile parameters . profiles whose central density is low either due to a shallow central cusp , or to a scale radius that is a reasonable fraction of the virial radius have cross - sections to multiple imaging that are reduced by a significant factor . moreover , the separation of multiple images is reduced ( by a factor of a few ) , although the total magnification is significantly enhanced . we find that the resulting magnification bias does not alter our conclusions . the zhao profile exhibits degeneracies between profile parameters with respect to lensing statistics . similarly , profile fits have parameter degeneracies which are a function of the minimization quantity adopted . if the properties of a profile core are accurately reproduced by an approximately degenerate profile , then the parameter degeneracies are nearly equivalent to those obtained from strong lensing statistics . however , a profile that attempts to fit the entire halo will introduce serious uncertainty into the inferred lensing rates ( up to an order of magnitude ) . this is particularly true in the region of parameter space where both @xmath23 and @xmath22 are small and will be an important consideration for detailed calculations of the lensing rate for sidm halos based on parametric results from n - body simulations . the lensing rate is a powerful probe of sidm in clusters of galaxies , particularly in the absence of a centrally dominant galaxy . we have obtained profile parameters for halos around clusters of galaxies by fitting zhao profiles to the simulations of yoshida , springel , white & tormen ( 2000 ) . we find that the optical depth to multiple imaging seriously constrains the sidm self - interaction cross - section . in particular , an interaction cross - section of @xmath178 is rarely capable of producing multiple images during its evolution . in the most highly concentrated phases , a halo composed of interacting dark matter with this cross - section has a rate of multiple imaging 1 to 2 orders of magnitude lower than the corresponding typical cdm halo . preliminary calculations show that centrally dominant galaxies increase the multiple imaging cross - section , but that the increase is small with respect to the variation in cross - section among different profiles . the presence of centrally dominant galaxies should not therefore inhibit the use of the multiple imaging rate as a probe of sidm . an important result from this study , with implications for cluster lensing studies is that the scatter in cdm / sidm profile parameters obtained from n - body studies describe a very large range of optical depth . the lensing statistics will therefore be dominated by the more concentrated members of the population , rather than by the typical halo . as a result , an estimate of the distribution of halo profiles must be included in future studies of cluster lensing statistics . the authors would like to thank romeel dav@xmath1 , bartosz pindor and daniel mortlock for helpful and stimulating discussions . we would also like to thank the anonymous referee whose comments led to the improvement of this work . this research was supported by nsf grant ast98 - 02802 to elt . jsbw acknowledges the support of an australian postgraduate award and a melbourne university overseas research experience award . avila - reese , v. , firmani , c. , donghia , e. , hernandez , x. , 2000 , in `` the seventh texas - mexico conference on astrophysics : flows , blows , and glows '' ( april 2000 ) , eds . w. lee & s. torres - peimbert , revmexaa ( serie deconferencias ) cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&0.000&0.008&0.174&&na & 26.52&46.64&&na & 75.41&18.46 + @xmath22&1.0&&0.001&0.199&0.918&&11.56&43.06&57.13&&100.1&16.11&8.105 + & 1.5&&0.226&1.715&4.263&&34.91&57.79&68.71&&12.33&5.732&3.861 + cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&na & 5.53 & 2.50 & & na & 35.57&7.35 & & na & 282.1&27.45 + @xmath22&1.0&&6.27 & 2.53 & 1.89 & & 45.32&7.50 & 4.13 & & 398.1&27.91&11.18 + & 1.5&&2.32 & 1.68 & 1.45 & & 6.24 & 3.22 & 2.38 & & 20.50&7.54 & 4.62 + cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&na & 3.76 & 1.70 & & na & 13.34&2.76 & & na & 49.51&4.82 + @xmath22&1.0&&4.27 & 1.73 & 1.29 & & 16.99&2.81 & 1.55 & & 69.88&4.90 & 1.96 + & 1.5&&1.57 & 1.15 & 0.99 & & 2.34 & 1.20 & 0.89 & & 3.60 & 1.32 & 0.81 + cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&na & 4.94 & 2.35 & & na & 29.23&6.55 & & na & 227.5&24.00 + @xmath22&1.0&&5.94 & 2.36 & 1.56 & & 41.98&6.59 & 2.97 & & 364.7&24.12&7.90 + & 1.5&&1.96 & 2.22 & 0.91 & & 4.61 & 1.87 & 1.10 & & 14.76&4.25 & 2.13 + cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&na & 6.73 & 3.20 & & na & 30.90&6.93 & & na & 101.7&10.74 + @xmath22&1.0&&8.10 & 3.21 & 2.13 & & 43.84&6.96 & 3.14 & & 163.1&10.79&3.53 + & 1.5&&2.67 & 1.66 & 1.24 & & 4.88 & 1.98 & 1.16 & & 6.60 & 1.90 & 0.95 + cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&0.000&0.030&0.296&&0.000&0.107&0.480&&0.000&0.396&0.839 + @xmath22&1.0&&0.004&0.344&1.184&&0.017&0.559&1.423&&0.070&0.975&1.799 + & 1.5&&0.355&1.972&4.220&&0.529&2.058&3.794&&0.814&2.264&3.453 + cccccccccccccc & & & & & & & + & & & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&0.000&0.054&0.557&&0.000&0.247&1.206&&0.000&0.814&1.869 + @xmath22&1.0&&0.008&0.639&1.955&&0.044&1.385&2.883&&0.163&2.147&3.240 + & 1.5&&0.603&2.847&5.286&&1.103&3.396&4.945&&1.492&3.259&4.050 + cccccccccc & & & & & + & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5&&0.008&0.026&0.233&&0.008&0.015&0.188 + @xmath22&1.0&&0.014&0.266&1.054&&0.010&0.217&0.952 + & 1.5&&0.306&1.916&4.581&&0.257&1.778&4.364 + cccccccccc & & & & & + & & & & & + & & & 5.0 & 12.5 & 20.0 & & 5.0 & 12.5 & 20.0 + & 0.5 & & na&3.250&1.339 & & na&1.875&1.080 + @xmath22&1.0&&14.00&1.337&1.148&&10.00&1.090&1.037 + & 1.5&&1.354&1.117&1.075&&1.137&1.037&1.024 +

a timelike metric space is a hausdorff topological space equipped with a partial order relation and a distance function satisfying a set of axioms including certain compatibility conditions between these two objects . the distance function is defined only on a certain subset ( whose definition uses the partial order ) of the product of the space with itself containing the diagonal . distances between triples of points , whenever they are defined , satisfy the so - called _ time inequality _ , which is a reversed triangle inequality . in the 1960s , herbert busemann developed an axiomatic theory of timelike metric spaces and of locally timelike metric spaces . his motivation comes from the geometry underlying the theory of relativity and the classical example he gives is the-dimensional lorentzian spaces . two other interesting classes of examples of timelike metric spaces introduced by busemann are the timelike analogues of the funk and hilbert geometries . in this paper , we investigate these geometries . we shall find new interactions among the euclidean , affine , projective and spherical timelike geometries . in particular , the de sitter metric is described as a special case of a timelike spherical hilbert metric . _ keywords. _ timelike space , timelike hilbert geometry , timelike funk geometry , time inequality , convexity , metric geometry , busemann geometry , lorentzian geometry , relativity . _ ams classification. _ 53c70 , 53c22 , 5cc10 , 53c23 , 53c50 , 53c45 .

1602.07072

a timelike metric space is a hausdorff topological space @xmath1 equipped with a partial order relation @xmath2 and a distance function @xmath3 which plays the role of a metric . this metric is asymmetric in the sense that the distance from @xmath4 to @xmath5 is not necessarily equal to the the distance from @xmath5 to @xmath4 . more drastically , the distance from @xmath4 to @xmath5 may be defined whereas the distance from @xmath5 to @xmath4 is not defined . more precisely , the distance @xmath6 is defined only for pairs @xmath7 satisfying @xmath8 ( that is , either @xmath9 ot @xmath10 ) . this distance function satisfies @xmath11 for every @xmath4 in @xmath1 , @xmath12 for every @xmath4 and @xmath5 such that @xmath10 , and the _ time inequality _ , namely , the reversed triangle inequality @xmath13 for all triples of points @xmath14 satisfying @xmath15 . the distance function @xmath3 and the order relation @xmath2 satisfy a set of axioms including compatibility conditions with respect to each other . for instance , it is required that every neighborhood of a point @xmath16 in @xmath1 contains points @xmath4 and @xmath5 satisfying @xmath17 . this axiom and others are stated precisely in the memoir @xcite by herbert busemann . the theories of timelike metric spaces , timelike @xmath18-spaces , locally timelike metric spaces and locally timelike @xmath18-spaces initiated by busemann generalize the geometric theories of metric spaces and of @xmath18-spaces that he developed in his book @xcite and in other papers and monographs . the motivation for the study of timelike metrics comes from the geometry underlying the theory of relativity . the classical example is the @xmath0-dimensional lorentzian space . as other interesting examples of timelike metric spaces , busemann introduced timelike analogues of the funk and hilbert geometries . in the present paper , we investigate these geometries . we establish several results concerning their geodesics , their convexity properties and their infinitesimal structure . we show in particular that they are finsler metrics in an appropriate sense ( we call them timelike finsler metrics ) . we also give a description of the de sitter space as a special case of a more general timelike hilbert geometry in a spherical geometry setting . we first introduce some preliminary notions and we establish some basic facts . with few exceptions , we shall use busemann s notation in @xcite , and we first recall it . let @xmath19 be a convex hypersurface in @xmath20 , that is , the boundary of an open ( possibly unbounded ) convex set @xmath21 . for any hypersurface @xmath19 which is not a plane , there is a well - defined associated open convex set @xmath22 . this is the convex connected components of @xmath23 . if @xmath19 is a hyperplane , the connected components of @xmath23 are both convex , and in this case we make a choice of one of the two connected components , that is , a half - space bounded by the hyperplane @xmath19 . we call the set @xmath22 associated to @xmath19 the _ interior _ of @xmath19 . we denote the closure @xmath24 of @xmath22 by @xmath25 . let @xmath26 be the set of supporting hyperplanes of @xmath19 , that is , the hyperplanes @xmath27 having nonempty intersection with @xmath19 and such that the open convex set @xmath22 is contained in one of the two connected components of @xmath28 . for each @xmath29 , we let @xmath30 be the open half - space bounded by the hyperplane @xmath27 and containing @xmath22 , and @xmath31 the open half - space bounded by @xmath27 and not containing @xmath22 . we have : @xmath32 furthermore , if @xmath33 is the complement of @xmath25 in @xmath34 , we have @xmath35 we now introduce a partial order relation on @xmath36 . for a point @xmath37 in @xmath36 , we let @xmath38 be the union of all the rays originating at @xmath37 and intersecting @xmath19 . clearly @xmath39 is convex and has nontrivial interior . depending on the shape of @xmath19 and the relative position of @xmath37 and @xmath19 , @xmath40 may or may not be closed in @xmath36 . we denote by @xmath41 the set of interior points of @xmath39 . for any two distinct points @xmath37 and @xmath16 in @xmath36 , we write @xmath42 and whenever this occurs , we say that @xmath16 _ lies in the future of _ we also say that @xmath37 _ lies in the past of _ @xmath16 ( see figure [ future ] ) . we write @xmath43 if either @xmath44 or @xmath45 . geometrically speaking , when @xmath46 , the ray @xmath47 from @xmath37 through @xmath16 intersects the hypersurface @xmath19 transversally . we denote by @xmath48 ( resp . @xmath49 ) the set of ordered pairs @xmath50 in @xmath51 satisfying @xmath44 ( resp . @xmath43 ) . since the set @xmath52 is open for each @xmath37 , the set @xmath48 is open in the product space @xmath51 . it is disjoint from the diagonal set @xmath53 . finally , for any a point @xmath37 in @xmath34 , we denote by @xmath54 the set of supporting hyperplanes @xmath55 that separate @xmath19 from @xmath37 . in other words , the set @xmath54 is characterized by @xmath56 we note that a supporting hyperplane @xmath27 which contains @xmath37 and is tangential to @xmath19 does not belong to the set @xmath54 . we have the following : [ p : inclusion ] for any two points @xmath37 and @xmath16 in @xmath36 , we have @xmath57 first suppose @xmath46 . we claim that every @xmath58 is an element of @xmath54 . indeed , if this does not hold , then there exists @xmath58 such that @xmath59 . therefore @xmath16 lies in @xmath31 by definition , @xmath37 lies in @xmath60 , and thus @xmath37 and @xmath16 lie across @xmath27 from each other . in particular the ray @xmath47 intersects @xmath19 transversally on the @xmath61 side , which contradicts the fact that @xmath62 . as for the converse , we argue by contraposition : if the ray @xmath47 does not intersect @xmath19 transversally ( hence @xmath63 ) then there exists some @xmath27 which lies in @xmath64 but not in @xmath54 . such a supporting hyperplane @xmath58 intersects the interior of the segment @xmath65 $ ] . by definition , @xmath16 lies on the other side of @xmath27 from @xmath19 . it follows that @xmath37 lies on the same side of @xmath19 with respect to @xmath27 . hence the hyperplane @xmath27 is an element of @xmath66 , namely @xmath67 . let @xmath68 be three points in @xmath36 . if @xmath46 and @xmath69 , then @xmath70 . the proof follows from proposition [ p : inclusion ] , since it gives : @xmath71 now we can define the timelike funk metric @xmath72 on the subset @xmath49 of @xmath73 . [ def : funk1 ] the function @xmath74 on pairs of distinct points @xmath75 in @xmath36 satisfying @xmath46 is given by the formula @xmath76 where @xmath77 is the point where the ray @xmath47 first hits the convex hypersurface @xmath19 and @xmath78 denotes the euclidean distance . note that @xmath79 exists since @xmath46 . as the ray @xmath80 intersects @xmath19 transversally , we have the strict inequality @xmath81 , hence the value of @xmath74 is strictly positive . for @xmath45 , we extend the definition by setting @xmath82 . we remark at this point that the metric space together with the partial order relation we have defined satisfies the three axioms @xmath83 and @xmath84 of busemann s @xcite . ( but to for the purpose of the present paper , the reader need not go through the general axiomatics of busemann since we are only concerned here with specific examples . ) let @xmath85 be a supporting hyperplane to @xmath19 at @xmath77 , namely @xmath86 . for @xmath37 in @xmath87 , let @xmath88 be the foot of the point @xmath37 on the hyperplane @xmath85 . in other words , @xmath89 is the nearest point projection map . from the similarity of the triangles @xmath90 and @xmath91 , we have @xmath92 using the convexity of @xmath19 , the quantity @xmath74 can be characterized variationally as follows . for any unit vector @xmath93 in @xmath34 and for any @xmath55 , define @xmath94 by @xmath95 . for @xmath96 in @xmath34 , consider the vector @xmath97 where the norm is the euclidean one . when the hyperplane supports @xmath36 at @xmath98 , we have @xmath99 and otherwise the point @xmath100 lies outside @xmath36 . when @xmath55 supports the convex set at some point other than @xmath37 , again by the similarity of the triangles @xmath101 and @xmath102 , we get @xmath103 note that the furthest point from @xmath37 on the ray @xmath104 of the form @xmath105 is @xmath98 . this in turn says that a hyperplane @xmath106 which supports @xmath36 at @xmath98 minimizes the ratio @xmath107 among all the elements of @xmath26 ; @xmath108 notice that there is an analogous formula for the classical ( non - timelike ) funk metric , where the infimum in the above formula is replaced by a supremum , see @xcite theorem 1 . as @xmath109 ( proposition [ p : inclusion ] ) , we have the following identification : @xmath110 [ rem : future ] the set of future points of a point @xmath37 , that is , the set of points @xmath16 satisfying @xmath44 , has the structure of a cone @xmath39 with nonempty interior @xmath111 this reminds us of the lorentzian space , which is the geometric setting of spacetime for the theory of relativity , where the cone of future points of @xmath37 is the ambient space of the physically possible trajectories of this point . the restriction of the distance function to the cone comes from the fact that a material particle travels at a speed which is less than the speed of light . thus , the set of points on the rays starting at @xmath37 that are on the boundary @xmath112 of the solid cone @xmath39 becomes an analogue of the light cone " of spacetime ( again using the language of relativity ) . however , in the case of lorentzian geometry , the distance between @xmath37 and a point on boundary of the light cone is zero , whereas in our setting it is undefined ( recall that the distance between two points @xmath37 is defined whenever we have @xmath46 ) . in conclusion , there are analogies between busemann s timelike theory and the geometric setting of relativity , but one has to be careful about several details . equation ( [ eq : ff ] ) may be used as a definition of the funk metric @xmath72 . one advantage would be that in the case where @xmath19 is unbounded , it allows us to naturally extend the domain of definition for @xmath113 so that the light cone of a point @xmath37 is defined to be the set of points @xmath114 satisfying @xmath115 . we prove that the function @xmath72 satisfies the reverse triangle inequality , which we call in this context , after busemann , the _ time inequality_. this inequality holds for mutually distinct points @xmath75 and @xmath116 in @xmath36 , satisfying @xmath117 : for any three points @xmath75 and @xmath116 in @xmath36 , satisfying @xmath117 , we have @xmath118 we use the formula ( [ eq : ff ] ) for the timelike funk distance . we have , from @xmath119 ( proposition [ p : inclusion ] ) : @xmath120 in the rest of this section , we study geodesics and spheres in timelike funk geometries . we shall refer to the paper @xcite for the corresponding results in the classical funk setting , and all the properties are motivated by busemann s work @xcite . first we consider geodesics for the timelike funk metric . we start with the definition of a geodesic . this definition is the same as in ordinary metric spaces , except that some care has to be taken so that the distances we need to deal with are always defined . a _ geodesic _ is a path @xmath121 , where @xmath122 may be an arbitrary interval of @xmath123 , such that for every pair @xmath124 in @xmath122 we have @xmath125 and for any triple @xmath126 in @xmath122 we have @xmath127 it follows easily from the definition that for any @xmath44 the euclidean segment @xmath65 $ ] joining @xmath37 to @xmath16 is the image of a geodesic . this makes the distance function @xmath113 projective " in the sense of hilbert s fourth problem @xcite if this problem is extended in an appropriate way to timelike spaces . in particular , the time inequality becomes an equality when @xmath37 , @xmath16 and @xmath116 satisfying @xmath128 are collinear in the euclidean sense . let us make an observation which concerns the non - uniqueness of geodesics and the case of equality in the time inequality . assume that the boundary of the convex hypersurface @xmath19 contains a euclidean segment @xmath129 . take three points @xmath68 in a small neighborhood @xmath130 of a point on @xmath129 , such that the euclidean distance of @xmath68 to @xmath129 taken in that order is decreasing ( figure [ segment ] ) . then , using the euclidean intercept theorem , we have @xmath131 applying the same reasoning to an arbitrary ordered triple on the broken euclidean segment @xmath65\cup[q , r]$ ] , we easily see that this segment is an @xmath113-geodesic . more generally , by the same argument , we see that _ any _ oriented arc in @xmath130 whose points get strictly closer to the segment @xmath129 when we move along this arc , is the image of an @xmath113-geodesic . in fact , the situation in the neighborhood @xmath130 is very special . it is a case where the timelike funk and the classical funk metrics coincide , distances being defined from a fixed point @xmath37 to points in an open cone centered at this point . ( notice that for this to happen , it is necessary that the time and the triangle inequalities hold , therefore they must both be equalities . ) from the preceding remarks on euclidean segments we can deduce the following : a timelike funk metric @xmath113 defined on a set @xmath132 associated to a convex hypersurface @xmath19 in @xmath34 satisfies the following properties : 1 . [ eucl1 ] the euclidean segments in @xmath36 that are of the form @xmath65 $ ] where @xmath44 are @xmath113-geodesics . any euclidean line @xmath133 from a point @xmath37 in @xmath36 to a point @xmath134 in @xmath135 , equipped with the metric induced from the timelike funk metric , is isometric to a euclidean ray . the euclidean segments in ( [ eucl1 ] ) are the unique @xmath113-geodesic segments if and only if the convex set @xmath22 is strictly convex . the proof is the same as that of the equivalence between ( 1 ) and ( 2 ) in corollary 8.7 of @xcite , up to reversing some of the inequalities ( i.e. replacing the triangle inequality by the time inequality ) . after the geodesics , we consider spheres . at each point @xmath37 of @xmath36 , given @xmath136 , a _ ( geometric ) future sphere _ of radius @xmath116 is the set of points in @xmath36 that are in the future of @xmath37 and situated at distance @xmath116 from this point . at each point @xmath37 of @xmath36 and for each @xmath136 , the geometric future sphere of center @xmath37 and radius @xmath116 is a piece of a convex hypersurface that is affinely equivalent to the portion @xmath137 , namely the part of the hypersurface @xmath19 that is visible from @xmath37 . the proof is analogous to that of proposition 8.11 of @xcite , and we do not repeat it here . this proposition implies that some affine properties of the hypersurface @xmath19 are local invariants of the metric . one consequence is the following strong local rigidity theorem , which is also an analogue of a property satisfied by the classical funk metric ( cf . the concluding remarks of the paper @xcite ) . let @xmath19 and @xmath138 be two hypersurfaces in @xmath34 and @xmath49 and @xmath139 the set of corresponding pairs of points for which the associated funk metrics @xmath113 and @xmath140 respectively are defined . if there exists subsets @xmath141 and @xmath142 and a map @xmath143 which is distance - preserving , then there is an open subset of @xmath19 which is affinely equivalent to an open subset of @xmath144 . the proof follows from the the fact that an isometry sends a future sphere to a future sphere . the corollary implies for instance that if @xmath145 is the boundary of a polyhedron and @xmath144 a strictly convex hypersurface , then there is no local isometry between the associated timelike metric spaces . the next statement is a simple observation in convex geometry , and it gives a monotonicity result for a pair of timelike funk geometries . [ lem : another ] given as before a convex set @xmath22 with associated funk distance @xmath113 , let @xmath146 be another convex set and @xmath147 the associated timelike distance function defined between appropriate pairs @xmath50 . suppose that @xmath37 and @xmath16 are in the domain of definition of both distances @xmath113 and @xmath148 , that is , @xmath44 with respect to the convex sets @xmath22 and @xmath149 . then we have @xmath150 using the notation of definition [ def : funk1 ] , we have @xmath151 with similar notation , we have @xmath152 since @xmath146 , we have @xmath153 and @xmath154 for some @xmath155 . the result follows from the fact that the function defined for @xmath155 by @xmath156 where @xmath157 are two constants is increasing . the timelike funk metric associated to a convex hypersurface @xmath36 in @xmath34 is finsler in an appropriate sense which we describe now . on the tangent space at each point of @xmath158 , there is a _ timelike norm - function _ which we also call a _ ( timelike ) minkowski functional _ , in analogy with the minkowski functional associated to a finsler structure in the classical sense . we show that the timelike funk distance @xmath72 between two points @xmath37 and @xmath16 is obtained by integrating this norm on tangent vectors along piecewise @xmath159 paths joining @xmath37 to @xmath16 and taking the supremum ( instead of the infimum , in the classical case ) of the lengths of such piecewise @xmath159 paths . we call such a structure a _ timelike finsler structure_. to make this more precise we start with the following definition ( cf [ def : min ] a _ timelike minkowski functional _ is a function @xmath160 ( which plays the role of a norm " in this timelike setting ) satisfying the following : 1 . @xmath160 is defined on an open convex cone @xmath161 having the origin as apex ; 2 . @xmath162 for all @xmath4 in @xmath163 ; 3 . @xmath164 for all @xmath4 in @xmath163 and @xmath165 ; 4 . @xmath166 for all @xmath167 . the _ unit sphere _ @xmath168 of such a norm function @xmath160 is the set of vectors @xmath4 in @xmath163 satisfying @xmath169 . in general , @xmath168 is a piece of a hypersurface in @xmath34 which is concave when viewed from the origin @xmath170 ( see figure [ concave ] ) . we allow the possibility that @xmath168 is asymptotic to the boundary of the cone @xmath163 . this surface is called the _ indicatrix _ of @xmath160 . the reason of the adjective _ timelike _ in the above definition is that in the lorentzian setting , the minkowski norm measures the lengths of vectors which are in the timelike cone , which is the part of spacetime where material particles move . in particular , there is a timelike minkowski functional @xmath160 for the standard minkowski space @xmath171 , equipped with the minkowski metric @xmath172 it is given by @xmath173 and it is defined for non - zero vectors @xmath4 in @xmath174 satisfying @xmath175 . we now study the timelike minkowski norms on tangent spaces of funk geometries . for every point @xmath37 in the timelike funk geometry of a space @xmath49 associated to a convex hypersurface @xmath19 , there is a _ timelike minkowski functional _ @xmath176 defined on the tangent space @xmath177 of @xmath36 at @xmath37 consisting of the non - zero vectors @xmath178 satisfying @xmath179 where we recall that @xmath52 is the ( nonempty ) interior of the cone spanned by @xmath19 with vertex @xmath37 . we denote the set of such vectors @xmath178 by @xmath180 . the function @xmath176 for a given point @xmath181 and for any vector @xmath182 is given by the following formula : @xmath183 for @xmath182 and @xmath184 when @xmath185 , where @xmath186 is as in ( [ eq : p ] ) above . by elementary geometric arguments ( see @xcite for a detailed discussion in the classical case which can be adapted to the present setting ) it is shown that @xmath187 for any nonzero vector @xmath182 . note that the quantity @xmath188 in the denominator is the length of the line segment from @xmath37 to the point where the ray @xmath189 hits the convex set @xmath25 for the first time . it is clear that the functional @xmath190 defined on the open cone @xmath191 in @xmath192 satisfies all the properties required by a timelike minkowski functional . we say that a piecewise @xmath159 curve @xmath193 in @xmath36 , defined on an interval @xmath122 of @xmath123 , is _ timelike _ if at each time @xmath194 the tangent vector @xmath195 is an element of @xmath196 . as the set @xmath191 is a subset of an open cone bounded by a concave hypersurface @xmath197 , the sub - level set @xmath198 is open in each tangent space @xmath199 , implying that the minkowski functional @xmath176 is upper semi - continuous . using this upper semi - continuity we define the _ length _ of a piecewise @xmath159 timelike curve @xmath200 by the lebesgue integral @xmath201 by standard arguments , again using the upper semi - continuity of @xmath202 , it follows that the length functional is upper semi - continuous in the @xmath203-topology on curves ( for detailed treatments , see @xcite or @xcite . ) we then define a distance function " @xmath204 for pairs of points @xmath205 satisfying @xmath206 by setting @xmath207 where the supremum is taken over all the timelike piecewise @xmath159 curves @xmath208\to \omega$ ] satisfying @xmath209 and @xmath210 . it is easy to check that the distance function @xmath204 is a timelike metric on the space @xmath49 . this timelike metric is the analogue of the so - called _ intrinsic metric _ in the classical ( non - timelike ) case . we call @xmath204 the _ timelike intrinsic metric_. we shall prove that this distance function coincides with the funk timelike metric . we state this as follows : [ th : distance ] the value of the distance @xmath211 for a pair @xmath212 coincides with @xmath74 . that is , we have @xmath213 in other words , we have the following [ funk - finsler ] the timelike funk metric is a timelike finsler metric with timelike finsler structure given by the minkowski functional @xmath176 . the minkowski functional @xmath176 which underlies a funk metric has a property which is analogous to the one noticed in @xcite which makes this metric the _ tautological finsler structure _ associated to the hypersurface @xmath19 ( or the convex body @xmath22 ) . tautological " is due to the fact that the indicatrix of the timelike minkowski functional at @xmath181 , that is the set @xmath214 is naturally identified with the subset @xmath197 of the hypersurface @xmath19 . recall that @xmath52 is the open cone centered at @xmath37 spanned by @xmath19 . we also note that with this identification , given a pair of points @xmath75 with @xmath46 , there always exists a distance - realizing ( length - maximizing ) geodesic from @xmath37 to @xmath16 , since the euclidean segment @xmath65 $ ] is an @xmath113-geodesic . in the theory of relativity , the existence of geodesics is established by imposing the so - called global hyperbolicity ( @xcite or @xcite ) of the spacetime . the global hyperbolicity , roughly put , consists of two conditions : the non - existence of closed causal curves and the sequential compactness of geodesics in @xmath203-topology . in our setting , the first condition is satisfied , while the second is not always guaranteed due to the lack of assumptions on the set @xmath112 . for a pair of points @xmath205 with @xmath44 , let @xmath47 be the ray from @xmath37 through @xmath16 and let @xmath79 be the first intersection point of this ray with the convex set @xmath25 . parameterize the euclidean segment @xmath65 $ ] by a path @xmath215 having euclidean arc - length parameter @xmath216 with @xmath217 then we have @xmath218 since @xmath219 by taking the supremum over the set of paths from @xmath37 to @xmath16 , this implies the inequality @xmath220 before continuing the proof of theorem [ th : distance ] , we show the following monotonicity for the intrinsic distance . [ lem : another ] for an open convex set @xmath22 bounded by @xmath19 , consider the associated timelike minkowski functional @xmath221 and the intrinsic distance @xmath204 induced by @xmath202 . also let @xmath146 be another open convex set bounded by @xmath222 , @xmath223 its associated timelike minkowski functional , and @xmath224 the intrinsic distance . suppose that @xmath37 and @xmath16 are in the domain of definition of both intrinsic distances @xmath204 and @xmath224 . then we have @xmath225 the set of admissible paths from @xmath37 to @xmath16 for the convex set @xmath149 is larger than the set of paths for @xmath149 . this is due to the inclusion @xmath226 for every @xmath4 in @xmath36 . also between the two timelike minkowski functionals @xmath202 and @xmath223 , we have the following inequality @xmath227 whenever the two quantities are defined . this follows from the definition of the minkowski functional : @xmath187 for any nonzero vector @xmath182 , as @xmath222 is closer to @xmath37 than @xmath19 . hence by integrating each functional along an admissible path and taking supremum over these paths , we have @xmath228 _ proof of theorem [ th : distance ] continued_. suppose that we have a convex hypersurface @xmath19 bounding an open convex set @xmath229 , and for @xmath230 , let @xmath231 , where @xmath232 is the open half - space bounded by a hyperplane @xmath106 supporting @xmath19 at @xmath98 . the open set @xmath149 comes with its intrinsic distance @xmath224 . we now apply lemma [ lem : another ] to this setting where a convex set @xmath149 contains @xmath22 , and obtain @xmath233 . now for the open half space @xmath234 , the values of @xmath74 , @xmath147 and @xmath211 all coincide , since under the hypothesis @xmath234 , the set @xmath235 of supporting hyperplanes consists of the single element @xmath106 , and the path @xmath236 from @xmath37 to @xmath16 is a length - maximizing path . also from the lemma above , we have @xmath237 . hence we have @xmath238 and the equality @xmath239 follows . we end this section by the following convexity result for the timelike funk distance : given a point @xmath4 in @xmath36 and @xmath240 , the set of points @xmath241 is a convex set in @xmath242 . given @xmath243 and @xmath244 in @xmath245 , parameterize the euclidean segment @xmath246 $ ] with an affine parameter @xmath247 $ ] . we will show that the function @xmath248 is convex in @xmath216 . fix a supporting hyperplane @xmath27 in @xmath249 . then @xmath250 and @xmath251 where @xmath252 is the unit vector at @xmath4 perpendicular to the hypersurface @xmath27 oriented toward @xmath27 . in particular @xmath253 is the gradient vector of the function @xmath254 . the sign of the second derivative says that @xmath255 is concave in @xmath216 for each @xmath256 . by taking the infimum over @xmath257 , the resulting function @xmath258 is concave in @xmath216 . this implies that the super - level set @xmath245 of the funk distance @xmath259 is convex . as an analogous situation in special relativity , the super - level set of the past - directed temporal distance measured from a fixed point in the minkowski space @xmath260 is convex . for example , the set below the past - directed hyperboloid : @xmath261 is convex . we define the timelike hilbert metric following the ideas of busemann in @xcite , with a slight modification . for this , we consider two disjoint open , possibly unbounded , convex sets @xmath262 and @xmath263 in @xmath20 , bounded by convex hypersurfaces @xmath145 and @xmath144 respectively . let @xmath264 be the complement of the closures @xmath265 and @xmath266 in @xmath34 . for @xmath267 , we denote by @xmath268 the set of supporting hyperplanes of @xmath269 . given a point @xmath37 in @xmath36 , for @xmath267 , we define @xmath270 to be the set of interior points of the cone @xmath271 spanned by @xmath272 with apex at @xmath37 . we denote by @xmath1 the set of points in @xmath36 that are contained in the interior of euclidean segments of the form @xmath273 $ ] joining a point @xmath274 in @xmath145 and @xmath275 in @xmath144 , with the open line segment @xmath276 entirely contained in @xmath36 . we define a partial order relation @xmath2 on @xmath1 by setting @xmath277 as usual , @xmath278 , we mean either @xmath46 or @xmath45 . we also introduce the subset @xmath48 ( resp . @xmath49 ) as the set of ordered pairs @xmath50 in @xmath51 satisfying @xmath44 ( resp . @xmath43 ) . since the set @xmath279 is open for each @xmath181 , and the set @xmath280 is open for each @xmath281 , it follows that the set @xmath48 is open in the product space @xmath51 . this set is disjoint from the diagonal set @xmath53 . if @xmath46 , then @xmath37 and @xmath16 are both in @xmath1 , and there exists a pair of points @xmath282 and @xmath283 , such that the open segment @xmath284 is disjoint from @xmath145 and @xmath144 and the quadruple @xmath285 are situated in that order on @xmath286 $ ] ( cf . figure [ time - hilbert ] ) . the converse does not hold ; namely for a quadruple @xmath285 aligned in that order on the segment @xmath286 $ ] with @xmath282 and @xmath283 , when either @xmath287 for some @xmath288 or @xmath289 for some @xmath290 , we have @xmath63 . note that @xmath145 and @xmath144 do not play a symmetric role in this definition . we have @xmath44 but not @xmath291 . we shall stress this fact when needed , talking for instance of an _ oriented _ euclidean line joining a point in @xmath145 to a point in @xmath144 when needed , but we shall not emphasize this fact when it is understood . [ def : hilbert ] the timelike hilbert metric on @xmath36 is a distance function @xmath292 defined on @xmath293 by the formula @xmath294 where for @xmath295 , @xmath296 is the timelike funk metric associated to the convex hypersurfaces @xmath272 . the timelike hilbert metric satisfies the time inequality . this follows from the definition of the timelike hilbert metric as a sum of two timelike funk metrics and the fact that the timelike funk metric satisfies the time inequality . in particular , the metric space together with the partial order relation thus defined satisfies the three axioms @xmath83 and @xmath84 of busemann s @xcite . the timelike hilbert metric satisfies additional properties which follow from those of the funk metric . in particular , we have : \(a ) in a timelike hilbert metric @xmath297 associated to two convex hypersurfaces @xmath145 and @xmath144 , the euclidean segments of the form @xmath286 $ ] where @xmath298 and @xmath299 , oriented from @xmath274 to @xmath275 such that @xmath300 for every @xmath27 in @xmath301 and @xmath302 for every @xmath303 in @xmath304 , are @xmath297-geodesics and such that the open segment @xmath284 is in the complement of @xmath305 . they are isometric to the real line . \(b ) the oriented euclidean segments contained in the segments of the form @xmath286 $ ] where @xmath298 and @xmath299 with @xmath300 for every @xmath27 in @xmath301 and with @xmath302 for every @xmath303 in @xmath304 , and their subsegments are the unique @xmath297-geodesics if and only if there are no segments @xmath286 $ ] in @xmath36 with @xmath274 in the interior of a segment @xmath306 in @xmath145 and @xmath275 in the interior of a segment @xmath307 in @xmath308 with @xmath306 and @xmath307 coplanar . the proof is an adaptation of that of the classical hilbert metric ( cf . @xcite or @xcite ) , and we omit it . we show that the corresponding timelike hilbert metric @xmath309 introduced above is a finsler metric , and we give its linear functional ( the analogue of the minkowski functional ) defined on a subset of each of the tangent space . consider a point @xmath37 in @xmath36 so that @xmath310 is nonempty ( where @xmath311 is the cone @xmath312 . ) denote by @xmath191 the set @xmath313 . using the notation of [ s : funk - finsler ] that concerns the infinitesimal finsler metric associated to a timelike funk metric , we define a linear functional on @xmath191 by the formula : @xmath314 where @xmath315 and @xmath316 are the timelike norms on tangent spaces associated to the funk metrics associated to @xmath145 and @xmath144 respectively ( see equations ( [ eq : p ] ) or ( [ eq : p1 ] ) ) . note that when @xmath178 is an element of @xmath191 , @xmath178 is in @xmath279 and @xmath317 is in @xmath318 . [ hilbert - finsler ] the timelike hilbert metric is a timelike finsler metric with timelike finsler structure given by the minkowski functional @xmath319 defined in ( [ eq : hilbert ] ) . let @xmath37 and @xmath16 be two points in the space @xmath264 associated to the convex sets @xmath145 and @xmath144 such that @xmath205 is in @xmath48 . in what follows , when we talk about a euclidean segment @xmath65 $ ] " joining @xmath37 to @xmath16 , we mean that the segment is oriented from @xmath37 to @xmath16 . recall that the euclidean segment @xmath65 $ ] is an @xmath320-geodesic , and the euclidean segment @xmath321 $ ] , is an @xmath322-geodesic . thus , we have @xmath323 } p_2(x , v)dx\ ] ] and @xmath324 } p_1(x , v)dx.\ ] ] since the segment @xmath321 $ ] is the interval @xmath65 $ ] traversed in the opposite direction , we have @xmath325 } p_1(x , v)dx=\int_{[p , q ] } p_1(x ,- v)dx.\ ] ] thus , we have @xmath326 } \left(p_2(x , v)+ p_1(x ,- v)\right)dx.\ ] ] furthermore , if @xmath327 is any path in the domain of definition of @xmath297 joining @xmath37 to @xmath16 , then we have @xmath328}p_2(x , v)\ ] ] and @xmath329}p_1(x ,- v).\ ] ] equations ( [ eq : h1 ] ) and ( [ eq : h2 ] ) show that @xmath297 is timelike finsler , with timelike minkowski functional at each point @xmath4 given by @xmath330 . the finsler structure @xmath319 is well - behaved in the sense that the linear functional @xmath331 is a timelike minkowski functional ( see [ def : min ] ) defined on the open cone @xmath332 in @xmath199 . since the sublevel set @xmath333 is open ( this follows from the same argument used for the timelike funk metric ) , @xmath319 is upper semi - continuous in the tangent bundle @xmath334 , making the length functional well - defined , and upper semi - continuous in @xmath203 topology of the path space . we now point out some two - dimensional examples of timelike hilbert metrics . analogous higher - dimensional analogues hold . let @xmath36 be a region contained by two parallel lines in the plane @xmath335 , namely the complement of two half - spaces @xmath336 bounded by a pair of parallel hyperplanes @xmath337 and @xmath338 , which without loss of generality , are assumed to be @xmath339 . then any timelike curve is a geodesic for the timelike hilbert metric . in this setting , a curve is timelike if at each point the tangent vectors are not vertical . consider the vertical projection @xmath340 . then the hilbert distance @xmath341 for @xmath342 is equal to @xmath343 where @xmath344 is the hilbert distance for the interval . this metric is sometimes called the one - dimensional hyperbolic metric " as this is the klein - beltrami model of the hyperbolic space @xmath345 . notice that @xmath36 is concave as well as convex in @xmath335 . the half space corresponds to the limiting case of the strip discussed above , @xmath346 , as @xmath347 . then the hilbert metric @xmath348 converges to the funk metric @xmath349 which is the timelike funk metric for the half - space @xmath350 . we will come back to this example later . leaving the setting of the euclidean space , we now work in the @xmath0-dimensional affine space @xmath351 as underlying space , then the sphere @xmath352 , and finally the projective space @xmath353 . we first recall a few facts . first , notice that the definition and the variational characterization of the timelike funk metric uses only the affine structure of the euclidean space . namely though the euclidean distance function appears in the definition of the timelike funk metric , only the ratio of the distances @xmath354 among the points @xmath355 on the ray @xmath47 is used . the corresponding remark is well known in the setting of the classical hilbert metric . also in the variational characterization of the funk distance , the ratio @xmath356 can be realized as the ratio of distances among points on the ray @xmath47 transversally intersecting the supporting hyperplane @xmath27 . this is an interesting fact , since it shows that perpendicularity , a concept which is euclidean but not affine , is not needed in the variational formulation . in the same trend , we recall that the convexity of a subset in affine space does not require a metric : a convex set in @xmath351 is a set whose intersection with any line is contractible . an affine space @xmath351 is usually thought of as a chart for the projective space @xmath353 . a convex set in @xmath353 can be regarded as a convex set in an affine space @xmath351 after choosing an appropriate chart . the notion of convexity in the euclidean setting extends to the spherical setting where one considers only subsets which are contained in an open hemisphere . an open convex subset @xmath22 of @xmath352 can be identified with the intersection of open half spaces indexed by the set @xmath26 of supporting hyperplanes @xmath27 , which are great hyperspheres . we have , in analogy with the euclidean setting , @xmath357 where @xmath30 is the open half space bounded by the hyperplane @xmath27 containing @xmath22 . as before , we denote the other half space by @xmath31 . the complement of the closure in @xmath352 is then @xmath358 note that the set @xmath26 is not empty , since the convex set lies in some hemisphere . for a point @xmath181 , denote the set of great circles by @xmath54 such that if @xmath359 . then @xmath37 lies on the other side of the convex set @xmath22 with respect to @xmath27 . we now define a timelike hilbert metric on the sphere @xmath352 . when talking about a convex open subset of @xmath352 , we shall always assume that it is contained in an open hemisphere . we consider two convex hypersurfaces @xmath145 , @xmath144 with non - intersecting closures @xmath360 , @xmath361 . the definition of the associated timelike hilbert metric is analogous to the euclidean one , with some extra care . namely , we let @xmath362 . we define a pre - order relation by first noting that given a great circle which intersects @xmath145 and @xmath144 at four distinct points , we can label them so that @xmath363 , @xmath364 and the two open arcs @xmath276 and @xmath365 are entirely contained in @xmath36 , while @xmath273 $ ] lies in @xmath366 , and @xmath367 $ ] in @xmath368 . this observation says that every spherical geodesic segment in @xmath36 connecting @xmath145 and @xmath144 is naturally oriented , where @xmath145 represents the past and @xmath144 the future . we denote by @xmath1 the set of points in @xmath36 that are contained in the interior of geodesic segments of the form @xmath273 $ ] joining a point @xmath274 in @xmath145 and @xmath275 in @xmath144 , with the future - oriented open line segment @xmath276 entirely contained in @xmath36 . given a point @xmath37 in @xmath1 , define @xmath279 to be the set of interior points of the future - directed cone @xmath369 spanned by @xmath144 with apex @xmath37 . similarly we set @xmath318 to be the set of interior points of the past - directed cone @xmath370 spanned by @xmath145 with apex @xmath37 . from the definition of @xmath1 , @xmath370 and @xmath369 are non - empty . we define a preorder relation @xmath2 on @xmath371 by setting @xmath277 we denote , as before , the set of points @xmath205 in @xmath372 satisfying @xmath46 by @xmath48 . we also write @xmath373 when @xmath44 or @xmath45 . we note that in the spherical setting , the cone @xmath270 is an open _ lune _ , with two vertices , @xmath37 and its antipodal point @xmath374 . unlike the situation studied in @xcite , it is not possible to define a timelike funk spherical metric , because given two distinct points in the complement of a convex subset of the sphere @xmath352 , there is no natural way of saying that one is in the future of the other ( the great circle through these points may intersect the convex set in two points ) . we recall that given four points @xmath375 situated in that order on a great circle on the sphere , the spherical cross ratio is defined by @xmath376 = \frac{\sin d(p_2 , p_4 ) \sin d(p_3 , p_1)}{\sin d(p_3 , p_4 ) \sin d(p_2 , p_1)}.\ ] ] its values are in @xmath377 . the spherical cross ratio is a projectivity invariant , cf . @xcite . now , for a pair of points @xmath205 in @xmath48 , let @xmath298 and @xmath299 be the intersection points between the great circle through @xmath37 and @xmath16 and the two hypersurfaces @xmath145 and @xmath144 , so that @xmath378 lie on the arc of great circle @xmath273 \subset \omega$ ] in that order . the timelike hilbert metric associated to the pair @xmath379 is a distance function @xmath292 defined on the subset @xmath380 of @xmath372 consisting of points @xmath37 and @xmath16 satisfying @xmath373 , using the above notation for @xmath381 , by the formula @xmath382.\ ] ] there is a timelike minkowski functional @xmath319 associated to the timelike hilbert metric @xmath297 as follows . first we denote by @xmath191 the cone in @xmath199 consisting of vectors tangent to great circles in @xmath310 . the set @xmath191 may be empty depending on @xmath37 . for a given point @xmath383 and for any vector @xmath182 , @xmath319 is given by the following formula : @xmath384 we also set @xmath385 when @xmath185 , where @xmath186 is as in ( [ eq : p ] ) above , and where @xmath386 for @xmath387 for @xmath267 . the fact that @xmath319 is the finsler structure for @xmath297 can be checked in the same way as for the euclidean setting ( theorem [ hilbert - finsler ] . ) likewise , it can be shown is that @xmath319 is upper semi - continuous , that it can be used to define lengths of curves , and that the timelike hilbert distance is the intrinsic timelike distance defined as the supremum of length of curves , as we noticed for the timelike funk distance . the fact that @xmath297 defines a timelike distance on @xmath36 can be proved as in the case of the euclidean timelike hilbert metric . the spherical geodesic segments are geodesics for this timelike spherical metric ; this can be proved as in the case of the spherical ( non - timelike ) hilbert metric considered in @xcite . in fact , there is a relation between the euclidean and the spherical timelike hilbert metrics which can be seen as follows . assume that the sets @xmath388 and @xmath389 are both contained in an open hemisphere of @xmath352 , which we may assume to be the northern hemisphere @xmath390 . ( we may consider this hemisphere as an affine chart of the projective space @xmath391 . ) there is an isomorphism between the timelike hilbert metric associated to the pair @xmath379 and a euclidean timelike hilbert metric whose associated convex sets are the images @xmath392 of @xmath393 @xmath394 in the euclidean space @xmath395 seen as the hyperplane @xmath396 via the projection @xmath397 centered at the origin of @xmath398 ( see @xcite for the details ) . there is a notable case of a timelike spherical metric . consider the case where the two convex hypersurfaces @xmath145 and @xmath144 are antipodal in @xmath352 , that is , @xmath399 where the minus sign refers to the antipodal map @xmath400 of @xmath352 modeled in @xmath398 . in particular , when @xmath145 is a small circle of radius @xmath401 in @xmath402 , then the resulting timelike hilbert metric is isometric to the de sitter metric restricted to the timelike vectors . below , we will explicitly identify the de sitter metric with the spherical timelike hilbert metric . taking quotients , this case is identified with a timelike hilbert metric on an open subset of the projective space @xmath403 , in which @xmath145 and @xmath144 become a single convex hypersurface @xmath19 under the antipodal quotient map @xmath404 . in this setting , there is a doubling phenomenon for the rays emitted from a point @xmath37 in the complement of @xmath405 : if such a ray intersects @xmath19 at @xmath406 in the future , then it also does so at a point @xmath407 in the past ( cf . figure [ future - past ] ) . in other words , for a pair of points @xmath75 satisfying @xmath46 , the ray @xmath47 from @xmath37 through @xmath16 intersects @xmath19 and the ray @xmath408 also intersects @xmath19 . in the light of the discussion about the spherical timelike hilbert metric above , this picture corresponds to the situation where the two convex subsets of @xmath352 are antipodally located ; @xmath399 . the line @xmath409 in @xmath353 is a great circle in @xmath352 , @xmath406 is the intersection point between the future - directed ray and @xmath145 , while @xmath407 is the intersection point between the ray and @xmath144 . so far , we have carefully avoided the issue of null vectors in timelike geometry . we did so because there is no coherent general treatment for the timelike funk and hilbert metrics . however , this setting where @xmath145 and @xmath144 are antipodally located on @xmath352 , is a particular situation worth being investigated . let @xmath36 be the complement of the set @xmath410 where @xmath411 . then a great circle intersecting @xmath145 at two points @xmath412 also intersect @xmath144 at two antipodal points @xmath413 . another way of saying this is that @xmath414 and also @xmath415 . recall that each @xmath270 is a lune whose two vertices are @xmath37 and @xmath374 . hence @xmath318 and @xmath279 are a pair of antipodal lunes . now consider the situation when a great circle @xmath416 intersect @xmath145 tangentially at @xmath16 , which can be thought as a limiting picture for @xmath417 and @xmath418 merging on @xmath145 . this circle also intersects @xmath144 tangentially at @xmath374 . we consider a pair of points @xmath419 on an arc of the great circle @xmath416 in @xmath36 , and the timelike hilbert distance @xmath341 , which is the logarithm of the cross ratio of the quadruple @xmath420 lying on the arc in that order . under the antipodal quotient map @xmath421 , @xmath422 and @xmath423 coincide , thus the value of the cross ratio is one , hence @xmath424 . as the choices of @xmath4 and @xmath5 on the great circle @xmath416 are arbitrary , we conclude that the ( naturally extended ) timelike minkowski functional evaluated along the great circle tangential to @xmath145 and @xmath399 is zero . in other words , given a point @xmath37 in @xmath36 , consider the cone @xmath425 consisting of great circles through @xmath37 which are tangential to the convex hypersurface @xmath144 . these great circles are automatically tangential to @xmath426 . the cone @xmath427 in @xmath199 on which the minkowski functional @xmath428 is defined is equal to @xmath279 . then the tangent vectors @xmath429 at @xmath37 constitute the future - directed light cone with respect to the timelike minkowski functional for the spherical timelike hilbert metric @xmath297 . in this way , we have demonstrated the existence of null vectors in the de sitter space within the framework of timelike hilbert geometry . the quotient space of the de sitter space is equipped with a _ locally timelike _ hilbert geometry , where the quotient is taken by the @xmath430 antipodal symmetry of @xmath431 , with @xmath145 a small circle of radius @xmath401 in @xmath432 . the timelike hilbert metric thus defined is only _ local _ , as the space @xmath433 is not time - orientable . namely consider the closed path from @xmath181 to itself , along the circle at infinity of @xmath434 . traversing the loop then reverses the orientation of the light cone ( cf . hawking - ellis @xcite , calabi - marcus @xcite ) . we now demonstrate an explicit isometry between the locally timelike hilbert metric and the de sitter metric . for this , we first recall that the sitter space is the unit sphere in the minkowski space @xmath260 @xmath435 equipped with the so - called de sitter metric , a lorentzian metric of type @xmath436 whose first fundamental form is induced from the ambient minkowski metric @xmath437 . the intersection between @xmath438 and the @xmath439-plane is a copy of @xmath440 which is totally geodesically embedded . by an element of @xmath441 , any pair of points @xmath205 , with @xmath16 lying in the future of @xmath37 in @xmath442 , can be isometrically transposed to a pair of points on @xmath443 so that the @xmath444 coordinates of the points are positive . hence we may assume without loss of generality that @xmath37 and @xmath16 belong to the upper hemisphere @xmath445 of @xmath443 in @xmath446 . we introduce a local parameter @xmath216 on @xmath443 so that @xmath447 note that @xmath216 is an arc - length parameter for the de sitter metric , as the tangent vector to @xmath448 has norm 1 . hence for @xmath449 and @xmath450 with @xmath451 , the de sitter distance @xmath211 is equal to @xmath452 . here the point @xmath16 lies in the future of @xmath37 ( @xmath46 . ) on the other hand , we can project a part of the hyperboloid @xmath453 onto the hyperplane @xmath454 along the rays from the origin of @xmath446 @xmath455 let @xmath456 and @xmath457 be the images of @xmath37 and @xmath16 by this correspondence , where @xmath458 . the asymptotic lines @xmath459 of the hyperboloid @xmath460 are sent to the points @xmath461 and @xmath462 . the cross ratio of those four points defines the hilbert metric @xmath297 for the convex set @xmath463 in the projective space @xmath464 , and for the pair of points @xmath465 and @xmath466 with @xmath467 , we have @xmath468 by noting the equality @xmath469 the hilbert distance @xmath470 is equal to @xmath471 . hence we have shown that @xmath472 for @xmath46 . by post - composing the map @xmath473 with the map @xmath474 where @xmath390 is the upper hemisphere @xmath475 , the geodesic through @xmath37 and @xmath16 in the de sitter space is identified with a great circle in the sphere , and the image of the map @xmath476 of the northern half of the de sitter space is @xmath477 where @xmath168 is the northern cap bounded by the small circle of radius @xmath478 . this demonstrates that the timelike geometry of the de sitter space is realized by the timelike hilbert metric modeled on the sphere . the maps @xmath473 and @xmath479 are _ perspectivities _ , namely they preserve the cross ratio ( see @xcite ) . we conclude that the de sitter distance is equal to the timelike hilbert distance . finally we note , as busemann mentions in his paper @xcite , that the funk metric associated to a convex hypersurface @xmath19 contained in an affine space @xmath351 appears as a special case of a hilbert metric , namely , the one associated to a pair @xmath480 where @xmath145 is the hyperplane at infinity @xmath481 in the projective space @xmath482 . the set @xmath145 is the collection of points which are `` infinite distance away '' from any pair of points in @xmath483 , in the sense that for any pair of points @xmath484 with @xmath485 , we have @xmath486 . in that case , and using the notation of definition [ def : hilbert ] , the hilbert distance from @xmath37 to @xmath16 associated to the pair @xmath480 is just the funk distance from @xmath37 to @xmath16 associated to the convex set @xmath144 . there is a funk metric associated with a convex hypersurface @xmath19 in the hyperbolic space @xmath487 . this was studied in @xcite . in the same way , one can define a timelike funk metric associated with convex subsets of @xmath487 . the pre - order @xmath46 is defined as in the case of the euclidean timelike funk metric , and the distance from @xmath37 to @xmath16 satisfying @xmath44 is given by the formula @xmath488 where @xmath98 is the point where the ray @xmath47 hits @xmath19 for the first time , and @xmath204 is the hyperbolic distance . several properties of the hyperbolic ( non - timelike ) funk metric proved in @xcite hold verbatim for this timelike hyperbolic funk metric . in particular , we have a variational formulation of the timelike hyperbolic funk metric : @xmath489 there is also a timelike hyperbolic hilbert metric , defined in an analogous way to the timelike hilbert metric defined in [ def : h ] , replacing , in the definition , the distance by the hyperbolic sine of the distance , as we did in the definition of the timelike hyperbolic metric in ( [ eq : hyp.ff ] ) . the hyperbolic segments are geodesics for the timelike hyperbolic funk and for the timelike hyperbolic hilbert metrics . the proofs use the same as the one of the analogous result for the hyperbolic ( non - timelike ) funk and hilbert metrics considered in @xcite . 1 . study the geometric properties of specific examples ( timelike geometries of individual or special classes of convex sets ) . 2 . study the infinite - dimensional case . 3 . study the reverse timelike funk metric . 4 . study the isometries of the timelike funk metric . 5 . study the horofunction boundaries of the timelike funk and hilbert metrics . find a characterization of the timelike hilbert geometries that can be embedded in a lorentzian space . complexify . _ acknowledgements _ the first author is supported by the french anr grant finsler . he is also grateful to the gakushuin university ( tokyo ) and to the tata institute of fundamental research ( bombay ) where part of this work was done . a. papadopoulos , hilbert s fourth problem athanase papadopoulos . in _ handbook of hilbert geometry _ ( a. papadopoulos and m. troyanov , ed . ) european mathematical society publishing house , zrich , 2014 , 391 - 431 . a. papadopoulos and m. troyanov , from funk to hilbert geometry . handbook of hilbert geometry ( a. papadopoulos and m. troyanov , ed . ) , european mathematical society publishing house , irma lectures in mathematics and theoretical physics , vol . 22 , p. 33 - 68 , 2014 . a. papadopoulos and s. yamada , on the projective geometry of constant curvature spaces , sophus lie and felix klein : the erlangen program and its impact in mathematics and physics , ( 2015 ) 237246 , european mathematical society .

we investigate the spin diffusion and transport in a graphene monolayer on sio substrate by means of the microscopic kinetic spin bloch equation approach . the substrate causes a strong rashba spin - orbit coupling field mev , which might be accounted for by the impurities initially present in the substrate or even the substrate - induced structure distortion . by surface chemical doping with au atoms , this rashba spin - orbit coupling is further strengthened as the adatoms can distort the graphene lattice from to bonding structure . by fitting the au doping dependence of spin relaxation from pi _ et al . _ [ phys . rev . lett . * 104 * , 187201 ( 2010 ) ] , the rashba spin - orbit coupling coefficient is found to increase approximately linearly from 0.15 to 0.23 mev with the increase of au density . with this strong spin - orbit coupling , the spin diffusion or transport length is comparable with the experimental values . in the strong scattering limit ( dominated by the electron - impurity scattering in our study ) , the spin diffusion is uniquely determined by the rashba spin - orbit coupling strength and insensitive to the temperature , electron density as well as scattering . with the presence of an electric field along the spin injection direction , the spin transport length can be modulated by either the electric field or the electron density . it is shown that the spin diffusion and transport show an anisotropy with respect to the polarization direction of injected spins . the spin diffusion or transport lengths with the injected spins polarized in the plane defined by the spin - injection direction and the direction perpendicular to the graphene are identical , but longer than that with the injected spins polarized vertical to this plane . this anisotropy differs from the one given by the two - component drift - diffusion model , which indicates equal spin diffusion or transport lengths when the injected spins are polarized in the graphene plane and relatively shorter lengths when the injected spins are polarized perpendicular to the graphene plane .

1012.0973

graphene is considered to be a promising candidate for the spintronic applications recently,@xcite partly due to the perfect two dimensionality , gate - voltage - tunable charge carrier type and density,@xcite high mobility@xcite as well as the potentially long spin relaxation time limited by the small intrinsic spin - orbit and hyperfine couplings.@xcite from the high mobility and long spin relaxation time , a long spin relaxation length , favorable to the spin information transport and manipulation , is anticipated . however , both the spin relaxation time and transport length were experimentally found to be much smaller than expected.@xcite this suggests that the spin relaxation in the experiments is most likely to be contributed by extrinsic factors such as the possible impurity - enhanced spin relaxation@xcite via the elliot - yafet@xcite mechanism or the enhanced rashba spin - orbit coupling field@xcite from the impurities.@xcite the former case may exist in a highly dirty graphene sample and causes the spin relaxation time @xmath4 to be proportional to the momentum relaxation time @xmath5.@xcite however , for the latter case , the dyakonov - perel ( dp ) spin relaxation mechanism@xcite dominates and the relation @xmath6 is absent . in fact , recently pi _ _ reported that @xmath4 increases with decreasing @xmath5 in the surface chemical doping experiment with au atoms on graphene,@xcite indicating that the dp spin relaxation mechanism is important there . however , the relation @xmath7 , valid when the dp spin relaxation mechanism is dominant and the scattering is strong enough , is not obeyed in their experiment.@xcite nevertheless , we will show that this deviation can be understood by taking account of the strengthening of the rashba spin - orbit coupling with the increasing coverage of au adatoms . the rashba spin - orbit coupling , referred to as an extrinsic one , is due to the broken of the inversion symmetry which can be caused by either a perpendicular electric field , the interaction with substrate , or the atoms adsorbed on the surface.@xcite the contribution of the electric field to the rashba spin - orbit coupling is small ( @xmath8ev under a perpendicular electric field as large as 1 v / nm ) , @xcite while the adatoms can effectively enhance the rashba spin - orbit coupling to be of order of 10 mev by distorting the graphene lattice from @xmath2 to @xmath3 bonding structure.@xcite in this work , we investigate the spin diffusion and transport limited by the dp mechanism in a graphene monolayer on sio@xmath0 substrate as presented by pi _ et al._.@xcite to account for the short spin relaxation time ( @xmath970 ps ) before au doping in the experiment,@xcite we assume that the impurities inevitably present in the substrate , as well as the other effects such as the substrate - induced structure distortion , cause a strong rashba spin - orbit coupling . when the surface chemical doping by au atoms@xcite is performed , the rashba spin - orbit coupling coefficient @xmath10 is further strengthened . by fitting the chemical doping dependence of spin relaxation time from the experimental data,@xcite we obtain the chemical doping dependence of @xmath10 . it is found that @xmath10 increases approximately linearly with the density of adatoms when the latter is not too high . with this essential information obtained , we then study the spin diffusion and transport in the graphene layer . the method utilized in our study is the kinetic spin bloch equation ( ksbe ) approach which has been successfully applied to the study of spin dynamics in semiconductors.@xcite in the framework of this approach , the spatial spin precession frequency during the steady - state scattering - free spin diffusion ( assumed to be along the @xmath11-axis ) is@xcite @xmath12 here @xmath13 is the dp term , @xmath14 is the external magnetic field and @xmath15 is the electron energy spectrum . the momentum dependence of @xmath16 leads to the inhomogeneous broadening in spin precession , with which any scattering ( including the coulomb scattering ) can cause an irreversible spin relaxation along with spin diffusion and transport.@xcite it is noted that different dp terms as well as different energy spectra lead to distinct momentum dependences of @xmath16 . for graphene , @xmath17 with @xmath18 m / s being the fermi velocity and @xmath19 with @xmath20 being the polar angle of momentum @xmath21 . therefore in the absence of any external magnetic field @xmath22 which depends on the angle @xmath20 rather than the magnitude of @xmath23 . this indicates that the inhomogeneous broadening is _ insensitive _ to temperature and electron density as long as @xmath10 is fixed . therefore the spin diffusion is only possible to be modulated effectively by the scattering.@xcite however , in this work it is revealed that when the scattering is strong enough ( just as in the graphene layer under study ) , the spin diffusion becomes insensitive to the scattering . as a result , the spin diffusion is uniquely determined by @xmath10 . moreover , the mean spin precession frequency @xmath24 shows a strong anisotropy which can also lead to the anisotropy of spin diffusion with respect to the spin polarization direction . this anisotropy is found to be quite different from the widely believed one predicted from the two - component drift - diffusion model.@xcite the discrepancy reveals the inadequacy of the two - component drift - diffusion model , especially for the cases with spin precession in spatial domain . this paper is organized as follows . in sec . ii , we present the model and introduce the ksbes . in sec . iii , we first investigate the spin relaxation by fitting the experimental data from pi _ et al._@xcite to obtain essential parameters and then study the spin diffusion and transport in graphene . both the analytical and numerical investigations are performed . by comparing the results from the analytical and numerical studies , we find that the analytical model depicts the zero - electric - field spin diffusion perfectly and the nonzero - electric - field spin transport with a small discrepancy which increases with the strength of the electric field . at last we summarize in sec . iv . the @xmath25-doped graphene monolayer under investigation is on a sio@xmath0 substrate perpendicular to the @xmath26-axis . the depth of the sio@xmath0 substrate is assumed to be @xmath27 nm and the electric field along the @xmath26-axis is @xmath28 with @xmath29 being the gate voltage . the spins are injected at @xmath30 and diffuse or transport along the @xmath11-axis . the external electric field , if applied , is along the @xmath11-axis , i.e. , @xmath31 . under the basis laid out in refs . and , the hamiltonian of electrons can be written as@xcite @xmath32 { c_{\mu{\bf k}s}}^{\dagger } c_{\mu{\bf k}s^\prime } \\ & & \mbox{}+ { h}_{\rm int}. \label{h_eff}\end{aligned}\ ] ] here @xmath33 labels the valley located at @xmath34 or @xmath35 point , @xmath36 denote the pauli matrices and @xmath37 ( @xmath38 ) is the annihilation ( creation ) operator of electron in @xmath33 valley with momentum @xmath21 ( relative to the valley center ) and spin @xmath39 ( @xmath40 ) . @xmath41 is the intrinsic spin - orbit coupling constant and @xmath42 is the electron charge ( @xmath43 ) . the coefficient in the rashba term @xmath44 [ eq . ( [ omega ] ) ] reads @xmath45 , with the first term contributed by the electric field along the @xmath26-axis and the second term by the substrate ( including the impurities initially present inside ) as well as the adatoms from surface chemical doping . the coefficient @xmath46 is @xmath47 mev@xmath48nm / v ( refs . and ) . the hamiltonian @xmath49 consists of the electron - impurity , electron - phonon as well as electron - electron coulomb interactions.@xcite we adopt the minimal model proposed by adam and das sarma@xcite to depict the electron - impurity scattering . within this model , only the intravalley electron - impurity scattering is important while the intervalley electron - impurity scattering is negligible due to the large momentum transfer from one valley to the other and the finite distance between the impurity layer and the graphene plane . the intravalley electron - impurity scattering matrix element reads @xmath50,@xcite where @xmath51 is the effective impurity density , @xmath52 is the effective distance of impurities from the graphene layer@xcite and @xmath53 is the coulomb potential under the random phase approximation.@xcite the electron - phonon scattering includes the intravalley electron - acoustic phonon scattering,@xcite the intervalley and intravalley electron - optical phonon scattering,@xcite as well as the intravalley electron - optical surface phonon scattering.@xcite the electron - electron coulomb scattering includes both the intervalley and intravalley scattering , with the screening under random phase approximation given in ref . . by using the nonequilibrium green function method,@xcite the ksbes are constructed as@xcite @xmath54 here @xmath55 represent the density matrices of electrons with relative momentum @xmath21 in valley @xmath56 at position @xmath11 and time @xmath57 . @xmath58 are the driving terms from the external electric field ( the fluctuation of electron density is neglected and thus the total electric field is taken to be the external one ) . the diffusion terms due to the spatial gradient are @xmath59 @xmath60 and @xmath61 are the coherent and scattering terms , respectively . their expressions can be found in ref . . in the steady - state scattering - free spin diffusion , the spatial spin precession frequency , given by eq . ( [ frequency ] ) , is immediately obtained according to the ksbes.@xcite in the following , we first study the spin relaxation in graphene by fitting the experimental data from pi _ et al._@xcite to obtain information on impurities ( including the effective density as well as the distance from the graphene layer ) and the chemical doping dependence of the rashba spin - orbit coupling coefficient . we then use the information to study the spin diffusion and transport in graphene , first analytically for the case with strong electron - impurity scattering only , and then numerically with _ all _ the scattering explicitly included . we fit the chemical doping dependence of spin relaxation time and diffusion coefficient from pi _ et al . _ 3(c ) in ref . ] to establish : ( i ) the density and typical distance from the graphene layer of charged impurities initially present in the substrate and those of the chemical doping adatoms ; and ( ii ) the dependence of @xmath10 on chemical doping . the electron density @xmath62 is @xmath63 @xmath64 and the temperature @xmath65 is 18 k.@xcite the electrons are initially polarized in the @xmath11-@xmath66 plane@xcite with the polarization @xmath67 assumed to be 0.05 . to perform the fitting , the ksbes are solved in the time domain under spatial uniform case , as carried out recently by zhou and wu in the ultraclean graphene monolayer.@xcite ( an analytical study of spin relaxation time in graphene is also given in appendix [ appa ] . ) the diffusion coefficient @xmath68 given by pi _ _ is actually for spin instead of charge , although it is treated as the charge diffusion coefficient in the experiment.@xcite in fact , these two coefficients are usually close to each other and jzsa _ et al . _ found this most likely to be the case in graphene when the electron density is high ( @xmath69 @xmath64)@xcite due to the weak electron - electron coulomb scattering.@xcite therefore we fit the experimental data with the charge diffusion coefficient @xmath70 , where @xmath71 is the electron mobility obtained under a small in - plane electric field.@xcite atom/(@xmath72s).@xcite dashed curve with closed squares : deposition time dependence of calculated diffusion coefficient , with the deposition time dependence of au density given by the solid curve ( the scale is on the right - hand side of the frame ) . crosses : experimental data from pi _ et al._.@xcite ( b ) dashed curve with closed squares : deposition time dependence of calculated spin relaxation time , with the deposition time dependence of @xmath10 shown by the solid curve with open squares ( the scale is on the right - hand side of the frame ) . crosses : experimental data from pi _ et al._.@xcite inset of ( b ) : dependence of @xmath10 on au density.,title="fig:",width=302 ] we first make use of the value of @xmath73 m@xmath74/s for the case without surface chemical doping given in ref . to explore the information of impurities initially present in the substrate . this single value of @xmath68 is not sufficient for us to fix both the effective density and distance from the graphene layer of these impurities . however , these details are not essential and we just choose two proper parameters , e.g. , @xmath75 @xmath64 and @xmath76 nm , to ensure @xmath77 m@xmath74/s . the surface chemical doping deposits au atoms on the graphene surface with a growth rate of 5@xmath78 atom/(@xmath72s).@xcite by fitting the deposition time ( adatom density ) dependence of @xmath68,@xcite the distance of adatoms from the graphene layer @xmath79 is obtained to be about 0.2 nm . nevertheless , the fitting does not confirm with the experimental data well when the deposition time exceeds 4 s [ compare the fitting data ( chain curve with triangles ) to the experimental data ( crosses ) in fig . [ figzw1](a ) ] . this indicates that the effective density of adatoms does not increase linearly with time any more when the doping has been performed for several seconds . therefore , when the doping time is longer than 4 s , we choose the proper density of adatoms to reproduce the experimental diffusion coefficient . in fig . [ figzw1](a ) , the deposition time dependence of au density is plotted by the solid curve with open squares ( the scale is on the right - hand side of the frame ) and that of the calculated diffusion coefficient is shown by the dashed curve with closed squares . with the parameters for two kinds of impurities obtained , we then fit the spin relaxation time @xmath4 to obtain @xmath10 under different deposition times . in fig . [ figzw1](b ) , the deposition time dependence of fitted @xmath10 is shown by the solid curve with open squares ( the scale is on the right - hand side of the frame ) and that of the calculated spin relaxation time is shown by the dashed curve with closed squares . the crosses represent the experimental spin relaxation times under different deposition times . in the inset of fig . [ figzw1](b ) , we also plot the dependence of @xmath10 on au density @xmath80 . it is shown that @xmath10 increases approximately linearly with au density when the latter is not so high . the fitted value of @xmath10 is comparable to the value estimated by ertler _ _ when taking account of the adatoms , i.e. , 0.3 mev.@xcite it is noted that @xmath81 has the largest value 0.027@xmath821 with @xmath83 ps ( ref . ) when @xmath84 . therefore , the electron system is in the strong scattering limit ( the electron - impurity scattering is dominant ) , let alone when the temperature is increased or the chemical doping is performed . it is necessary to point out that in the experiment the gate voltage @xmath29 is adjusted to keep @xmath62 constant during chemical doping as adatoms also donate electrons to the graphene layer.@xcite however , @xmath29 does not exceed 200 v and the term @xmath85 is at least two orders of magnitude smaller than @xmath10 . therefore @xmath86 and is solely determined by the impurities . when @xmath87 , @xmath88 mev and is contributed by the impurities in the substrate . in this section we study the spin diffusion in graphene analytically for the case with only the electron - impurity scattering . no external electric field is present . we first perform the fourier transformation of the steady - state ksbes with respect to the polar angle @xmath20 and then retain the equations involving the lowest three orders.@xcite the neglect of the higher orders will not lose much information in the strong scattering limit where the electron distribution approaches isotropy in the momentum space . as a result the following second - order differential equation about @xmath89 [ @xmath90 and @xmath91 is obtained : @xmath92-\frac{\alpha_{\rm r}^2}{\hbar^2v_{\rm f}^2}[\sigma_x,[\sigma_x,\rho_{\mu k}^0(x)]]\\ \hspace{-0.3 cm}&&\mbox{}-\frac{\alpha_{\rm r}^2}{\hbar^2v_{\rm f}^2}[\sigma_y,[\sigma_y,\rho_{\mu k}^0(x)]]=0 . \label{dif}\end{aligned}\ ] ] it is noted that with only the lowest three orders of @xmath93 considered from the beginning , the electron - impurity scattering time is actually absent from the above equation ( refer to appendix [ appb ] for detail ) . this indicates that in the strong scattering limit the spin diffusion becomes insensitive to scattering in graphene . we define the `` spin vector '' as @xmath94 $ ] and @xmath95 can be solved from eq . ( [ dif ] ) with boundary conditions ( refer to appendix [ appb ] for detail ) . then one can calculate the total spin signal contributed by all the different electron states in two valleys as @xmath96\\ & = & \frac{1}{\pi}\int_0^{+\infty}dkk{\bf s}_{\mu k}^0(x ) . \label{sum}\end{aligned}\ ] ] in the following we present the solutions of @xmath97 under three typical boundary conditions . for boundary condition ( i ) @xmath98 and @xmath99 , which corresponds to the case with the injected spins polarized along the @xmath11-axis , @xmath100 for boundary condition ( ii ) @xmath101 and @xmath99 , @xmath102 for boundary condition ( iii ) @xmath103 and @xmath99 , @xmath104 in the above equations @xmath105 and @xmath106 @xmath107 , @xmath108 , @xmath109 and @xmath110 . it is noted that the spin precession frequency given by the simplified model is @xmath111 , a little smaller than @xmath112 due to the approximations made here . from eqs . ( [ sx])-([sz ] ) one notices that in the strong scattering limit , the spin diffusion is not only insensitive to the scattering , but also unrelated to temperature @xmath65 and electron density @xmath62 . nevertheless , the coefficient @xmath10 may depend on @xmath65 and/or @xmath62 , with the relation unclear so far . for simplicity we assume @xmath10 to be independent of @xmath65 and @xmath62 in this work . as a result , the spin diffusion in the strong scattering limit is uniquely determined by @xmath113 , which is only modulated by chemical doping . ( [ sx])-([sz ] ) indicate a strong anisotropy of spin diffusion with respect to the spin - polarization direction . for the cases with the injected spins polarized along the @xmath11- and @xmath26-axis , both the spin signals show an exponential decay in the magnitude accompanying with the precession in the @xmath11-@xmath26 plane . the spin precessions have the same frequency @xmath114 except for a phase difference . however , when the injected spins are polarized along the @xmath66-axis , the spin signal decays exponentially without any precession , i.e. , it is bound along the @xmath66-axis . the above phenomena are understood by noticing that the mean effective magnetic field felt by the diffusing electrons is along the @xmath66-axis as @xmath115 . in the non - local spin valve experiments , the spin diffusion length is usually measured from the exponential decay of spin signal with the increasing spacing between the central spin - injector and -detector ferromagnetic electrodes.@xcite in these experiments , the ferromagnetic electrodes happen to be magnetized along the @xmath66-axis and therefore the injected and detected spin polarizations are both along the @xmath66-axis . with such configuration , the exponential decay of spin signal with increasing spacing between the electrodes can be well observed . however , if the injected spins are polarized in the @xmath11-@xmath26 plane , the spatial spin precession is expected to be detected . besides the anisotropy of spin precession , the spin diffusion length also shows an anisotropy as @xmath116 with @xmath117 . in fact , when the injected spins are polarized along any other direction in the @xmath11-@xmath26 plane , the spin diffusion length is all the same as @xmath118 ( @xmath119 ) [ for this case the solution of @xmath97 is the combination of eqs . ( [ sx ] ) and ( [ sz ] ) ] . however , based on the widely utilized two - component drift - diffusion model@xcite which gives @xmath120 [ eq . ( [ d6 ] ) in appendix [ dde ] ] , one may expect that the spin diffusion lengths satisfy @xmath121 as the spin relaxation times in time domain follow ( refer to appendix [ appa ] ) @xmath122 and @xmath123 . it is noted that only when the injected spins are polarized along the @xmath66-axis , for which no spin precession exists , the two - component drift - diffusion model gives the result in consistence with that from the ksbes , i.e. , @xmath124 the discrepancy in the anisotropies given by the ksbes and the two - component drift - diffusion model strongly indicates the inadequacy of the two - component drift - diffusion model . due to the loss of the off - diagonal spin components , i.e. , the spin coherence , the two - component drift - diffusion model not only fails to predict the spin precession in spatial domain in the absence of an external magnetic field , but also incorrectly inherits the anisotropy from the spin relaxation in time domain . we emphasize that the reason for the different anisotropic properties of spin diffusion in spatial domain and spin relaxation in time domain is that the inhomogeneous broadening is quite different in these two cases . in spatial domain the inhomogeneous broadening governing the spin diffusion arises from the @xmath21 dependence of @xmath16 , while in time domain from that of @xmath125 . _ reported the relationship between the in - plane and out - of - plane spin relaxation times directly from the anisotropy of spin diffusion via the two - component drift - diffusion model.@xcite however , based on the above discussion , one may realize that studying the anisotropy of spin relaxation time in such a way can be incorrect . finally , from another point of view , if the two - component drift - diffusion model is still used , then in order to reflect the correct anisotropy of spin diffusion , the spin diffusion coefficient has to differ from the charge diffusion coefficient and shows an anisotropy as @xmath126 with @xmath127 . it should be pointed out that all the above analysis and conclusion also apply to the electron system where the energy spectrum is parabolic in momentum and the linear rashba spin - orbit coupling term @xmath128 is dominant , such as that in the asymmetric inas quantum wells.@xcite that is because the steady - state scattering - free spatial spin precession frequency @xmath16 in this system has the similar momentum dependence as shown in eq . ( [ frequency]).@xcite however , for electron system in the absence of the dp term but under a magnetic field perpendicular to both the spin polarization and spin transport directions such as in bulk silicon@xcite and symmetric silicon quantum wells,@xcite or with the dresselhaus term@xcite containing the cubic dependence on momentum such as in gaas quantum wells,@xcite the situation is quite different as @xmath16 depends on the magnitude of momentum . in fact , it has been revealed in the symmetric silicon quantum wells under an in - plane magnetic field that the scattering can suppress spin diffusion effectively in the strong scattering limit.@xcite we further take account of the electric field along the @xmath11-axis to study the spin transport . still only the strong electron - impurity scattering is included . the second - order differential equation about @xmath89 , corresponding to eq . ( [ dif ] ) but including the driving term , reads ( refer to appendix [ appc ] ) @xmath92-\frac{\alpha_{\rm r}^2}{\hbar^2v_{\rm f}^2}[\sigma_x,[\sigma_x,\rho_{\mu k}^0(x)]]\\ \nonumber \hspace{-0.3 cm}&&\mbox{}-\frac{\alpha_{\rm r}^2}{\hbar^2v_{\rm f}^2}[\sigma_y,[\sigma_y,\rho_{\mu k}^0(x)]]-ee\partial_x\partial_{\varepsilon_{\bf k}}\rho_{\mu k}^0(x)\\ \hspace{-0.3 cm}&&\mbox{}-i\frac{\alpha_{\rm r}ee}{\hbar v_{\rm f}}[\sigma_y,\partial_{\varepsilon_{\bf k}}\rho_{\mu k}^0(x)]=0 . \label{transport}\end{aligned}\ ] ] it should be pointed out that when the electric field is so large that the electron density matrices @xmath129 become strongly anisotropic due to the driving of the electric field , retaining only the lowest three orders of @xmath93 to obtain the above equation of @xmath89 may not be sufficient . the second - order differential equation about @xmath95 is obtained from the above equation and that about @xmath97 can be obtained by further summing over @xmath21 and @xmath33 ( refer to appendix [ appc ] ) . with the same three different typical boundary conditions presented in the previous section , @xmath97 is solved to have the same form as eqs . ( [ sx])-([sz ] ) except that the parameters are now electric - field dependent . explicitly , @xmath130 @xmath131 , @xmath132 , @xmath133 and @xmath134 . in the above equations , @xmath135 here @xmath136 and @xmath137 ( @xmath138 ) is the chemical potential of electrons with spin parallel ( antiparallel ) to the spin - polarization direction . it is noted that when the electric field is absent , i.e. , @xmath139 , all the above solutions recover those presented in the previous section . in most conditions ( such as in the present work ) electrons in graphene are highly degenerate . in the degenerate limit with small spin polarization , @xmath140 , where @xmath141 is the magnitude of the fermi momentum of unpolarized electrons with density being @xmath62 ( appendix [ appc ] ) . differing from the spin diffusion without electric field , the spin transport becomes sensitive to electron density as @xmath142 depends on the electron density . in the nondegenerate limit , @xmath143 ( appendix [ appc ] ) and the spin transport becomes sensitive to temperature rather than electron density . moreover , with this value of @xmath142 , eq . ( [ lytrans ] ) becomes @xmath144^{-1 } , \label{ddly}\ ] ] where @xmath145 is the spin diffusion length without electric field [ eq . ( [ ly ] ) ] . this result recovers that from the two - component drift - diffusion model , which apparently fails to correctly reflect the anisotropy of spin transport.@xcite therefore , our investigation again indicates that only when the spatial spin precession is absent , the two - component drift - diffusion model gives the appropriate depiction of spin transport . in fig . [ figzw2 ] we plot the dependence of @xmath146 , @xmath147 and @xmath148 on @xmath142 . from fig . [ figzw2](a ) , one notices that the spin transport length decreases with increasing @xmath142 ( @xmath149 ) . on one hand , this means that when the electron density is fixed ( e.g. , @xmath150 @xmath64 , for which the variation of @xmath142 from @xmath151 to 8 corresponds to a variation of @xmath152 from about @xmath153 to 2.2 kv / cm ) , the spin transport is suppressed ( enhanced ) by increasing the electric field parallel ( antiparallel ) to the spin injection direction . on the other hand , this also means that when the non - zero electric field parallel ( antiparallel ) to the spin injection direction is fixed , the spin transport is enhanced ( suppressed ) by increasing electron density . [ figzw2](b ) and ( c ) indicate that the spin precession frequency @xmath147 and the phase angle @xmath148 vary with @xmath142 marginally ( with a variation @xmath9 2 % ) . in fact , when @xmath154 becomes even larger , both @xmath147 and @xmath148 quickly saturate ( @xmath147 approaches @xmath155 and @xmath148 approaches @xmath156 ) . therefore , the spin precession pattern in spatial domain is insensitive to the electric field or the electron density . the ksbes need to be solved numerically in order to take full account of all the different kinds of scattering as well as the large electric field . to numerically solve the ksbes , the initial conditions are set as @xmath157/\sum_{\mu\bf k}\mbox{tr}[\rho_{\mu\bf k}(0,0)]=p_0,\end{aligned}\ ] ] and the two - side injection boundary conditions@xcite are @xmath158 here the injected spins at left boundary @xmath30 are assumed to be polarized along @xmath159 with polarization @xmath160 . @xmath161 stands for the right boundary with @xmath162 much longer than the spin diffusion or transport length . @xmath163 are the fermi distributions of electrons at the two boundaries when the external electric field is absent . when the electric field is present , @xmath163 then stand for the drifted fermi distributions of hot electrons.@xcite in the previous analytical study the boundary conditions are in fact approximated as the single - side injection case . this approximation works well when the scattering is strong.@xcite by numerically solving the ksbes , the steady - state distribution of spin polarization along @xmath159 is obtained as @xmath164/\sum_{\mu\bf k}\mbox{tr}[\rho_{\mu\bf k}(x,+\infty)]$ ] and then the spin diffusion or transport length is determined from the exponential decay of @xmath165 ( or its envelope ) along the @xmath11-axis . versus position @xmath11 with the injected spins polarized along the @xmath11- , @xmath66- and @xmath26-axis , respectively . the squares , circles and triangles are obtained by numerically solving the ksbes with @xmath166 k , @xmath167 @xmath64 , @xmath168 @xmath64 and @xmath169 . the solid , dashed and chain curves are calculated from eqs . ( [ sx])-([sz ] ) with @xmath170 , @xmath171 and @xmath172 , respectively . , title="fig:",width=302 ] as revealed by the analytical model , the spin diffusion shows anisotropic properties with respect to the polarization direction of injected spins . in fig . [ figzw3 ] , we show the spatial distribution of the absolute value of the steady - state spin polarization @xmath173 for the cases with the injected spins polarized along the @xmath11- , @xmath66- and @xmath26-axis , respectively . @xmath169 , with which @xmath174 mev . the squares , circles and triangles are obtained by numerically solving the ksbes while the solid , dashed and chain curves are calculated by eqs . ( [ sx])-([sz ] ) . when @xmath174 mev the analytical model gives @xmath175 @xmath33 m , @xmath176 @xmath33 m and @xmath177 @xmath33m@xmath178 . the anisotropy of spin diffusion is clearly shown in this figure . it is noted that the simplified analytical model almost perfectly recovers the numerical results [ except that the spin precession frequencies for both cases with the injected spins polarized along the @xmath11- and @xmath26-axis are numerically shown to be closer to @xmath179 rather than @xmath180 given by the analytical study ( the difference is expected from the approximations made in the analytical analysis ) ] . in fact , further numerical calculations show that varying @xmath65 from 18 to 300 k and/or @xmath62 from 0.5 to @xmath63 @xmath64 changes the numerical results marginally . this is consistent with the conclusion from the analytical model , i.e. , the spin diffusion of electrons in graphene is _ insensitive _ to @xmath65 and @xmath62 in the strong scattering limit . as a result , in the strong scattering limit , one can depict the spin diffusion quite well with the _ single _ parameter @xmath10 via eqs . ( [ sx])-([sz ] ) . in fig . [ figzw4 ] , we plot the deposition time dependence of spin diffusion length with @xmath181 , @xmath182 and @xmath183 respectively by the solid curves . the spin diffusion lengths are directly obtained from eqs . ( [ lx])-([ly ] ) . it is shown that with the increase of chemical doping time , @xmath10 increases and the spin diffusion length decreases . for comparison , we also plot the deposition time dependence of spin diffusion length given by the two - component drift - diffusion model ( chain curves ) , i.e. , @xmath184 , with @xmath68 and @xmath185 given in fig . [ figzw1 ] . the comparison between these two sets of results shows that , only when the injected spins are polarized along the @xmath66-axis , the two - component drift - diffusion model yields the same result as that from the ksbes , just as revealed in the analytical study [ refer to eq . ( [ consistence ] ) and the discussion there ] . - , @xmath66- and @xmath26-axis , respectively . the results from the ksbes ( solid curves ) and the two - component drift - diffusion model ( chain curves ) are both plotted for comparison.,title="fig:",width=321 ] the electron system under investigation is always in the strong scattering limit and therefore the spin diffusion becomes insensitive to scattering . however , the properties of spin diffusion in the weak scattering limit can be different . in order to investigate the spin diffusion with scattering strength ranging from the weak to strong scattering limit , we artificially vary the impurity density in the substrate from 0 to 10@xmath186 @xmath64 . at the same time , the chemical doping is absent ( no adatom ) and @xmath10 is kept as a constant , e.g. , 0.153 mev . we choose @xmath187 k , @xmath188 @xmath64 and @xmath189 . in fig . [ figzw5 ] we plot the dependence of spin diffusion length @xmath145 on the impurity density by the dashed curve . for comparison , we also plot the corresponding dependence of spin relaxation time @xmath190 on the impurity density by the solid curve ( the scale is on the right - hand side of the frame ) . it is seen that with the increase in @xmath191 , while @xmath145 decreases obviously in the weak scattering limit ( @xmath192 @xmath64 ) and then saturates in the strong scattering limit , @xmath190 first decreases in the weak scattering limit ( refer to the inset for detail ) and then increases almost linearly in the strong scattering limit.@xcite the two - component drift - diffusion model is able to capture the dependence of spin diffusion length on @xmath191 by means of the relation @xmath193 : while @xmath194 , @xmath190 decreases with @xmath191 in the weak scattering limit and @xmath195 in the strong scattering limit ; therefore @xmath145 first decreases with @xmath191 and then becomes insensitive to @xmath191 ( the insensitivity of @xmath145 to @xmath191 in the strong scattering limit is revealed previously by the analytical study ) . it should be emphasized that in fig . [ figzw5 ] the results are shown with @xmath10 being a constant . in reality , when one further takes account of the increase of @xmath10 with increasing @xmath191 , @xmath145 should always decrease with increasing @xmath191 , from the weak scattering limit to the strong scattering limit . -axis . @xmath187 k , @xmath188 @xmath64 and @xmath174 mev.,title="fig:",width=321 ] versus position @xmath11 under different electric fields . the electric field dependence of spin transport length @xmath119 is also plotted with the scale on the right - hand side and top of the frame , where the squares and double - dotted chain curve are obtained from the numerical calculation and from eq . ( [ lxtrans ] ) , respectively . @xmath196 k , @xmath150 @xmath64 , @xmath197 @xmath64 and @xmath169.,title="fig:",width=302 ] under electric fields with opposite directions : ( a ) @xmath198 kv / cm and ( b ) @xmath199 kv / cm . the squares are from the numerical calculation while the curves are from eq . ( [ lxtrans ] ) . @xmath196 k , @xmath197 @xmath64 and @xmath169.,title="fig:",width=302 ] at last we investigate the spin transport under an electric field along the @xmath11-axis . @xmath196 k , @xmath150 @xmath64 , @xmath200 @xmath64 and @xmath169 . the injected spins are polarized along the @xmath26-axis . in fig . [ figzw6 ] the position dependence of @xmath173 under different electric fields as well as the @xmath152 dependence of @xmath119 ( squares with the scale on the right - hand side and top of the frame ) are plotted . it is shown that while the spin - precession pattern almost keeps the same with varying @xmath152 , the spin transport length is increased ( decreased ) by increasing the electric field along the @xmath201 ( @xmath11)-direction.@xcite these results are in consistence with the analytical study presented in sec . [ stanalytical ] . for comparison , we further plot the @xmath152 dependence of @xmath119 from eq . ( [ lxtrans ] ) by the double - dotted chain curve with the scale also on the right - hand side and the top of the frame . it is shown that the analytical model depicts the spin transport in the low electric - field regime well except when the electric field antiparallel to the spin - injection direction is large ( e.g , a discrepancy reaches 20% when @xmath152 reaches @xmath202 kv / cm ) . the electron density dependence of spin transport is also investigated . in fig . [ figzw7 ] , we plot the density dependence of spin transport length under the electric field parallel ( @xmath198 kv / cm ) and antiparallel ( @xmath199 kv / cm ) to the spin transport direction in ( a ) and ( b ) , respectively . the squares are from the numerical calculation and the curves are from eq . ( [ lxtrans ] ) . it is clearly shown that for the cases with opposite directions of the electric field , the density dependences of spin transport length have opposite tendencies . in conclusion , we have investigated the spin diffusion and transport in graphene monolayer on sio@xmath0 substrate as presented by pi _ et al._,@xcite by means of the ksbe approach . the substrate ( including the impurities initially present ) contributes a rashba spin - orbit coupling field much stronger than the one modulated by the electric field perpendicular to the graphene layer . by surface chemical doping with au adatoms , the rashba spin - orbit coupling coefficient @xmath10 is increased . by fitting the chemical doping dependence of diffusion coefficient and spin relaxation time,@xcite we obtain the information on impurities as well as the chemical doping dependence of @xmath10 . our fitting finds that @xmath10 increases linearly from 0.15 to 0.23 mev with increasing au density when the latter is not so high . with the necessary parameters obtained from fitting , we investigate the spin diffusion and transport in graphene both analytically and numerically . the analytical study with only the electron - impurity scattering included reveals that in the strong scattering limit ( just as the situation under investigation in the present work ) , the spin diffusion is _ uniquely _ determined by @xmath10 . when the injected spins are polarized along the @xmath11- , @xmath66- and @xmath26-axis , the spin diffusion lengths are given by the analytical study with an anisotropy as @xmath203 and @xmath204 . meanwhile , the spatial spin precession is present when the injected spins are polarized in the @xmath11-@xmath26 plane but absent when the injected spins are polarized along the @xmath66-axis . further numerical calculations with all the scattering explicitly included show that the analytical model depicts the spin diffusion pretty well . it is noted that the anisotropy of spin diffusion length from the ksbes differs from the one from the two - component drift - diffusion model where @xmath205 . the qualitative discrepancy indicates the inadequacy of the two - component drift - diffusion model due to the neglect of the off - diagonal spin components , i.e. , the spin coherence . in fact , only when the injected spins are polarized along the @xmath66-axis and the spatial spin precession is absent , the two - component drift - diffusion model gives the same spin diffusion length as the ksbe approach does . the analytical and numerical study of spin transport under an electric field parallel or antiparallel to the spin injection direction is also investigated . in the presence of the electric field , the analytical model depicts the spin transport with a small discrepancy which increases with the strength of the electric field . it is shown that when the electric field is applied , the spin precession in spatial domain for the cases with the injected spins polarized along the @xmath11- and @xmath26-axis remains almost unchanged . however , the spin transport length is increased ( decreased ) by increasing the magnitude of the electric field when it is antiparallel ( parallel ) to the spin transport direction . moreover , in the presence of the electric field , the spin transport becomes sensitive to the electron density , differing from the case of spin diffusion . the spin transport is enhanced ( suppressed ) by increasing electron density when the electric field is parallel ( antiparallel ) to the spin injection direction . this work was supported by the national natural science foundation of china under grant no . one of the authors ( mww ) acknowledges valuable discussions with j. fabian and b. j. van wees . we consider spin relaxation in graphene under the spatial uniform case in the absence of the electric field . we only include the electron - impurity scattering . the ksbes , eq . ( [ ksbee ] ) , are then simplified to be @xmath206-\frac{2\pi}{\hbar}\sum_{{\bf k}^\prime } m_{\bf k - k^\prime}i_{\bf kk^\prime}\\ & & \mbox{}\times\delta(\varepsilon_{\bf k}-\varepsilon_{\bf k^\prime})[\rho_{\mu { \bf k}}(t)-\rho_{\mu { \bf k^\prime}}(t ) ] . \label{a1}\end{aligned}\ ] ] here @xmath207 is the total electron - impurity scattering matrix element contributed by impurities in the substrate and au adatoms . @xmath208 $ ] is the form factor.@xcite by expanding @xmath209 as @xmath210 , one comes to @xmath211+\frac{\alpha_{\rm r}}{2\hbar}[\sigma_-,\rho_{\mu k}^{l-1}(t)]\\ & & \mbox{}-\frac{\rho_{\mu k}^{l}(t)}{\tau^l_k},\end{aligned}\ ] ] where @xmath212 , and @xmath213 with @xmath214 depending only on @xmath215 . it is noted that @xmath216 . retaining the lowest three orders of @xmath217 , i.e. , @xmath218 , @xmath219 , and using the initial conditions @xmath220 , one obtains the second - order differential equation about @xmath221 as @xmath222\\&&\mbox{}+\frac{\alpha_{\rm r}^2}{2\hbar^2}[\sigma_y , [ \sigma_y , \rho_{\mu k}^0(t)]]=0 \label{relax}\end{aligned}\ ] ] with an affiliated initial condition @xmath223 . defining the spin vector as @xmath224 $ ] , one can obtain an equation satisfied by @xmath225 directly from the above one , which reads @xmath226s_{\mu k\alpha}^0(t)=0 \end{aligned}\ ] ] with @xmath227 . with the initial condition @xmath228 , @xmath229 is solved to be @xmath230,\end{aligned}\ ] ] where @xmath231 . when the scattering is strong enough and hence @xmath232 , @xmath233 as a result , for spins polarized along the @xmath11- and @xmath66-axis the spin relaxation times are @xmath234 , while for spins polarized along the @xmath26-axis @xmath235 . from eq . ( [ taukl ] ) one notices that @xmath236 is in fact the momentum relaxation time @xmath237 . for the highly degenerate electron system in graphene , @xmath238 therefore we have @xmath239 . the spin diffusion in the absence of an electric field is also investigated for the case with only the electron - impurity scattering included . performing angle expansion on the steady - state ksbes in a way similar to that shown in appendix [ appa ] , one arrives at @xmath240 -\gamma[\sigma_-,\rho_{\mu k}^{l-1}(x)]\\ & & \mbox{}+\frac{2}{v_{\rm f}}\frac{\rho_{\mu k}^{l}(x)}{\tau_k^l}=0\end{aligned}\ ] ] with @xmath241 . retaining the lowest three orders of @xmath242 one obtains three equations involving @xmath243 as @xmath244-\gamma[\sigma_-,\rho_{\mu k}^{-1}(x)]=0,\label{l0}\\ & & \hspace{-1 cm}\partial_x\rho_{\mu k}^{0}(x)-\gamma[\sigma_-,\rho_{\mu k}^{0}(x)]+\frac{2}{v_{\rm f}}\frac{\rho_{\mu k}^{1}(x)}{\tau_k^1}=0,\label{l1}\\ & & \hspace{-1 cm}\partial_x\rho_{\mu k}^{0}(x)+\gamma[\sigma_+,\rho_{\mu k}^{0}(x)]+\frac{2}{v_{\rm f}}\frac{\rho_{\mu k}^{-1}(x)}{\tau_k^1}=0 . \label{l-1}\end{aligned}\ ] ] from these equations one immediately arrives at eq . ( [ dif ] ) with @xmath236 being irrelevant . by multiplying @xmath36 and performing trace on both sides of eq . ( [ dif ] ) , one gets the equation satisfied by @xmath245 which can be written as @xmath246 with specified boundary conditions , @xmath95 is solved and the total spin signal @xmath97 is obtained by eq . ( [ sum ] ) , as presented in sec . [ sdanalytical ] . explicitly , taking the boundary condition ( i ) given in sec . [ sdanalytical ] as an example , one obtains @xmath95 as @xmath247 with the parameters @xmath118 , @xmath114 , @xmath248 , @xmath249 and @xmath250 given in sec . [ sdanalytical ] . by further summing over @xmath21 and @xmath33 one arrives at eq . ( [ sx ] ) . the analytical study of spin transport is carried out analogly . the driving term from the electric field in the steady state is approximated as @xmath251 then the fourier transformation of the steady - state ksbes reads @xmath240 -\gamma[\sigma_-,\rho_{\mu k}^{l-1}(x)]\\ & & \mbox{}-ee\frac{\partial \rho_{\mu k}^0(x)}{\partial \varepsilon_{\bf k}}(\delta_{l-1}+\delta_{l1})+\frac{2}{v_{\rm f}}\frac{\rho_{\mu k}^{l}(x)}{\tau_k^l}=0.\end{aligned}\ ] ] from this equation one comes to eq . ( [ transport ] ) by retaining the lowest three orders of @xmath242 . the equation satisfied by @xmath245 is then @xmath252 having the experience of solving eq . ( [ seq1 ] ) , we assume that @xmath95 has the solution as @xmath253 and therefore @xmath254 . performing summation over @xmath33 and @xmath21 on both sides of the above equation and using the trick @xmath255{\bf t}(x)&=&-\frac{\int_0^{+\infty}d\varepsilon_{\bf k}s^0_{\mu k}(0)}{\hbar^2v_{\rm f}^2}{\bf t}(x ) \\ & = & -\frac{\frac{1}{\beta}\ln\frac{1+e^{\beta\mu_{\uparrow}}}{1+e^{\beta\mu_{\downarrow}}}}{s(0)\hbar^2v_{\rm f}^2}{\bf s}(x),\end{aligned}\ ] ] one obtains the equation satisfied by @xmath97 as @xmath256 in which @xmath142 is given by eq . ( [ chi ] ) . with the three different typical boundary conditions presented in sec . [ sdanalytical ] , @xmath97 is solved to have the same form as eqs . ( [ sx])-([sz ] ) except that the parameters are now given in sec . [ stanalytical ] . we now calculate @xmath142 in both the degenerate and nondegenerate limits . in the degenerate limit , @xmath257 . making use of the relation @xmath258 , one has @xmath259 where @xmath141 , the magnitude of fermi momentum of unpolarized electrons with density being @xmath62 . in the nondegenerate limit , @xmath260 and @xmath261\\ & = & \frac{1}{\pi(\beta\hbar v_{\rm f})^2}(e^{\beta\mu_{\uparrow}}-e^{\beta\mu_{\downarrow}}),\end{aligned}\ ] ] therefore @xmath262 the two - component drift - diffusion equation can be derived from the ksbes in the collinear spin space@xcite with the @xmath26-axis along the initial spin - polarization direction @xmath159 , by neglecting the spin coherence ( i.e. , the off - diagonal components of the density matrices ) . the density matrices then have the diagonal form as @xmath263 $ ] . in the following we present a brief derivation of the two - component drift - diffusion equation from the ksbes with only the strong electron - impurity scattering considered . other kinds of scattering can also be incorporated similarly under elastic scattering approximation . the spin relaxation time is @xmath264 and the momentum relaxation time is @xmath5 ( both are given in appendix [ appa ] ) and we neglect their momentum dependence hereafter . the two - component drift - diffusion equation is obtained from the equation of continuity and the equation of current , both to be derived from the ksbes . the equation of continuity is derived as follows . by multiplying the ksbes [ eq . ( [ ksbee ] ) ] with @xmath265 and then performing the trace , one obtains the simplified ksbes for each spin band ( @xmath266 , @xmath267 ) as @xmath268 the right - hand side of the above equation comes from the term @xmath269\}$ ] , which can be calculated with the aid of the ksbes in the time domain [ eq . ( [ a1 ] ) ] and the corresponding solution [ eq . ( [ a7 ] ) ] . performing summation over @xmath33 and @xmath21 on both sides of eq . ( [ d1 ] ) in the steady state , one comes to @xmath270 where @xmath271 is the electron density with spin @xmath272 at position @xmath11 . up to the first order of the electric field @xmath152 , the first summation over @xmath21 in the above equation leads to zero when @xmath273 is approximated by @xmath274 , the distribution in equilibrium . defining the charge current along the @xmath11-axis with spin @xmath272 as @xmath275 with @xmath276 , one has the equation of continuity @xmath277 we then calculate the current @xmath278 from the diagonal part of the ksbes @xmath279 where the right - hand side of the equation comes from the electron - impurity scattering ( appendix [ appa ] ) . in the steady state , multiplying @xmath280 on both sides of the equation and then summing over @xmath33 and @xmath21 , one comes to @xmath281 again , up to the first order of the electric field , one has @xmath282 where the mobility @xmath283 and the charge diffusion coefficient @xmath284 . for the case with small spin polarization , @xmath285 in which @xmath286 is the total electron density . finally , the two - component drift - diffusion equation is obtained by combining eqs . ( [ d2 ] ) and ( [ d3 ] ) @xmath287 this equation is consistent with that in the literature.@xcite the equation of @xmath288 then reads @xmath289 with boundary condition @xmath290 , @xmath291 is solved as @xmath292 where the spin transport length@xcite @xmath293^{-1}. \label{d5}\ ] ] when the electric field is zero , @xmath294 as indicated by eqs . ( [ d5 ] ) and ( [ d6 ] ) , in the frame of the two - component drift - diffusion model , the anisotropy of the spin transport or diffusion is solely determined by that of the spin relaxation and no spatial spin precession can be obtained . a. k. geim and k. s. novelia , nature mater . * 6 * , 183 ( 2007 ) . n. tombros , c. jzsa , m. popinciuc , h. t. jonkman , and b. j. van wees , nature ( london ) * 448 * , 571 ( 2007 ) . a. h. c. neto , f. guinea , n. m. r. peres , k. s. novoselov , and a. k. geim , rev . phys . * 81 * , 109 ( 2009 ) . e. h. hang , s. adam , and s. das sarma , phys . * 98 * , 186806 ( 2007 ) . k. s. novoselov , a. k. geim , s. v. morozov , d. jiang , m. i. katsnelson , i. v. grigorieva , s. v. dubonos and a. a. firsov , nature * 438 * , 197 ( 2005 ) . k. s. novoselov , a. k. geim , s. v. morozov , d. jiang , y. zhang , s. v. dubonos , i. v. grigorieva , and a. a. firsov , science * 306 * , 666 ( 2004 ) . k. i. bolotin , k.j . sikes , z. jiang , m. klima , g. fudenberg , j. hone , p. kim and h. l. stormer , sol . state commun . * 146 * , 351 ( 2008 ) . j. fischer , b. trauzettel , and d. loss , phys . b * 80 * , 155401 ( 2009 ) . c. l. kane and e. j. mele , phys . rev . lett . * 95 * , 226801 ( 2005 ) . d. h. hernando , f. guinea , and a. brataas , phys . b * 74 * , 155426 ( 2006 ) . h. min , j. e. hill , n. a. sinitsyn , b. r. sahu , l. kleinman , and a. h. macdonald , phys . b * 74 * , 165310 ( 2006 ) . m. gmitra , s. konschuh , c. ertler , c. a. draxl , and j. fabian , phys . b * 80 * , 235431 ( 2009 ) . c. jzsa , m. popinciuc , n. tombros , h. t. jonkman , and b. j. van wees , phys . rev . lett . * 100 * , 236603 ( 2008 ) . h. goto , a. kanda , t. sato , s. tanaka , y. ootuka , s. odaka , h. miyazaki , k. tsukagoshi , and y. aoyagi , appl . . lett . * 92 * , 212110 ( 2008 ) . e. w. hill , a. k. geim , k. novoselov , f. schedin , and p. blake , ieee trans . mag . * 42 * , 2694 ( 2006 ) . s. cho , y. f. chen , and m. s. fuhrer , appl . lett . * 91 * , 123105 ( 2007 ) . h. haugen , d. h. hernando , and a. brataas , phys . b * 77 * , 115406 ( 2008 ) . w. h. wang , k. pi , y. li , y. f. chiang , p. wei , j. shi , and r. k. kawakami , phys . b * 77 * , 020402(r ) ( 2008 ) . m. ohishi , m. shiraishi , r. nouchi , t. nozaki , t. shinjo , and y. suzuki , jpn . j. appl . phys . * 46 * , l605 ( 2007 ) . l. brey and h. a. fertig , phys . b * 73 * , 195408 ( 2006 ) . v. p. gusynin and s. g. sharapov , phys . lett . * 95 * , 146801 ( 2005 ) . c. l. kane and e. j. mele , phys . lett . * 95 * , 226801 ( 2005 ) . s. v. morozov , k. s. novoselov , m. i. katsnelson , f. schedin , d. c. elias , j. a. jaszczak , and a. k. geim , phys . lett . * 100 * , 016602 ( 2008 ) . x. du , i. skachko , a. barker , and e. y. andrei , nat . nanotechnol . * 3 * , 491 ( 2008 ) . w. han , k. pi , w. bao , k. m. mccreary , y. li , w. h. wang , c. n. lau , and r. k. kawakami , appl . 94 * , 222109 ( 2009 ) . m. popinciuc , c. jzsa , p. j. zomer , n. tombros , a. veligura , h. t. jonkman , and b. j. van wees , phys . b * 80 * , 214427 ( 2009 ) . n. tombros , s. tanabe , a. veligura , c. jzsa , m. popinciuc , h. t. jonkman , and b. j. van wees , phys . * 101 * , 046601 ( 2008 ) . c. jzsa , t. maassen , m. popinciuc , p. j. zomer , a. veligura , h. t. jonkman , and b. j. van wees , phys . b * 80 * , 241403(r ) ( 2009 ) . k. pi , w. han , k. m. mccreary , a. g. swartz , y. li , and r. k. kawakami , phys . * 104 * , 187201 ( 2010 ) . c. ertler , s. konschuh , m. gmitra , and j. fabian , phys . b * 80 * , 041405(r ) ( 2009 ) . r. j. elliott , phys . rev . * 96 * , 266 ( 1954 ) . a. h. c. neto and f. guinea , phys . lett . * 103 * , 026804 ( 2009 ) . a. varykhalov , j. s. barriga , a. m. shikin , c. biswas , e. vescovo , a. rybkin , d. marchenko , and o. rader , phys . * 101 * , 157601 ( 2008 ) . m. i. dyakonov and v. i. perel , zh . fiz . * 60 * , 1954 ( 1971 ) [ sov . jetp * 33 * , 1053 ( 1971 ) ] . m. q. weng and m. w. wu , phys . rev . b * 66 * , 235109 ( 2002 ) . j. l. cheng and m. w. wu , j. appl . phys . * 101 * , 073702 ( 2007 ) . j. l. cheng , m. w. wu , and i. c. da cunha lima , phys . rev . b * 75 * , 205328 ( 2007 ) . p. zhang and m. w. wu , phys . rev . b * 79 * , 075303 ( 2009 ) . b. y. sun , p. zhang , and m. w. wu , semicond . . technol . * 26 * , 075005 ( 2011 ) . , edited by m. ziese and m. j. thornton ( springer , berlin , 2001 ) . z. g. yu and m. e. flatt , phys . rev . b * 66 * , 201202(r ) ( 2002 ) . i. uti , j. fabian , and s. das sarma , phys . lett . * 88 * , 066603 ( 2002 ) . j. fabian , i. uti , s. das sarma , phys . b * 66 * , 165301 ( 2002 ) . j. fabian , a. m. abiague , c. ertler , p. stano , i. uti , acta phys . slov . * 57 * , 565 ( 2007 ) . s. adam and s. das sarma , solid state commun . * 146 * , 356 ( 2008 ) . m. r. ramezanali , m. m. vazifeh , r. asgari , m. polini , and a. h. macdonald , j. phys . a : math . theor . * 42 * , 214015 ( 2009 ) . e. h. hwang and s. das sarma , phys . b * 77 * , 115449 ( 2008 ) m. lazzeri , s. piscanec , f. mauri , a. c. ferrari , and j. robertson , phys . rev 95 * , 236802 ( 2005 ) . s. fratini and f. guinea , phys . rev . b * 77 * , 195415 ( 2008 ) . h. haug and a. p. jauho , _ quantum kinetics in transport and optics of semiconductors _ ( springer , berlin , 1998 ) . y. w. tan , y. zhang , k. bolotin , y. zhao , s. adam , e. h. hwang , s. das sarma , h. l. stormer , and p. kim , phys . lett . * 99 * , 246803 ( 2007 ) . i. appelbaum , b. huang , and d. j. monsma , nature ( london ) * 447 * , 295 ( 2007 ) . g. dresselhaus , phys . rev . * 100 * , 580 ( 1955 ) . j. l. cheng and m. w. wu , j. appl . phys . * 99 * , 083704 ( 2006 ) .

we use detailedradiative transfer calculations to further test the claim of rauch et al . ( 2008 ) that they have detected spatially extended faintemission from the elusive host population of dampedabsorption systems ( dlas ) in their recent ultra - deep spectroscopic survey . we investigate the spatial and spectral distribution ofemission due to star - formation at the centre of dlas , and its dependence on the spatial and velocity structure of the gas . our model simultaneously reproduces the observed properties of dlas and the faintemitters , including the velocity width and column density distribution of dlas and the large spatial extent of the emission of the faint emitters . our modelling confirms previous suggestions that dlas are predominately hosted by dark matter ( dm ) halos in the mass range , and are thus of significantly lower mass than those inferred for lyman break galaxies ( lbgs ) . our modelling suggests that dm halos hosting dlas retain up to 20% of the cosmic baryon fraction in the form of neutral hydrogen , and that star formation at the centre of the halos is responsible for the faintemission . the scattering of a significant fraction of theemission to the observed radii , which can be as large as 50kpc or more , requires the amplitude of the bulk motions of the gas at the centre of the halos to be moderate . the observed space density and size distribution of the emitters together with the incidence rate of dlas suggests that theemission due to star formation has a duty cycle of . [ firstpage ] quasars : absorption lines galaxies : formation

0912.1345

* hereafter ) recently reported the results of an ultra - deep spectroscopic survey for low surface brightness ly@xmath4 emitters at redshift @xmath5 . a 92 hour long exposure with the eso vlt fors2 instrument yielded a sample of 27 faint line emitters with fluxes of a few times @xmath6 erg s@xmath7@xmath8 , which they argue are likely to be dominated by @xmath0 . they further conclude that the large comoving number density , @xmath9 , and the large covering factor @xmath10 suggest that the emitters can be identified with the elusive host population of damped ly@xmath4 systems ( dlas ) and high column density lyman limit systems . * hereafter ) , building on the successful model for dlas of @xcite , presented a simple model that simultaneously accounts for the kinematic properties and incidence rate of the observed dlas _ and _ the luminosity function and the size distribution of the emitters in the context of the @xmath11cdm model for structure formation . the model assumes a simple relation between the size of the damped absorption and @xmath0emission regions , and proposes that the @xmath0luminosity is proportional to the total halo mass . further corroborated the suggestion that cooling radiation is not expected to contribute significantly to the observed @xmath0emission , and that the emitters are most likely powered by star formation . in the model , dlas are small galaxies hosted by dm halos with masses in the range @xmath12 to @xmath13 @xmath14and have rather large low surface brightness @xmath0 halos which extend to radii of up to 50kpc or larger . in order to fit the observed size distribution of the faint @xmath0emitters , assumed that the @xmath0emission extends to radii somewhat larger than is required to reproduce the incidence rate for dlas . however , no modelling of the gas distribution and @xmath0radiative transfer was done . we present such modelling here to investigate whether the sizes , surface brightness profiles and spectral line shapes can be reproduced with simple but plausible assumptions for the distribution and the physical properties of the gas in the dm halos suggested by our previous modelling to be the hosts of the faint emitters . studies of the radiative transfer of @xmath0have a long history . the first investigations employed approximate calculations and simple physical arguments @xcite . later , analytic solutions were found for simple geometries in the limit of large optical depth @xcite . it was realised early , however , that monte carlo simulations provided the most flexible method for investigating arbitrary geometry , density distributions and velocity structures . many investigations have employed such techniques - , ( * ? ? ? * hereafter ) , . our modelling is most similar to that of and , who modelled @xmath0radiative transfer in collapsing proto - galaxies and high - redshift galaxies respectively . in many instances we make similar assumptions as these authors , but our code was developed independently . the paper is structured as follows . in section 2 we give a brief summary of the salient features of our monte - carlo code for @xmath0radiative transfer and show results for standard test problems . in section 3 we discuss our assumptions for the distribution and the physical properties of the gas . we also show the dependence of the surface brightness profile and the spectral line shapes on these assumptions . in section 4 we present the results for a consistent model of the size distribution and the luminosity function of the faint @xmath0emitters . the technical details of the implementation of @xmath0radiative transfer monte - carlo algorithm are described in an appendix . @xmath0is a resonant transition in hydrogen . the optical depth at line centre @xmath15 of a typical @xmath16region is thus generally very large . @xmath0photons , which are expected to be produced copiously in star - forming galaxies , will therefore undergo many scatterings . escape from the @xmath16region usually requires diffusion in both frequency and space . the properties of the emergent radiation depend sensitively on not only the spatial distribution of the gas , but also on its kinematics , ionization state , temperature and dust content . with modern computers monte - carlo sampling of the diffusion process has become the method of choice . we employ such a code for simple spatial and kinematical configurations . our code makes a number of well - tested assumptions and approximations . throughout , we will use a dimensionless frequency variable @xmath17 , the frequency displacement from line centre in units of the doppler frequency width ( see appendix a for details ) . as the photons experience a large number of scatterings before they escape , we can inject all our photons at line centre in the fluid frame of the gas ( @xmath18 ) for simplicity . the photon scattering is partially coherent in the rest frame of the scattering atom , the final frequency of the photon differs from the initial frequency only by the ( often negligible ) effect of atomic recoil . the direction of the photon post - scattering is chosen from a dipole distribution , although choosing an isotropic distribution makes little difference . we also incorporate a cosmic abundance of deuterium ( see ) . finally , we give all results in the rest frame of the centre of mass of the emitter . monte carlo @xmath0rt codes like the one used here can be significantly accelerated by skipping scatterings in the core of the line - profile [ @xcite , ] . in the doppler core , the optical depth is so large that the photon experiences very little spatial diffusion while it awaits an encounter with a rapidly moving atom that will scatter it in frequency space into the wings of the line profile . the gas distributions we consider here are spherically symmetric . we exploit this symmetry by using an arrangement of spherical shells . a detailed description of the implementation and relevant formulae of our code can be found in appendix [ a : lyart ] . there are a number of analytical solutions to @xmath0radiative transfer problems against which we have tested our code . @xcite calculated the redistribution function for the case of coherent scattering in the atom s frame with radiation damping ( i.e. incorporating the lorentzian natural line width ) . the result ( equation 3.12.2 of @xcite ) is @xmath19 , which is defined as the probability that a photon , whose frequency prior to absorption was @xmath17 , is re - emitted with frequency in the range @xmath20 $ ] . in the top left panel of figure [ fig : plottest ] we compare the analytic formula with the output of our code . the thin , smooth line is the analytic solution and the thick histogram is the output from our code for @xmath21 ( blue , red , black respectively ) for @xmath22 k ( @xmath23 ) . the two lines are almost indistinguishable . @xcite and @xcite derived an analytic expression for the spectrum @xmath24 of radiation emerging from an optically thick ( @xmath25 ) , uniform , static slab of neutral hydrogen , where line - centre photons are injected at the centre of the slab , atomic recoil is neglected and @xmath15 is the line - centre , centre - to - edge optical depth . the solution normalised to @xmath26 is given by , @xmath27}.\ ] ] the comparison of our code with this analytic solution is shown for @xmath22k in the top right panel of figure [ fig : plottest ] . the agreement is again very good , especially as @xmath15 increases . next we compare the results of our code to the analogue of the above solution for a uniform , static sphere . as the bottom left panel of figure [ fig : plottest ] shows , the agreement is also very good . to test the code for the case of bulk motions in the gas , we compare our code with the modelling of @xcite . loeb & rybicki investigated @xmath0rt for a uniform @xmath16distribution undergoing hubble expansion with @xmath28 , where @xmath29 is the hubble constant at redshift @xmath30 . for photons blueward of line - centre in the fluid frame , the expansion of the universe will eventually redshift the photon back into resonance . only photons redward of line centre can propagate to infinity . the optical depth to infinity is @xmath31 where @xmath32 is the frequency in the fluid frame . this defines a critical frequency @xmath33 . photons with @xmath34 are redshifted enough to stream freely . we follow here @xcite and use a new frequency variable @xmath35 . we refer the reader to appendix b2 of for a careful discussion of the modifications necessary to permit a meaningful comparison between the two codes . in the bottom right panel of figure [ fig : plottest ] we compare the results of our code with those of @xcite , and . the relevant parameters are : @xmath36 , @xmath37 @xmath38 and @xmath22k . note that @xmath39 . the agreement is again excellent . our knowledge of the spatial distribution and kinematics of the neutral hydrogen in the dlas / faint @xmath0emitters is still somewhat limited . the statistics of the occurrence dlas and their column density distribution give us integral constraints on the spatial distribution of the gas , while the velocity distribution of low ionization species tracing neutral hydrogen gives us some indication of the velocity range of bulk motions . the bulk motions of the gas appear to have velocities that range from a few tens of km / s to several hundred km / s @xcite . the relative contribution of ordered and random motions and the role of gas inflow and galactic winds is , however , still very uncertain @xcite . if the identification of the faint @xmath0emitters as dla host galaxies is indeed correct , then this gives us for the first time constraints on the spatial extent of the gas distribution in individual objects for a sizable ( @xmath0 ) emission selected sample . while there is still significant ambiguity due to the unknown `` duty cycle '' for @xmath0emission ( as discussed by ) , theoretical modelling gives us a handle on the halo masses and virial velocities expected to host the dlas / faint @xmath0emitters . by duty cycle , we mean the fraction of dm host haloes detectable through their @xmath0emission at any given time . there are a number of reasons why this fraction may be smaller than unity . the haloes , for example , may not continuously form stars , but geometrical factors due to a preferential escape of the @xmath0emission in certain directions due to variations in the neutral hydrogen column density ( and/or the dust column density ) should also contribute @xcite . given these uncertainties , we have decided to follow and investigate @xmath0radiative transfer for sources at the centre of dm halos with a range of masses , a duty cycle for @xmath0emission , a simple spherically symmetric gas distributions without dust , and either inflow or outflow with velocities which vary as a power law with radius . this already leads to a rich variety in the predicted surface brightness profiles and spectral shapes and allows us to study the influence of important physical parameters . we begin by specifying the radial distribution of neutral hydrogen in a given halo . the total amount of hydrogen is set relative to the cosmic . ] mass fraction of hydrogen @xmath40 . throughout we assume a helium fraction of @xmath41 . there are a number of reasons to suspect that the hydrogen mass fraction in a typical halo is lower than @xmath42 . firstly , baryons are subject to the smoothing effects of gas pressure . secondly , gas that forms stars is both ionised and extremely compact . stars will also ionise the neutral gas around them this is a source of @xmath0but also reduces the amount of @xmath16that remains to scatter photons . finally , stellar and agn driven galactic winds are expected to drive gas out of galaxies back into the igm . as a first attempt at modelling this effect , we reduce the total mass of baryons in the halo to a fraction @xmath43 of the cosmic value . on top of reducing the amount of neutral hydrogen in a typical halo , it is known that the uv background will significantly ionise the gas in halos too small to self - shield . it is also easier for galactic winds to drive gas out of small , shallow halos . in , this effect was implemented via an exponential suppression of the cross - section of neutral hydrogen below a critical circular velocity @xmath44 . this was necessary in order to fit the observed velocity width distribution of associated low ionization metal absorption . here , we will implement this suppression by reducing the total amount of neutral hydrogen in halos below @xmath44 , such that the total mass of neutral hydrogen in a halo of mass @xmath45 is @xmath46 ( see for a detailed discussion ) . we use the fiducial parameters @xmath47 , @xmath48 . for the radial distribution of the gas , we assume an nfw profile at @xmath49 , characterised by a scale radius @xmath50 defined in @xcite . following the simulations of ( * ? ? ? * equation ( 9 ) ) , we alter the nfw profile to give the halo a thermal core at @xmath51 . the profile is then specified by the total mass of the halo @xmath45 and the concentration parameter @xmath52 . for dependence of the concentration parameter on the mass , we take the mean value of the @xmath53 correlation as given by @xcite , @xmath54 for dark matter , @xcite found that @xmath55 , with a log - normal distribution and a scatter around this mean value of @xmath56 , in agreement of the results of @xcite and @xcite . as we will find later , a significantly larger @xmath57 is appropriate for the baryons ; we will use the column density distribution of dlas to constrain @xmath57 in section [ fnx ] . as we discuss further in section [ limits ] , the gas in the dlas / faint emitters can be expected to self - shield against the meta - galactic ionizing uv background at @xmath58 . the corresponding self - shielding radius in the dm halos we are studying here is generally smaller than the virial radius . we therefore set the outer radius of the @xmath16to be the virial radius in our modelling and also ignore radiative transfer through the igm . we set @xmath59 k as a fiducial temperature . the biggest uncertainty in our modelling is probably the kinematical state of the gas . we follow and for our fiducial model we assume the gas to be infalling with a power - law radial velocity profile given in equation 10 of . ] parameterised by @xmath60 and @xmath4 @xmath61 where @xmath62 is the virial radius , for which we follow the definition of @xcite . values in the range @xmath63 $ ] should be reasonable . the upper limit describes a spherical top - hat collapse , while the lower limit represents the accretion of massless shells onto a point mass ( @xmath64 ) . a spherical top - hat would have @xmath65 . while gas infall will certainly be an important feature of the gas kinematics , star formation driven outflows are also likely to play a role . note that for a spherically symmetric gas distribution , the red and blue side of the line profiles will just be interchanged if the gas is assumed to be outflowing instead of inflowing with the same velocity profile ( ignoring the very slight effects of recoil and deuterium ) . for the more massive and more actively star - forming lbgs , galactic winds have been suggested to sweep up an expanding shell . we investigate such a configuration in section [ sect : shells ] . as shown in the legend . _ bottom _ : the column density profiles for the fiducial model with the total mass as given in the legend in units of @xmath66 . the column density is calculated along a sightline that passes all the way through the halo at a distance @xmath67 from its centre . note that the @xmath68 line has been boosted by a factor of @xmath69 to make it visible on the given axes . the horizontal dashed line indicates the minimum column density of a dla.,title="fig:",scaledwidth=40.0% ] as shown in the legend . _ bottom _ : the column density profiles for the fiducial model with the total mass as given in the legend in units of @xmath66 . the column density is calculated along a sightline that passes all the way through the halo at a distance @xmath67 from its centre . note that the @xmath68 line has been boosted by a factor of @xmath69 to make it visible on the given axes . the horizontal dashed line indicates the minimum column density of a dla.,title="fig:",scaledwidth=40.0% ] and identified star - formation as the most likely source for the faint emitters . @xcite used continuum emission to place stringent limits on _ extended _ star formation in dlas . we have thus assumed a centrally - peaked emissivity all photons are created at @xmath70 . before we attempt to model the data of in detail , we will consider the effect on the spectra and surface brightness distribution of altering the parameters of our model . our fiducial model parameters are @xmath71 . the values of @xmath57 and @xmath43 that we have chosen will be justified in section [ fnx ] . the surface brightness @xmath72 scales with the total luminosity @xmath73 , which will be given in the caption to each figure . the values of the luminosity chosen will be justified in section [ sizelum ] . the velocity and column density profiles for the fiducial model are shown in figure [ fig : nrvr ] , where the column density is as seen along a sightline that passes all the way through the halo at an impact parameter @xmath74 . in this section , we will consider the effects of changing the most important parameters in the model : mass , concentration , velocity profile ( parameterized by the power law index @xmath4 ) and baryon fraction . further parameters ( temperature and amplitude of the bulk motions of the gas ) are discussed in appendix [ a : tvampz ] . the models of give us a handle on the masses of the halos that host dlas . figure [ fig : fbmass ] considers halos with masses of @xmath75 , which at @xmath76 correspond to @xmath77 , @xmath78 kpc , @xmath79 arcsec , where @xmath80 is the angular radius corresponding to @xmath62 . we see that , as the mass increases , the emerging photons emerge bluer . this is because , as we add more gas , the central column densities increase and the photons must shift further from line centre in order to escape . the @xmath81 profiles are increasingly double - peaked for lower masses , while the @xmath82 profiles only ever have one , blue peak . this is because the innermost region of the @xmath81 halo has the smallest bulk velocity , and thus most resembles the uniform , static sphere of figure [ fig : plottest ] . as the bulk velocities increase at the centre of the halo , the amount of energy transferred between the gas and photons in each scattering is increased , favouring one of the two peaks ( the blue / red peak for inflowing / outflowing gas respectively ) . the surface brightness profile shows that the dominant effect in increasing the mass is that the virial radius ( which we have assumed to be the outer radius of the @xmath16 ) increases . the @xmath82 profile is much more centrally peaked than the @xmath81 profile . the reason is that the larger bulk velocities at the centre of the @xmath82 halo can shift the photon into the wings of the spectral line , resulting in reduced spatial diffusion . for the concentration parameter , we have considered the values @xmath83 and @xmath84 . the spectra and surface brightness profiles for these models are shown in figure [ fig : fbcv ] . as the concentration increases , the photons generally emerge bluer . the @xmath81 profile becomes _ more _ double - peaked as the concentration increases , because the @xmath16column density increases at smaller radii , where the bulk velocity is lower . the surface brightness profile is also more centrally peaked for higher concentrations , as more scatterings occur at smaller radii . the effect of changing the baryonic fraction @xmath43 is shown in figure [ fig : fbfe ] . as baryons are removed from the halo , the spectrum shifts toward @xmath85 and the surface brightness profile becomes more centrally peaked , as there is less gas in the outer parts of the halo to scatter the photons . the baryon fraction where the photons are not scattered efficiently anymore to the virial radius ( where we have the gas distribution assumed to cut - off ) corresponds to a hi column density of about @xmath86 . for very small @xmath43 and @xmath87 , some of the photons can escape the halo without scattering at all , creating a very narrow peak at @xmath85 . some of the spectra also show a trough at @xmath88 due to deuterium . note , however , that for @xmath89 the gas in the halos would not be able to self - shield anymore against the meta - galactic uv background at @xmath58 so the model would be internally inconsistent for such small baryonic fractions . for the much brighter @xmath0emission seen in many lbgs , the @xmath0emission is systematically red - shifted by several hundred @xmath90 . this is generally attributed to backscattering of the @xmath0radiation from a wind - driven expanding shell . used a @xmath0rt code similar to ours to calculate the emergent spectrum from an expanding shell . we will investigate here in a similar fashion if this picture could apply to the emitters . we consider a shell of @xmath16 , where the inner radius is a fraction @xmath91 times the outer radius @xmath92 . the shell is expanding at a uniform velocity @xmath93 , and has a column density of @xmath94 . the temperature is set by the doppler velocity @xmath95 . our fiducial model has the parameters : @xmath96 , similarly to the fiducial model of verhamme et al . for the outer radius of the shell @xmath92 we have a chosen a value similar to the models of the previous sections . figure [ fig : fbsh ] shows the effect of altering the parameters that verhamme et al . found to be most important , the column density and the expansion velocity . note that , for small column densities ( upper panels ) , a certain fraction of photons were able to escape directly without scattering . these photons form a very sharp peak at @xmath85 , and at small impact parameter , which has been removed from these plots for clarity . as the column density increases ( top panels ) , the spectrum becomes redder , and the small blue peak disappears , as the photons must scatter further from line centre in order to escape . the most significant change in the surface brightness profile occurs for small column densities , where some of the photons are able to escape directly , leaving fewer to scatter at large radii . as the expansion velocity increases ( bottom panels ) , the back - scattering mechanism becomes more pronounced . photons which scatter off the far side of the shell back through its interior are far enough from line centre to escape through the front side of the shell . this mechanism creates the two red peaks ( the reddest peak comes from photons that backscatter more than once ) . at small velocities , we approach a profile similar to the static sphere . the surface brightness profiles are very flat when compared to the infall / outflow models of the gas in an nfw halo which we had considered before . qualitatively , they appear much flatter than the profiles of the emitters . almost all of the emitters show a central peak . one way to produce a central peak for a shell geometry is for the column density to be low enough for photons to be able to escape directly ; the resolution of the instrument then broadens the delta function at @xmath97 into a central peak . this would limit the column density of the shell to @xmath98 . we conclude that the emitters appear unlikely to be halo - scale expanding shells of @xmath16around a central @xmath0source . the radial ( column ) density distribution and the velocity field are the physical properties which most strongly affect the spectral distribution of the @xmath0emission in our modelling . for a spherical density distribution with neither outflow nor infall , the distance between the peaks increases with increasing optical depth in the same way as the uniform static slab solution . infall and outflow lead to a suppression of the red / blue peak and an increased shift of the opposite peak . the suppression increases with increasing velocity amplitude and optical depth . the surface brightness profile depends strongly on the radial velocity profile . with increasing amplitude of the bulk motion , the diffusion in frequency space is accelerated and the emission becomes more centrally peaked . in our model with increasing amplitude towards decreasing radius ( @xmath82 ) , the average bulk motions are larger and the effect is more pronounced . in the model with the expanding shell , the photons can travel unimpeded until they encounter the shell which leads to a rather flat surface brightness profile . the photons diffuse radially until the column density of neutral hydrogen density drops to values around @xmath86 somewhat dependent on the spatial profile of neutral hydrogen and the velocity field . the faint extended @xmath0emission should thus have a rather sharp edge which is defined by the surface inside which the gas is able to self - shield and the optical depth rises rapidly . we begin with a qualitative summary of the properties of the emitters . note , however , that due to the faintness of the sources the spectral and spatial profiles are rather noisy , making it difficult to identify signatures of inflow ( more prominent blue peak ) or outflow ( more prominent red peak ) . as discussed in , for 12 of the 27 spectra only a single emission peak is visible while six / three of the spectra show a weak secondary blue / red counter - peak . the remaining spectra are extended in frequency space without a clear peak structure . the widths of the spectral peaks ranges from @xmath99 , which corresponds to @xmath100 for gas with a temperature of @xmath101 k. the surface brightness profiles are predominantly centrally peaked with wings that often extend well beyond the gaussian core of the psf . this is particularly true of the brightest sources , while the fainter sources are more difficult to characterise due to the noise . modelling the radial distribution of neutral hydrogen in dm halos allows us to calculate the column density distribution along sightlines that intersect the halos . comparison with the observed column distribution of dlas provides a useful constraint in this regard . we will use here the same notation as . the column density distribution is defined such that the number of systems ( @xmath102 ) intersected by a random line of sight between absorption distance ] . ] @xmath103 and @xmath104 , with @xmath16column density between @xmath105 and @xmath106 is , @xmath107 for our model we calculate this quantity using the ps . we need two ingredients . the first is @xmath108 , the mass function of dark matter halos , as calculated by @xcite . the second ingredient is the column density of neutral hydrogen in a given halo ( of mass @xmath109 at absorption distance @xmath103 ) , as a function of the ( physical ) impact parameter @xmath74 , @xmath110 . this is calculated from the neutral hydrogen density as a function of radius . given that @xmath105 is a monotonically decreasing ( and thus invertible ) function of @xmath74 , the region between @xmath105 and @xmath106 is an annulus with cross - sectional area @xmath111 . hence , we can write the column density distribution as , @xmath112 the results are shown in figure [ fig : fnx ] . the most important parameters ( given that we integrate over all m ) are the normalisation of the concentration parameter for the baryons @xmath57 , and the baryon fraction @xmath43 relative to the cosmic value . a good fit to the observed column density distribution is obtained for @xmath113 and @xmath114 . the figure demonstrates the effect of altering these two parameters . decreasing @xmath43 decreases the overall normalisation , while @xmath57 mostly affects the high column density end of the distribution . changing @xmath44 has a comparatively small effect . note that our model somewhat overpredicts the number of absorption systems with column densities below @xmath115 , in the regime of lyman limit systems ( lls ) . in this column density range the observed @xmath116 flattens significantly @xcite , an effect which can be attributed to the inability of ( super ) lls to self - shield completely against the meta - galactic uv background @xcite . since we have not attempted to model the ionization of the gas and in particular the self - shielding of the gas in any detail , it is not surprising that our model does not reproduce this . the uv background should , however , not effect our results for the population of predominantly neutral dlas . figure 5 shows that in the lls regime the inferred size of the emission region depends very weakly on column density . for our model ; the black crosses show the data of @xcite . the upper panel shows the effect of changing the concentration parameter of the radial gas density profiles . the lower panel shows the effect of altering the baryon fraction @xmath43 ( defined relative to the cosmic value).,title="fig:",scaledwidth=47.0% ] for our model ; the black crosses show the data of @xcite . the upper panel shows the effect of changing the concentration parameter of the radial gas density profiles . the lower panel shows the effect of altering the baryon fraction @xmath43 ( defined relative to the cosmic value).,title="fig:",scaledwidth=47.0% ] similarly to , we can also calculate the predicted probability distribution of the velocity width @xmath117 of the associated low ionization metal absorption of dlas for our model as follows , @xmath118 where @xmath119 is the conditional probability distribution as discussed in section 2 of . the dla cross - section is given by @xmath120 , where @xmath121 . the result is shown in figure [ fig : vw ] , along with the observational data of of the associated low - ionization metal absorption of dlas . the black crosses show the observational data compiled in figure 10 of . the legend shows the parameter @xmath122 , below which the baryonic fraction is assumed to be suppressed due to the effect of photo - heating and/or galactic winds.,scaledwidth=47.0% ] as in our model fits the data well with values of @xmath122 in the range @xmath123 . we should point out that the probability distribution @xmath124 was originally derived from simulations which do not include the effect of galactic winds and where the distribution of gas in a given halo is somewhat different from what we have assumed here ( see @xcite for a more detailed discussion ) . we should also emphasize that , rather than using a simple power - law scaling for the absorption cross section of dlas as in , we have used here a radial distribution of neutral hydrogen which is simultaneously consistent with the column density distribution of dlas _ and _ the size distribution of the rauch et al . emitters , as we will see in the next section . we will not attempt to fit the individual rather noisy spectral and surface brightness profiles of the emitters here , but we will instead focus on the statistical properties of the population of emitters . we calculate the cross - section weighted size distribution @xmath125 expected from our model as follows . the observations of @xcite achieved a 1 @xmath126 surface brightness detection limit of @xmath127 . we calculated the expected observed size of our model emitters by determining the radius @xmath67 ( or equivalently , impact parameter ) at which the surface brightness drops below the limit , @xmath128 . this procedure gives the radius of the emitter as a function of the mass of the halo . to do this , we need to specify the intrinsic @xmath0luminosity @xmath73 as a function of mass . similarly to , we assume that the luminosity is proportional and with @xmath129 , @xmath130 defined below has the same value as in . ] to the total mass of neutral hydrogen , @xmath131 this means that the luminosity is subject to the same exponential suppression as the ( neutral ) gas content of dm halos for small circular velocities . of the @xmath0emitters , compared with the observations of rauch et . al ( black curve ) . the coloured solid curves are for @xmath82 , while the dashed curves are for @xmath81 . the red and green lines are for the @xmath132 model , with the values of @xmath130 as given in the legend . the blue curve assumes that @xmath133 , that is , assuming we can see emission all the way to the virial radius . the lines have been normalised to @xmath134 assuming a duty cycle of the @xmath0emission with @xmath135 , such that @xmath136 . the values of @xmath135 for each model ( in the order they appear in the legend ) are @xmath137 for @xmath82 and @xmath138 for @xmath81 . the models with the two values in bold correspond best to the data.,scaledwidth=48.0% ] we now calculate the size distribution in the form of the inferred cumulative incidence rate @xmath139 and compare it with the data of in figure [ fig : dndz_r ] . the curves have been normalised to @xmath140 by assuming the emission occurs with a duty cycle @xmath135 , @xmath141 . the values of @xmath135 for each model are given in the caption to the figure . the solid curves are for @xmath82 , while the dashed curves are for @xmath81 . the red and green curves assume @xmath142 , with the values of @xmath130 as given in the legend . the blue curve shows the size distribution assuming that @xmath133 , that is , assuming that the emission is detected all the way to the virial radius . as we have defined the size of the emitters at a fixed flux level @xmath143 , the observed size of the halos becomes larger as @xmath130 increases . for @xmath82 the region with emission above this flux level is smaller than in the model with @xmath81 halos , due to the more centrally concentrated surface brightness profile . the size distribution flattens at small @xmath67 due to the exponential suppression of the luminosity at low @xmath144 . figure [ fig : dndz_r ] shows that we can find values of @xmath145 that fit the observed cumulative size distribution well , where the value of @xmath145 depends on @xmath4 . the required duty cycle is @xmath146 , and is fairly insensitive to @xmath4 . the value of @xmath145 for @xmath82 is rather high , due to the very peaked surface brightness profiles in this case . of our model , along with the rauch et al . data ( black solid curve with errors indicated by the grey dot - dashed curves ) . the same models are shown as in figure [ fig : dndz_r ] , except that the duty cycle is chosen to match the luminosity function data . the parameters for each model ( in the order they appear in the legend ) are ( with negligible dependence on @xmath4 ) : @xmath147 . , scaledwidth=48.0% ] figure [ fig : ngm_l ] shows the corresponding luminosity distribution @xmath148 , along with the data ( black solid curve ) . note that the luminosity @xmath73 predicted by our model is calculated by integrating the surface brightness inside the radius @xmath67 where the surface brightness is above the observational limit . photons that are scattered to radii where the surface brightness falls below this limit are lost in the noise ; the observed luminosity is thus always less than the intrinsic luminosity . note further that the observed luminosities have not been corrected for `` slit losses '' . the actual @xmath0luminosities may thus be a factor two or more larger . note finally , that the intrinsic @xmath0emission could be significantly larger due to absorption by dust . the figure shows the same models as in figure [ fig : dndz_r ] , except that the duty cycle is chosen to fit the luminosity function data . the solid and dashed curves are very close , meaning that there is practically no dependence on @xmath4 . the corresponding values of @xmath135 are given in the caption to the figure . the observed lumininosity function is reproduced reasonable well by our model if we assume @xmath149 , the value required to match the observed size distribution for our model with @xmath150 this model predicts a somewhat steeper faint end slope than appears to be observed , but as discussed in it is uncertain whether the turn - over in the observed luminosity function is real or due to incomplete identification of emitters close to the detection threshold . should the turnover consolidate with deeper data then this may suggest a somewhat sharper cut - off of the efficiency for @xmath0emission in shallow potential wells as we had e.g. assumed in . the required duty cycle is @xmath151 and thus comparable to the value required to fit the observed size distribution . with @xmath152 , @xmath149 , @xmath153 our model therefore fits both the observed size and luminosity distribution of the emitters . as previously discussed , the model with @xmath154 has strongly centrally peaked emission . as is apparent from figure [ fig : ngm_l ] this leads to a mismatch with the observed luminosity function if we fix @xmath155 to fit the observed size distribution for this value of @xmath4 . even for a significantly smaller duty cycle of @xmath156 we can not match the shape of the observed luminosity function , which in this case is significantly steeper than predicted by the model over the full range of luminosities . we therefore conclude that we can successfully reproduce both the absorption properties of dlas and the @xmath0emission data of , using a self - consistent model with centrally concentrated , star - formation powered @xmath0production with a duty cycle of @xmath157 , coupled with radiative transfer effects that set the observed size of the emitters _ if _ the velocity field of the gas facilitates the scattering of the photons to large radii with moderate central bulk velocities as in the model with @xmath81 . our model s success in simultaneously reproducing the absorption properties of dlas significantly strengthens the assertion of @xcite that the faint emitters are in fact the host galaxies of dlas . ) . the same models are shown as in figure [ fig : dndz_r ] . the majority of dlas / emitters have masses in the range @xmath158 . we refer the reader to for a comparison of our prediction of @xmath159 to that of numerical simulations . , scaledwidth=48.0% ] finally we calculate @xmath159 , the contribution to the incidence rate from halos of different masses . the result is shown in figure [ fig : dndxm ] . the majority of dlas / emitters have masses in the range @xmath160 for all models . this is a similar range of masses to that found in the numerical simulations of @xcite . we refer the reader to for a more detailed comparison of our prediction of @xmath159 with that of numerical simulations . the success of @xcite in reproducing the observed metalicity distribution is additional evidence in favour of this mass range , as are the observed spatial correlation of dlas and lbgs @xcite . in , we presented modelling in which we assumed simple scaling relations for the sizes and luminosities of the rauch et al . emitters with the masses / circular velocities of their dm host haloes . the scaling relation for the sizes was motivated by the modelling of the velocity width distribution of the associated low ionization metal absorption of dlas by @xcite . no attempt to model the spatial and spectral distribution of the @xmath0emission was made in . it is thus gratifying that our modeling here with @xmath0radiative transfer through inflowing / outflowing distributions of neutral hydrogen in the emitters for which the radial profile is constrained by the observed column density ditribution is able to still fit the luminosity function and size distribution of the emitters as well as the velocity width distribution of dlas for a specific choice of the scaling of the inflow / outflow velocity with radius . this is a non - trivial result and means that the radiative transfer calculations as constrained by the observed properties of the rauch et al . emitters and dlas produce a scaling relation of the sizes of emitters with the masses / circular velocities of their dm host haloes very simular to that assumed by . our modelling has a number of limitations and ( over- ) simplifications that we will discuss in this section . firstly , we have assumed spherical symmetry throughout . deviations from a uniform spherical configuration will most likely make escape easier for @xmath0photons in certain directions . similarly , our simple velocity field is certainly an oversimplification . more complex velocity fields may bring photons passing through gas in the outer regions of the halo back toward line centre . this may also increase the spatial extent of the emitter . we have also ignored radiative transfer outside the virial radius of the halo . this effectively assumes not only that scattering in the outer parts of dm halos is negligible but that the same is true for scattering by the neutral hydrogen contained in the igm . whether these are reasonable assumptions will depend both on the poorly constrained distribution of neutral hydrogen in the outer parts of low - mass dm halos and on the bulk motion of the scattering gas relative to the igm ( see e.g. @xcite ) . we have furthermore not modelled the effect of ionising radiation , either from external or internal sources of ionising photons , except by reducing the amount of neutral hydrogen within the virial radius of the dm halos below its cosmic value , ostensibly due to the effects of the uv background . more sophisticated modelling should take into account the self - shielding of the gas against the meta - galactic uv background self - consistently . we have checked here that the radius at which the gas is able to maintain neutrality by self - shielding is @xmath161 , the size of the emission region at the flux level of the rauch et al . emitters as defined in section [ sizelum ] . the meta - galactic uv background should thus have little effect on the size of the @xmath0emission region predicted by our modelling . note , however , that this will not be the case anymore at fainter flux levels as the neutral hydrogen density / column density will drop sharply outside the region able to self - shield . internal sources of ionising stars such as stars would reduce the neutral fraction of the gas , and in an inhomogeneous manner . neither stars nor the uv background are expected to significantly ionise the bulk of the neutral gas in the proto - galaxies studied here these are the sites of dlas after all . spatially extended star formation is likely to increase the production of @xmath0 at large radii , possibly increasing the observed size of the emitters . finally , we have also ignored the effects of dust . our assumption of a negligible effect of dust may not be unreasonable as dlas are known to have a rather small dust content and the escape fraction for @xmath0emission appears to increase strongly towards unity with decreasing stellar mass of the emitters @xcite . the recent modelling of @xmath0radiative transfer by @xcite including dust also suggest that for the majority of dlas , with their rather low metallicity which are hosted by rather low mass halos , the effect of dust is probably not too important . note however that there are counter - examples . as shown by , models with almost complete absorption of dust can explain the fact that the expected @xmath0emission from the star formation in izw 18 appears to be entirely absorbed despite a rather modest reddening . as discussed in section 3.1.1 , the presence of dust may actually be ( partially ) responsible for the smaller than unity duty cycle which we require to fit the luminosity function . note also that the effect of dust should depend sensitively on the clumpiness of the neutral gas @xcite . we have used a monte - carlo radiative transfer code to model the spatial and frequency distribution of the @xmath0emission due to star formation in ( proto-)galaxies at the centre of dm halos with masses of @xmath12 to @xmath162 for a range of assumptions for the spatial distribution and the dynamical state of neutral hydrogen . dm halos in this mass range had been previously identified as the likely hosts of dlas and the recently detected population of faint spatially extended @xmath0emitters . our main results are the following : * as previously found by other authors , the spectral shape of the @xmath0emission from star formation in galaxies for which the dynamics of neutral hydrogen is dominated by infall / outflow is characterized by a strong blue / red peak , occasionally accompanied by a weaker red / blue peak . the spectral shape is very sensitive to the spatial distribution , velocity structure and ( to a smaller extent ) the temperature structure of the gas . the larger central column densities in the more massive galaxies / halos make escape of the @xmath0photons more difficult . the photons must then scatter further in frequency and space and thus emerge with a larger frequency shift and the emission extends to larger radii . larger bulk motions lead to more energy per scattering being transferred between the photons and the gas . this results in a larger frequency shift of the dominant spectral peak and a larger contrast between the strong and weak spectral peak . * the surface brightness profiles for photons emitted at the centre of the halos show a central peak with wings extending as far as our assumed neutral hydrogen distribution as long as the column density of neutral hydrogen exceeds about @xmath163 . the spatial profile of the emission is likewise sensitive to the spatial distribution , velocity and temperature structure of the gas . the spatial distribution is significantly more centrally peaked when the amplitude of the bulk motions increase toward the centre of the halo . * expanding shells of neutral hydrogen similar to those invoked to explain the @xmath0emission from lbgs produce spectra with one or more prominent red peaks . the surface brightness profiles are very flat , remaining essentially constant for 75% of the radius of the shell . this appears at odds with the observed profiles of the emitters , almost all of which show a central peak . * our modelling simultaneously reproduces the column density distribution of the neutral hydrogen and the velocity width of the associated low ionisation metal absorption of dlas , as well as the size distribution and the luminosity function of the rauch et al . emitters if we assume i ) that absorbers and emitters are hosted by dm halos that retain about 20% of the cosmic baryon fraction in the form of neutral hydrogen , with a spatial distribution which follows a nfw profile with concentration parameter @xmath164 times larger than that of the dark matter , ii ) that absorbers and emitters are hosted by dm halos with virial velocities @xmath165 , and iii ) that the central @xmath0emission due to star formation has a duty cycle of @xmath166 and the luminosity is proportional to the mass of neutral hydrogen in the dm halos . * the dm halos that contribute most to the incidence rate of dlas have masses in the range @xmath167 - @xmath168 and virial velocities in the range of 35 to 230 @xmath90 . the lower cut - off is mainly determined by the rather sharp decrease in the velocity width distribution of the associated low - ionization metal absorption in dlas at velocity width @xmath169 but may also be reflected in the turn - over of the @xmath0luminosity function at the faintest fluxes . the dm host halo masses are significantly smaller than those inferred for @xmath170 lbgs , consistent with the much higher space density of the faint emitters . the success of our detailed @xmath0radiative transfer modelling in explaining the observed properties of both dlas and the faint rauch et al emitters with a consistent set of assumptions further strengthens the suggestion that the faint emitters are indeed the long - searched for host galaxies of dla / lls . together with our modelling , the observed properties of the faint emitters should thus provide robust estimates not only of the space density , @xmath0luminosity and extent of the gas distribution but also of the masses of the dm halos , the duty cycle of star formation and the spatial profile and kinematics of the gas distribution of a statistically representative sample of dla host galaxies . the current ultra - deep spectroscopic surveys in the hudf and hdf should soon provide important additional information on the stellar content and possibly also dust content of these objects which will allow to further test the nature of what are almost certainly the building blocks of typical present - day galaxies like our own . lab is supported by an overseas research scholarship and the cambridge commonwealth trust . we thank richard bower and bob carswell for comments and the referee for a helpful report which has improved the manuscript . 99 adams t. f. , 1971 , apj , 168 , 575 adams t. f. , 1972 , apj , 174 , 439 ahn s .- , lee h .- w . , lee h. m. , 2000 , jkas , 33 , 29 ahn s .- , lee h .- w . , lee h. m. , 2001 , apj , 554 , 604 ahn s .- , lee h .- w . , lee h. m. , 2002 , apj , 567 , 922 atek h. , schaerer d. , kunth d. , 2009 , a&a , 502 , 791 auer l. h. , 1968 , apj , 153 , 783 avery l. w. , house l. l. , 1968 , apj , 152 , 493 barnes l. a. , haehnelt m. g. , 2009 , mnras , 397 , 511 birnboim y. , dekel a. , 2003 , mnras , 345 , 349 bonilha j. r. m. , ferch r. , salpeter e. e. , slater g. , noerdlinger p. d. , 1979 , apj , 233 , 649 bullock j. s. , kolatt t. s. , sigad y. , somerville r. s. , kravtsov a. v. , klypin a. a. , primack j. r. , dekel a. , 2001 , mnras , 321 , 559 cantalupo s. , porciani c. , lilly s. j. , miniati f. , 2005 , apj , 628 , 61 caroff l. j. , noerdlinger p. d. , scargle j. d. , 1972 , apj , 176 , 439 cooke j. , wolfe a. m. , gawiser e. , prochaska j. x. , 2006 , apj , 652 , 994 dijkstra m. , haiman z. , spaans m. , 2006 , apj , 649 , 14 dijkstra m. , loeb a. , 2009 , arxiv , arxiv:0902.2999 gould a. , weinberg d. h. , 1996 , apj , 468 , 462 haehnelt m. g. , steinmetz m. , rauch m. , 1998 , apj , 495 , 647 haehnelt m. g. , steinmetz m. , rauch m. , 2000 , apj , 534 , 594 hansen m. , oh s. p. , 2006 , mnras , 367 , 979 harrington j. p. , 1973 , mnras , 162 , 43 hummer d. g. , 1962 , mnras , 125 , 21 kere d. , katz n. , weinberg d. h. , dav r. , 2005 , mnras , 363 , 2 kere d. , katz n. , fardal m. , dav r. , weinberg d. h. , 2009 , mnras , 375 laursen p. , razoumov a. o. , sommer - larsen j. , 2009 , apj , 696 , 853 laursen p. , sommer - larsen j. , andersen a. c. , 2009 , arxiv , arxiv:0907.2698 loeb a. , rybicki g. b. , 1999 , apj , 524 , 527 macci a. v. , dutton a. a. , van den bosch f. c. , moore b. , potter d. , stadel j. , 2007 , mnras , 378 , 55 maller a. h. , bullock j. s. , 2004 , mnras , 355 , 694 nagamine k. , wolfe a. m. , hernquist l. , springel v. , 2007 , apj , 660 , 945 navarro j. f. , frenk c. s. , white s. d. m. , 1996 , apj , 462 , 563 natta a. , beckwith s. , 1986 , a&a , 158 , 310 neufeld d. a. , 1990 , apj , 350 , 216 omeara j. m. , prochaska j. x. , burles s. , prochter g. , bernstein r. a. , burgess k. m. , 2007 , apj , 656 , 666 ono y. , et al . , 2009 , arxiv , arxiv:0911.2544 osterbrock d. e. , 1962 , apj , 135 , 195 panagia n. , ranieri m. , 1973 , a&a , 24 , 219 pettini m. , steidel c. c. , adelberger k. l. , dickinson m. , giavalisco m. , 2000 , apj , 528 , 96 pettini m. , rix s. a. , steidel c. c. , adelberger k. l. , hunt m. p. , shapley a. e. , 2002 , apj , 569 , 742 pontzen a. , et al . , 2008 , mnras , 390 , 1349 prochaska j. x. , wolfe a. m. , 1997 , apj , 487 , 73 prochaska j. x. , wolfe a. m. , 2009 , apj , 696 , 1543 quider a. m. , pettini m. , shapley a. e. , steidel c. c. , 2009 , mnras , 398 , 1263 rauch m. , et al . , 2008 , apj , 681 , 856 razoumov a. o. , norman m. l. , prochaska j. x. , sommer - larsen j. , wolfe a. m. , yang y .- j . , 2008 , apj , 683 , 149 rybicki g. b. , lightman a. p. , 1979 , rpa .. book santos m. r. , 2004 , mnras , 349 , 1137 schaerer , d. , verhamme , a. , 2008 , a&a , , 480 , 369 sheth r. k. , tormen g. , 2002 , mnras , 329 , 61 tasitsiomi a. , 2006 , apj , 645 , 792 tescari e. , viel m. , tornatore l. , borgani s. , 2009 , mnras , 397 , 411 unno w. , 1952 , pasj , 4 , 100 verhamme a. , schaerer d. , maselli a. , 2006 , a&a , 460 , 397 verhamme a. , schaerer d. , atek h. , tapken c. , 2008 , a&a , 491 , 89 veilleux s. , cecil g. , bland - hawthorn j. , 2005 , ara&a , 43 , 769 wechsler r. h. , bullock j. s. , primack j. r. , kravtsov a. v. , dekel a. , 2002 , apj , 568 , 52 wolfe a. m. , gawiser e. , prochaska j. x. , 2005 , ara&a , 43 , 861 wolfe a. m. , chen h .- w . , 2006 , apj , 652 , 981 zanstra h. , 1949 , ban , 11 , 1 zheng z. , miralda - escud j. , 2002 , apj , 568 , l71 zheng z. , miralda - escud j. , 2002 , apj , 578 , 33 the details of each step in our monte carlo code for @xmath0rt are outlined below throughout , @xmath171 ( for @xmath172 ) denotes a random number generated uniformly between 0 and 1 . 1 . we begin by specifying , as a function of position , @xmath173 * @xmath174 , the number density of hi , * @xmath175 , the @xmath0emissivity ( in photons / s/@xmath176 ) , * @xmath177 , the bulk velocity field of hi , and * @xmath178 , the temperature of hi . + for the modelling in this paper , these quantities are discussed in section [ s : gasmodel ] . [ generate ] we generate a photon at an initial position @xmath179 according to the emissivity . we then choose the photon s initial direction @xmath180from an isotropic distribution , @xmath181 where @xmath182 + we generate the photon s initial frequency as follows . the emission ( and absorption ) profile in the rest frame of the emitting atom is assumed to be lorentzian , @xmath183 where @xmath184 is the central @xmath0frequency and @xmath185 is the natural line width . we take into account the thermal doppler broadening with an maxwell - boltzmann ( mb ) distribution of the velocity of the scattering atoms . the line profile can then be written as an average of the lorentz profile over the atoms velocities , @xmath186 where the argument of @xmath187 takes into account the doppler shift in the frequency of the photon as seen by an atom with velocity component @xmath188 in the direction of the photon s motion . the other quantities are @xmath189 the thermal velocity dispersion , relative line width , doppler frequency width and frequency displacement in units of @xmath190 , respectively . we will use @xmath17 as our frequency variable . we have also made use of the _ voigt function _ , @xmath191 the voigt function can be approximated as a doppler core and lorentz wings , @xmath192 the transition between the two occurs at a ` critical ' frequency @xmath193 , defined as the solution to @xmath194 . for @xmath59 k , we have @xmath195 . we will use the approximation to the voigt function given in @xcite . + equation assumes that the `` laboratory frame '' ( in which we are measuring the photon s frequency ) is the same as the fluid rest frame . in order to take account of the bulk fluid velocity , we replace the frequency in the lab frame , @xmath17 , with the frequency as seen by the fluid , @xmath196 + in practice , the photons are very likely to be emitted close to line centre , and because the photons suffers a large number of resonant scatterings , any `` memory '' of the initial frequency is quickly erased . thus , we will usually inject all photons at line centre in the fluid frame ( @xmath197 ) . [ choosedist ] the distance travelled depends on the optical depth . the probability that the photon propagates a physical distance corresponding to an optical depth between @xmath198 and @xmath199 is @xmath200 . we choose an optical depth for the photon from this distribution , @xmath201 + to find the physical distance travelled , we perform the following line integral , solving for @xmath202 , @xmath203 where @xmath204 is the path travelled by the photon , and @xmath205 is the scattering cross - section of @xmath0photons , @xmath206 and @xmath207 is the @xmath0oscillator line strength while @xmath208 is the classical electron radius . it is worth noting how the integrand depends on position . the most obvious dependence is that of a spatially varying density @xmath174 . the dependence of @xmath209 is two - fold . firstly , in case of a spatially - varying bulk velocity , @xmath210 . secondly , if @xmath211 depends on position , then so will both @xmath17 and @xmath212 in @xmath213 via @xmath190 and @xmath214 . + once @xmath202 is found , the position of the next scattering is @xmath215 . if this is outside the hi region , then the algorithm is terminated . next we choose the velocity of the scattering atom . naively , one might think that this step involves generating the three components of the atom s velocity from the maxwell - boltzmann distribution . however , we are choosing the velocity of an atom _ given _ that it scatters a photon with frequency @xmath17 . we can therefore divide the velocity of the atom into one component parallel ( @xmath216 ) and two components perpendicular to the direction of motion of the photon ( @xmath217 , @xmath218 ) . the two perpendicular components do not alter the frequency of the photon as seen by the atom , and are thus chosen from a maxwell - boltzmann distribution . from equations - it can be seen that the probability that a photon with frequency @xmath17 scatters off an atom with velocity ( along the direction of propagation of the photon ) between @xmath216 and @xmath219 is , @xmath220 where @xmath221 . a scheme for generating random numbers from this distribution is given in @xcite . now that we have calculated @xmath222 , the velocity of the atom that scatters the photon , we perform a lorentz transform into the rest frame of the atom . in this frame , we assume that the frequency of the scattered photon differs from the frequency of the incident photon only by the recoil effect ( * ? ? ? we choose a new direction for the photon . our code can incorporate either an isotropic or a dipole distribution for the direction of the re - emitted photon . the results are insensitive to the choice . we then transform back into the laboratory frame . at speeds much less than @xmath223 we have , for an initial photon direction @xmath180 , and having chosen a final photon direction @xmath224 , that the final frequency of the photon is , @xmath225 in general , the final term , known as the recoil term @xcite , is negligible for the modelling in this paper . + we now return to step [ choosedist ] and repeat it until the photon escapes the hi region . once the photon escapes the region , its properties ( frequency , angle of escape etc . ) are recorded . we then return to step [ generate ] and generate another photon . our code incorporates the presence of a cosmic abundance of deuterium , following the method presented in . adding deuterium has only a minimal effect . monte carlo @xmath0rt codes like the one used here can be significantly accelerated by skipping scatterings in the core of the line profile . we define the core - to - wing transition to occur at a critical frequency @xmath226 , which is not the same as @xmath193 . whenever a photon is in the core @xmath227 , we force it into the wing by choosing the scattering atom s velocity to be large . we follow the method presented in @xcite , , choosing the perpendicular components of the atom s velocity such that @xmath228 . we choose the value of @xmath226by requiring that , for a uniform sphere , less than a fraction @xmath229 of the photons that emerge have @xmath230 . using the analytic solution for the emergent spectrum of a sphere , this gives , @xmath231 where @xmath15 is the line - centre optical depth from the centre to the edge . we find that setting @xmath232 up to max@xmath233 gives an acceptable compromise between speed and accuracy . this procedure can accelerate the code by several orders of magnitude , e.g. for a static sphere with @xmath234 , the code is accelerated by a factor of more than a thousand . in cases where @xmath235 , @xmath211 and/or @xmath236 have a complicated dependence on @xmath237 , solving for @xmath202 in equation can be computationally expensive . in spherical symmetry , we use a similar approach to , dividing the sphere into shells of uniform density . within each shell , the integral becomes trivial , @xmath238 . if the edge of a shell is reached , the optical depth to the edge of the shell is subtracted from @xmath198 and a new @xmath202 is calculated using the new value of @xmath239 . we space the shells so that each shell has equal column density . we choose the number of shells to keep the frequency dispersion within each shell small compared to the change in frequency as the photon crosses a shell ; 1000 shells are usually sufficient . in this section , we will consider the effect of altering the gas temperature and amplitude of the bulk velocity . for this section , the fiducial model is different to that in section [ fidmod ] : @xmath240 . in this section , we consider the effect of decreasing @xmath60 . the results are shown in figure [ fig : fbvaf ] . setting @xmath241 would result in @xmath242 being symmetric about @xmath243 ( ignoring the small effects of deuterium and recoil ) . this tendency is seen clearly in both of the left plots . these plots also show that photons emerge bluer as the velocity of infall increases . photons gain energy from head - on collisions with atoms , and the more energetic the atoms , the more energy is transferred between photons and gas . the surface brightness plots show that the emission becomes more extended as the velocity is reduced . a rather flat surface brightness profile is characteristic of a uniform , static @xmath16sphere . this is because the bulk velocity can give photons a `` free ride '' through the halo , doppler shifting their frequency away from line centre in the fluid frame without changing @xmath17 in the laboratory frame . in the @xmath87 case , the peak surface brightness is reduced by almost 2 orders of magnitude by this effect . it is also worth noting that the smaller @xmath60 is , the less the dependence of the profile on @xmath4 . the effect of altering the temperature is shown in figure [ fig : fbt ] . note that the spectrum is given as a function of velocity ( @xmath244 ) , as @xmath17 itself is temperature dependent ( @xmath245 ) . for the @xmath82 case , there is no dependence on the temperature because the photon begins its flight with a frequency ( in the fluid frame ) in the scattering wings , and is unlikely to return to the doppler core . remember that the scattering cross - section does not depend on @xmath211 in the lorentz wing of the profile . in the @xmath81 case , the effect of temperature is minimal . the red peak disappears at lower temperatures due to the corresponding increase in the scattering cross - section , @xmath246 , when @xmath17 is small .

we have performed a large cosmological hydrodynamics simulation tailored to the deep survey with the hubble space telescope made in 2012 , the so - called udf12 campaign . after making a light - cone output , we have applied the same color selection criteria as the udf12 campaign to select galaxies from our simulation , and then , have examined the physical properties of them as a proxy of the real observed udf12 galaxies at . as a result , we find that the halo mass is almost linearly proportional to the observed ultraviolet ( uv ) luminosity ( at ) . the dust attenuation and uv slope well correlates with the observed uv luminosity , which is consistent with observations quantitatively . the star formation rate ( sfr ) is also linearly proportional to the stellar mass and the specific sfr shows only a weak dependency on the mass . we also find an increasing star formation history with a time - scale of myr in the high- galaxies . an average metallicity weighted by the lyman continuum luminosity reaches up to solar even at , suggesting a rapid metal enrichment . we also expect mjy at 350 ghz of the dust thermal emission from the galaxies with , which can be detectable with the atacama large milimetre - submilimetre array . the galaxies selected by the udf12 survey contribute to only 5212% of the cosmic sfr density from to , respectively . the james webb space telescope will push the detection fraction up to 7772% . [ firstpage ] cosmology observations ; galaxies evolution ; galaxies formation ; galaxies high - redshift ; galaxies luminosity function , mass function ;

1310.0114

structure formation in the universe proceeds from a small scale to a large scale through accretion and merging processes of dark matter halos in the context of the current concordance @xmath11 cold dark matter ( @xmath11cdm ) cosmology . yet baryonic processes making luminous objects such as galaxies are not well understood and this is one of the most important questions in the modern observational cosmology . in order to understand the processes of galaxy formation and evolution , it is important to look into the site where galaxies are forming in the very high-@xmath5 universe . they are intriguing objects as the potential sources causing the cosmic reionization which is also one of the hottest topics in the current cosmology research . high-@xmath5 galaxy candidates upto @xmath10 have been detected by ground - based large telescopes @xcite . the redshifts of some of the galaxies are confirmed by spectroscopy @xcite . however , it is difficult to detect galaxies at @xmath0 using the dropout ( lymam break ) method by the ground - based telescopes because the ly@xmath12 break shifts to the near - infrared ( nir ) wavelength where the observation is strongly affected by the earth s atmosphere . therefore , a space telescope which is free from the atmospheric effect is needed to detect highest-@xmath5 galaxies @xcite . since the new wide field camera ( wfc3 ) was installed on the hubble space telescope ( hst ) in 2009 , its nir channel has revolutionized the search for highest-@xmath5 galaxies @xcite . in 2012 , a new campaign to take the deepest images of the sky ever obtained was conducted with hst / wfc3 @xcite . this is called the udf12 campaign . a number of candidates of galaxies beyond @xmath13 have been discovered in the udf12 data @xcite . the data set is the best one to investigate such highest-@xmath5 galaxies even though their redshifts have not been confirmed by spectroscopy yet ( see @xcite about a spectroscopic follow - up strategy with alma ) . in order to study the physical properties of high-@xmath5 galaxies , the so - called spectral energy distribution ( sed ) fitting is often used @xcite . when applying the method to the udf12 galaxies , however , the narrow coverage of the seds causes a problem . the hst / wfc3 data , in fact , cover only a narrow range of ultraviolet ( uv ) in the rest - frame of the galaxies . the galaxies at very high - redshifts are detected only one longest wavelength band at about 1.6 @xmath14 m in the observer s frame . spitzer / irac data sometimes help the situation for brightest galaxies , but it is limited because of shallowness of the irac data . in the preset paper , we examine the physical properties of the udf12 galaxies by utilizing a cosmological hydrodynamics simulation in order to overcome the limitation of the narrow coverage of the seds with observational data currently available for highest-@xmath5 galaxies . galaxy formation models based on cosmological hydrodynamics simulations are now well developed and can reproduce reasonably well statistical quantities such as luminosity functions ( lfs ) , stellar mass functions ( smfs ) , and cosmic star formation history ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in such a simulation , one can select simulated galaxies which would be detected by the udf12 campaign and would satisfy the color selection criteria . then , we can investigate the physical properties such as halo mass , stellar mass , star formation rate of the selected `` mock - udf12 galaxies '' . a similar approach has been proposed by @xcite but their method is based on a semi - analytic model of galaxy formation and did not discuss the highest-@xmath5 galaxies sampled by the udf12 campaign . in section 2 , we describe our numerical simulations and calibration parameters . then , we virtually observe the galaxies produced by our simulation through a light - cone output and select simulated galaxies by the same criteria as those in the udf12 survey in section 3 . in section 4 , we present physical properties of the selected mock - udf12 galaxies in our simulation . in section 5 , we discuss a few implications and predictions . the final section is devoted for our conclusion . throughout this paper , we adopt a @xmath11cdm cosmology with the matter density @xmath15 , the cosmological constant @xmath16 , the hubble constant @xmath17 in the unit of @xmath18 and the baryon density @xmath19 . the matter density fluctuations are normalised by setting @xmath20 @xcite . all magnitudes are quoted in the ab system @xcite . the assumed initial mass function ( imf ) in the observational data and in our simulation is always the salpeter imf with the mass range of 0.1100 @xmath21 . we first describe our cosmological hydrodynamics simulation . we calibrate parameters in the code by comparing the smf observed at @xmath13 . then , we introduce the sed model of the simulated galaxies . to do this , we need to adopt a recipe for the dust attenuation which has additional parameters . to calibrate them , we compare the uv lfs in our simulation with the observed ones at @xmath22 and @xmath23 . these observational data are compiled from literatures including the udf12 results . our simulation code is based on an updated version of the tree - pm smoothed particle hydrodynamics ( sph ) code gadget-3 which is a successor of tree - pm sph code gadget-2 @xcite . we have implemented relevant physical processes such as star formation , supernova ( sn ) feedback and chemical enrichment following @xcite . as introduced by @xcite , we have assumed that the initial wind speed of gas around stellar particles in which sne occur is proportional to the local velocity dispersion of dark matter . the feedback model is motivated by recent observations of @xcite . we employ a total of @xmath24 particles for dark matter and gas in a comoving volume of @xmath25 cube . the mass of a dark matter particle is @xmath26 and that of a gas particle is initially @xmath27 , respectively . the softening length for the gravitational force is set to be @xmath28 in comoving unit . gas particles can spawn star particles when they satisfies a set of standard criteria for star formation , and then , their mass is reduced . on the other hand , their mass increases when the stellar population evolve and return a part of the mass including metal elements into the interstellar medium . a star particle represents a cluster of stars with a range of stellar masses because individual stars can not be resolved in the cosmological hydrodynamic simulation . for each snapshot of the simulation , we run a friends - of - friends ( fof ) group finder with a comoving linking length of 0.2 in units of the mean particle separation to identify groups of dark matter particles as halos . gas and star particles near dark matter particles which compose a fof group are also regarded as the member of the group . then , we identify gravitationally bound groups of at least 32 total ( dark matter+gas+star ) particles as substructures ( subhalos ) in each fof group using subfind algorithm developed by @xcite . we regard substructures that contain at least 5 star particles as bona - fide galaxies . above this threshold of the star particle number , the discreteness of the stellar mass begins to disappear . we note that the minimum halo mass of our simulated galaxies is around @xmath29 which depends on the relative fraction of dark matter , gas and star particles . the minimum stellar mass of our simulated galaxies is around @xmath30 which depends on the age of individual star particle in galaxies . henceforth , we only use galaxies satisfying these conditions for our study . this treatment of the simulated galaxies is different from our previous work @xcite and thus affects the shape of the stellar mass function at lower-@xmath5 because the fof halos may contain many substructures . however , the overall effect is not significant at higher-@xmath5 where we discuss in this paper . some recent theoretical work claims that the effect of the so - called chemical feedback ( i.e. , the transition from popiii to popii / popi stars ) that affects the early evolution of very high-@xmath5 galaxies and in particular their metal enrichment history ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in this study , however , we do not consider this effect ( i.e. , we consider only popii / popi star population ) because our simulation resolution is not enough to discuss the effect . in order to calibrate parameters in the code such as sn feedback efficiency , we compare our model with observations of the stellar mass function ( smf ) of galaxies , assuming the observed smfs to be complete and correct . [ fig1 ] represents our smf at @xmath13 ( left panel ) and those at @xmath13 to @xmath23 ( right panel ) . in the left panel , we also plot the observational results ( the points with error bars ) at @xmath13 taken from @xcite . our model reproduce the observed smf reasonably well . there is an issue of the definition of the stellar mass for low-@xmath5 galaxies : the mass including remnant mass or not @xcite . however , the effect is negligible at the high-@xmath5 . when comparing our simulated galaxies with observed galaxies , we calculate the spectral energy distribution ( sed ) of the simulated galaxies . the sed of each star particle which has its own age , metallicity , and mass is calculated by using the population synthesis code pgase @xcite . we assume a universal salpeter initial mass function ( imf ) with mass range @xmath31 to @xmath32 for all star particles . we then sum the seds of individual star particles composing a simulated galaxy to obtain the total intrinsic sed of the galaxy . @xcite argued that nebular components ( continuum + line ) affect the rest - frame optical band magnitude for high-@xmath5 galaxies . in this study , however , we consider only the uv range @xmath33 , where ly@xmath12 emission line is the only important nebular line . some spectroscopic surveys of ly@xmath12 emission line at @xmath34 have been performed , but all failed @xcite , probably because of heavy absorption by intergalactic neutral hydrogen before the completion of the cosmic reionization . the nebular continuum does not contribute to the seds in the uv range if the age of galaxies is larger than @xmath35 @xcite . thus , we do not consider the contribution of nebular components in this study . we treat the dust attenuation similarly to @xcite and @xcite . an important difference is that , in this study , we simply adopt the calzetti law @xcite rather than using q values of a specific material and size . the optical depth @xmath36 at @xmath37 , which is used for the normalization of the attenuation law , is calculated by the following equation : @xmath38 where @xmath39 and @xmath40 are a typical size and the material density of dust grains , respectively . we set @xmath41 and @xmath42 following sne dust production models @xcite . note that any effects of different choices of these values are absorbed in one parameter determining the dust surface mass density , @xmath43 , as described below . the surface density is defined by the following equation : @xmath44 where @xmath45 and @xmath46 are the total dust mass and the effective radius of the dust distribution in a galaxy , respectively . we assume @xmath45 and @xmath46 to be proportional to the metal mass , @xmath47 , and the half mass radius , @xmath48 , of the dark matter distribution in a galaxy , respectively . the latter two values are directly calculated in our simulation . therefore , @xmath45 and @xmath46 are expressed as @xmath49 and @xmath50 , respectively , where @xmath51 and @xmath52 are the proportional constants of the dust mass and the effective radius , respectively . these two parameters , in fact , can be reduced to one parameter . equation ( [ eq : dsmd ] ) is reduced to @xmath53 where @xmath54 , which is a global constant for all the galaxies in our simulation . then , we calculate the escape probability of uv photons at @xmath37 ( @xmath55 ) using the following equation : @xmath56 where @xmath57 is a parameter whose value is ranging from @xmath58 to @xmath59 . in equation ( [ eq : ep ] ) , a uniform slab where stars and dust are well mixed is sandwiched in the middle of two stellar layers , where dust is negligible @xcite . the parameter @xmath57 is the fraction of the thickness of the central star+dust slab in the total thickness ; it is the well - mixed dusty slab geometry when @xmath57 is unity , and it is the case with a central infinitely thin dusty sheet when @xmath57 is zero . such a sandwich geometry is favored to explain the observed relation between the uv color and the infrared - to - uv flux ratio of nearby galaxies @xcite . in addition , we have found a similar situation in our simulation : star particles are sometimes found at the periphery of subhalos and the dust ( or gas ) density at such region is very low . the escape probability @xmath55 is saturated at @xmath60 when @xmath61 . we calibrate @xmath62 and @xmath57 so as to reproduce the observed uv luminosity function ( lf ) at @xmath13 , and we keep these values even at higher redshifts . for the cosmological simulation in this study , we adopt @xmath63 and @xmath64 , respectively . if we set @xmath65 , with which we can reproduce the number count of the submm galaxies @xcite , @xmath52 is uniquely determined as @xmath66 . interestingly , this value results in the dust distribution radius , @xmath46 , very similar to the half - mass radius of the stellar mass distribution . we also apply the igm absorption for the blue side of @xmath67 following @xcite with an extrapolation for @xmath0 , which should not cause a significant uncertainty because the spectrum below ly@xmath12 is completely damped by the igm absorption for such high - redshifts . since our simulation box size has the angular size of about @xmath68 at @xmath6910 , we can have about 130 hst / wfc3 fields - of - view in one side of the box . then , we calculate the uv lfs in the 130 rectangular parallelepiped sub - boxes which has the approximately udf12-sized area in one side . figure [ fig2 ] shows the average of the uv lfs ( solid lines ) and their standard deviations ( shaded areas ) of our simulation results at @xmath22 and @xmath23 . the points with error - bars represent a compilation of several observational results @xcite . our models reproduce the observed lfs reasonably well within the standard deviations which express the cosmic variance of the udf12-sized pencil beam survey . we also show the average uv lfs for the case without dust attenuation as the dashed lines , indicating a certain amount of dust attenuation even at @xmath0 . we need to select galaxies which would be detected and selected in the udf12 campaign . the construction of the mock - udf12 sample is important for this study . it should be close similarly to the real observation . first , we make a light - cone output from a number of snapshots of the simulation . then , we apply the same color selection criteria as for the real udf12 galaxies . in general , the survey depth along the line - of - sight is very wide of the lyman break method using broad - band filters . thus , we can not cover all the survey volume by one snapshot of our simulation . in order to directly compare our model with such observations , we generate a light - cone output which extends from @xmath70 to @xmath71 using a number of simulation outputs . we coordinate the output redshifts to fill the volume of a light - cone from @xmath70 to @xmath71 without any gap . we then randomly shift and rotate each simulation box so that the same objects do not appear multiple times on a single line - of - sight at different epochs . in order to directly compare our model with observations , we calculate broad - band magnitudes of the simulated galaxies on the light - cone and choose dropout galaxies with the same manner of the observations . we use the same set of broad - band filters as the udf12 which has wfc3/ir ( @xmath72 , @xmath73 , @xmath74 and @xmath75 ) and acs ( @xmath76 , @xmath77 , @xmath78 , @xmath79 and @xmath80 ) filters . we first apply the detection limit of the udf12 survey . the conditions are @xmath81 and @xmath82 for the @xmath10 selection , @xmath82 and @xmath83 for the @xmath84 selection and @xmath85 and @xmath83 for the @xmath86 and the @xmath7 selections , respectively . then , we apply the color selection criteria for @xmath10 as @xmath87 and for @xmath84 as @xmath88 as in @xcite . the color selection criteria for @xmath89 are @xmath90 and for @xmath91 are @xmath92 as in @xcite . note that we consider no effect occurring in real observations such as observational noise , surface brightness dimming , object detection incompleteness , etc . , in the selection procedure . therefore , we have assumed these effects to be small . we reserve to examine these effects in future work . [ gal_dist ] shows the spacial distribution of the color selected simulated galaxies . the red , green , blue and magenta points represent the simulated galaxies selected by the criteria for @xmath93 and @xmath7 , respectively . the brown points are the galaxies doubly selected by the criteria for @xmath84 and @xmath86 . the point size represents @xmath75 band magnitude and the largest ( smallest ) one corresponds to @xmath94 . we note that almost all simulated galaxies selected by the criteria of @xmath7 are also selected by the criteria of @xmath86 . we find that only a small number of galaxies are selected at @xmath95 in our simulation . we also check the selection efficiency of the color selections . [ sel_efficiency ] presents the number fraction of galaxies brighter than the observational limiting magnitude of 30 mag and satisfying the color criteria among all the galaxies in our simulation ( see sec.2 for the definition of the galaxies ) as a function of redshift . the solid , dashed , dash - dotted and dotted lines are the fractions for @xmath93 and @xmath7 selections , respectively . the redshift ranges selected are @xmath96 ( @xmath10 color selection ) , @xmath97 ( @xmath84 color selection ) , @xmath98 ( @xmath86 color selection ) and @xmath99 ( @xmath7 color selection ) . we find that the color selection works reasonably well . next , we note a remarkably small number fraction of the galaxies selected as mock - udf12 galaxies from all galaxies in our simulation . interestingly , the number fraction of the color selected galaxies is less than @xmath100 even at @xmath10 and less than @xmath101 at @xmath7 . this means that the galaxies detected in the current deepest survey are still just the tip of iceberg and we probably miss almost all star forming galaxies at @xmath0 . we revisit this issue later in section 5.3 . we have constructed a mock - udf12 galaxy sample from our simulation in the previous section and we are now ready to examine their physical properties . first , we examine the halo and stellar masses of the mock - udf12 galaxies . second , the dust attenuation and uv slope of the mock - udf12 galaxies are discussed . third , we look into the correlation between the stellar mass and star formation rate ( sfr ) and specific sfr . we also check the cosmic sfr density at @xmath0 . then , the ages and the star formation history of the galaxies are examined . finally , the metallicity , morphology and size are investigated . in this subsection , we study the relation between the uv magnitude and the halo or stellar mass . the left and right panels in fig . [ uv_hm_sm ] represent the halo masses and the stellar masses of the mock - udf12 galaxies , respectively , as a function of the absolute uv magnitude calculated with the dust attenuation . hereafter , we denote the absolute uv magnitude with the dust attenuation as just the absolute uv magnitude or the uv magnitude . the small triangle , square , circle and star points represent the each mock - udf12 galaxy at @xmath93 and @xmath7 , respectively . the large points with error - bars are the median values and the central @xmath102 ranges in halo mass bins . interestingly , the relation between the halo or stellar mass and the uv magnitude is not significantly different for different redshifts . this indicates that the mass luminosity relation for the mock - udf12 galaxies is independent of the redshift and the mass - luminosity ratio in a certain halo mass at different epoch is almost constant . we here present simple formulae to estimate the halo or stellar mass from the uv magnitude . let us assume a functional form as @xmath103 where @xmath104 and @xmath105 are the fitting parameters . if @xmath105 is equal to @xmath106 , then the uv luminosity is linearly proportional to the host halo ( or stellar ) mass . for the mock - udf12 galaxies , we have obtained @xmath107 for the halo mass relation and @xmath108 for the stellar mass relation therefore , the uv luminosity is almost linearly proportional to the halo mass for the mock - udf12 galaxies , but the stellar mass scales more strongly with the magnitude . using equation ( [ eq : mhalo - muv ] ) and ( [ eq : mstellar - muv ] ) , we have obtained @xmath109 for the relation between the halo mass and the stellar mass when the halo mass is less than @xmath110 @xmath21 . interestingly , this is similar to the results at @xmath13 and 8 of @xcite in which they have performed a multi - epoch abundance matching method . according to @xcite , the udf12 galaxies at @xmath111 are little affected by dust . however , as already shown in fig . [ fig2 ] , we need a certain amount of dust attenuation to reproduce the observed uv lfs . let us compare the dust attenuation in the mock - udf12 galaxies with the observations to test whether our treatment for the dust attenuation works well quantitatively . the top left panel of fig . [ auv - beta ] represents the dust attenuation @xmath112 at @xmath37 as a function of the uv magnitude . the small points are the uv attenuation of each simulated galaxy , while the large points with error - bars are the median values and the central @xmath102 ranges in uv magnitude bins . the range of the dust attenuation is @xmath113 to @xmath114 . this upper limit of @xmath112 comes from our recipe of the uv escape probability described in equation ( 4 ) ; we have a lower limit of the probability as @xmath60 when @xmath115 and we set @xmath64 . the @xmath112 range corresponds to @xmath1160.76 under the calzetti law we assumed ( i.e. , @xmath117 ) . interestingly , the typical value of the dust attenuation is proportional to the uv magnitude . we also find that there are no luminous ( @xmath118 ) galaxies with less attenuation ( @xmath119 ) but faint galaxies are distributed over all the range of @xmath112 . @xcite argue @xmath120 for the udf12 galaxies with the calzetti law , which corresponds to @xmath121 . a similar attenuation is found for @xmath122 mock - udf12 galaxies , whereas a larger attenuation is found for more uv luminous galaxies . recently , @xcite argue a possibility of @xmath123 which is more than a factor of 2 larger than our result . we also find that the dust attenuation is not zero even in galaxies observed at @xmath7 in our simulation . note that we have solved the chemical evolution ( i.e. , dust amount evolution ) consistently with the star formation activity in galaxies in our simulation although we have assumed a constant dust - to - metal ratio which may evolve depending on the star formation history @xcite . next we explore the uv slope . in order to compare our simulations with the observation , we adopt exactly the same formulae as @xcite to calculate the uv slope @xmath3 : @xmath124 for @xmath10 and @xmath84 galaxies , and @xmath125 for @xmath126 galaxies . the top right , the bottom left and the bottom right panels show the uv slope @xmath3 for the mock - udf12 galaxies at @xmath127 and @xmath86 , respectively . the small points are the uv slope @xmath3 of each simulated galaxy , while the large points with error - bars are the median values and the central @xmath102 ranges in uv magnitude bins . we also plot the observed uv slope @xmath3 ( cross points ) taken from @xcite . the uncertainty on @xmath3 for each measurement reaches 1.2 for @xmath128 and @xmath129 ( eq . [ [ eq : betaz78 ] ] ) and 2.6 for @xmath130 ( eq . [ [ eq : betaz9 ] ] ) at the faintest magnitude . at @xmath10 , the mock - udf12 galaxies with @xmath131 have @xmath132 , but less luminous galaxies are distributed over @xmath133 to @xmath134 . this upper limit corresponds to the upper limit of @xmath112 in our recipe . on the other hand , the observed data are distributed over @xmath133 to 0 . probably this larger scatter is partly because of the observational error , while we may underestimate the scatter in the simulated galaxies because we employ a unique dust attenuation model in which we adopt the calzetti extinction law with a single dust geometry for all simulated galaxies . the mean value @xmath135 for @xmath136 reported by @xcite is consistent with our expectation ( @xmath137 ) at the same magnitude . a similar relation is also found by @xcite . at @xmath84 and @xmath86 , the @xmath3 distribution of the mock - udf12 galaxies shows a large redward scatter . this is because the ly@xmath12 break affects the bluer band magnitude in the formulae to estimate @xmath3 presented above . overall the @xmath3 distribution of the mock - udf12 galaxies reproduce the observations very well . the star formation rate ( sfr ) is very important to study how the stellar components in galaxies were build up and how many ionizing photons were produced for the reionization of the universe . [ sm_sfr_ssfr ] represents the sfr ( left panel ) and the specific sfr ( ssfr ; right panel ) as a function of the stellar mass for the mock - udf12 galaxies . the small symbols are the same meanings as in fig . [ uv_hm_sm ] and large symbols with error - bars are the median values and the central 68% ranges in stellar mass bins . the mock - udf12 galaxies show an almost linear correlation between their sfr and stellar mass and form the so - called ` main - sequence ' for star - forming galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in the left panel of figure 7 , the correlations at different redshifts are roughly overlapped each other , while a weak evolution of sfrs from @xmath7 to @xmath10 can be seen . this evolution is more clearly seen in the right panel showing ssfrs : higher ssfrs as higher-@xmath5 . this is probably because objects collapsed at higher-@xmath5 have generally more dense gas , thus the sfr becomes higher in such objects . on the other hand , the dispersions in ssfrs of the different redshifts are in fact similar to each other as found from the 68% ranges shown . the apparent wider distributions for the lower-@xmath5 samples is just because of a larger number of the galaxies . the ssfrs of the mock - udf12 galaxies are higher than those of ` main sequence ' galaxies at @xmath138 @xcite but are similar to those of lbgs at @xmath139 @xcite as shown by the dashed lines in fig . [ sm_sfr_ssfr ] . this indicates that the udf12 galaxies at @xmath0 are in a similar starburst phase found in lbgs at @xmath139 rather in the normal quiescent phase found at low-@xmath5 . however , we find that most of our mock - udf12 galaxies lie in the lower part than the @xmath140 observed relations . let us explore the physical process controlling such relations ( i.e. , ` main sequence ' ) for the udf12 galaxies . suppose that the sfr is proportional to the dark matter ( dm ) accretion rate estimated by @xcite : @xmath141 @xmath142 and @xmath143 where @xmath144 is the star formation efficiency . if we introduce a constant efficiency as @xmath145 , then we obtain @xmath146 which is shown as the shaded area in the left panel of fig . [ sm_sfr_ssfr ] . the redshift dependence of eq . ( [ eq : dmac ] ) from @xmath10 to @xmath7 is as small as our simulation results . we also obtain @xmath147 which is the shaded area in the left panel of fig . [ sm_sfr_ssfr ] . we find a very good agreement between our simulation and the simple estimation by equations ( [ eq : sfr ] ) and ( [ eq : ssfr ] ) at the massive end of the stellar mass , but a small difference can be seen . this implies that @xmath148 has a weak stellar ( halo ) mass dependence and the baryon physics is also necessary to explain the trend . the star formation rate density ( sfrd ) is also useful to check the calibration of our simulation as well as the stellar mass function and the uv lfs . [ sfrd ] shows the evolution of the sfrd from @xmath7 to @xmath10 . the triangles represent observational estimates based on galaxies with @xmath149 @xcite . the all observational data are corrected for dust attenuation . thus , the simulated data do not include any dust attenuation . the circle and square points are the sfrd of the galaxies with @xmath149 as the observations and the total sfrd in our simulation , respectively . our results for @xmath149 reasonably agree with the observations within the errors , suggesting that our calibration is good enough . one important finding from the comparison is that the sfrd of the galaxies with @xmath149 is only a minor part of the total sfrd even at @xmath10 . therefore , we still miss a major part of the sfrd even in the current deepest survey . we compute the mass weighted age @xmath150 of each mock - udf12 galaxy to characterize the stellar population . the definition of the age is given by the following equation : @xmath151 where @xmath152 and @xmath153 are the age and mass of the _ i_th star cluster in a simulated galaxy . [ hm_age ] shows the mass weighted age of the mock - udf12 galaxies as a function of their halo mass . the symbols are the same as fig . [ uv_hm_sm ] . the typical ages of the mock - udf12 galaxies are @xmath154 , @xmath155 , @xmath156 and @xmath157 , respectively . the ages decrease with increasing redshift . on the other hand , there is a large dispersion found at halo masses less than about @xmath158 . this is because the sn feedback effectively works in such smaller mass galaxies and then the star formation occurs stochastically in such galaxies ( see also the next subsection ) . similar discussion can be seen in @xcite . in addition , the mass resolution of our simulation may not be enough to resolve such smaller mass galaxies and may enlarge the dispersion . one has to assume a star formation history to perform the sed fitting method which is often used to explore the physical properties such as the stellar mass and the sfr of high-@xmath5 galaxies . a constant sfr or exponentially decreasing sfr are usually assumed . however , recent studies suggest an increasing sfr with time for high-@xmath5 galaxies ( e.g. , * ? ? ? thus , it is very important to examine the sfh from theoretical point of view . [ sm_sfh ] represents average sfhs of the mock - udf12 galaxies for a certain mass range and a redshift . the lines in each panel at @xmath10 and @xmath84 show the sfhs of simulated galaxies within the stellar mass range of @xmath159@xmath160 , @xmath161@xmath162 , @xmath163@xmath164 and @xmath165@xmath166 from the top to the bottom . the lines at @xmath86 and @xmath7 show the sfhs within the range of @xmath161@xmath162 , @xmath163@xmath164 and @xmath165@xmath166 from the top to the bottom . the time resolution of the sfhs is set to be @xmath167 yr to see the stochastic star formation activity . as found in fig . [ sm_sfh ] , the sfr clearly increases with time . this suggests that it is strongly preferable to choose increasing sfr with the time ( e.g. , exponentially increasing ) for the sed fitting method . this result is consistent with previous theoretical work @xcite . in order to examine the typical time - scale of the star formation , we adopt an exponential function : @xmath168 where @xmath104 and @xmath169 are the normalization parameter and the typical star formation time - scale , respectively . we obtain these values with a @xmath170 method . the typical star formation time - scales @xmath169 obtained are @xmath171 ( @xmath10 ) , @xmath172 ( @xmath84 ) , @xmath173 ( @xmath86 ) and @xmath174 ( @xmath7 ) . these values are almost constant among different mass ranges if the redshift is the same and are somewhat longer than but still similar to the mass weighted age obtained in the previous subsection . observationally , the metallicity of galaxies is determined by measuring a ratio ( or ratios ) of emission or absorption lines . emission lines are usually emit from the hii regions and the photo - dissociation regions ( pdrs ) . thus , these lines come from the star forming region which is relatively dense region in a galaxy . in order to compare such observations with our model properly , we need to calculate the metallicity in star forming regions in a simulated galaxy rather than that in the whole of the galaxy . we here introduce a new definition of the metallicity for simulated galaxies as following equation : @xmath175 where @xmath176 and @xmath177 are the metallicity and the intrinsic lyman continuum luminosity of @xmath178th star particle in a simulated galaxy . we have assumed that the gas metallicity around a star particle is the same as that of the star particle . in this definition , only the metal around the very young star particles is taken into account and the metal around aged star particles and the metal far from star forming regions do not contribute to the metallicity . the metallicity of this definition is usually higher than that averaged over the whole of a simulated galaxy . we call this metallicity the nebular metallicity [ hm_metal ] represents the nebular metallicity of the mock - udf12 galaxies as a function of their host halo mass . the symbols are the same as fig . [ uv_hm_sm ] . the nebular metallicity is roughly proportional to their host halo mass although the dispersion is large . we also find that the metallicity evolution from @xmath7 to @xmath10 is weak as with the sfr . interestingly , the metallicity of some galaxies reaches at @xmath179 solar metallicity ( @xmath180 ) even at @xmath7 . these suggest that the metal enrichment is a rapid process . in order to explore the size and the morphology of the mock - udf12 galaxies , it is necessary to resolve the inner structure of the simulated galaxies . thus , we discuss only massive ( uv bright : @xmath131 ) galaxies which are composed of at least about @xmath181 star particles . [ morphology ] shows the surface brightness distributions at the rest - frame @xmath37 of the uv brightest galaxy among the mock - udf12 galaxies at @xmath10 ( the left panel ) and @xmath84 ( the right panel ) , respectively . we have applied a gaussian smoothing with @xmath182 in order to take the point - spread - function of the hst / wfc3 approximately into account . the images of the uv brightest mock - udf12 galaxies are roughly round but still show somewhat clumpy and elongated structure , in particular in the @xmath10 case . interestingly , such a shape has been found in the observational data @xcite . to compare with observational studies ( the brightness profile ) , we make a stacked uv images of the mock - udf12 galaxies . to do so , we stack the distributions of star particles projected on the sky using the light - cone output of the mock - udf12 galaxies . in the stacking process , we regard the luminosity center of each galaxy as the galaxy center and we do not choose any specific direction of the galaxy ( i.e. , the position angles in the stacking are random ) . the spatial unit of the galaxies is the physical unit not comoving unit . in our simulation box , the numbers of galaxies with @xmath183 and @xmath184 are @xmath185 and @xmath186 at @xmath187 , respectively . [ size ] shows the azimuthally summed up and normalized cumulative profiles of the stacked images of the simulated galaxies at @xmath10 ( left panel ) and @xmath84 ( right panel ) . the solid and dashed lines represent the profiles for @xmath183 and @xmath184 , respectively . the half light radii of the stacked images with @xmath183 @xmath188 are @xmath189 @xmath190 at @xmath10 and @xmath191 @xmath190 at @xmath84 , respectively . these results are also similar to the results of the udf12 campaign @xcite . we have presented some physical properties of the udf12 galaxies at @xmath1010 expected from our cosmological hydrodynamic simulation . here we discuss a few additional issues derived from the simulation . so far , we have studied only upto @xmath7 selected galaxies . @xcite ( see also * ? ? ? * ; * ? ? ? * ) presented one @xmath192 galaxy candidate ( xdfjh-39546284 ) . thus , it is also interesting to study the physical properties of such @xmath95 galaxies not only for hst deep survey but also for future observations with the james webb space telescope and @xmath193 telescopes . here , we use the same color selection criterion as @xcite : @xmath194 moreover , we adopt the detection limit criterion of @xmath195 . as a result , we find @xmath196 simulated galaxies from @xmath197 to @xmath198 in our simulation box . all their @xmath75 band magnitudes are fainter than @xmath199 . we note that our field - of - view ( fov ) is about @xmath200 at the redshift and then the expected number is about two galaxies per one fov of the udf12 survey . therefore , the detection of one such highest-@xmath5 galaxy in the udf12 campaign is consistent with our simulation if we consider the poisson error . the halo and stellar masses of these galaxies are @xmath201 and @xmath202 , respectively . the stellar mass weighted age and the nebular metallicity @xmath203 are @xmath204 and @xmath205 , respectively . clearly , @xmath95 galaxies selected by the drop - out technique are very early evolutionary phase objects . as discussed in section 4.2 , the mock - udf12 galaxies suffer a modest amount of dust attenuation . therefore , these galaxies also emit the thermal radiation from dust in the far - infrared wavelength . they could be detectable with the atacama large milimetre - submilimetre array ( alma ) . submm observations of very high-@xmath5 galaxies is also important to know the chemical enrichment and the dust formation at very high-@xmath5 . according to @xcite , the submm flux of star - forming galaxies with @xmath206 is about @xmath207 . indeed , some of our simulated galaxies have @xmath206 and the galaxies would be detectable with alma . we calculate the submm flux of the simulated galaxies by the same procedure as @xcite . the dust temperature for our sample galaxies is typically @xmath208 . [ submm ] shows the expected submm flux density at @xmath209 in the observer s frame as a function of the observed @xmath75 band magnitude . we find a tight correlation between the two quantities independent of the redshift . if we set the detection limit as 0.1 mjy , galaxies brighter than @xmath210 may be detected . however , most of the galaxies with @xmath210 actually have much fainter submm flux density . in order to increase the probability of the detection , we should observe galaxies with @xmath211 or even 26.5 . actually , the brightest @xmath75 magnitude in the udf12 survey is @xmath212 . however , other wide - field surveys like the brightest of reionizing galaxies ( borg ) survey @xcite and surveys for gravitationally - lensed galaxies like the cluster lensing and supernova survey with hubble ( clash ) survey @xcite have reported detection of some galaxies with @xmath213 at @xmath0 . these galaxies would be good targets for alma . so far we have discussed the galaxies detected and color - selected by the udf12 campaign . as shown in fig . [ sel_efficiency ] , however , most of the simulated galaxies are not selected owing to either detection limit or the color criteria . indeed , the number fraction of the selected galaxies is about @xmath214 for @xmath10 and about @xmath215 for @xmath7 . what about the corresponding fraction in terms of the cosmic star formation density ( sfrd ) ? [ sfrd ] shows the evolution of the sfrd from @xmath7 to @xmath10 . there , we consider an additional condition of @xmath149 in order to directly compare our simulation with the observations . this condition is about one magnitude brighter than the detection limit of the udf12 campaign . in this subsection , we measure the contribution of the udf12 galaxies to the cosmic sfrd quantitatively . we find that the contributions to the total sfrd of the mock - udf12 galaxies are @xmath216 and @xmath217 for @xmath93 and @xmath7 , respectively . this means that we do not see faint galaxies which contributed to the dominant part of the sfrd . note that we can resolve only halos with @xmath218 in our simulation . thus , the relative contribution of the detectable galaxies could be even smaller if there are a significant amount of less massive galaxies than our simulation . let us study the properties of galaxies unselected by the udf12 survey . there are four cases : ( 1 ) galaxies satisfy the color selection criteria and are bright enough to be detected , ( 2 ) galaxies satisfy the color selection criteria but are too faint to be detected , ( 3 ) galaxies do not satisfy the color criteria but are bright enough to be detected and ( 4 ) galaxies do not satisfy neither color criteria nor the brightness limit . table [ fraction_table ] represents the contributions of each case to the sfrd . we find that the simulated galaxies satisfying the color criteria ( cases [ 1 ] and [ 2 ] ) contribute to a major fraction of the sfrd ( 8595% ) but most of it comes from galaxies fainter than the current detection limit ( i.e. , case [ 2 ] ) , except at @xmath13 where the detected and not - detected galaxies have similar contributions . many of these faint galaxies in the case ( 2 ) will be detected by the next generation telescopes . on the other hand , the simulated galaxies of the cases ( 3 ) and ( 4 ) contribute to only a minor fraction of the sfrd ( 515% ) . interestingly , all these galaxies reside around the redshift boundaries defined by the color selection and there are no galaxies around middle of the redshift ranges . the deviation from color selection criteria of the cases ( 3 ) and ( 4 ) galaxies is less than 0.4 mag . thus , if the selection criteria are somewhat relaxed , the galaxies bright enough ( i.e. , case [ 3 ] ) can be selected by the color selection . in other words , there are no galaxies with very red uv colors at @xmath93 and @xmath7 in our simulation . finally , we provide a prediction after the james webb space telescope ( jwst ) starts operation . the values in parentheses of table [ fraction_table ] represent the star formation contribution of each case with the detection limit of @xmath219 which corresponds to the jwst one . we find that the jwst can detect more than half ( 7277% ) of star formation activity . yet , much fainter galaxies accounting for @xmath220 of the @xmath7 star forming activity will still remain undetectable even with the jwst . [ cols="<,^,>,>,>,>",options="header " , ] we have made a cosmological hydrodynamic simulation tailored to examine the physical properties of very high-@xmath5 galaxies detected by the deep survey with hst / wfc3 performed in 2012 , the so - called udf12 campaign @xcite . our simulation code follows the formation and the evolution of star forming galaxies by new feedback models based on @xcite and can reproduce the statistical properties of star forming galaxies such as lyman break galaxies ( lbgs ) , lyman @xmath12 emitters and sub - mm galaxies @xcite . first , we calibrate our model parameters so as to reproduce not only the stellar mass function at @xmath10 but also the uv luminosity function from @xmath10 to @xmath7 . then , we generate a light - cone output which extends from @xmath70 to @xmath71 using a number of simulation outputs in order to directly compare our model with the observations . it is the most important point of this study that we adopt the same color selection criteria as the udf12 campaign to select galaxies from our simulation . we call the selected galaxies the mock - udf12 galaxies . using these simulated galaxies as a proxy of the observed udf12 galaxies , we can examine the physical properties such as the halo and stellar masses , the specific star formation rate , the star formation history , etc . of the real galaxies . this is a new approach to reveal the physical properties of the highest-@xmath5 galaxies for which the so - called sed fitting method may not be reliable because we have few photometric data points only in the rest - frame uv . we find that the observed uv luminosity of the mock - udf12 galaxies is almost linearly proportional to their host halo mass and the correlation does not change so much along the redshift from @xmath10 to @xmath7 . for example , the halo mass at @xmath2 is about @xmath221 . the stellar mass of the galaxies is also proportional to the uv luminosity but the power - law slope is about 1.4 . we have derived the fitting formulae for the halo and stellar masses as a function of the uv absolute magnitude not corrected for the dust attenuation . we also find that many mock - udf12 galaxies are affected by the dust attenuation . the amount of the attenuation well correlates with the uv magnitude but shows a large dispersion ranging from @xmath113 to 1.9 ( or @xmath116 to 0.76 ) which is larger than that argued by @xcite . nevertheless , our simulation reproduces the observed uv slope @xmath3 distribution reported by @xcite . the mean uv slope also correlates with the uv magnitude . such a trend consistent with the observational results @xcite . there is no luminous ( @xmath183 ) and less attenuation ( @xmath222 ) galaxies in our simulation . the star formation rate ( sfr ) is proportional to the stellar mass with a power - law slope about unity . the specific star formation rate ( ssfr ) shows a very weak dependency on the stellar mass and the value is about @xmath223 which is higher than that of the main - sequence for star - forming galaxies at @xmath224 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) but is similar to that of lbgs at @xmath225 ( e.g. , * ? ? ? this indicates that lbgs from @xmath139 to @xmath7 are in a similar starburst phase . with the hypothesis of the sfr proportional to the dark matter accretion rate , these relations between the stellar mass and the sfr or the ssfr are reproduced reasonably well . this supports the hypothesis . however , the slight gap between our model and the simple model means that the baryon physics is also necessary to reproduce the trend . the evolution of the cosmic sfr density from @xmath10 to @xmath7 in our simulation reasonably matches with the recent observed results @xcite . we have also examined the star formation history ( sfh ) of the mock - udf12 galaxies and found an increasing sfr with time . this trend is independent of the redshift and the mass . such an increasing sfh is also suggested by recent observational studies ( e.g. , * ? ? ? therefore we conclude that an increasing sfh should be assumed in the sed fitting work where usually a constant or decreasing sfh has been assumed so far . if we fit the increasing sfh by an exponential function , the typical time - scales for @xmath93 and @xmath7 galaxies are @xmath171 , @xmath172 , @xmath173 and @xmath174 , respectively . we also find that the mass - weighted average ages of the mock - udf12 galaxies are @xmath154 , @xmath155 , @xmath156 and @xmath157 . these time - scales are roughly consistent with each other . we have introduced an average metallicity weighted by lyman continuum luminosity . we call it the nebular metallicity which is the one usually measured from emission line ratios observationally . interestingly , the nebular metallicity of some mock - udf12 galaxies is @xmath226 to @xmath227 solar metallicity . the result implies that the metal enrichment proceeds rapidly even at @xmath7 . we have also explored the size and morphology of the mock - udf12 galaxies . we find that the uv brightest galaxies at @xmath10 and @xmath228 are roughly round but still show clumpiness and asymmetry . after stacking the distributions of star particles projected to the observer s sky in the light - cone output and produced the stacked uv images for galaxies within the two brightest uv magnitude ranges , we find that the half light radii of the stacked images at @xmath10 ( @xmath84 ) are @xmath189 ( @xmath191 ) for the @xmath183 cases and @xmath229 ( @xmath229 ) for the @xmath184 cases , respectively . these results are consistent with the recent observational results @xcite . finally , we have discussed a few additional issues derived from our simulation . there is one candidate galaxy at @xmath192 in the udf @xcite . in our simulation , we expect two galaxies brighter than the current detection limit and satisfying the color selection for @xmath95 per one field - of - view of the udf12 survey . therefore , the number density is consistent with each other . we also present the expected submm flux density of the mock - udf12 galaxies . the galaxies with @xmath9 , the expected submm flux density is @xmath8 mjy at 350 ghz in the observer s rest - frame . this will be detectable with the alma . we have examined the fraction of galaxies selected or not selected by the udf12 survey . the detected and color - selected galaxies contribute to only @xmath216 and @xmath217 of the cosmic star formation rate density ( sfrd ) at @xmath230 and @xmath7 , respectively . a significant part ( 3583% ) of the sfrd comes from galaxies satisfying the color selection criteria but not bright enough to be detected with the current detection limit . if we lower the detection limit of @xmath219 which corresponds to that of the jwst , the fraction of the detected and color - selected galaxies would drastically increases to @xmath231 and @xmath232 at @xmath230 and @xmath7 , respectively . yet , @xmath233 star forming activity at @xmath7 remains undetectable even with the jwst . the fraction of the galaxies not satisfying the color criteria is as small as 514% and they reside around not far from the color boundary . therefore , we conclude that no galaxy with a very red uv color exists at @xmath0 in our simulation . we thank an anonymous referee for her / his constructive comments . we also thank tomoaki ishiyama for his comments on the halo accretion rate of galaxies . numerical simulations have been performed with the eup , primo and sgi cluster system installed at the institute for the physics and mathematics of the universe , university of tokyo and with t2k tsukuba at centre for computational sciences in university of tsukuba . is and aki acknowledge the financial support of grant - in - aid for young scientists ( a : 23684010 ) by mext , japan . this work is partially supported by grant - in - aid for young scientists ( s ) ( 20674003 ) and by the first program subaru measurements of images and redshifts ( sumire ) by the council for science and technology policy . to acknowledges the financial support of grant - in - aid for young scientists ( b : 24740112 ) . 99 behroozi , p. s. , wechsler , r. h. , conroy , c. , 2013 , apj , 770 , 57 biffi , v. , maio , u. , 2013 , mnras , 436 , 1621 bouwens , r. j. , et al . , 2004 , apj , 616 , l79 bouwens , r. j. , illingworth , g. d. , franx , m. , ford , h. 2007 , apj , 670 , 928 bouwens , r. j. , et al . , 2010 , apj , 709 , l133 bouwens , r. j. , et al . , 2011 , apj , 737 , 90 bouwens , r. j. , et al . , 2011 , nature , 469 , 504 bouwens , r. j. , et al . , 2012a , arxiv:1211.2230 bouwens , r. j. , et al . , 2012b , apj , 752 , l5 bouwens , r. j. , et al . , 2012c , apj , 754 , 83 bouwens , r. j. , et al . , 2013 , arxiv:1306.2950 bowler , r. a. a. , et al . , 2012 , mnras , 426 , 2772 bradley , l. d. , et al . , 2012 , apj , 760 , 108 brammer , g. b. , et al . , 2013 , apj , 765 , l2 brinchmann , j. , charlot , s. , white , s. d. m. , tremonti , c. , kauffmann , g. , heckman , t. , brinkmann , j. , 2004 , mnras , 351 , 1151 bunker , a. j. , et al . , 2013 , mnras , 430 , 3314 calzetti , d. , armus l. , bohlin , r. c. , kinney , a. l. , koornneef , j. , storchi - bergmann , t. , 2000 , apj , 533 , 682 capak , p. l. , faisst , a. , vieira , j. d. , tacchella , s. , carollo , m. , scoville , n. z. , 2013 , apj , 773 , 14 castellano , m. , et al . , 2010b , a&a , 524 , a28 coe , d. , et al . , 2013 , apj , 762 , 32 daddi , e. , et al . , 2007 , apj , 670 , 156 dayal , p. , dunlop , j. s. , maio , u. , ciardi , b. , 2013 , mnras , 434 , 1486 draine , b. t. , lee , h. m. , 1984 , apj , 285 , 89 draine , b. t. , et al . , 2007 , apj , 663 , 866 dunlop , j. s. , et al . , 2013 , mnras , 432 , 3520 ellis , r. s. , et al . , 2013 , apj , 763 , l7 ferland , g. j. , koristra , k. t. , verner , d. a. , ferguson , j. w. , kingdon , j. b. , verner , e. m. , pasp , 110 , 761 finkelstein , s. l. , et al . , 2013 , nature , 502 , 524 fioc , m. , rocca - volmerange , b. , 1997 , a&a , 326 , 950 gonzlez , v. , labb , i. , bouwens , r. j. , illingworth , g. , franx , m. , kriek m. , 2011 , apj , 735 , l34 jaacks , j. , thompson , r. , nagamine , k. , 2013 , apj , 766 , 94 jaacks , j. , choi , j .- h . , nagamine , k. , thompson , r. , varghese , s. , 2012 , mnras , 420 , 1606 jaacks , j. , nagamine , k. , choi , j .- h . , 2012 , mnras , 427 , 403 inoue , a. k. , 2003 , pasj , 55 , 901 inoue , a. k. , 2011a , mnras , 415 , 2920 inoue , a. k. , 2011b , earth , planets and space , 63 , 1027 inoue , a. k. , buat , v. , burgarella , d. , panuzzo , p. , takeuchi , t. t. , iglesias - pramo , j. , 2006 , mnras , 370 , 380 inoue , a. k. , shimizu , i. , tamura , y. , matsuo , h. , okamoto , t. , yoshida , n. , 2014 , apj , 780 , 18 ishiyama t. , et al . , 2013 , apj , 767 , 146 kimm , t. , cen , r. , 2013 , apj , 776 , 35 koekemoer , a. m. , 2013 , apjs , 209 , 3 komatsu , e. , et al . , 2011 , apjs , 192 , 18 labb , i. , et al . , 2010 , apj , 716 , l103 mcbride , j. , fakhouri , o. , ma , c .- p . 2009 , mnras , 398 , 1858 madau , p. , 1995 , apj , 441 , 18 martin , c. l. , 2005 , apj , 621 , 227 mclure , r. j. , et al . , 2010 , mnras , 403 , 960 mclure , r. j. , et al . , 2013 , mnras , 432 , 2696 nagamine , k. , ouchi , m. , springel , v. , hernquist , l. , 2010 , pasj , 62 , 1455 noeske , k. g. , et al . , 2007 , apj , 660 , l43 nozawa , t. , kozasa , t. , umeda , h. , maeda , k. , nomoto , k. , 2003 , apj , 598 , 785 oesch , p. a. , et al . , 2010 , apj , 709 , l16 oesch , p. a. , et al . , 2012 , apj , 759 , 135 oesch , p. a. , et al . , 2013 , apj , 773 , 75 okamoto , t. , eke , v. r. , frenk , c. s. , jenkins , a. , 2005 , mnras , 363 , 1299 okamoto , t. , frenk , c. s. , jenkins , a. , theuns , t. 2010 , mnras , 406 , 208 okamoto , t. , frenk , c. s. , 2009 , mnras , 399 , l174 okamoto , t. , nemmen , r. s. , bower , r. g. , 2008 , mnras , 385 , 161 oke , j. b. , 1990 , aj , 99 , 1621 ono , y. , et al . , 2012 , apj , 744 , 83 ono , y. , et al . , 2013 , apj , 777 , 155 ota , k. , et al . , 2008 , apj , 677 , 12 ouchi , m. , et al . , 2009 , apj , 706 , 1136 ouchi , m. , et al . , 2010 , apj , 723 , 869 overzier , r. , et al . , 2013 , mnras , 428 , 778 postman , m. , et al . , 2012 , apjs , 199 , 25 reddy , n. a. , pettini , m. , steidel , c. c. , shapley , a. e. , erb , d. k. , law , d. r. , 2012 , apj , 754 , 25 robertson , b. e. , ellis , r. s. , dunlop , j. s. , mclure , r. , 2010 , nature , 468 , 49 robertson , b. e. , et al . , 2013 , apj , 768 , 71 salvaterra , r. , ferrara , a. , dayal , p. , 2011 , mnras , 414 , 847 salvaterra , r. , maio , u. , ciardi , b. , campisi , m. a. , 2013 , mnras , 429 , 2718 sawicki , m. , yee , h. k. c. , 1998 , aj , 115 , 1329 schaerer , d. , de barros , s. , 2008 , a&a , 502 , 423 schenker , m. a. , et al . , 2013 , apj , 768 , 196 shibuya , t. , et al . , 2012 , apj , 752 , 114 shimizu , i. , yoshida , n. , okamoto , t. , 2011 , mnras , 418 , 2273 shimizu , i. , yoshida , n. , okamoto , t. , 2012 , mnras , 427 , 2866 shimizu , i. , inoue , akio . k. , 2013 , pasj , 65 , 96 springel , v. , white , s. d. m. , tormen , g. , kauffmann , g. , 2001 , mnras , 328 , 726 springel , v. , 2005 , mnras , 364 , 1105 todini , p. , ferarra , a. , 2001 , mnras , 325 , 726 trenti , m. , et al . , 2011 , apj , 727 , l39 trenti , m. , et al . , 2012 , apj , 746 , 55 treu , t. , schmidt , k. b. , trenti , m. , bradley , l. d. , stiavelli , m. , 2013 , apj , 775 , 29 vanzella , e. , et al . , 2011 , apj , 730 , l35 wilkins , s. m. , et al . , 2010 , mnras , 403 , 938 xu , c. , buat , v. , 1995 , a&a , 293 , l65 yan , h. , windhorst , r. a. , 2004 , apj , 612 , l93 zheng , w. , et al . , 2012 , nature , 489 , 406

the congruence lattices of all algebras defined on a fixed finite set ordered by inclusion form a finite atomistic lattice . we describe the atoms and coatoms . each meet - irreducible element of being determined by a single unary mapping on , we characterize completely those which are determined by a permutation or by an acyclic mapping on the set . using these characterisations we deduce several properties of the lattice ; in particular , we prove that is tolerance - simple whenever .

1612.07648

in 1963 g. grtzer and e.t . schmidt proved that every algebraic lattice is isomorphic to the congruence lattice of some algebra ( @xcite ) . since the algebras constructed by them were infinite , the result immediately raised the question : does every finite lattice occur as the congruence lattice of a _ finite _ algebra ? the problem remained open till today , and it is usually referred as the _ finite lattice representation problem_. it is an abstract representation problem because it asks for a solution up to isomorphism . the concrete version is the more involved question : given a sublattice @xmath3 of the the partition lattice @xmath4 of all equivalence relations on a set @xmath0 , does there exist an algebra on the same base set @xmath0 , such that @xmath3 _ equals _ the congruence lattice of this algebra ( in @xcite such lattices @xmath3 are characterized by closure properties ) . the subject of the present paper is related to the finite representation problem in its concrete version . for a fixed finite set @xmath0 we consider all possible congruence lattices of algebras with base set @xmath0 . these congruence lattices ( ordered by inclusion ) form itself a lattice @xmath1 and we are going to investigate this lattice . an important tool is our knowledge about the lattice @xmath5 of all quasiorder lattices of algebras defined on the set @xmath0 described in @xcite ( using some techniques developed previously in the papers @xcite and @xcite ) . these two lattices are strongly interrelated : there is a residual mapping from @xmath5 to @xmath1 . therefore , in section [ sec:2 ] , we investigate on abstract level , how lattice properties ( which are relevant for us ) behave under residual mappings ( for instance , the coatoms of @xmath1 directly can be obtained from the coatoms of @xmath5 , see [ b1 ] ) . based on preliminary results from section [ sec:1 ] and the results of @xcite and section [ sec:2 ] , we describe the atoms ( @xmath6-irreducible elements ) , coatoms ( section [ sec:3 ] ) and further @xmath7-irreducible elements ( sections [ sec:4 ] and [ sec:4a ] ) of the lattice @xmath1 . finally , in section [ sec:5 ] , we investigate several lattice theoretic properties of @xmath1 , e.g. , it is tolerance simple , but has no properties related with modularity . throughout the paper we fix a base set @xmath0 ( if not stated otherwise , @xmath0 is assumed to be finite ) . further , let @xmath8 and @xmath9 . for a mapping @xmath10 , we write @xmath11 for the image of an element @xmath12 , and @xmath13 ( @xmath14 ) denotes the @xmath15-fold composition of @xmath16 ( by convention , @xmath17 is the identity mapping @xmath18 ) . [ a00 ] let @xmath4 and @xmath19 denote the set of all _ equivalence relations _ ( reflexive , symmetric and transitive ) and _ quasiorders _ ( reflexive and transitive relations ) , respectively , on a set @xmath0 . the least and the greatest quasiorders ( which are in fact equivalences ) are @xmath20 and @xmath21 . a unary mapping @xmath10 _ preserves _ a quasiorder @xmath22 ( in particular , an equivalence @xmath23 ) , notation @xmath24 , if @xmath25 this fact is also expressed by the following notions and notation : @xmath16 is an _ endomorphism _ of @xmath26 ( @xmath27 ) , @xmath26 is _ invariant for _ or _ compatible with _ @xmath16 , or @xmath26 is a _ quasiorder of @xmath28 _ ( @xmath29 ) , or @xmath30 is a _ congruence of @xmath28 _ ( @xmath31 ) . the identity @xmath32 as well as all constant mappings @xmath33 are called _ trivial _ because just they preserve every quasiorder @xmath34 . for a unary algebra @xmath35 , @xmath36 , let @xmath37 and @xmath38 be its _ congruence _ and _ quasiorder lattice _ , respectively , i.e. the lattice of all equivalences or quasiorders that are compatible with each @xmath39 . moreover , let @xmath40 denote the lattice of all such congruence lattices and quasiorder lattices , respectively , on @xmath0 , ordered by inclusion . instead of @xmath38 and @xmath37 we sometimes write simply @xmath41 and @xmath42 . since congruences of an algebra are characterized by the unary polynomial functions of the algebra , the lattice @xmath1 is in fact the lattice of all congruence lattices of _ arbitrary _ ( not necessarily unary ) algebras on the set @xmath0 ( the same holds for quasiorders and @xmath5 ) . [ a01 ] the relation @xmath43 induces ( via the operators @xmath44 and @xmath45 ) a galois connection between unary mappings and equivalence relations on @xmath0 . the galois closures are just the elements ( congruence lattices ) @xmath46 and monoids of the form @xmath47 ( for some set @xmath48 ) , in particular , we have @xmath49 the meet in @xmath1 is the intersection while the join of elements @xmath50 ( @xmath51 ) is given by @xmath52 . clearly , @xmath53 implies @xmath54 . thus @xmath7-irreducibles in @xmath1 must be of the form @xmath55 for a single function @xmath16 because @xmath37 is the intersection of all @xmath55 with @xmath39 . analogously , @xmath6-irreducible ( in case of infinite @xmath0 , completely @xmath6-irreducible ) elements of @xmath1 must be of the form @xmath56 for a single equivalence relation @xmath57 , because , for @xmath58 , @xmath59 is the intersection of all @xmath60 and thus @xmath61 is the join ( in @xmath1 ) of all @xmath62 with @xmath63 . [ a1a ] for @xmath64 consider the corresponding partition @xmath65 into equivalence classes . if @xmath66 , @xmath67 , , @xmath68 are the equivalence classes of @xmath30 with at least two elements , then we use the notation @xmath69\,[b_{1},b_{2},\dots]\,\dots\ , [ c_{1},c_{2},\dots]\text { or } \\ \kappa&=[c_{1}]\,[c_{2}]\,\dots\,[c_{m}].\end{aligned}\ ] ] all other elements which do not appear in the list form one - element equivalence classes . [ a0 ] here we introduce some special notions for monounary algebras ; for a more general view to monounary algebras we refer to @xcite . let @xmath28 be a finite monounary algebra . let @xmath70 be the subalgebra of @xmath28 generated by an element @xmath71 . obviously , we have @xmath72 . we write @xmath73 if @xmath74 is ( the carrier set of ) a subalgebra of @xmath28 . considering the graph @xmath75 of @xmath16 , one can use a graph theoretic terminology . for @xmath12 , let @xmath76 denote the _ connected component _ of @xmath77 to which @xmath78 belongs ( note that two vertices @xmath79 are connected w.r.t . @xmath16 , iff there exist @xmath80 with @xmath81 ) . a component @xmath82 of @xmath16 is called _ nontrivial _ if it contains at least two elements ( thus a trivial component is just a fixed point ) . for a monounary algebra @xmath28 , the least quasiorder and congruence , resp . , containing a pair @xmath83 is denoted by @xmath84 and @xmath85 ( _ principal congruence _ ) , resp . , and we have @xmath86 here @xmath87 denotes the symmetric closure and @xmath88 the transitive closure of a binary relation @xmath89 . we now collect some properties for functions @xmath90 with @xmath91 . [ a2 ] let @xmath90 be nontrivial and @xmath91 . then we have 1 . [ a2i ] @xmath92 , + in particular we have @xmath93 and @xmath94 . [ a2ii ] let @xmath74 be a subalgebra of @xmath28 . then either @xmath74 is also a subalgebra of @xmath95 or @xmath96 is constant on @xmath74 , where the constant does not belong to @xmath74 . is clear since @xmath97 implies @xmath98 what follows from the assumption @xmath91 . : for a subalgebra @xmath74 , @xmath99 belongs to @xmath55 . let @xmath100 . if @xmath96 is not constant on @xmath74 , then there exists @xmath101 such that @xmath102 . because @xmath103 , by ( i ) we have @xmath104 , in particular @xmath105 . thus @xmath74 is closed under @xmath96 . if @xmath96 is constant on @xmath74 and @xmath74 is not a subalgebra of @xmath95 , then the constant can not be an element of @xmath74 . [ a2a ] the property in [ a2 ] completely characterizes the containment of the congruence lattices . we have for @xmath90 : @xmath106 in preparation of the next proposition we need the following lemma . [ a3a ] let @xmath16 be a permutation of prime power order @xmath107 with at least two cycles of length @xmath107 . then @xmath108 . the inclusion `` @xmath109 '' is always true . to show `` @xmath110 '' , let @xmath111 . thus there exist @xmath112 with @xmath113 and therefore some principal quasiorder @xmath84 which is not preserved by @xmath114 for some @xmath115 . we must show @xmath116 . assume on the contrary that @xmath117 or , equivalently , @xmath118 . we can assume @xmath119 ( because there must exist @xmath120 with @xmath121 , one can use @xmath122 instead of @xmath123 ) . if @xmath124 belong to the same cycle of the permutation @xmath16 , then @xmath125 ( cf . * lemma 3.1 ) ) and we have @xmath126 , a contradiction . thus we may assume @xmath127 , @xmath128 where @xmath129 are different cycles of @xmath16 of length @xmath130 and @xmath131 , resp . moreover , w.l.o.g . . then we have @xmath133 and @xmath134 where @xmath135 ; note that @xmath136 . we distinguish the following cases ( recall @xmath137 and hence @xmath138 ) : case 1 : @xmath139 , i.e. , @xmath140 for some @xmath141 . by lemma [ a2 ] , @xmath114 is constant @xmath142 on @xmath143 and constant @xmath144 on @xmath145 . if there exists @xmath146 , then @xmath147 , a contradiction . thus @xmath148 , i.e. , @xmath149 and @xmath150 must be two cycles of length @xmath107 . but then @xmath114 does not preserve @xmath151 since @xmath152[fx , f^{2}y]\dots[f^{i}x , f^{i+1}y]\dots$ ] , a contradiction . case 2 : @xmath153 , hence @xmath154 . in particular we have @xmath155 , i.e. , @xmath156 . further , by [ a2 ] , @xmath114 is constant @xmath157 on @xmath158 . if @xmath159 , then @xmath160 , a contradiction . if @xmath161 , then there exists a cycle @xmath162 of length @xmath107 which is distinct from @xmath149 . let @xmath163 . then each block of @xmath164 contains exactly one element of @xmath149 , hence @xmath165 , a contradiction to @xmath166 . thus the assumption @xmath117 must fail , i.e. , @xmath116 . [ a3 ] let @xmath90 be nontrivial such that @xmath91 and let @xmath16 be a permutation of prime power order @xmath107 with at least two cycles of length @xmath107 . then there exists @xmath167 such that @xmath168 . @xmath91 is equivalent to @xmath169 , thus @xmath170 by lemma [ a3a ] , consequently @xmath171 . from ( * proposition 2.5(b ) ) we conclude @xmath172 . clearly @xmath173 can be chosen less than @xmath107 since @xmath174 . we shall strongly use results about the lattice @xmath5 of quasiorder lattices for the investigation of the lattice @xmath1 of congruence lattices . however , we want to seperate those connections which are of pure lattice theoretic nature ( and which are from our point of view of its own interest ) . this is done in this section . based on the observation that @xmath175 is a residual mapping , we consider this case in a general setting . let @xmath176 and @xmath3 be arbitrary lattices which for simplicity here are assumed to be finite , the least and largest elements are denoted by @xmath177 and @xmath178 . a mapping @xmath179 is called _ residual _ if it is a @xmath7-homomorphism ( and therefore also monoton with respect to the lattice orders ) and @xmath180 ( cf . e.g.@xcite or @xcite ) . the following proposition shows that then the @xmath7-irreducible elements of @xmath1 , in particular coatoms , can be constructed from the @xmath7-irreducible elements of @xmath5 . [ b1 ] let @xmath179 be a surjective residual mapping . 1 . [ b1i ] let @xmath181 be a @xmath7-irreducible element in @xmath3 . then @xmath182 and each @xmath183 which is maximal in @xmath184 is @xmath7-irreducible in @xmath176 . [ b1ii ] let @xmath185 and let each @xmath183 with @xmath186 be @xmath7-irreducible in @xmath176 . then @xmath187 is @xmath7-irreducible in @xmath3 . [ b1iii ] assume @xmath188 + then for each coatom @xmath185 there exists a coatom in @xmath183 such that @xmath186 . [ b1iv ] assume condition @xmath189 above and @xmath190 then @xmath191 is a coatom in @xmath3 if @xmath26 is a coatom in @xmath176 . moreover , the set of all coatoms of @xmath3 is @xmath192 . : let @xmath185 be @xmath7-irreducible . then @xmath193 is nonempty because @xmath194 is surjective . let @xmath26 be maximal in @xmath184 . then @xmath26 is the meet of @xmath7-irreducible elements , say @xmath195 with @xmath7-irreducible @xmath196 ( @xmath197 ) . it follows @xmath198 . because @xmath187 is @xmath7-irreducible there exists @xmath197 such that @xmath199 . since @xmath200 and @xmath26 was chosen maximal with respect to @xmath186 , we have @xmath201 , i.e. , it is @xmath7-irreducible . : let @xmath202 for some @xmath203 . since @xmath194 is surjective , there exist @xmath196 with @xmath204 , @xmath205 . . then @xmath207 and @xmath26 must be @xmath7-irreducible by assumption . consequently , there is @xmath205 with @xmath201 , thus @xmath208 , i.e. , @xmath187 is @xmath7-irreducible . : since @xmath176 is finite , by there exists a maximal @xmath7-irreducible @xmath183 with @xmath186 . if @xmath26 were not a coatom then there would exist a @xmath209 with @xmath210 . by the maximality property of @xmath26 , we get @xmath211 , thus @xmath212 ( since @xmath187 is coatom ) and by the assumption from we would get @xmath213 , a contradiction . : let @xmath183 be a coatom . then @xmath214 because of @xmath189 . thus there exists some coatom @xmath187 in @xmath3 with @xmath215 . by there exists a coatom @xmath216 in @xmath176 such that @xmath217 . then @xmath218 and with ( ) we get that @xmath219 is a coatom in @xmath3 . this together with shows that @xmath192 is the set of all coatoms of @xmath3 . [ b1a ] concerning [ b1 ] , since @xmath176 is finite , for any @xmath220 there exists a maximal ( and therefore @xmath7-irreducible ) @xmath26 with @xmath221 . [ b2 ] there are many applications of residual mappings in various contexts , in particular in connection with the unique corresponding residuated mapping ( establishing a `` covariant galois connection '' ) . however , for this paper the only example which we need is the above mentioned residual mapping @xmath222 ( recall @xmath223 and @xmath224 from [ a00 ] ) . clearly , @xmath225 . the next lemma shows that the assumptions ( ) and ( ) in proposition [ b1 ] , are satisfied for this example . notice that @xmath4 and @xmath19 are the greatest elements of the lattices @xmath1 and @xmath5 , respectively . [ b3 ] * for @xmath226 , @xmath227 implies @xmath228 . * for coatoms @xmath229 in @xmath5 , @xmath230 implies @xmath231 . ( i ) : it is well - known that trivial functions ( @xmath18 and the constants , say @xmath162 ) are the only mappings which preserves all equivalence relations . thus we have @xmath232 . ( ii ) : this will follow immediately from proposition [ p2 ] proved in section [ sec:3 ] . note that the coatoms of @xmath1 and @xmath5 are of the form @xmath55 and @xmath233 for some specific @xmath16 ( they are of type ( i)-(iii ) as we shall see in theorem [ thm1 ] ) . thus from [ b1 ] we immediately get : [ b4 ] @xmath234 is the set of all coatoms of @xmath1 . we close this section with two results which shall turn out to be useful later . for an element @xmath162 of a lattice @xmath1 let @xmath235 denote the principal filter generated by @xmath162 . [ b5 ] for @xmath236 and @xmath237 we have @xmath238 . recall @xmath239 and @xmath240 . now , if @xmath241 then @xmath242 since @xmath243 is order preserving and @xmath244 . thus @xmath245 . conversely , if @xmath246 , then @xmath247 and we get @xmath248 , i.e. , @xmath241 . [ b6 ] for @xmath36 we have @xmath249 if and only if @xmath250 . let @xmath251 and @xmath252 . then we have @xmath253 . if @xmath254 then @xmath255 and we get @xmath238 from lemma [ b5 ] . conversely , assume @xmath238 and let @xmath256 . then @xmath257 , thus @xmath258 . consequently , @xmath259 , i.e. , @xmath260 , and we get @xmath261 ( the latter inclusion follow from @xmath262 ) , hence @xmath254 . in this section we are going to describe the atoms ( @xmath6-irreducibles ) and coatoms of @xmath1 . the case @xmath263 is trivial . then @xmath1 consists only of one lattice , namely @xmath264 . therefore , in the following we assume always @xmath265 . the @xmath6-irreducibles are easily described . in the following theorem , @xmath0 may be an arbitrary , not necessarily finite , set . [ thm2 ] the completely @xmath6-irreducibles of @xmath1 are exactly the congruence lattices of the form @xmath266 where @xmath57 is an arbitrary equivalence relation . moreover , each @xmath6-irreducible is an atom in @xmath1 , i.e. the lattice @xmath1 is atomistic . completely @xmath6-irreducibles must be of the form @xmath267 as noted in [ a01 ] . the characterization ( * ) follows immediately from ( * ? ? ? * corollary 2.5 ) , where it is shown that @xmath268 ( and therefore equals @xmath269 ) for any equivalence relation @xmath30 . clearly , such lattices are atoms in @xmath1 and therefore @xmath6-irreducible , since @xmath270 is the only proper sublattice . [ a4 ] for @xmath271 let @xmath272 be the set of atoms contained in @xmath176 . it is natural to ask which sets of atoms are of the form @xmath273 for some @xmath271 . equivalently , for given @xmath274 , we may ask for @xmath275 . formally we put @xmath276:=\{\delta,\nabla\}\cup\operatorname{at}(\operatorname{con}(a,\operatorname{end}e))$ ] because then @xmath277 $ ] is a closure operator which is well - known : @xmath276 $ ] coincides with the galois closures of the galois connection @xmath278 as well as of @xmath279 restricted to equivalence relations ( because @xmath280 ) , cf . @xcite or @xcite , and can be explicitly described by so - called graphical compositions as shown by h. werner in @xcite . now we describe the coatoms . [ thm1 ] the coatoms of @xmath1 are exactly the congruence lattices of the form @xmath55 where @xmath281 satisfies * @xmath16 is nontrivial and @xmath282 , or * @xmath16 is nontrivial , @xmath283 is a constant , say @xmath284 , and @xmath285_{\ker f}}|\geq 3 $ ] , or * @xmath286 for some prime @xmath287 such that the permutation @xmath16 has at least two cycles of length @xmath287 . remark : it can happen that different functions @xmath16 from the theorem give the same coatom . this explicitly will be clarified in proposition [ p2 ] . the theorem describes three types of functions , the graphs of which are shown in figure [ fig:1 ] ( all labeled elements are mandatory , all others are optional ) . moreover , there are functions which define coatoms but which are of none of the types ( i)-(iii ) , e.g. , on @xmath288 we get a coatom @xmath289 for @xmath290 , @xmath291 and @xmath292 where @xmath16 is of type ( i ) , but @xmath96 is of no type . the coatoms in the lattice @xmath5 of quasiorder lattices are known from ( * ? ? ? * theorem 3.1 ) and can be described exactly as @xmath233 for the nontrivial functions @xmath16 of the three types ( i)-(iii ) . thus the theorem immediately follows from corollary [ b4 ] . note that the proof is based on corollary [ b4 ] which follows from [ b1 ] and needs the condition ( ) in its concrete form in lemma [ b3](ii ) . this will be proved with proposition [ p2 ] below . moreover , in [ p2 ] it will be clarified when two different functions of type ( i)-(iii ) give the same congruence lattice @xmath55 ( by [ b1 ] and [ b3](i ) we already know that each coatom in @xmath1 must be of the form @xmath293 for some function @xmath16 of type ( i)-(iii ) ) . however , before stating [ p2 ] we need some more notions , notations and a lemma . [ n1 ] if @xmath294 is of type ( i ) and has exactly one nontrivial component @xmath295 with fixed point @xmath296 , then let @xmath297 be defined by @xmath298 see figure [ fig:1a ] . if @xmath294 is of type ( ii ) with two - element image @xmath299=\{u , z\}$ ] , where @xmath296 shall denote the fixed point , then let @xmath297 be defined by @xmath300 in all other cases we put @xmath301 . from the figure [ fig:1a ] is clear that @xmath302 , moreover @xmath28 and @xmath303 have the same principal congruences , thus @xmath304 . for a function @xmath305 of type ( i ) or ( ii ) , respectively , a triple @xmath306 of three different elements is called _ essential _ for @xmath96 ( or _ @xmath96-essential _ ) if @xmath307 and @xmath308 , or @xmath309 and @xmath310 , respectively , see figure [ fig : es ] . [ l : es ] let @xmath91 for functions @xmath311 , and let three different elements @xmath312 satisfy @xmath313 . then we have : * if @xmath16 is of type ( i ) , then @xmath306 or @xmath314 is @xmath16-essential . * if @xmath16 is of type ( ii ) , then @xmath315 or @xmath306 or @xmath316 or @xmath314 is @xmath16-essential ( note that @xmath296 never appears in the first component of these triples ) . because of lemma [ a2]([a2i ] ) we have @xmath317 . for functions of type ( i ) or ( ii ) the only possibilities that all three elements @xmath318 belong to the same block of @xmath85 are those mentioned in ( a ) and ( b ) , cf . figure [ fig : es ] and figure [ fig:1 ] . if @xmath305 is of type ( i ) or ( ii ) , then the @xmath96-essential triples @xmath319 or @xmath320 ( notation as in figure [ fig:1 ] ) , respectively , satisfy the condition @xmath313 from lemma [ l : es ] . [ p2 ] let @xmath311 be nontrivial operations of one of the types ( i)-(iii ) such that @xmath321 . then @xmath322 if @xmath16 is of type ( i ) or ( ii ) , and @xmath323 for some @xmath324 if @xmath16 is of type ( iii ) . in particular we always have @xmath289 . if @xmath16 is of type ( iii ) , then from proposition [ a3 ] ( where functions @xmath16 of prime power order @xmath107 are considered , here one has to take @xmath325 ) we conclude that @xmath326 for some @xmath327 . corresponding to the types of @xmath16 and @xmath96 , there remain altogether 6 cases to consider , denoted by @xmath328 if @xmath16 is of type @xmath329 and @xmath96 of type @xmath330 , where @xmath331 and @xmath332 . we start with the cases where @xmath16 and @xmath96 have different types . for these cases we shall indicate elements @xmath79 with @xmath333 , a contradiction to [ a2]([a2i ] ) ; e.g. by using lemma [ l : es ] or by finding @xmath334 . take the @xmath96-essential triple @xmath335 . by [ l : es](a ) we have that @xmath336 or @xmath337 is @xmath16-essential . thus @xmath338 $ ] , in contradiction to @xmath339 . since @xmath96 is of type ( iii ) , there exist elements @xmath340 with @xmath341[a_{1},b_{1}]\ldots[a_{p-1},b_{p-1}]$ ] ( see figure [ fig:1 ] ) . because @xmath342 we get that @xmath16 ( as function of type ( i ) or ( ii ) ) must satisfy @xmath343 ( moreover , @xmath344 ) and @xmath345 $ ] ( because @xmath346 for type ( i ) , and @xmath347 for @xmath16 of type ( ii ) ) ; in particular @xmath348 can not contain @xmath349 , a contradiction . take the @xmath96-essential triple @xmath319 . by [ l : es](b ) there exists an @xmath16-essential triple @xmath350 with @xmath351 . further , there must exist @xmath352 with @xmath353 . note that @xmath354 consists only of 2-element blocks for each @xmath355 . therefore , @xmath356 $ ] if @xmath357 , or @xmath358 $ ] if @xmath359 , can not be contained in @xmath360 or @xmath361 , respectively , a contradiction . now we continue with the cases where @xmath16 and @xmath96 are of the same type . let @xmath362 be an @xmath96-essential triple . according to lemma [ l : es](a ) , we have only the following two possibilities : @xmath363 clearly this must hold for _ each _ essential triple of @xmath96 . we show that ( * ) implies @xmath364 and ( * * ) implies @xmath365 . let @xmath366 arbitrary . consider case ( * ) : if @xmath367 , take the @xmath96-essential triple @xmath368 for which only case ( * ) is possible ( since @xmath369 ) , i.e. @xmath370 . if @xmath371 , take the @xmath96-essential triple @xmath372 ( provided that @xmath373 ) or @xmath374 ( if @xmath375 ) for which again case ( * ) must hold ( since @xmath376 and @xmath377 ) , i.e. @xmath378 , too . altogether @xmath379 . now consider case ( * * ) : if @xmath367 , take the @xmath96-essential triple @xmath368 for which only case ( * * ) is possible ( since @xmath380 ) , i.e. @xmath381 . if @xmath382 , take the essential triple @xmath383 for which again case ( * * ) must hold ( since @xmath384 ) , i.e. @xmath385 . the case @xmath386 can not appear because then we would have @xmath387[z , gc]=\theta_{g}(a , c)\subseteq \theta_{f}(a , c)=[a , c , fc]$ ] , a contradiction . thus altogether we have @xmath365 . let @xmath336 be a @xmath96-essential triple . then the triples @xmath388 and @xmath389 satisfy the assumptions of lemma [ l : es](b ) , consequently neither @xmath296 nor @xmath390 can be in the first component of the corresponding @xmath16-essential triples . thus there remain only the following two possibilities : @xmath391 clearly this must hold for _ each _ essential triple of @xmath96 . we show that ( * ) implies @xmath364 and ( * * ) implies @xmath365 ( equivalentyl , @xmath392 ) . consider case ( * ) : since @xmath377 is the unique fixed point of @xmath16 , for each @xmath96-essential triple @xmath393 we get case ( * ) , i.e . @xmath393 is @xmath16-essential and thus @xmath16 and @xmath96 agree on all essential triples . if @xmath394 does not belong to an @xmath96-essential triple , then @xmath395 and @xmath396[u , z]$ ] . from @xmath397[u , fb]$ ] we conclude @xmath384 ; altogether @xmath379 . consider now case ( * * ) : since @xmath398 is the unique fixed point of @xmath16 but @xmath399 is not a fixed point , for each @xmath96-essential triple @xmath393 we must have case ( * * ) , i.e. @xmath400 is @xmath16-essential , in particular @xmath401 and @xmath402 . if @xmath394 does not belong to an @xmath96-essential triple , then @xmath395 and @xmath396[u , z]$ ] . from @xmath397[z , fb]$ ] we conclude @xmath403 ; altogether @xmath365 . the coatoms are @xmath7-irreducible in @xmath1 . now we want to deal with @xmath7-irreducible congruence lattices in general . they all must be of the form @xmath55 for a single @xmath16 and we have : [ p3 ] let @xmath55 be a @xmath7-irreducible element in @xmath1 . then there exists @xmath407 such that @xmath408 is @xmath7-irreducible in @xmath5 and @xmath289 , @xmath171 . the proof directly follows from proposition [ b1 ] and remark [ b1a ] applied to the residual mapping @xmath243 , cf . [ b2 ] ( the role of @xmath409 in [ b1 ] and [ b1a ] here is played by @xmath410 ) . we shall describe first the @xmath7-irreducibles for permutations @xmath16 and in the next section for acyclic @xmath16 . for permutations @xmath16 the @xmath7-irreducible quasiorder lattices @xmath233 are known . they are described in ( * ? ? ? * theorem 3.2 ) : @xmath16 is either a transposition or of the form as given in theorem [ thm3 ] below . we first exclude the transpositions : [ p4 ] let @xmath294 be a transposition @xmath411 . then @xmath55 is not @xmath7-irreducible . if @xmath294 is a transposition , then there are elements @xmath412 such that @xmath413 , @xmath414 and @xmath415 for @xmath416 . let @xmath417 be the nontrivial functions defined by @xmath418 , @xmath419 and all @xmath416 are fixed point for @xmath420 and @xmath421 . then , for the principal congruences , we have @xmath422 $ ] for all @xmath79 with the only exceptions @xmath423 $ ] for all @xmath424 . note @xmath425 and @xmath426 . therefore @xmath427 , @xmath428 and @xmath429 . the following proposition deals with those functions which will play the crucial role in the next theorem . [ p1a ] let @xmath265 and let @xmath430 be a permutation of prime power order @xmath107 with at least two cycles of length @xmath107 . then the principal filter @xmath431 is a chain . moreover , each element of this chain is @xmath7-irreducible ( except the top element @xmath4 ) and is of the form @xmath432 , where @xmath326 for some @xmath433 . given @xmath16 as indicated we know from ( * ? ? ? * theorem 4.2 and 4.3 ) that the principal filter @xmath434 is a chain . from lemma [ a3a ] and corollary [ b6 ] we conclude @xmath435 , consequently @xmath436 is also a chain , therefore each element ( except @xmath4 ) of this chain is @xmath7-irreducible and thus of the form @xmath432 . by [ a3 ] each nontrivial @xmath96 with @xmath91 is of the form @xmath326 . [ thm3 ] a congruence lattice @xmath55 with a nontrivial permutation @xmath16 is @xmath7-irreducible in @xmath1 if and only if @xmath16 is of prime power order @xmath107 with at least two cycles of length @xmath107 . `` @xmath437 '' was proved in [ p1a ] . `` @xmath438 '' : let @xmath16 be a permutation such that @xmath55 is @xmath7-irreducible . by [ p3 ] there exists @xmath305 such that @xmath408 is @xmath7-irreducible in @xmath5 and @xmath439 as shown in ( * ? ? ? * lemma 2.4(iv ) ) ) , from ( * ) follows that each subalgebra of @xmath28 is also a subalgebra of @xmath95 , while from ( * * ) and lemma [ a2 ] ( interchange the roles of @xmath16 and @xmath96 ) follows that each subalgebra of @xmath95 with at least two elements is also a subalgebra of @xmath28 ( since @xmath16 is a permutation , it can not be constant on two elements ) . at first we show that @xmath96 is also a permutation . let @xmath440 such that @xmath441 . since cycles ( what coincides with components ) of @xmath16 are subalgebras of @xmath28 and thus also of @xmath95 , @xmath442 implies @xmath443 , @xmath444 , hence @xmath102 . thus let @xmath124 belong to the same cycle of @xmath16 and assume @xmath445 . then from ( * * ) and lemma [ a2 ] ( here the roles of @xmath16 and @xmath96 are interchanged ) we conclude @xmath446 $ ] , thus @xmath447 is a subalgebra of @xmath28 and therefore @xmath448 ( as mentioned above ) . thus w.l.o.g . we can assume @xmath449 . let @xmath450 and let @xmath162 be the cycle of @xmath16 which contains @xmath296 . then @xmath451 is a subalgebra(with at least @xmath452-elements ) of @xmath95 but not of @xmath28 ( since @xmath453 ) , a contradiction . thus @xmath96 is a permutation . therefore , from ( * ? ? ? * proposition 2.5(b ) ) and @xmath7-irreducibility of @xmath408 we get that @xmath96 is a permutation of prime power order @xmath107 with at least two cycles of length @xmath107 or that @xmath96 is a transposition . since @xmath454 is @xmath7-irreducible , @xmath96 can not be a transposition by lemma [ p4 ] . from ( * * ) and proposition [ a3 ] ( interchange the role of @xmath16 and @xmath96 ) we get @xmath455 for some @xmath433 . from ( * * ) we further conclude that @xmath287 can not divide @xmath173 ( since @xmath456 ) . therefore @xmath16 and @xmath96 generate the same cyclic subgroup , in particular @xmath16 also has order @xmath107 and at least two cycles of length @xmath107 , and we are done . in this section we deal with acyclic algebras @xmath28 . for an acyclic @xmath294 and @xmath71 let @xmath457 denote the so - called _ depth _ of @xmath458 ( after @xmath15 times applying @xmath16 to @xmath458 one gets a fixed point ) and let @xmath459 ( this is the length of a longest `` tail '' in the graph of @xmath16 ) . [ d0 ] for a nontrivial , acyclic function @xmath294 we consider the following conditions ( cf . figure [ fig:3 ] ) : * there exist distinct elements @xmath460 such that @xmath461 and @xmath462 , * @xmath16 is connected ( i.e. , only one component ) and there exist distinct elements @xmath463 such that @xmath464 . [ d1a ] let @xmath294 be nontrivial and acyclic such that @xmath16 is not of type ( i ) , not of type ( ii ) and satisfies neither condition [ d0](a ) nor ( b ) . then @xmath55 is @xmath7-reducible . if @xmath465 then @xmath16 is of type ( i ) . thus we can assume @xmath466 . we distinguish the following cases : case 1 : @xmath16 has at least two components . + then ( b ) trivially does not hold . if ( a ) fails to hold , then @xmath16 has exactly one component , say @xmath82 , with elements of depth @xmath452 while all other components have elements of depth at most @xmath467 . in particular there are at least two fixed points , say @xmath468 and @xmath469 . therefore @xmath16 is of the form as given in figure [ fig:5a ] ( the shadowed part is @xmath82 ) . we define the functions @xmath421 and @xmath470 as follows ( see figure [ fig:5a ] ) : @xmath471 case 2 : @xmath16 has only one component with fixed point , say @xmath284 . + then ( a ) trivially does not hold . if ( b ) fails to hold then all elements @xmath458 with @xmath472 map to the same element , say @xmath467 , of depth @xmath467 . case 2a : if @xmath473 , then there is only one element of depth @xmath467 , because otherwise @xmath16 would be of type ( ii ) what is excluded by assumption . therefore @xmath16 is of the form as given in figure [ fig:5b ] . we define the functions @xmath421 and @xmath470 as follows ( see figure [ fig:5b ] ) : @xmath474 case 2b : if @xmath475 then @xmath16 must be of the form as given in figure [ fig:5c ] . we define the functions @xmath421 and @xmath470 as follows ( see figure [ fig:5c ] ) : @xmath476 in all cases the functions @xmath477 are nontrivial and it is easy to check that these functions satisfy @xmath427 ( @xmath205 ) and @xmath478 . in fact , using the figures [ fig:5a ] , [ fig:5b ] and [ fig:5c ] , we can check @xmath479 , therefore @xmath480 by remark [ a2a ] ; moreover the inclusions are strict ( e.g. for case 1 we have @xmath481=\theta_{g_{1}}(1,b_{1})\neq \theta_{f}(1,b_{1})=[1,b_{1}][0,0']$ ] ) . further , @xmath482 for all @xmath79 ( e.g. for case 1 we have @xmath483[1,0,0']=[2,b_{1}][1,0 ] \lor [ 2,b_{1}][0,0']= \theta_{g_{1}}(2,b_{1})\lor \theta_{g_{2}}(2,b_{1})$ ] , cf . figure [ fig:5a ] ) . consequently , @xmath484 . thus @xmath28 is @xmath7-reducible . [ d1b ] let @xmath294 be acyclic such that @xmath16 is not of type ( i ) , not of type ( ii ) and does satisfy either condition [ d0](a ) or ( b ) . let @xmath485 for @xmath305 . then we have @xmath486 for the equivalence relation @xmath487 $ ] . assume @xmath488 and we shall show that this leads to a contradiction . if @xmath16 satisfies ( b ) , it is convenient to put formally @xmath489 . we have @xmath466 , because @xmath465 means that @xmath16 is of type ( i ) . : _ @xmath490_. + in fact , if @xmath491 were not a subalgebra , then ( by [ a2]([a2ii ] ) ) @xmath96 would be constant on @xmath491 . thus @xmath492=\theta_{g}(0,2)\in \operatorname{con}(a , g)$ ] , a contradiction . : _ @xmath493_. + assume that @xmath494 is not a subalgebra . then @xmath96 must be constant on @xmath494 ( by [ a2]([a2ii ] ) ) where the constant is outside @xmath494 . because of claim 1 we get @xmath495 and @xmath496 . the values @xmath497 can not appear ( otherwise @xmath492\in\operatorname{con}(a , g)$ ] ) , thus it remains @xmath498 . since @xmath499 , again by [ a2]([a2ii ] ) @xmath96 must be constant @xmath452 on these elements , in particular @xmath500 . consequently @xmath501[1,0',0]$ ] , a contradiction ( see figure [ fig:3 ] ) . : we have _ @xmath502 _ , i.e. @xmath503 . + if @xmath504 , then @xmath505 ( by claim 2 ) . because @xmath506 ( recall @xmath507 in case ( b ) ) and by [ a2]([a2ii ] ) @xmath96 must be constant @xmath467 on these elements , in particular @xmath508 . thus @xmath509[1,1'][0,0']$ ] implies @xmath498 according to claim 1 . consequently @xmath510=\rho_{1}$ ] , a contradiction . thus @xmath503 . : @xmath96 and @xmath16 agree on @xmath491 . + because @xmath503 ( by claim 3 ) , the values @xmath497 can not appear ( otherwise @xmath492\in\operatorname{con}(a , g)$ ] ) . thus @xmath498 ( by claim 1 ) . it remains to prove @xmath511 . if @xmath512 then @xmath513 ( by claim 2 ) . thus @xmath514[0,1',0']$ ] what implies either @xmath515 or @xmath508 , the former gives the contradiction @xmath516[1,1'][0,0']$ ] and the latter gives the contradiction @xmath517 $ ] . thus @xmath511 . : @xmath96 and @xmath16 agree on @xmath518 . + from @xmath519[1,1'][0,0']$ ] we conclude @xmath520 . however @xmath508 gives the contradiction as seen in claim 4 . consequently @xmath521 . thus @xmath518 is a subalgebra of @xmath95 ( since @xmath522 ) . further , @xmath523[0,0']$ ] gives @xmath524 , thus @xmath525 ( note @xmath526 must belong to the subalgebra @xmath518 ) . finally , @xmath527 $ ] implies @xmath528 ( note @xmath529 must belong to the subalgebra @xmath530 ) . : we have @xmath531 , i.e. @xmath532 , for each @xmath71 . + if @xmath16 satisfies [ d0](b ) , then @xmath533 , and with [ a2]([a2ii ] ) and @xmath503 we get @xmath531 . + if @xmath16 satisfies [ d0](a ) , then @xmath534 and @xmath535 are subalgebras of @xmath28 and @xmath503 , @xmath536 . thus they are also subalgebras of @xmath95 ( due to [ a2]([a2ii ] ) ) , consequently the intersection @xmath537 is also a subalgebra of @xmath95 . : @xmath364 . + because of claim 4 and 5 we have to show @xmath538 for each @xmath539 . if @xmath458 is a fixed point of @xmath16 , then @xmath540 and from claim 6 we get @xmath541 . thus let @xmath542 . we can assume @xmath543 ( otherwise interchange the role of @xmath467 and @xmath544 ) . we have @xmath545[1,fx][0,f^{2}x,\dots , f^{k}x ] & \text { if } t_{f}(x)\geq 2,\\ [ 2,x][0,1,fx ] & \text { if } t_{f}(x)=1 . \end{cases } \end{aligned}\ ] ] if @xmath546 we conclude @xmath547 and get @xmath538 ( since @xmath548 ) . if @xmath549 , then we conclude @xmath550 . moreover , by assumption we have @xmath551 and , by claim 6 we get @xmath552 . thus @xmath553 . from claim 7 we get @xmath289 , a contradiction . [ thm4 ] a congruence lattice @xmath554 with an acyclic @xmath294 is @xmath7-irreducible in @xmath1 if and only if @xmath16 is of type ( i ) or ( ii ) or satisfies the condition [ d0](a ) or ( b ) . note that @xmath55 is a coatom for functions @xmath16 of type ( i ) or ( ii ) , and therefore @xmath7-irreducible . so we need not consider these cases in the following . `` @xmath438 '' : follows from proposition [ d1a ] `` @xmath437 '' : from proposition [ d1b ] we conclude that @xmath555 since @xmath556 $ ] we have @xmath557 and the above intersection can not be equal to @xmath55 . therefore @xmath55 is @xmath7-irreducible . at first we consider the problem how many coatoms ( atoms , resp . ) do we need such that their meet ( join , resp . ) in @xmath1 gives the least ( greatest , resp . ) element of @xmath1 . we assume throughout that @xmath265 . [ p7 ] there are two or three coatoms in the lattice @xmath1 whose meet is @xmath558 . more precisely , for @xmath559 , there are two coatoms @xmath55 and @xmath432 such that @xmath560 . for @xmath561 , three coatoms are necessary ( and sufficient ) for this property . since @xmath562 is a coatom in @xmath1 iff @xmath233 is a coatom in @xmath5 ( cf . [ b4 ] , [ p2 ] ) , the result mainly follows from the corresponding result in ( * ? ? ? * proposition 6.2 ) . in that paper for @xmath563 there are indicated two permutations @xmath16 and @xmath96 of type ( iii ) , for @xmath564 three permutations of type ( iii ) , for @xmath565 one permutation of type ( iii ) and two functions of type ( i ) and for @xmath566 three functions of type ( i ) . now , with @xmath1 instead of @xmath5 , in case @xmath567 for @xmath568 also two functions ( e.g. , of type ( ii ) ) suffice , e.g. , @xmath569 , @xmath570 , @xmath571 and @xmath572 , @xmath573 , @xmath574 . it can be checked easily that two coatoms are not sufficient for @xmath575 . now we look for tolerances of the lattice @xmath1 . because tolerance simplicity implies interesting properties of a lattice ( see , e.g. , @xcite ) , we looked for this property for the lattice @xmath1 and got an affirmative result in theorem [ p6b ] below . [ p6ba ] let @xmath581 be a lattice with the order and covering relation denoted by @xmath582 and @xmath583 , respectively . if @xmath581 is a bounded lattice ( in particular , if it is finite ) , its least and greatest elements are denoted by @xmath584 and @xmath585 . a _ tolerance _ of @xmath581 is a reflexive and symmetric binary relation @xmath586 compatible with the lattice operations @xmath7 and @xmath6 . let @xmath587 denote all tolerances of @xmath581 . with respect to set - theoretic inclusion the tolerances form an algebraic lattice @xmath588 with least element @xmath589 and greatest element @xmath590 ( called _ trivial _ tolerances ) . a lattice @xmath581 is called _ tolerance simple _ if it has no nontrivial tolerances , i.e. , @xmath591 . for @xmath592 , let @xmath593 denote the least tolerance in @xmath587 containing the pair @xmath123 . clearly , for each @xmath594 , we have @xmath595 the following properties are known ( see , e.g. , @xcite ) for @xmath592 : @xmath596 a lattice @xmath581 is called _ atomistic _ if every element @xmath597 is the join of some atoms of @xmath581 . the atoms of @xmath581 , denoted by @xmath598 in the following , play an important role also in connection with tolerance simplicity . from @xcite we deduce ( see also @xcite ) the following : a finite atomistic lattice @xmath581 satisfying @xmath599 for every atom @xmath600 , is tolerance simple . : since @xmath610 is the same tolerance for each atom @xmath611 , we will denote it by @xmath612 . we have @xmath613 for all @xmath611 , consequently @xmath614 . from we get @xmath615 , i.e. , @xmath616 for all @xmath611 . as mentioned above in [ p6ba ] , this implies tolerance - simplicity of @xmath581 . now , instead of the abstract lattice @xmath581 , we return to the concrete lattice @xmath1 . recall that @xmath1 is atomistic and @xmath617 where @xmath618 ( theorem [ thm2 ] ) . the least and greatest elements are @xmath619 and @xmath620 . as defined in [ a1a ] , @xmath621 $ ] denotes the equivalence relation ( on @xmath0 ) with one nontrivial block @xmath622 . since @xmath30 is nontrivial there exist distinct elements @xmath626 such that @xmath623 but @xmath627 . clearly @xmath628\subseteq \kappa$ ] . define @xmath294 via @xmath629 if @xmath630 , and @xmath415 otherwise . then @xmath631 is of type ( i ) and therefore @xmath55 is a coatom in @xmath1 ( cf . theorem [ thm1 ] ) . since @xmath632 , the function @xmath16 preserves neither @xmath30 nor @xmath621 $ ] . hence @xmath633\notin\operatorname{con}(a , f)$ ] , consequently the atoms @xmath634}$ ] are not contained in @xmath55 and therefore , by applying lemma [ p6bb ] , we obtain @xmath625})$ ] . in view of lemma [ p6bb ] it is sufficient to show @xmath635 i.e. , for all @xmath636 . due to lemma [ p6bc ] we even may restrict to equivalence relations of the form @xmath637 $ ] , @xmath638 $ ] for @xmath639 . if @xmath640 , then @xmath641}=e_{[a_{2},b_{2}]}$ ] and we are done . hence , w.l.o.g , we can restrict to the following two cases : _ case _ ( a ) : @xmath642 and @xmath643 . + since @xmath2 there exists an element @xmath644 . define @xmath294 by @xmath645 if @xmath646 and @xmath415 otherwise . then @xmath16 is of type ( i ) and @xmath55 is a coatom ( cf . theorem [ thm1 ] ) . from the definition immediately follows that @xmath16 does not preserve neither @xmath637 $ ] nor @xmath638 $ ] . consequently @xmath647 , @xmath648 and from lemma [ p6bb ] we conclude @xmath649 . _ case _ ( b ) : @xmath650 . + consider the permutation @xmath651 ( with two cycles of length @xmath452 ) . by theorem [ thm1 ] , @xmath55 is a coatom ( type ( iii ) ) and we have @xmath652 , @xmath653 and as in case ( a ) above we get @xmath649 . if @xmath0 has at least @xmath654 elements , say @xmath655 , then consider the nontrivial equivalence relations @xmath656 $ ] , @xmath657[2,3]$ ] , @xmath658 $ ] and the function @xmath294 defined by @xmath659 if @xmath660 , and @xmath415 otherwise . we get the sublattice as shown in figure [ fig:7 ] , e.g. , @xmath55 is a coatom by theorem [ thm1 ] and it is easy to check that for @xmath661 ( @xmath662 ) we have @xmath663 and @xmath664 ( @xmath665 ) . note that @xmath16 does not preserve neither @xmath666 nor @xmath667 nor @xmath668 . obviously @xmath669 is a sublattice isomorphic to @xmath670 . thus @xmath1 is neither @xmath284- nor @xmath467-modular . further , @xmath671 ( blue line ) but the meet with @xmath672 ( dashed blue line ) is not a covering ; likewise @xmath673 ( red line ) but the join with @xmath674 ( dotted red line ) is not a covering . hence @xmath1 is neither lower nor upper semimodular . for @xmath566 , theorem [ p6b ] and proposition [ p6a ] do not remain valid . in this case , @xmath1 is the lattice of all subsets of @xmath675 containing @xmath676 , where @xmath677 , @xmath678 , are the nontrivial equivalence relation on @xmath0 . thus it is a boolean lattice with @xmath679 elements . therefore it is modular and it is not tolerance - simple ( it is even not congruence - simple ) . we dedicate our paper to the memory of e. tams schmidt who passed away in 2016 . the final version will appear in a special issue of algebra universalis . each of the authors is grateful to have known tams and to profit from numerous inspiring and friendly discussions with this outstanding mathematician . in particular , the third author is indepted to tams as phd supervisor introducing him to contemporary lattice theoretical investigations and giving constructive hints concerning the finite representation problem . d. jakubkov - studenovsk , r. pschel , and s. radeleczki , _ the lattice of compatible quasiorders of acyclic monounary algebras_. order 28(3 ) , ( 2011 ) , 481497 , ( online publication doi 10.1007/s11083 - 010 - 9186 - 9 ) . d. jakubkov - studenovsk , r. pschel , and s. radeleczki , _ irreducible quasiorders of monounary algebras_. j. aust . 93(3 ) , ( 2013 ) , 259276 , ( online publication doi 10.1017/s1446788712000328 ) . m. kindermann , _ ber die quivalenz von ordnungspolynomvollstndigkeit und toleranzeinfachheit endlicher verbnde_. in : _ contributions to general algebra ( proc . klagenfurt conf . , klagenfurt , 1978 ) _ , heyn , klagenfurt , 1979 , pp . 145149 . r. pschel , _ galois connections for operations and relations_. in : k. denecke , m. ern , and s.l . wismath ( eds . ) , _ galois connections and applications _ , vol . 565 of _ mathematics and its applications _ , kluwer academic publishers , dordrecht , 2004 , pp . 231258 . r. pschel and s. radeleczki , _ endomorphisms of quasiorders and related lattices_. in : g. dorfer , g. eigenthaler , h. kautschitsch , w. more , and w.b . mller ( eds . ) , _ contributions to general algebra 18 _ , verlag heyn gmbh & co kg , 2008 , pp . 113128 , ( proceedings of the klagenfurt conference 2007 ( aaa73+cya22 ) , febr . 2007 ) . h. werner , _ which partition lattices are congruence lattices ? _ in : _ lattice theory ( proc . , szeged , 1974 ) _ , north - holland , amsterdam , 1976 , pp . 433453 . colloq . jnos bolyai , vol .

correlated variations in the line and continuum emission from active galactic nuclei ( agn ) can be used to determine the size and geometry of the broad emission line regions ( blrs ) . we have spectrophotometrically monitored a well - defined sample of 28 palomar - green quasars in order to obtain measurements of their blrs and to investigate the relationships between quasar luminosity , central black hole mass , and blr size in agn . spectrophotometry was obtained every 14 months for 7.5 years , yielding 2070 observing epochs per object . both the continuum and emission line fluxes of all of the quasars were observed to change during the duration of the observing program . seventeen of the 28 objects were observed with adequate sampling ( independent observing epochs ) to search for correlated variations between the balmer emission lines and the continuum flux . for each of these 17 objects , a significant correlation was observed , with the balmer line variations lagging those of the continuum by days ( rest frame ) . our work increases the available luminosity range for studying the size mass luminosity relations in agn by two orders of magnitude and doubles the number of objects suitable for such studies . combining our results with comparable published data available for seyfert 1 galaxies , we find the blr size scales with the rest - frame 5100 luminosity as . this determination of the scaling of the size of the blr as a function of luminosity is significantly different from those previously published , and suggests that the effective ionization parameter in agn may be a decreasing function of luminosity . we are also able to constrain , subject to our assumption that gravity dominates the motions of the blr gas , the scaling relationship between the mass of the central black holes and the luminosity in agn . we find that the central mass scales with 5100 luminosity as . this is inconsistent with all agn having optical luminosity that is a constant fraction of the eddington luminosity .

astro-ph9911476

reverberation mapping , observing the degree and nature of the correlation between continuum and emission line flux variations , is one of the major tools for studying the distribution and kinematics of the gas in the broad line region ( blr ) of active galactic nuclei ( agn ; see reviews by peterson 1993 and by netzer & peterson 1997 ) . about 17 low luminosity agn ( seyfert 1 galaxies ) have been successfully monitored and produce statistically meaningful results ( see wandel , peterson , & malkan 1999 , and references therein ) . best studied among these is the seyfert 1 galaxy , ngc 5548 , which was monitored from the ground for over eight years , and from space for several long periods ( peterson et al . 1999 , and references therein ) . several other seyfert 1s were observed for periods of order 1 year or less , and nine seyfert 1s were studied over a period of eight years ( peterson et al . the measured time lags between the emission lines and the continuum light curves in these objects can be interpreted in terms of the delayed response of a spatially - extended blr to a variable , compact source of ionizing radiation . while the observations do not uniquely determine the geometry of the blr , they give its typical size which , for seyfert 1 galaxies , is of the order of light - days to several light - weeks ( @xmath4@xmath5 cm ) . recent studies have shown that the time lags determined in ngc 5548 for different observing seasons correlate with the seasonal luminosity of the object ( peterson et al . 1999 ) , and have presented evidence for keplerian motions of the blr gas ( peterson & wandel 1999 ) . while seyfert 1 galaxies have been studied successfully using reverberation mapping , few similar studies of the more luminous agn the quasars have been presented . those that have been published have been limited in their success at determining the properties of the quasar blrs . past attempts to monitor quasars spectrophotometrically have generally suffered from temporal sampling and/or flux calibrations that are not sufficient for the determination of the blr size . zheng and collaborators ( zheng & burbidge 1986 , zheng et al . 1987 , and zheng 1988 ) reported results of a monitoring program executed on 30 quasars for several years , with a sampling interval of about one year . they found the emission - line flux to change in response to the continuum changes , but because of the limited time resolution , only an upper limit of about one light - year for the blr size could be deduced . several other groups ( prez , penston , & moles 1989 , korista 1991 , jackson et al . 1992 , and erkens et al . 1995 ) observed samples of quasars for approximately one year sampled with monthly observations . each of these groups reported line variations on time scales of a few months , but because of the short duration of these programs the data were insufficient for more specific conclusions . wisotzki et al . ( 1998 ) monitored the gravitationally lensed quasar he1104@xmath61805 for five years . although the quasar continuum varied considerably , the emission line fluxes appeared to remain constant . researchers have also attempted to use spectra obtained with the _ international ultraviolet explorer ( iue ) _ and reverberation mapping techniques to study agn . in particular , several studies ( gondhalekar et al . 1986 , gondhalekar 1990 , sitko 1990 , obrien & gondhalekar 1991 , and koratkar et al . 1998 ) deduced blr sizes of 0.012 light - years for agn . however , these results are controversial due to insufficient sampling and/or uncertainty regarding the absolute flux calibrations of the iue observations ( e.g. , bohlin & grillmair 1988a , 1988b ) . even the quasar best studied by iue , 3c273 , has yielded disputed results when different researchers have analyzed similar iue data sets . both obrien & harries ( 1991 ) and koratkar & gaskell ( 1991a ) found a measurable and similar lag between continuum and blr variations , while ulrich , courvoisier , & wamsteker ( 1993 ) argue that the line variations reported in the earlier studies were only marginally significant . thus , even in the the best - studied case , past results have proven controversial . to obtain more definite results on the blr size in quasars , we began monitoring a well - defined sub - sample of 28 quasars from the palomar - green ( pg ) sample ( schmidt & green 1983 ) in 1991 march at the wise and steward observatories . results from the first 1.5 years were presented in maoz et al . ( 1994 ; paper i ) where it was shown that most quasars underwent continuum variations ( at 4800 rest wavelength ) with amplitudes of 10% 70% . balmer - line variability , correlated with the continuum variations , was detected in several objects . the preliminary data showed that the typical response time of the emission lines is @xmath76 months . reverberation mapping of such objects therefore requires several years , with sampling intervals of less than a few months . further results , based on four years of monitoring two quasars from our sample , pg0804 + 761 and pg0953 + 414 , were presented in kaspi et al . ( 1996a ; paper the measured time lags between the balmer lines and the continuum variations of the quasars suggested that the blr size grows roughly as the square root of the luminosity of the nucleus . wandel et al . ( 1999 ) confirmed this result using a larger sample of agn . during the period of our spectrophotometric project we also monitored several quasar samples photometrically in the @xmath8 and @xmath9 bands . the broad - band results for radio- and optically - selected samples were presented in netzer et al . ( 1996 ) and giveon et al . ( 1999 ; hereafter paper iii ) , respectively . the optical sample in paper iii consists of 42 pg quasars , including all 28 objects discussed in the present paper . the present paper presents final results from 7.5 years of spectroscopic monitoring of our sample . in 2 we describe the sample , observations , data reduction , and present the light curves . in 3 we perform a time series analysis to determine the response time of the emission - line flux to continuum variations . we estimate the blr size , continuum luminosity , and central mass for each quasar in 4 , and in 5 we discuss the relations among these properties . the sample , observing technique , and the reduction procedure , are described in detail in papers i , ii , & iii . for the sake of completeness , we briefly repeat them here , with emphasis on the few procedures which , in the final reduction , were modified with respect to previous descriptions . the sample consists of 28 pg quasars . objects with a northern declination , @xmath10 mag ( based on the magnitudes given in schmidt & green 1983 ) , redshift @xmath11 , and a bright comparison star within @xmath12 of the quasar , were selected . table [ tsample ] presents detailed information about the sample . throughout this paper we will use only the first four digits in the pg object name to identify individual objects . cccccccccccc pg0026 + 129 & 00 26 38.1 & 12 59 30 & 0.142 & 15.3 & -23.7 & 0.13 & 38 & 56 & @xmath13 & 95 & 42.0 pg0052 + 251 & 00 52 11.1 & 25 09 24 & 0.155 & 14.7 & -24.1 & 0.12 & 41 & 56 & @xmath14 & 92 & 153.4 pg0804 + 761 & 08 04 35.4 & 76 11 32 & 0.100 & 14.5 & -23.7 & 0.11 & 38 & 70 & @xmath15 & 182 & 315.6 pg0844 + 349 & 08 44 33.9 & 34 56 09 & 0.064 & 15.1 & -22.1 & 0.08 & 31 & 49 & @xmath16 & 124 & 36.8 pg0953 + 414 & 09 53 48.3 & 41 29 58 & 0.239 & 15.4 & -24.5 & 0.00 & 38 & 36 & @xmath17 & 104 & 31.7 pg1048 + 342 & 10 48 56.1 & 34 15 23 & 0.167 & 17.7 & -21.9 & 0.00 & 30 & 1 & @xmath18 & 186 & 94.7 pg1100 + 772 & 11 00 27.4 & 77 15 09 & 0.313 & 15.9 & -24.7 & 0.06 & 46 & 9 & @xmath19 & 68 & 84.1 pg1202 + 281 & 12 02 08.9 & 28 10 54 & 0.165 & 16.2 & -23.0 & 0.03 & 37 & 9 & @xmath20 & 73 & 219.8 pg1211 + 143 & 12 11 44.8 & 14 19 53 & 0.085 & 14.4 & -23.4 & 0.13 & 52 & 38 & @xmath21 & 162 & 352.2 pg1226 + 023 & 12 26 33.4 & 02 19 42 & 0.158 & 12.8 & -26.3 & 0.00 & 26 & 39 & @xmath22 & 171 & 171.2 pg1229 + 204 & 12 29 33.1 & 20 26 03 & 0.064 & 15.5 & -21.7 & 0.00 & 32 & 33 & @xmath23 & 75 & 291.5 pg1307 + 085 & 13 07 16.2 & 08 35 47 & 0.155 & 15.6 & -23.5 & 0.02 & 35 & 23 & @xmath24 & 138 & 186.5 pg1309 + 355 & 13 09 58.5 & 35 31 15 & 0.184 & 15.5 & -24.0 & 0.00 & 30 & 9 & @xmath25 & 167 & 74.2 pg1322 + 659 & 13 22 08.5 & 65 57 25 & 0.168 & 16.0 & -23.3 & 0.05 & 27 & 8 & @xmath26 & 104 & 277.5 pg1351 + 640 & 13 51 46.3 & 64 00 28 & 0.087 & 14.7 & -23.2 & 0.03 & 40 & 30 & @xmath27 & 109 & 196.5 pg1354 + 213 & 13 54 11.5 & 21 18 30 & 0.300 & 17.3 & -23.2 & 0.04 & 25 & 3 & @xmath28 & 125 & 133.4 pg1402 + 261 & 14 02 59.2 & 26 09 51 & 0.164 & 15.6 & -23.6 & 0.00 & 27 & 6 & @xmath29 & 171 & 334.8 pg1404 + 226 & 14 04 02.5 & 22 38 03 & 0.098 & 16.5 & -21.8 & 0.01 & 27 & 8 & @xmath30 & 77 & 223.0 pg1411 + 442 & 14 11 50.1 & 44 14 12 & 0.089 & 15.0 & -22.9 & 0.00 & 34 & 24 & @xmath31 & 128 & 347.0 pg1415 + 451 & 14 15 04.3 & 45 09 57 & 0.114 & 16.3 & -22.3 & 0.00 & 29 & 3 & @xmath32 & 189 & 95.2 pg1426 + 015 & 14 26 33.8 & 01 30 27 & 0.086 & 15.7 & -22.2 & 0.12 & 33 & 20 & @xmath33 & 106 & 341.4 pg1444 + 407 & 14 44 50.2 & 40 47 37 & 0.267 & 17.4 & -23.0 & 0.00 & 27 & 5 & @xmath34 & 158 & 155.4 pg1512 + 370 & 15 12 46.9 & 37 01 56 & 0.371 & 16.4 & -24.5 & 0.05 & 35 & 10 & @xmath35 & 63 & 284.9 pg1613 + 658 & 16 13 36.3 & 65 50 38 & 0.129 & 14.9 & -23.8 & 0.04 & 42 & 48 & @xmath36 & 101 & 164.2 pg1617 + 175 & 16 17 56.2 & 17 31 34 & 0.114 & 15.5 & -22.8 & 0.15 & 39 & 35 & @xmath37 & 147 & 253.0 pg1700 + 518 & 17 00 13.4 & 51 53 37 & 0.292 & 15.4 & -25.2 & 0.02 & 41 & 39 & @xmath38 & 104 & 183.5 pg1704 + 608 & 17 04 03.5 & 60 48 31 & 0.371 & 15.6 & -25.5 & 0.00 & 39 & 25 & @xmath39 & 114 & 188.2 pg2130 + 099 & 21 30 01.2 & 09 55 01 & 0.061 & 14.7 & -22.4 & 0.17 & 47 & 64 & @xmath40 & 86 & 68.2 cccccccccc pg0026 & 47904830 & 49205045 & 50705100 & * 54085468 * & 55085654 & 58185868 & 69987138 & 73047728 & 77307790 pg0052 & * 48644914 * & 49265118 & 54185488 & 54185488 & 55185698 & 59085978 & 70287108 & 73987724 & 77257800 pg0804 & 46804705 & 47104850 & 48804905 & * 52245264 * & 52665474 & 55985630 & 67466814 & 69407412 & 74727520 pg0844 & 45034535 & 45364690 & * 47104750 * & 50485078 & 50985248 & 54145440 & 67086764 & 68487120 & 72487348 pg0953 & * 52385278 * & 52845450 & 54725506 & 58465892 & 59066086 & 62886320 & & & pg1211 & 45254560 & 46254755 & 47804810 & * 51205160 * & 51805340 & 54905530 & 66706720 & 69007310 & 74807540 pg1226 & 49254945 & 49465130 & * 51355165 * & 54945522 & 55305756 & 58885928 & 70347085 & 72507788 & 77897808 pg1229 & 45004560 & 45604690 & 46944758 & 50525088 & 50945244 & * 54125456 * & 65626654 & 68387148 & 72447320 pg1307 & 48704915 & 49385088 & 51045154 & * 54585488 * & 55085698 & 58645914 & 70507100 & 74007740 & 77507800 pg1351 & 45854645 & 46464781 & 48154850 & * 50485128 * & 51965352 & 55455605 & 66386770 & 68987322 & 73627432 pg1411 & 45704600 & 46504790 & 48054835 & * 51555185 * & 52015382 & 55205555 & 67006760 & 68707360 & 74907560 pg1426 & 45654605 & 46054775 & 47304796 & * 51055145 * & 51705360 & 55155555 & 66806730 & 69307270 & 74907550 pg1613 & & & & * 53005350 * & 53605566 & 57405790 & 70007050 & 70137742 & 77437803 pg1617 & 46504700 & 47454950 & 49304960 & * 52505295 * & 53105550 & 56725722 & 68986968 & 71187498 & 75407580 pg1700 & & & & * 60306080 * & 61206410 & 65206570 & 77608015 & 82458689 & 86908730 pg1704 & 58005865 & 59006025 & * 60406100 * & 64606520 & 65756735 & 70207100 & & & pg2130 & 44604490 & 45334710 & 47104735 & * 50485084 * & 50945240 & 53945440 & 64986598 & 67987118 & 72467318 -1 in -0.7 in -1 in -0.7 in the observations were carried out using the steward observatory ( so ) 2.3 m telescope and the wise observatory ( wo ) 1 m telescope . throughout the years different detectors were used . at so we used the boller & chivens spectrograph with a 300 l / mm grating , a @xmath41-wide slit and a ti ccd . in 1992 september this ccd was replaced with a 1200@xmath42800 loral ccd . at wo the faint object spectroscopic camera ( fosc ; kaspi et al . 1995 ) was used with a 15@xmath43-wide slit and a front - illuminated 1024@xmath421024 ti ccd as a detector . in 1994 february the slit was replaced by a 10@xmath43-wide slit and the detector used was a 1024@xmath421024 back - illuminated tektronix ccd . the typical wavelength coverage , at both observatories , was @xmath140008000 , with a dispersion of @xmath13.8 / pixel , and a spectral resolution of @xmath110 . spectrophotometric calibration for every quasar was accomplished by rotating the spectrograph slit to the appropriate position angle so that a nearby comparison star ( columns [ 11 ] & [ 12 ] in table [ tsample ] ) was observed simultaneously with the object . a wide slit was used to minimize the effects of atmospheric dispersion at the non - parallactic position angle . this technique provides excellent calibration even during poor weather conditions , and accuracies of order 1% 2% can easily be achieved . observations typically consisted of two consecutive exposures of the quasar / star pair . total exposure times were usually 40 minutes at the so 2.3 m telescope and 2 hours at the wo 1 m telescope . the spectroscopic data were reduced using standard iraf routines . the consecutive quasar / star flux ratios were compared to test for systematic errors in the observations and to clean cosmic rays . the ratio usually reproduced to 0.52.5% at all wavelengths and observations with ratios larger than 5% were discarded . we verified that none of our comparison stars are variable to within @xmath12% by means of differential photometry with other stars in each field ( paper iii ) , spectra were calibrated to an absolute flux scale using observations of spectrophotometric standard stars on one or more epochs . the absolute flux calibration has an uncertainty of @xmath110% , which is not shown in the error bars in our light curves . the error bars reflect only the measurement and differential uncertainties , which are of order 1%2% . = 8.5 cm for each quasar we used all the spectra to produce an average spectrum and a root - mean - square ( rms ) spectrum , defined as : @xmath44 ^ 2\right\}^{1/2},\ ] ] where the sum is taken over the @xmath45 spectra , and @xmath46 is the average spectrum ( peterson et al . the mean spectra of all the quasars in our sample are shown in fig . [ spectra1 ] . an example of a typical rms spectrum and comparison with the mean spectrum is shown in fig . [ pg0844spec ] for pg0844 . we used the average and rms spectra to choose line - free spectral bands suitable for setting the continuum underlying the emission lines , and the wavelength limits for integrating the line fluxes . the spectral intervals for the balmer lines and the continuum bands on both sides of each line are given in table [ tabwave ] . the line and continuum fluxes were measured algorithmically for all epochs by calculating the mean flux in the continuum bands , and summing the flux above a straight line in @xmath47 that connects the continuum bands straddling the emission line . this process measures the total emission in each line , i.e. , the flux of the broad component of the line together with its narrow component . the uncertainty in the line flux was estimated by propagating the uncertainty in the setting of the continuum levels , determined from the standard deviation of the mean in the continuum bands . to this we added in quadrature our estimate for the uncertainty in the differential spectrophotometry . for the purpose of time series analysis , we generally chose a region just blue of for the continuum band . in several cases where the blue side of has an inferior light curve ( due to , e.g. , atmospheric absorption bands and/or superposed emission line blends ) we chose continuum bands around . table [ tabwave ] lists in bold font the continuum band chosen for each object . both the continuum bands and the emission lines can be affected by the contribution of variable feii lines . however , correcting for this effect is beyond the scope of the present study . similarly , our balmer line measurements might be contaminated by narrow forbidden emission - lines . though we measure without the [ oiii]@xmath484995 , 5007 lines ( since there is a clear separation between them ) we can not do so for the [ nii]@xmath486548 , 6583 lines blended with the line . however , we note that contamination from narrow lines should not affect our time series analysis since this analysis relies on the variable parts of the lines , and the narrow lines have never been observed to vary in agn over relevant time scales ( e.g. , kaspi et al . 1996b ) . the agn continuum dominates the emission - line flux and host galaxy starlight contribution to the broad - band photometry . to improve the sampling of the continuum light curves , we therefore merged the photometric @xmath8 band light curves of paper iii into the spectrophotometric continuum light curves ( see paper i ) . the photometric campaign lasted 7 years and we extended the observations of the present 28 objects for six more months to better overlap with the spectrophotometric campaign . those last photometric observations were reduced and added to the light curves as explained in paper iii . the photometric and spectrophotometric light curves of each object were intercalibrated by comparing all pairs of spectrophotometric and photometric observations separated in time by less than a few days ( typically @xmath15 days , and in some cases up to @xmath110 days ) . this produced 10 to 15 pairs of points per object . a linear least - squares fit between the spectrophotometric continuum fluxes and the photometric fluxes ( the magnitudes of paper iii translated to fluxes ) was used to merge the two light curves . for each object , the total number of photometric and spectrophotometric observations is given in table [ tsample ] , columns ( 8) & ( 9 ) ( the sum of the two columns gives the total number of points in the continuum light curve ) . note that for particular emission - lines in a few objects , some data points are missing because of cosmic - rays contaminations , insufficient wavelength coverage , or low s / n . light curves for the 17 objects that have more than 20 spectrophotometric data points are presented in fig . [ flcs ] with the data listed in table [ contab ] and table [ linetab]shai / pg/ ] . each of the remaining 11 objects in our sample have less than 10 spectrophotometric observations over the 7.5 year period . such rare sampling is insufficient to determine the line - to - continuum lag and these objects will not be discussed further in this paper . -1 in -0.7 in -1 in -0.5 in -1 in -0.7 in the continuum variability properties of our quasar sample are extensively discussed in paper iii . all 17 objects underwent optical continuum variability ( at 5100 rest wavelength ) of 25%150% , measured relative to the minimum flux . in order to compare the line variability to the continuum variability in each quasar , we define an intrinsic normalized variability measure , @xmath49 , where @xmath50 and @xmath51 are the average and the rms of the flux in a given light curve , respectively , and @xmath52 is the mean uncertainty of all data points in that light curve . this measure is less sensitive to noise and outliers than the minimum - to - maximum flux ratio . the two cases where @xmath53 is slightly negative ( in pg1426 and pg1617 ) are consistent with no variability . table [ variability ] summarizes @xmath54 for the 5100 rest frame continua band and all of the available emission - line light curves . in general we find that all of the balmer emission lines in a specific object have about the same level of variability . in all objects , the variability of the emission lines is smaller than the continuum variability . in the few cases where those trends do not hold , it appears that they may be masked by noise ( e.g. , most cases in which the emission line varies with a larger amplitude than the continuum involve , which has lower s / n than and ) . the main goal of this project is to determine the time lag between the continuum and the balmer - line flux variations in high - luminosity agn . to quantify this lag , we use two methods for correlating the line and continuum light curves . the first method is the interpolated cross - correlation function ( iccf ) of gaskell & sparke ( 1986 ) and gaskell & peterson ( 1987 ) , as implemented by white & peterson ( 1994 ; see also a review by gaskell 1994 ) . the second method is the @xmath55-transformed discrete correlation function ( zdcf ) of alexander ( 1997 ) which is an improvement on the discrete correlation function ( dcf ; edelson & krolik 1988 ) . the zdcf applies fisher s @xmath55 transformation to the correlation coefficients , and uses equal population bins rather than the equal time bins used in the dcf . the two independent methods are in excellent agreement for these data and we will use only the iccf results in the following analyses . the uncertainties in the cross correlation lag determination can be estimated using the model - independent fr / rss monte carlo method described by peterson et al . ( 1998b ) . in this method , each monte carlo simulation is composed of two parts : the first is a `` random subset selection '' ( rss ) procedure which consists of randomly drawing , with replacement , from a light curve of @xmath45 points a new sample of @xmath45 points . after the @xmath45 points are selected , the redundant selections are removed from the sample such that the temporal order of the remaining points is preserved . this procedure reduces the number of points in each light curve by a factor of @xmath56 and accounts for the effect that individual data points have on the cross - correlation . the second part is `` flux randomization '' ( fr ) in which the observed fluxes are altered by random gaussian deviates scaled to the uncertainty ascribed to each point . this procedure simulates the effect of measurement uncertainties . peterson et al . ( 1998b ) demonstrate that under a wide variety of realistic conditions , the combined fr / rss procedure yields conservative uncertainties compared to the true situation . cc 2448461.5 & @xmath57 2448490.6 & @xmath58 2448520.3 & @xmath59 2448545.8 & @xmath60 2448769.6 & @xmath61 2448836.5 & @xmath62 2448869.4 & @xmath63 2448890.4 & @xmath64 2448925.2 & @xmath65 2448949.4 & @xmath66 cc 2448836.5 & @xmath67 2448890.4 & @xmath68 2448949.4 & @xmath69 2448988.6 & @xmath70 2449218.6 & @xmath71 2449220.5 & @xmath72 2449248.4 & @xmath73 2449250.5 & @xmath74 2449278.6 & @xmath75 2449281.7 & @xmath76 crrrr pg0026 & 14.7 & 9.3 & 8.0 & 8.4 pg0052 & 17.6 & 10.1 & 11.4 & 20.2 pg0804 & 18.2 & 5.8 & 6.2 & 8.7 pg0844 & 10.1 & 6.7 & 9.7 & 10.0 pg0953 & 13.3 & & 5.5 & 6.2 pg1211 & 16.2 & 10.3 & 12.2 & 12.8 pg1226 & 11.9 & 7.7 & 7.8 & 12.3 pg1229 & 10.9 & 8.1 & 12.5 & 17.9 pg1307 & 9.8 & 7.9 & 12.8 & 14.3 pg1351 & 9.8 & 4.0 & 4.7 & pg1411 & 8.5 & 3.4 & 5.1 & 2.8 pg1426 & 15.3 & 7.0 & 8.8 & pg1613 & 12.4 & 5.6 & 6.1 & pg1617 & 17.3 & 7.4 & 11.5 & pg1700 & 6.8 & & 3.2 & pg1704 & 14.5 & & 10.7 & 18.0 pg2130 & 9.3 & 6.2 & 8.9 & 8.9 ccllccllc pg0026 & & @xmath77 & @xmath78 & 0.81 & 55 & @xmath79 & @xmath80 & @xmath81 pg0026 & & @xmath82 & @xmath83 & 0.84 & 53 & @xmath84 & @xmath85 & @xmath86 pg0052 & & @xmath87 & @xmath88 & 0.68 & 53 & @xmath89 & @xmath90 & @xmath91 pg0052 & & @xmath92 & @xmath93 & 0.75 & 56 & @xmath94 & @xmath95 & @xmath96 pg0052 & & @xmath97 & @xmath98 & 0.74 & 56 & @xmath99 & @xmath100 & @xmath101 pg0804 & & @xmath102 & @xmath103 & 0.83 & 70 & @xmath104 & @xmath105 & @xmath106 pg0804 & & @xmath107 & @xmath108 & 0.85 & 70 & @xmath109 & @xmath110 & @xmath111 pg0804 & & @xmath112 & @xmath113 & 0.74 & 67 & @xmath114 & @xmath115 & @xmath116 pg0844 & & @xmath117 & @xmath118 & 0.74 & 46 & @xmath119 & @xmath120 & @xmath121 pg0844 & & @xmath122 & @xmath123 & 0.79 & 48 & @xmath124 & @xmath125 & @xmath126 pg0844 & & @xmath127 & @xmath128 & 0.58 & 45 & @xmath129 & @xmath130 & @xmath131 pg0953 & & @xmath132 & @xmath133 & 0.72 & 35 & @xmath134 & @xmath135 & @xmath136 pg0953 & & @xmath137 & @xmath138 & 0.54 & 34 & @xmath139 & @xmath140 & @xmath141 pg1211 & & @xmath142 & @xmath143 & 0.87 & 37 & @xmath144 & @xmath145 & @xmath146 pg1211 & & @xmath147 & @xmath148 & 0.87 & 36 & @xmath149 & @xmath150 & @xmath151 pg1211 & & @xmath152 & @xmath153 & 0.86 & 34 & @xmath154 & @xmath155 & @xmath156 pg1226 & & @xmath157 & @xmath158 & 0.71 & 34 & @xmath159 & @xmath160 & @xmath161 pg1226 & & @xmath162 & @xmath163 & 0.78 & 39 & @xmath164 & @xmath165 & @xmath166 pg1226 & & @xmath167 & @xmath168 & 0.79 & 39 & @xmath169 & @xmath170 & @xmath171 pg1229 & & @xmath172 & @xmath173 & 0.78 & 32 & @xmath174 & @xmath175 & @xmath176 pg1229 & & @xmath177 & @xmath178 & 0.80 & 33 & @xmath179 & @xmath180 & @xmath181 pg1229 & & @xmath182 & @xmath183 & 0.81 & 29 & @xmath184 & @xmath185 & @xmath186 pg1307 & & @xmath187 & @xmath188 & 0.67 & 20 & @xmath189 & @xmath190 & @xmath191 pg1307 & & @xmath192 & @xmath193 & 0.59 & 23 & @xmath194 & @xmath195 & @xmath196 pg1307 & & @xmath197 & @xmath198 & 0.64 & 23 & @xmath199 & @xmath200 & @xmath201 pg1351 & & @xmath202 & @xmath203 & 0.49 & 29 & @xmath204 & @xmath205 & @xmath206 pg1411 & & @xmath207 & @xmath208 & 0.87 & 24 & @xmath209 & @xmath210 & @xmath211 pg1411 & & @xmath212 & @xmath213 & 0.64 & 24 & @xmath214 & @xmath215 & @xmath216 pg1426 & & @xmath217 & @xmath218 & 0.87 & 20 & @xmath219 & @xmath220 & @xmath221 pg1426 & & @xmath222 & @xmath223 & 0.83 & 20 & @xmath224 & @xmath225 & @xmath226 pg1613 & & @xmath227 & @xmath228 & 0.43 & 44 & @xmath229 & @xmath230 & @xmath231pg1613 & & @xmath232 & @xmath233 & 0.43 & 48 & @xmath234 & @xmath235 & @xmath236 pg1617 & & @xmath237 & @xmath238 & 0.80 & 32 & @xmath239 & @xmath240 & @xmath241 pg1617 & & @xmath242 & @xmath243 & 0.75 & 34 & @xmath244 & @xmath245 & @xmath246 pg1617 & & @xmath247 & @xmath248 & 0.67 & 32 & @xmath249 & @xmath250 & @xmath251 pg1700 & & @xmath252 & @xmath253 & 0.54 & 37 & @xmath254 & @xmath255 & @xmath256 pg1704 & & @xmath257 & @xmath258 & 0.43 & 24 & @xmath259 & @xmath260 & @xmath261 pg2130 & & @xmath262 & @xmath263 & 0.59 & 64 & @xmath264 & @xmath265 & @xmath266 pg2130 & & @xmath267 & @xmath268 & 0.66 & 64 & @xmath269 & @xmath270 & @xmath271 pg2130 & & @xmath272 & @xmath273 & 0.72 & 54 & @xmath274 & @xmath275 & @xmath276 all emission - line light curves presented in fig . [ flcs ] were cross correlated with their corresponding continuum light curves . the results of the cross - correlation analysis are presented in table [ tccfs ] . the object name is given in column ( 1 ) and the specific balmer line is listed in column ( 2 ) . for the iccf we list the centroid time - lag , @xmath277 , ( computed from all points within 80% of the peak value , @xmath278 ) in column ( 3 ) . we define the iccf peak as the point of maximum correlation and list the peak position , @xmath279 , in column ( 4 ) , and the peak value , @xmath278 , in column ( 5 ) . in column ( 6 ) we list the total number of points in the emission - line light curve ( the total number of points in the continuum light curve is given in table [ tsample ] columns [ 8 ] & [ 9 ] ) . the uncertainties given for @xmath277 and @xmath279 were computed with the fr / rss method using @xmath280 monte carlo realizations which were used to build up a cross - correlation peak distribution ( ccpd ; maoz & netzer 1989 ) . the range of uncertainties contains 68% of the monte carlo realizations in the ccpd and thus would correspond to 1@xmath51 uncertainties for a normal distribution . in addition to the cross - correlation results , we also list in table [ tccfs ] velocity measurements and the average of the observed flux for each line . column ( 7 ) lists the mean full - width at half maximum ( fwhm ) of the emission - line , column ( 8) lists the line fwhm measured from the rms spectrum ( see [ secmass ] below ) , and column ( 9 ) lists the average of the observed line flux and its rms ( calculated from the light curves of table [ linetab ] ) . [ fccfs ] presents the cross - correlation functions ( ccfs ) for the 40 out of 46 emission - line / continua pairs having peak correlation coefficients above 0.4 . all 40 ccfs indicate a positive time lag of the balmer lines with respect to the optical continuum . the time lags are of order a few weeks to a few months , and the ccf peaks are highly significant for most lines . in a few cases , @xmath277 and @xmath279 are consistent with a zero time lag and we attribute this to the fact that these emission - line light curves are noisy , and to the conservative error estimate of the fr / rss method . choosing our best lag determination ( see below ) , i.e. , avoiding noisy results obtained from weak lines ( generally ) , we conclude that a time lag has been detected in one or more of the balmer lines for _ all _ 17 quasars . when comparing the time - lags for pg0804 and pg0953 obtained in paper ii to the ones deduced here , we find an increase of about 65% for both objects , although the results are consistent at the 2@xmath51 level . the luminosities of both objects also increased during the second half of the monitoring period . changes in the time lag between different observing seasons is seen in lower - luminosity agn ( e.g. , ngc4151 and ngc5548 ) and is thought to be a real effect , rather than a result of measurement error . for instance , the blr may become larger when the object becomes brighter . converting the observable balmer emission - line time lags , line widths , and continuum fluxes into interesting physical parameters , namely , blr sizes , and agn central masses and luminosities , is not straightforward . we therefore discuss each in some detail below . we analyze our data together with data for other agn . wandel et al . ( 1999 ) have uniformly analyzed the reverberation mapping data of 17 seyfert 1s , and deduced time lags using the same techniques described in the previous section . combining their results with the present ones , we obtain , for the first time , reliable size mass luminosity relations for 34 agn spanning over 4 orders of magnitudes in continuum luminosity . our analysis is similar to that of wandel et al . ( 1999 ) , and the few significant differences are noted below . we apply below linear regression analysis to the data . unless otherwise noted , the method is the one described by press et al . ( 1992 ) , which is based on an iterative process to minimize @xmath281 , taking into account the uncertainties in both coordinates . in some of our data , the uncertainties are asymmetric ( e.g. , as a result of the fr / rss method and the use of the ccpd ) , and since the fitting method does not account for this , we use the mean of the positive and negative uncertainty estimates in a given coordinate for each data point . we find that similar results are obtained when using the pearson linear correlation and the spearman rank - order correlation . we present below only pearson coefficients . -1,9 in -0.8 in caution must be exercised when using time lags to estimate blr sizes because there are various assumptions and complications involved : 1 . the ccf is sensitive to the characteristics of the continuum variation , since it is a convolution of the transfer function and the continuum autocorrelation function ( penston 1991 ; peterson 1993 ) . hence , there is a dependence of the peak and centroid of the ccf on the nature of the continuum variability . an illustration of this dependence is given by netzer & maoz ( 1990 ) . a real change in blr size and geometry , over the 7 year period , is not out of the question . using the time - lags measured here , and the observed fwhms , we deduce a dynamical ( crossing ) time of the order of 30 years for the more luminous objects and about a year for the lowest luminosity seyfert galaxies . the measured time lag can be affected by the nonlinear response of some emission - line intensities to continuum variations . nonlinear responses are suggested by various studies ( obrien , goad , & gondhalekar , 1995 ; shields , ferland , & peterson , 1995 ) and discussed by kaspi & netzer ( 1999 ) . 4 . the ionizing continuum variations may behave differently from the observed optical continuum . due to anisotropic continuum emission , the emission line gas may be illuminated by an ionizing continuum that is poorly represented by the observed optical continuum ( e.g. netzer 1987 ) . furthermore , variable beaming or obscuration may be different along the different lines of sight from the continuum source to the observer and to the gas . most studies show the continuum varies in phase with little or no time delay ( albeit , with smaller amplitude at longer wavelengths ) over a wide wavelength range . there are , however , some exceptions ( e.g. , peterson et al . 1998b ) and it is clear that a full understanding of the continuum emission processes is yet to be obtained . an extreme example is shown by nandra et al . ( 1998 ) who found that the x - ray and uv continuum are not strongly correlated at zero time lag in ngc 7469 , with some x - ray emission peaks lagging behind the uv peaks by @xmath14 days . recently , maoz , edelson , & nandra ( 2000 ) have demonstrated even weaker correlation at zero lag between x - ray and optical variations in ngc 3516 . 5 . for a given object , different lines from different ions have different time lags . thus , considering the entire sample , the same emission lines should be used in all objects . balmer lines are useful since their lag has been measured in most agn where reverberation mapping has been attempted . however , as argued by kaspi & netzer ( 1999 ) , the theoretical modeling of these lines is very uncertain and they are useful mostly because they provide a uniform base for comparing low luminosity seyferts with high luminosity quasars . . there may be flux contributions by other emission lines blended with the balmer lines . these emission lines may have a different time lag and affect the measured emission line lag . an example of this potential situation is the variable feii emission lines near . a possible manifestation of some of these difficulties is the fact that for some objects , different monitoring campaigns result in different time lags . the effect was first noted in the time lags found by peterson et al . ( 1991 ; @xmath17 days ) and netzer et al . ( 1990 ; @xmath120 days ) for the seyfert galaxy ngc 5548 . other examples are the longer time lag deduced for pg0804 and pg0953 in this paper , compared with paper ii ( see above ) , and the different time lags in each observing season for ngc 5548 ( peterson et al . 1999 ) . bearing in mind the above points , we convert the time lags ( deduced from the iccf centroid , @xmath277 from table [ tccfs ] ) of the balmer emission lines directly into blr sizes ( hereafter ) after applying a cosmological @xmath282 factor . we find that time lags derived from multiple balmer lines are consistent with each other for individual quasars . the derived linear relations are : @xmath283 and @xmath284 i.e. , both are consistent with a slope of 1.0 and a zero intercept . in view of this , we adopt the mean time lag measured for and as the best estimate of . the relatively low equivalent width of makes it more difficult to measure and we do not include it in the estimate of blr size . in several objects only one of or is available and we thus use a single line . for the seyfert 1s we use the time lags from wandel et al . ( 1999 ) and correct them by the @xmath282 factor . we have used the same method for calculating the time lags as that used by wandel et al . values for for all objects are listed in table [ trlm ] , column ( 2 ) . lcccc 3c120 & @xmath285 & @xmath286 & @xmath287 & @xmath288 3c390.3 & @xmath289 & @xmath290 & @xmath291 & @xmath292 akn120 & @xmath293 & @xmath294 & @xmath295 & @xmath296 f9 & @xmath297 & @xmath298 & @xmath299 & @xmath300 ic4329a & @xmath301 & @xmath302 & @xmath303 & @xmath304 mrk79 & @xmath305 & @xmath306 & @xmath307 & @xmath308 mrk110 & @xmath309 & @xmath310 & @xmath311 & @xmath312 mrk335 & @xmath313 & @xmath314 & @xmath315 & @xmath316 mrk509 & @xmath317 & @xmath318 & @xmath319 & @xmath320 mrk590 & @xmath321 & @xmath322 & @xmath323 & @xmath324 mrk817 & @xmath325 & @xmath326 & @xmath327 & @xmath328 ngc3227 & @xmath329 & @xmath330 & @xmath331 & @xmath332 ngc3783 & @xmath333 & @xmath334 & @xmath335 & @xmath336 ngc4051 & @xmath337 & @xmath338 & @xmath339 & @xmath340 ngc4151 & @xmath341 & @xmath342 & @xmath343 & @xmath344 ngc5548 & @xmath345 & @xmath346 & @xmath347 & @xmath348 ngc7469 & @xmath349 & @xmath350 & @xmath351 & @xmath352 pg0026 & @xmath353 & @xmath354 & @xmath355 & @xmath356 pg0052 & @xmath357 & @xmath358 & @xmath359 & @xmath360 pg0804 & @xmath361 & @xmath362 & @xmath363 & @xmath364 pg0844 & @xmath365 & @xmath366 & @xmath367 & @xmath368 pg0953 & @xmath369 & @xmath370 & @xmath371 & @xmath372 pg1211 & @xmath373 & @xmath374 & @xmath375 & @xmath376 pg1226 & @xmath377 & @xmath378 & @xmath379 & @xmath380 pg1229 & @xmath381 & @xmath382 & @xmath383 & @xmath384 pg1307 & @xmath385 & @xmath386 & @xmath387 & @xmath388 pg1351 & @xmath389 & @xmath390 & @xmath391 & @xmath392 pg1411 & @xmath393 & @xmath394 & @xmath395 & @xmath396 pg1426 & @xmath397 & @xmath398 & @xmath399 & @xmath400 pg1613 & @xmath401 & @xmath402 & @xmath403 & @xmath404 pg1617 & @xmath405 & @xmath406 & @xmath407 & @xmath408 pg1700 & @xmath409 & @xmath410 & @xmath411 & @xmath412 pg1704 & @xmath413 & @xmath414 & @xmath415 & @xmath416 pg2130 & @xmath417 & @xmath418 & @xmath419 & @xmath420 = 8.5 cm the exact shape of the spectral energy distribution ( sed ) of most objects in our sample is unknown . furthermore , much of this luminosity is emitted in the unobservable far uv and there are still unsolved fundamental issues concerning the shape of the continuum ( e.g. , zheng et al . 1997 , laor et al . 1997 , and references therein ) . even in those cases where multiwavelength data exist , the data are generally obtained at different epochs , and variability may render a simple integration over a wide wavelength range inappropriate . ( see also a discussion by dumont , collin - souffrin , & nazarova ; 1998 , that demonstrates how poorly the sed is determined for even one of the best - studied agn , ngc 5548 . ) intrinsic reddening and possible orientation effects complicate the situation and it is thus unclear what fraction of the total luminosity we are sampling . an additional complication is the contribution of the host galaxy to the luminosity of the nucleus . while in high - luminosity agn this contribution is negligible , this is not the case in low - luminosity objects . determining the host galaxy flux within a given spectrograph aperture is a complicated task , attempted only in a few cases ( e.g. , peterson et al . the implementation of such corrections is beyond the scope of this paper . following wandel et al . ( 1999 ) , we use @xmath421 at 5100 ( rest frame ) as our luminosity measure . this luminosity is calculated assuming zero cosmological constant , deceleration parameter @xmath422 , and hubble constant @xmath423kms@xmath424mpc@xmath424 ( note that these values are different from the ones used by wandel et al . this monochromatic luminosity can be translated into an integrated 0.11 @xmath425 m luminosity , assuming a power - law continuum of the form @xmath426 , as in papers i & ii , by applying a correction factor of 3.1 . for the 5100 point , we use the continuum measurement on the red side of ( see wavelength ranges in column ( 7 ) of table [ tabwave ] ) . the uncertainty in this value is taken to be only due to the variation range of each object , and is represented by the rms of the light curve . the observed flux has been corrected for galactic extinction using the @xmath427 values from ned ( see table [ tsample ] column [ 7 ] ) and a standard extinction curve ( savage & mathis 1979 ) . monochromatic luminosities , @xmath428(5100 ) , are listed in table [ trlm ] , column ( 3 ) . we have also tried an alternative process to determine the optical luminosity . using line - free bands in the optical spectra , we have fitted a continuum to the mean and rms spectra of all objects . we integrated those continua over the rest frame wavelength of 45006000 to obtain the mean luminosity and its range . we have found the results to be very similar to those obtained by using the monochromatic luminosity . we have also examined the consequences of using the luminosity ( rather than the optical luminosity ) to represent the ionizing luminosity of each agn . here too , we have found the results presented below to be mostly unchanged by this choice . below we use only @xmath429 ) as our luminosity measure . to estimate the central masses of the quasars we need to assume gravitationally dominated motions of the blr clouds ( @xmath430 ) . this assumption is presented in gaskell ( 1988 ) for the seyfert 1 galaxy ngc4151 and is also discussed and justified by wandel et al . ( 1999 ) and peterson & wandel ( 1999 ) for the seyfert 1 galaxy ngc 5548 ( though see gaskell 1996 for a more general approach ) . there are , however , several complications regarding the velocity field . although it is generally believed that the line widths represent a keplerian velocity , it is not clear how to convert the observed profile into a velocity measure , since the observed profiles are different from the ones expected from simple keplerian orbits of arbitrary inclinations . in the absence of such information , it is common to use the rest - frame fwhm of the emission line . because the broad emission lines of agn are composed of a narrow component superposed on a broader components , a unique fwhm determination is not straightforward . in particular , deblending the components is complicated since the line profiles can not be represented by simple analytical functions . also , the line profiles and widths are variable as is evident , for example , in gilbert et al . ( 1999 ) and from the spectra obtained in this study . these variations can be due to a real change in the blr velocity field , or other reasons , such as blending with weak lines , or observational effects ( e.g. , inaccurate wavelength calibration , or changes in the seeing ) . = 8.5 cm we use two different methods to estimate a typical velocity . in the first method , for each quasar we measure the fwhm of each line in all the spectra and calculate the mean fwhm , @xmath431(mean ) , and its uncertainty ( rms ) . this empirical way of estimating the scatter accounts for all the uncertainties involved , such as spectral resolution , pixelation of the line profile , and uncertainties in continuum determination . the values are listed in table [ tccfs ] , column ( 7 ) . the second method , proposed by peterson et al . ( 1998a ) , uses the rms spectrum to compute the fwhm of the lines . in principle , constant features in the mean spectrum ( such as narrow forbidden emission lines , narrow components of the permitted emission lines , galactic absorptions , and constant continuum and broad - line features ) are excluded in this method . the fwhm measures only the part of the line that varied and thus corresponds to the measured from the reverberation mapping . for the rms spectrum of each object we have listed the emission lines fwhm , @xmath431(rms ) , in column ( 8) of table [ tccfs ] . there is no simple way to determine the uncertainty of @xmath431(rms ) , and instead we have used the relative uncertainty in @xmath431(mean ) . like the time lags derived using and , the fwhm of all lines for a given object are consistent with each other . thus , for the pg quasars we use the average fwhm of and . for the seyfert 1 galaxies we have used the mean and rms fwhm quoted in wandel et al . the fwhm from the rms spectrum versus the mean fwhm from all spectra is plotted in fig . [ ffvsf ] . we find the correlation to be highly significant with correlation coefficient of 0.888 . a linear fit gives the relation @xmath432 which is plotted as the solid line in fig . [ ffvsf ] . the two velocity estimates are similar and peterson et al.s ( 1998a ) preference for the fwhm determined from the rms spectra is not empirically justified . however , in the subsequent analysis we will use both estimates of the broad - line fwhm for each object . to determine @xmath433 we correct @xmath431 by a factor of @xmath434 , to account for velocities in three dimensions and for using half of the fwhm . the virial `` reverberation '' mass is then : @xmath435 the results of this calculation are listed in columns ( 4 ) and ( 5 ) of table [ trlm ] for @xmath431(mean ) and @xmath431(rms ) , respectively . the luminosity relation is presented in fig . [ frvsl ] . the correlation coefficient is 0.827 , and its significance level is 1.7@xmath436 . a linear fit to the points gives @xmath437 ( solid line plotted in fig . [ frvsl ] ) . considering the seyfert nuclei ( @xmath438(5100 )@xmath439 ) , or the pg quasars alone , we find only marginally significant correlations , probably because of the narrow luminosity ranges . a significant correlation emerges only when using the whole luminosity range . the present result is remarkable for two reasons . first , earlier studies of this kind found a smaller power - law index in the blr size agn luminosity relation ( closer to 0.5 , e.g. , koratkar & gaskell 1991b ; wandel et al . a line with this slope was fit to the data and is shown as a dashed line in fig . [ frvsl ] . our result using the combined sample is not consistent with previous results . second , under the assumptions that the shape of the ionizing continuum in agn is independent of @xmath440 , and that all agn are characterized by the same ionization parameter and blr density ( as suggested by the similar line ratios in low- and high - luminosity sources ) , one expects @xmath441 . the theoretical prediction is based on the assumption that the gas distribution , and hence the mean blr size , scales with the strength of the radiation field . our present result suggests that those assumptions should be re - examined . this is also implied from recent models ( e.g. , kaspi & netzer 1999 and references therein ) which show a wide distribution of properties ( such as blr density , column density , and ionization parameter ) across the blr of a single active nucleus . therefore , a range of properties may also exist among different agn , and the above assumptions of uniform ionization parameter and blr density for all agn is likely incorrect . if an effective ionization parameter can be defined , our result suggests that it may be a decreasing function of luminosity . = 8.5 cm our sample allows us to to re - address the issue of the velocity luminosity relation in agn . shuder ( 1984 ) noted that the fwhm of the balmer lines increases with luminosity for a sample of 25 agn . wandel & yahil ( 1985 ) found a correlation coefficient of 0.5 between the full width at zero intensity ( fw0i ) and the 4000 continuum luminosity for a literature compilation of 94 agn . joly et al . ( 1985 ; and references therein ) also reported a weak correlation between the fw0i of and the optical luminosity in a similar collection of objects . boroson & green ( 1992 ) note the fw0i of is quite sensitive to noise and depends strongly on the quality of the feii subtraction . these authors found an anticorrelation coefficient of @xmath60.275 between the fwhm and absolute @xmath442 magnitude , significant at the 99% confidence level , in a sample of 87 pg quasars . in a recent work stirpe , robinson , & axon ( 1999 ) measured velocities for 126 agn and found them to weakly correlate with the luminosity . = 8.5 cm in addressing the issue of the velocity luminosity relation we note that our sample is _ not _ complete , especially with the inclusion of the heterogeneous seyfert sample . the results must therefore be treated with caution . the fwhms from the mean spectrum are plotted against the luminosity in the bottom panel of fig . [ ffvsllog ] . there is no significant correlation between the variables . however , if we omit the data point for the narrow line seyfert 1 , ngc 4051 , which clearly deviates from the other points and highly influences the fit , we do obtain a marginally significant correlation . the correlation coefficient is @xmath60.386 , and its significance level is 2.7@xmath443 . a linear fit gives @xmath444 fwhms from the rms spectrum versus luminosity are plotted in the top panel of fig . [ ffvsllog ] . the correlation coefficient is @xmath60.540 has a significance level of 1.2@xmath445 and the best linear fit to the data is @xmath446 thus , we find significant anticorrelations , in our incomplete sample , between the fwhm of the balmer lines and the luminosity of the objects such that @xmath447 . our result indicates a stronger anticorrelation coefficient when @xmath448 is used . while the above studies used only one epoch for each quasar in their measurements , our study uses the means of the luminosities and velocities for each object over 7.5 years , and averages both and . the pg quasars @xmath431 found in this study agrees with the ones found by boroson & green s ( 1992 ) to within @xmath110% , except for two objects ( pg1351 and pg1704 ) for which we measure significantly narrower lines . if we exclude those two objects , the anticorrelation becomes weaker and its significance decreases . as the correlation we find is opposite to those found by previous studies , the issue of the velocity luminosity relation needs further investigation , which is beyond the scope of this paper . our two velocity estimates produce two mass estimates for each object ( see table [ trlm ] and [ secmass ] ) . our mass estimates based on the determination of the fwhm from the mean spectra are plotted versus luminosity in the bottom panel of fig . [ fmvsl ] . the correlation coefficient between these two parameters is 0.646 and has a significance level of 3.7@xmath449 . a linear fit gives @xmath450 and is plotted as a solid line in the diagram . the mass estimates based on the determination of the fwhm measured from the rms spectra are plotted in the top panel of fig . [ fmvsl ] . we find correlation coefficient of 0.473 with a significance level of 4.7@xmath451 . a linear fit to this relation gives @xmath452 and is also plotted as a solid line . the results of the above two methods are not consistent . moreover , while it is arguable that using the rms spectra to determine @xmath433 is a better method ( see [ secmass ] ) , the mass luminosity correlation based on this measure is less significant . this can perhaps be attributed to the fact that the line fluxes in the rms spectra are weaker and hence the uncertainty in the corresponding fwhm larger . l relation does not agree with the one found by koratkar & gaskell ( 1991b ) of @xmath453 , nor does it agree with the one found by wandel et al . ( 1999 ) of @xmath454 . however , when wandel et al . ( 1999 ) use an unweighted linear fit they find a slope of 0.54 which is in good agreement with our result . we have used a linear regression analysis which takes into account the uncertainties in both coordinates ( see [ sml ] ) . the fact that the scatter in the mass luminosity relation is larger than that of the size luminosity and velocity luminosity relations may indicate that luminosity , rather than mass , is the variable that mainly determines the blr size . in fact , from the individual size and velocity relations we expect : @xmath455 , i.e. , a weak dependence of mass on luminosity , with all agn having similar masses . in practice , we have found a somewhat stronger dependence , @xmath456 , but with a large scatter . l relation can be expressed in terms of the eddington luminosity , @xmath457 . roughly estimating the bolometric luminosity as @xmath458(5100 ) , we obtain an eddington ratio of @xmath459 the eddington limit , based on this rough estimate for @xmath460 , is plotted as a dashed line in fig . [ fmvsl ] . some of the quasars in our sample appear to be approximately at or beyond the eddington limit . according to fig . [ fmvsl ] , the eddington limit traces an envelope in the mass luminosity plane ( if we ignore the two lower right hand points , pg1700 and pg1704 , which have very large errors ) . this depends , somewhat , on the factor chosen to estimate the bolometric correction , i.e. the rather uncertain slope of the unobserved uv continuum . however , if our chosen bolometric correction is realistic , we may be seeing a direct indication that agn energy is generated by gas accretion . wandel ( 1999 ) reviews three classes of agn mass estimation methods . while the x - ray variability method ( using the shortest time scale for global luminosity variations as the light travel time across the schwarzschild radius ) and the accretion disk modeling method ( deriving an accretion disk model that best fits the observed agn continuum ) suggest that the eddington ratio increases with luminosity , wandel ( 1999 ) notes that the kinematic methods ( such as reverberation mapping ) have yet to show a similar trend . our reverberation mapping result ( equation [ eqeddr ] ) indicates for the first time that the eddington ratio increases with luminosity . models that suggest the bulk of the luminosity is due to energy release via mechanisms which radiate up to a set fraction of the eddington luminosity are not consistent with the derived mass luminosity relation . one such model is a geometrically thin , optically thick , accretion disk that requires @xmath461 to be self - consistent ( laor & netzer 1989 ) . our finding suggests that the mass accretion rate grows with luminosity much faster than the central mass , which would mean very different disk properties in low- and high - luminosity sources . our luminosity determination is based on the monochromatic flux at 5100 . in the thin disk model , this monochromatic flux may represents a different fraction of the object s bolometric luminosity in agn of different masses . thus , it is not clear that the thin accretion disk model can be ruled out by the new results . spectrophotometric monitoring of a large , optically - selected quasar sample has shown clear correlations and well - defined time lags between the optical continuum and the balmer - line light curves . while the seyfert 1 galaxies that have been studied in this manner all have optical luminosities @xmath462 1.5@xmath4210@xmath463 ergs s@xmath424 , the new sample allows us to measure time lags in agn with luminosities up to @xmath464 ergs s@xmath424 . our work increases the available luminosity range for studying the size luminosity relations in agn by two orders of magnitude and doubles the number of objects suitable for these studies . we have combined our results for 17 quasars with data for 17 seyfert 1 galaxies having reliable time lag measurements , and derived uniform estimates of blr size , central masses , and luminosities for the combined sample . our main finding is that the blr size scales with the 5100 luminosity as @xmath465 . this is significantly different from wandel et al.s ( 1999 ) analysis and is also in contradiction with simple theoretical expectations , both suggesting @xmath466 . we have also found that the velocity field of the blr scales inversely with the luminosity , @xmath467 . combining the measured with the observed fwhms , we have obtained a mass luminosity relation for agn , @xmath468 , which , however , has a large intrinsic scatter . l correlations are based on two different estimates ( mean and rms ) of the fwhm of the balmer lines and are not consistent with each other , despite the fact that the two measured values for the fwhm are generally consistent . empirically , at least , it is not obvious which method of fwhm measurement is preferable . our results show the usefulness of long - term monitoring of high - luminosity agn . there is a need to expand the luminosity range to include the highest luminosity quasars and this will require some 510 years of observations . follow - up studies are also needed for some of the results obtained here , such as better determinations of the gas distribution in the blr and the exact sed of the quasars . future work using this sample will include the study of time - variable line profiles , lags between continuum bands , the intrinsic baldwin relation , and more . we are grateful to brad peterson for supplying us with the seyfert 1 data and for many enlightening comments and discussions , and to the referee , martin gaskell , for his constructive advice . ari laor is acknowledged for very useful discussions . we thank john dann and the wo staff for their expert assistance with the observations throughout the years . research at the wo is supported by grants from the israel science foundation . monitoring of pg quasars at so was supported by nasa grant nag 5 - 1630 . h.n . & s.k . acknowledge financial support by the the jake adler chair of extragalactic astronomy . s.k . acknowledges financial support by the colton scholarships .

thanks to the availability of high - resolution high - sensitivity telescopes such as the very large array , the _ hubble space telescope _ , and the _ chandra x - ray observatory _ , there is now a wealth of observational data on relativistic jets from active galactic nuclei ( agn ) as well as galactic sources such as black - hole x - ray binaries . since the jet speeds can not be constrained well from observations , but are generally believed to be relativistic , physical quantities inferred from observables are commonly expressed in terms of the unknown beaming parameters : the bulk lorentz factor and the line - of - sight angle , usually in their combination as relativistic doppler factor . this paper aims to resolve the discrepancies existing in the literature about such `` de - beaming '' of derived quantities , in particular regarding the minimum - energy magnetic field estimate . the discrepancies arise because the distinction is not normally made between the case of a fixed source observed with different beaming parameters and the case where the source projection on the sky is held fixed . the former is usually considered , but it is the latter that corresponds to interpreting actual jet observations . furthermore , attention is drawn to the fact that apparent superluminal motion has a spatial corollary , here called `` retardation magnification '' , which implies that most parts of a relativistic jet that are actually present in the observer s frame ( a `` world map '' in relativity terminology ) are in fact hidden on the observer s image ( the `` world picture '' in general , or `` supersnapshot '' in the special case of astronomy ) . galaxies : jets ism : jets and outflows relativity

0806.3798

for over 50 years from the appearance of the seminal _ zur elektrodynamik bewegter krper _ @xcite , only the lorentz transformations of the 4-coordinates of `` events '' were considered in the literature , but not how relativistically moving bodies would _ appear _ when looked at or photographed . this was first done independently and ( relatively ) simultaneously by @xcite and @xcite ; the former showed that the projected outline of a relativistically moving sphere is always a circle , while the latter provided a more extensive discussion of the appearance of moving bodies and pointed out the key features of observing relativistically moving bodies : they appear both _ rotated _ and _ scaled _ ( more details will be given below ) . the motivation for writing the present paper is work on interpreting observations of relativistic jets ( * ? ? ? * e.g. , ) , where the need arises to infer physical properties of the jet fluid in its own rest frame from observations , subject to corrections due to relativistic beaming , whose magnitude is , however , not known from observations . a particular quantity of interest is the rest - frame minimum - energy magnetic - field estimate for a synchrotron source @xcite , and there are different opinions in the literature about how the true rest - frame minimum - energy field scales with the relativistic doppler factor compared to that inferred assuming a non - relativistic source ( compare eqn . a3 of * ? ? ? * to eqn . some of the argument revolves around whether the observed morphological features of jets are `` blobs '' or `` jets '' , and their ( apparently ) different beaming properties . given these differences of opinion on how to `` de - beam '' properly , it is perhaps surprising that nasa s astrophysics data system lists only three papers on interpreting jet observations as citing @xcite , and his results do not seem to be part of the common knowledge of jet researchers . one of the citing papers is @xcite , who consider the implications of relativistic beaming on the difference between observed and intrinsic source counts and give detailed formulae for relating observed and jet - frame fluxes and emissivities . some of these formulae had already been presented in the seminal paper by @xcite . it appears that the difficulties in interpreting jet observations arise because the problem under consideration is ill - posed . as will be argued in detail below , what matters for interpreting jet observations subject to unknown beaming paramters is that we have observed the 2-dimensional projection of a source s appearance onto the plane of the sky and try to infer the source s rest - frame properties from this projection . confusion arises because most formulae in the literature consider what happens to the observed quantities when a _ fixed source _ moves with different lorentz factors and at different line - of - sight angles to the observer , while in observations , it is the _ projection _ of the source which is held constant . furthermore , the effects of light - travel time delays along the line of sight are typically only mentioned explicitly in work comparing jet simulations to observations , e.g. , in @xcite and @xcite , but not in the observational literature . this and the preference for adopting the fixed - source view may be related to the fact that lorentz transformations are usually covered in great detail in a typical course on special relativity , but @xcite and @xcite are hardly mentioned in relativity textbooks . this leads to the present paper with the following outline : the remainder of the introduction summarizes the results by @xcite and @xcite and sets out some basic definitions and terminology . the appearance of relativistic objects , and in particular of astrophysical jets , is discussed in [ s : appearance ] , both from a theoretical point of view and using a simple ray - tracer . ready - to - use formulae for relating jet - frame quantities to observables are given in [ s : beaming ] , including the minimum - energy field . the discussion and summary are given in [ s : disc ] , while appendix [ s : illustr.lor ] describes some _ gedankenexperimente _ on non - conventional world - map measurements that lead to length expansion and time acceleration . as first noted by @xcite , there is a fundamental difference in relativity between the _ locations _ ( 4-coordinates ) of events as judged by observers that are local to the events and equipped with sets of clocks that are synchronized in their rest frame , and the _ appearance _ of relativistically moving bodies as judged by distant observers by means of photons that are received simultaneously ; a little earlier , @xcite had considered the special case of the observed outline of a relativistically moving sphere . the set of event locations is a _ world map _ , while the picture that is taken of the events is a _ world picture_. in the special case of photons arriving at right angles to the detector taking the world picture , it is called a _ supersnapshot astronomical observations clearly fall under the definition of a supersnapshot . the appearance of relativistically moving objects in a supersnapshot is governed by two aspects of photon paths in special relativity @xcite : 1 . two photons traveling abreast with a separation @xmath0 in one frame ( i.e. , photons traveling `` alongside each other '' with @xmath0 measured perpendicular to their direction of motion ) do so in _ all _ frames . this is the case because @xmath1 is invariant under lorentz transformations and @xmath0 is a space - like interval . 2 . if a photon is traveling at an angle @xmath2 to the direction of motion of some frame that is moving with speed @xmath3 and lorentz factor @xmath4 with respect to an observer , the angle between the direction of motion and the photon direction in that frame is related to the angle @xmath5 between the direction of motion and the photon direction in the observer s frame by @xmath6 where @xmath7 etc . , or , equivalently , @xmath8 where @xmath9 is the relativistic doppler factor @xmath10^{-1}. \label{eq : dopplerdef}\ ] ] the latter phenomenon is the well - known angle aberration ; the former is perhaps less well - known , but essential for the analysis of images of relativistically moving objects , and implies that the supersnapshot is a _ scaled _ version of the rest - frame image . taken together , they yield terrell s result that the appearance of such an object in a supersnapshot is simply the object s appearance as seen from the aberrated angle @xmath2 in its rest frame , with its apparent size along the direction of motion scaled by the doppler factor @xmath11 . as a consequence of eq . ( [ eq : costrans ] ) , even approaching objects appear to be seen `` from behind '' unless @xmath12 , i.e. , @xmath13 ; in the limiting case @xmath14 , a relativistic object is seen exactly side - on in its rest frame and with exactly its rest - frame length as its `` projected '' length . it is useful clearly to set out the terminology for the remainder of the paper , because the brightness pattern observed in astrophysical jets can be at rest in frames that are different from both the observer frame , and the fluid rest frame , as discussed in detail by @xcite . their discussion and notation is adopted here . it distinguishes between `` blob '' , `` jet '' and `` shock '' features , which are defined by being at rest in one of three frames relevant to the problem . thus , it is useful to give the definitions of the relevant frames together with those of the morphological terms : 1 . the `` observer frame '' is that in which the astronomer is at rest . once appropriate cosmological corrections are applied , the observer frame is conceptually identical to the frame in which the jet source and its host are at rest . + a `` jet '' feature is then a _ resolved _ brightness pattern whose outline is at rest in the observer frame . observer - frame quantities have no primes , e.g. @xmath15 for volume emissivity . the `` fluid frame '' is the rest frame of the emitting fluid , which is taken to be moving through the observer frame at relativistic speed . the term `` rest frame '' is used interchangeably with `` fluid frame '' . + a `` blob '' or `` plasmoid '' is a brightness pattern whose outline is at rest in the fluid frame . fluid - frame quantities will be designated by double primes , e.g. @xmath16 . a `` pattern '' or `` shock '' feature is a brightness pattern whose outline is at rest neither in the fluid nor in the observer frame , e.g. , a shock traveling through the jet fluid . it defines the third frame , the frame in which this pattern is at rest . pattern - frame quantities have single primes , e.g. @xmath17 . the emissivity of the fluid traveling through such a `` shock '' transforms according to the fluid s doppler factor @xmath18 , while its projected appearance and morphology are governed by the shock s doppler factor @xmath19 . these are fairly intuitive definitions . nevertheless , the difference between the `` blob '' and `` jet '' formulae in eqn . c7 of @xcite is just one of _ choice of integration boundaries _ , and in particular whether the integration boundaries are held fixed in the observer frame when @xmath11 is changed ( jet case ) or are allowed to vary according to the different projected morphology of a `` blob '' under changes of @xmath11 . thus , it is possible to apply a `` jet '' formula to a small segment of a blob as long as the integration boundaries are held fixed in the observer frame . in [ s : beaming.blobs.restframe ] and [ s : beaming.jets.restframe ] below , i will present detailed formulae for converting observed to fluid - frame properties in each case , with expressions for the minimum - energy field in [ s : disc.bmin.obsfixed ] . this section summarizes the basic definitions of surface brightness / intensity , flux density and luminosity of astronomical sources , as well as the beaming properties of blobs , jets , and shocks . i will give explicit formulae for observed surface brightness and total flux in terms of source parameters for simple geometries , as well as ray - tracing images showing the appearance of such sources in supersnapshots . for the computation of surface brightness and flux , i use the notation and formulae as given by @xcite and @xcite , assuming an optically thin , isotropic emission with a power - law emissivity @xmath20 that is constant within the emitting region . all observables will be expressed in terms of the emissivity @xmath16 in the fluid rest frame and the source size in the pattern frame @xmath21 , which is identical to the fluid and observer frame for a `` blob '' and `` jet '' , respectively . cosmological transformations , however , are not always given explicitly in order to simplify the notation ; they can be re - incorporated in the usual way by inserting appropriate powers of @xmath22 for cosmological redshifts , and using the appropriate cosmological distance measures . the surface brightness or intensity @xmath23 , flux density @xmath24 and luminosity @xmath25 of a source are given by @xmath26 where @xmath27 is the luminosity distance to the source , which has specific emissivity @xmath15 , volume v and projected surface area a. the transformation properties of these quantities then follow from the relativistic invariance of @xmath28 and the volume transformation ( taken from appendix c of * ? ? ? * ) : @xmath29 assuming optically thin emission makes the discussion appropriate for arcsecond - scale jets , where sources are not compact enough for self - absorption to become important . the difficulties of interpreting observations of optically thick sources , such as compact cores and milli - arcsecond scale jets , have been highlighted by @xcite and @xcite . the essential point here is that the appearance of optically thick sources varies as function of viewing direction , and the relativistic angle aberration implies that the beamed appearance is governed by this intrinsic viewing angle dependence in addition to the flux and surface brigthness beaming . the volume transformation deserves separate consideration . the fact that the outline of the different kinds of brightness pattern is at rest in different frames has led some authors to write down different volume transformation formulae for astronomical observations of `` jets '' and `` blobs '' ( see * ? ? ? * , appendix a , and ; * ? ? ? * , appendix a , e.g. ) . however , what matters for the volume transformation of a feature identified in an astronomical image or radio map is only that the image is a supersnapshot . what matters for the interpretation of the supersnapshot is the volume of fluid whose photons arrive simultaneously on the _ supersnapshot _ , not the volume of fluid that is located within the jet volume in the _ world map_. hence , the correct volume transformation for any fluid volume @xmath30 observed by means of a supersnapshot is @xmath31 where @xmath18 is the doppler factor of the fluid in the observer frame . if the decisive criterion was not the fact that astronomical observations are supersnapshots , one could argue with equal justification that the correct volume transformation formula for the fluid in a jet section is @xmath32 because the jet volume is at rest in the observer frame and hence appears contracted in the rest frame of the fluid , or alternatively that the correct transformation is @xmath33 because the fluid is moving through the observer frame , and therefore _ it _ is contracted . both can of course be correct , depending on whether one is judging the jet volume with the help of events that are simultaneous in the fluid or the observer frame . however , a supersnapshot corresponds to neither world - map case the supersnapshot criterion is photons _ arriving _ simultaneously at the observer , which nearly always does not correspond to photons _ being emitted _ simultaneously in any frame . that eq . ( [ eq : vtrans ] ) is correct for both the `` jet '' and `` blob '' case can be seen also by considering a section of a `` jet '' as a collection of infinitesimal blobs that are each at rest in the fluid frame . alternatively , a `` jet '' can be considered as a section of a `` blob '' that is moving through a transparent gap in obscuring material that is at rest in the observer frame if 90% of a blob s volume is covered in the observer frame , the rest - frame volume of the visible part is 10% of the blobs s total observer - frame volume , and hence must also be 10% of the blob s rest - frame volume . as an alternative derivation of eq . ( [ eq : vtrans ] ) , consider that the observer - frame volume of a `` jet '' or `` blob '' ( or an infinitesimal element of it ) is given by @xmath34 where @xmath35 is its extent transverse to the line of sight in the plane of its motion , @xmath36 is the extent perpendicular to both the line of sight and the direction of motion , and @xmath37 is along the line of sight . the individual factors of @xmath38 transform into the fluid rest frame as follows . first , since @xmath36 is perpendicular to the direction of motion , it is not affected by relativity in any way , and @xmath39 . next , recall from [ s : intro.worldsnap ] above that the transverse separation @xmath0 of two photon paths , i.e. , light rays , is lorentz - invariant . the transverse extent @xmath35 is defined by two such parallel light rays and therefore it is also lorentz - invariant , hence @xmath40 . finally , to determine the transformation properties of @xmath37 , consider the following argument the optical depth @xmath41 along @xmath37 has to be lorentz - invariant since it encodes the fraction @xmath42 of photons that are absorbed by the jet material , which is independent of the motion of any observer @xcite . by definition , the optical depth is @xmath43 where @xmath44 the absorption coefficient of the material . the lorentz invariance of @xmath41 therefore implies that @xmath37 transforms inversely to @xmath44 . from the lorentz invariance of @xmath45 ( again see * ? ? ? * ) , it follows that @xmath37 transforms as @xmath46 , i.e. , @xmath47 . hence @xmath48 , again yielding eq . ( [ eq : vtrans ] ) . thus , the relation between rest - frame and observer - frame volume for supersnapshots is always given by eq . ( [ eq : vtrans ] ) , no matter whether we are considering a `` jet '' , `` blob '' or even `` shock '' feature . as noted at the end of the preceding section , the well - known apparent difference between the beaming formulae for a blob and a jet ( @xmath49 versus @xmath50 , such as in app . c7 of * ? ? ? * ) is in fact just a difference of _ integrands _ ; since the _ integration boundaries _ differ depending on whether an object is considered as blob or jet , the final answer is independent of the assumed geometry . in other words , * jets and blobs have the same beaming properties if identical source volumes are considered*. the equivalence of jet and blob formulae will be shown explicitly in [ s : beaming.jets.restframe ] and [ s : disc.bmin.obsfixed ] below . while eq . ( [ eq : vtrans ] ) appears straightforward to interpret , the supersnapshot is merely a projection of the observer - frame volume onto the plane of the sky , so that * the observer - frame volume @xmath38 is not a direct observable * ( see fig . [ f : ray_sph ] below ) . therefore , the volume formula can only be used for interpreting astronomical images if an assumption is made about the geometry of the source . however , its use in determining observables from known rest - frame quantities is straightforward . this section attempts to give an intuitive pictorial representation of how relativistic jets appear in supersnapshots . the first part considers jet observations as supersnapshots of infinitesimally thin relativistic rods . this approach is appropriate for demonstrating how the scale change in a supersnapshot , referred to below as _ retardation magnification _ , arises as spatial corollary to the well - known temporal phenomenon of apparent superluminal motion @xcite . the second part presents results from a ray - tracer that demonstrate the differences between world maps and world pictures . those effects are particularly important which arise from the extent of actual jets perpendicular to the direction of motion , as they results in extra light - travel delays between the near and far side of the jet that are not present in the case of an idealized , thin rod . consider a relativistically moving rod , i.e. , an object for which the light - travel time _ across _ its extent is negligible compared to that _ along _ its extent . the rod is subject to lorentz contraction , so that its rest - frame length @xmath51 is related to its length in the observer frame by @xmath52 where @xmath53 is inferred from a world - map analysis . for the interpretation of observations , where the observer is sufficiently distant from the moving rod to be able to take a _ supersnapshot _ , we want to relate the apparent size of the object on the supersnapshot to its actual length in the world map . to clarify terminology , the term `` projected size '' ( symbol @xmath54 ) will be used to mean the size that corresponds to the projected extent as measured on the supersnapshot , while the term `` apparent size '' ( @xmath55 ) is the size that is inferred from the projected size by deprojecting with the line - of - sight angle @xmath5 , i.e. , @xmath56 . from the constancy and finite value of the speed of light , @xmath57 where i have defined a magnification @xmath58 substituting eq . ( [ eq : ltrans ] ) recovers the well - known relation @xmath59 ( e.g. , * ? ? ? equivalently , @xmath60 not by coincidence , the projected velocity @xmath61 ( usually called `` apparent transverse velocity '' ) in apparent superluminal motion is related to the actual velocity @xmath62 by exactly the same magnification factor that relates the projected length to the actual length : @xmath63 from eqs . ( [ eq : ltrans_mag ] ) , ( [ eq : ltrans_dproj ] ) and ( [ eq : ltrans_dproj_obs ] ) , we can read off all the relevant implications for the interpretation of supersnapshots : 1 . for an approaching rod with @xmath64 , @xmath65 and therefore @xmath66 for any value of @xmath67 . thus , the _ apparent _ , deprojected length @xmath55 of a relativistic approaching object is always greater than its observer - frame length @xmath53 as inferred from a world map . in other words , * in a supersnapshot , any approaching relativistic object appears magnified compared to its actual observer - frame size*. only for @xmath68 , @xmath69 and @xmath70 , and the lorentz contraction of a thin rod becomes observable . the projected length @xmath54 is always less than or equal to the rest - frame length @xmath51 . the limiting case @xmath71 occurs for @xmath72 , which implies @xmath73 and @xmath74 . in this case , the `` projected '' appearance of the rod in a supersnapshot is identical to the view of an observer looking at the rod from 90 in its own rest frame , without any lorentz contraction . this fact was already derived at the end of [ s : intro.worldsnap ] , there based on eq . ( [ eq : sintrans ] ) . the second point implies that it is possible to constrain the orientation and speed of a moving relativistic object if its rest - frame size is known . in the case of jets , this may be possible if jet features are known to have a certain ratio of length to width , since the width is not affected by the beaming . figure [ f : magnif ] gives a more intuitive illustration of why distant relativistic objects appear magnified in a supersnapshot compared to their true observer - frame extent . as in apparent superluminal motion , the cause of this effect is the time delay between light signals reaching the observer from the end of the object that is closest to the observer and those from the end of the object furthest from the observer . therefore , i call this effect _ retardation magnification_. in analogy to the first point above , any relativistic object s apparent velocity is _ always magnified _ compared to its true velocity , without necessarily appearing to be superluminal . for astrophysical jets , however , there is a catch : jets are produced by accretion disks around a compact object , and they terminate in a `` hot spot '' ( this can be a shock terminating an fr ii jet , or a flaring point at which an fr i jet decelerates substantially ) . both are moving through the observer frame much more slowly than the jet material itself . hence , the apparent size of any jet feature can not be larger than the separation between the core and the hot spot . thus , if jet features appear magnified , it means that not all of the features that are actually _ present _ in the observer frame can be _ visible simultaneously _ to the observer , as the total apparent length would then need to be larger than the actual separation between core and hot spot . hence , the magnification implies that parts of the jet are hidden from the observer s view . this retardation _ hiding _ occurs because light signals from the further end of the jet have not yet had time to reach the distant observer , even though they have already emerged from the core . as the object appears magnified by a factor @xmath75 , it follows that only a fraction @xmath76 of the object is visible ; since time delays are greater for the parts of the object that are further from the observer , the visible part of the jet is that closest to the observer , and the hidden part is that furthest from the observer . between the observer s line of sight and the jet s direction of motion , for different lorentz factors @xmath77 as given . the visible fraction is the inverse of the magnification factor @xmath78 defined in the text . relativistic flux beaming affects the detectability of jets ; for @xmath79 , the visible fraction of a jet at the critical beaming angle @xmath80 is given by @xmath81 ; as the jet becomes significantly de - beamed for larger line - of sight angles , this is the maximum visible fraction of the jet material in a relativistic jet.,width=317 ] figure [ f : visfrac ] shows the visible fraction as function of lorentz factor and line - of - sight angle . figure [ f : angles ] illustrates which parts of a relativistic rod are visible to a distant observer . the relativity - textbook analogue would be a relativistic train emerging from one tunnel and disappearing into a second one . if the distance between the end of one tunnel and the beginning of the second one is sufficiently short , a distant observer looking at the train from a small angle to the train s direction of motion will observe fewer railway carriages between the two tunnels than actually fit between them as judged by observers creating a world map . the magnification obviously applies to blobs , as their outlines are at rest in the _ fluid _ frame . however , it also applies to individual fluid elements that make up a `` jet '' feature whose outline is at rest in the _ observer _ frame . hence , a given section of an approaching jet actually _ contains _ more fluid elements than are _ visible _ simultaneously to an observer . this is another way of deriving that eq . ( [ eq : vtrans ] ) is always the correct volume transformation for supersnapshots , even in the jet case . the discussion so far has dealt exclusively with 1-dimensional rods , with negligible light travel time in the direction transverse to the direction of motion compared to the light travel time along the direction of motion . this situation does not apply to features of real astrophysical jets , which have comparable extent along and across the direction of motion . the next section illustrates the effect of light - travel delays across a relativistically moving object . to give a visual illustration of the difference between a world map and a supersnapshot , this section contains images of the same emitting regions , once in the world map view , identical to what would be seen if the speed of light was infinite , and once in a supersnapshot view , appropriate for astronomical observations . as just noted , the discussion in the preceding section applies directly to `` blob '' features , whose outline is at rest in the fluid frame , but also to individual fluid elements making up a `` jet '' feature , whose outline is at rest in the observer frame . the main difference between a blob and a jet for the ray - tracing images is that the _ outline _ of blobs experiences the magnification effects . by contrast , the outline of stationary jet features does not get magnified and behaves according to our intuition , which is formed by observing bodies moving at velocities much less than the speed of light . hence , this section concentrates on the aspect of how the observed ( projected ) and true geometry of objects in supersnapshots are related , including the question how to infer the volume of a relativistic object from a supersnapshot . the discussion here and in the entire the paper is restricted to optically thin objects . the effects of relativistic beaming on flux measurements and quantities derived from them will be examined in the following section . figures [ f : ray_sph ] and [ f : ray_slab ] shows some simple ray - tracing pictures illustrating the difference between world map and supersnapshot for blobs , and the different projection properties for relativistically moving and stationary features . the following points about observations blobs or jets can be inferred from figs . [ f : ray_sph ] and [ f : ray_slab ] : 1 . as expected from @xcite , blobs that are spherical in their rest frame are always observed as spherical blobs , even though they are in fact lens - shaped in the observer frame ( due to the lorentz contraction along the direction of motion ) . they are still magnified in the supersnapshot , but the magnification is now _ along _ the line of sight ; this is why their observer - frame volume still scales as @xmath82 , even though their projected appearance is identical to their rest - frame appearance . 2 . the observed _ shape _ of spherical blobs is not affected by changes in the line - of - sight angle , and hence the doppler factor ; however , their _ spacing _ changes by a factor @xmath18 with respect to their rest - frame spacing , by @xmath75 with respect to their true observer - frame spacing , resulting in an observed projected spacing of @xmath83 times their true observer - frame spacing . thus , relativistic blobs ( and relativistic objects in general ) behave very unintuitive under changes of the observation angle the closer the line - of - sight angle is to the critical angle given by @xmath74 to the direction of motion , the further they appear to be apart . 3 . by contrast , stationary spherical jet features behave as expected by our everyday intuition , with smaller projected spacings between individual features for smaller line - of - sight angles . the geometry of individual blobs ( spherical vs. slab ) does not affect _ how many _ of the blobs are seen , or their total beamed luminosity , but it does affect how the _ projected appearance _ of individual blobs varies with line - of - sight angle . the figures do not consider moving shocks , features which are at rest neither in the frame of the observer nor in that of the fluid . however , their projected shape as function of doppler factor is identical to the projected shape of blobs , i.e. , it is governed by the shock s doppler factor @xmath19 . the fluid s emissivity in the observer frame , on the other hand , is governed by the fluid s doppler factor @xmath18 . beaming formulae have been given in numerous places in the literature . the usual approach is one that considers a _ fixed source volume _ and then determines the changes in the received flux density , surface brightness etc . resulting from changes in the doppler factor , i.e. , the line - of - sight angle and lorentz factor . however , in astronomical observations , it is not the _ source _ volume that is fixed , but the source s _ projected appearance _ in the supersnapshot . the fact that the source volume differs when deprojecting the observed appearance with different line - of - sight angles is not considered often in the beaming literature . it is therefore worthwhile to restate the beaming prescriptions for both directions , inferring changing observables for fixed source properties , and changing source properties from fixed observables . as above , the equations assume optically thin , isotropic emission with a power - law emissivity @xmath20 that is constant within the emitting region . in the case of a blob , the pattern frame is identical with the fluid frame , so that @xmath84 and @xmath18 is the relevant doppler factor . to get observed quantities in terms of source parameters , we need to specify a geometry for the emission region . i consider simple spherical emission regions as well as `` slab''-shaped regions of either square or circular cross - section . the beamed total flux density of a blob is given by @xmath85 where emissivity and source volume are held fixed in the fluid rest frame . the beamed bolometric luminosity is obtained by integrating @xmath24 over the appropriate range of _ rest - frame _ frequencies , leading ( after substitution using eqs . [ eq : ldef ] and [ eq : nutrans ] ) to @xmath86 in general , the beaming properties of bolometric quantities are obtained from the frequency differentials by adding @xmath87 to the exponent of the doppler factor , and the exponent to the doppler factor of beamed bolometric quantities does not contain any spectral - shape parameter such as @xmath88 . surface brightness beaming is important because the detectability of jet features in optical and radio observations is determined by their peak surface brightness , not by their total flux . the observed surface brightness is given by the rest - frame volume emissivity integrated along the observer s line of sight after rotation into the rest frame . [ [ spherical - blobs ] ] spherical blobs + + + + + + + + + + + + + + + for a spherical blob , the beamed surface brightness is @xmath89 times the surface brightness that would be perceived by an observer at rest with respect to the blob , but at the same distance . since the surface brightness of a spherical blob varies according to the different length of the line of sight as function of sky coordinates , i do not give an explicit formula here . [ [ slab - blobs ] ] slab blobs + + + + + + + + + + for a slab blob with square cross - section of side length @xmath90 , the observed surface brightness far enough from the edges is independent of sky coordinates and given by @xmath91 ignoring edge effects is appropriate for slab blobs with length : width ratios greater than about 2 . the same expression applies to the surface brightness along the projected axis of a cylindrical blob with observed diameter @xmath90 , and the scaling of observed surface brightness with the beaming parameters @xmath92 applies to the entire cylindrical blob . equations ( [ eq : blobfluxdens ] ) and ( [ eq : bloblum ] ) by themselves give the dependence of received flux and inferred luminosity on observation angle and lorentz factor . in order to infer rest - frame quantities such as @xmath16 from observations , it is necessary to infer the rest - frame volume from the source s projected appearance . the prescription for this depends on source geometry . [ [ spherical - blobs-1 ] ] spherical blobs + + + + + + + + + + + + + + + as discussed above , a spherical blob will appear as the same spherical blob to any observer , including the rest observer . hence , the rest - frame volume is simply @xmath93 , where @xmath94 is the observed radius of the blob . this apparently contradicts eq . ( [ eq : vtrans ] ) , but there is in fact no contradiction , as can be seen from fig . [ f : ray_sph ] : the @xmath95 view of the `` supersnapshot '' column shows that for @xmath96 , the ( unobservable ) `` observer - frame volume '' of a spherical blob has a larger extent _ along _ the line of sight than can be inferred from its projected appearance , and a smaller one for @xmath97 . thus , the rest - frame emissivity of the material is given by @xmath98 [ [ slab - blobs-1 ] ] slab blobs + + + + + + + + + + next , i consider a slab - shaped blob with square or circular cross - section @xmath99 , i.e. , a blob whose intrinsic size perpendicular to the direction of motion can be inferred directly from its transverse angular size in the plane of the sky . the rest - frame length of the slab can be inferred from the projected length @xmath54 and the fact that our line of sight crosses the blob in its rest frame at an angle given by eq . ( [ eq : sintrans ] ) , i.e. , @xmath100 . thus , @xmath101 and @xmath102 again ignoring edge effects . comparing to the relation between observed flux and rest - frame volume emissivity , eq . ( [ eq : blobfluxdens ] ) , it is appears to be a contradiction that eq . ( [ slabblob_jtrans ] ) has a different scaling with lorentz factor and observation angle . the difference arises because eq . ( [ eq : blobfluxdens ] ) applies to observations of the same _ source _ from different directions ( @xmath30 is kept fixed ) , while eq . ( [ slabblob_jtrans ] ) describes the situation where the _ observables _ are kept fixed , so that different line - of - sight angles correspond to different inferred values for @xmath30 . the decisive aspect that is often neglected is the length of the sight line across the blob changes due to the relativistic angle aberration ( eq . [ eq : sintrans ] ) , leading to the @xmath103 term in the final expression . because of this line - of - sight deprojection , the `` de - beaming '' formula can not be expressed as function of the doppler factor @xmath18 only . the beamed flux density of a jet feature is given by @xmath104 this arises simply from integrating the fluid - frame emissivity transformed into the observer frame over the observer - frame volume . though comparison with the blob case , eq . ( [ eq : blobfluxdens ] ) , seems to reveal the usual difference in the exponent of the doppler factor between `` jet '' and `` blob '' case , the expression is multiplied by _ different _ volumes , fluid - frame @xmath30 for the blob case , observer - frame @xmath38 for the jet case , that are at rest in _ different _ frames . therefore , merely comparing the exponent of the doppler factor does not contain the full information about beaming properties of blobs versus jets . if the emitting volume is a slab jet with transverse width @xmath90 , the volume of a section with projected length @xmath105 is just @xmath106 . hence , the observed surface brightness of such a slab jet is @xmath107 this is identical to the beamed surface brightness of a slab _ blob _ ( eq . [ eq : slabblobsurfbright ] ) , as it must , because a continuous jet can be considered as a section of a long blob . asking about the `` rest - frame '' volume of a jet feature is not directly meaningful , since the outline of the jet feature is at rest in the observer s frame , while the emitting fluid is not , so there is nt a single frame in which both the fluid and its outline are at rest . however , by substituting eq . ( [ eq : vtrans ] ) , we can of course express eq . ( [ eq : jetfluxdens ] ) in terms of the volume @xmath108 of jet fluid that contributes photons to the observer s supersnapshot , obtaining the same expression as for the `` blob '' case , eq . ( [ eq : blobfluxdens ] ) , and hence also the same expression for the relation between observer - frame and fluid - frame bolometric luminosity , eq . ( [ eq : bloblum ] ) . if the supersnapshot nature of astronomical observations were ignored , an alternative definition of `` fluid - frame luminosity '' of a jet could be derived by noting the following : according to the world map of an observer at rest in the fluid frame , the volume of the jet is _ contracted_. hence , @xmath109 and the fluid - frame bolometric luminosity would be given by @xmath110 i.e. , exactly the opposite scaling of @xmath30 and hence @xmath111 with @xmath77 from that given by ( * ? ? ? the reason for this apparent contradiction , and that it is only apparently a contradiction , is again that result of a relativistic experiment depends on what is held constant in which frame ( compare appendix [ s : illustr.lor ] ) . all that matters for jet observations is the supersnapshot with constant photon arrival times . equation ( [ eq : jetfluxdens ] ) of course correctly describes the dependence of the received flux for a fixed jet feature when varying the fluid s lorentz factor and the line - of - sight angle . as long as we infer the correct observer - frame @xmath38 from the fixed apparent size of a jet feature as function of the unknown fluid lorentz factor and the line - of - sight angle , we can infer the fluid - frame emissivity by inverting that equation . for a spherical jet feature , the observer - frame volume can of course be inferred directly from the projected radius @xmath94 of the sphere , leading to @xmath112 for a `` slab '' feature , we can measure its cross - sectional area @xmath99 and assume some rotational symmetry . for fixed projected length @xmath54 , the deprojected length is then just @xmath113 , so that the relation between emissivity and observed flux of a jet section is @xmath114 again ignoring edge effects , i.e. , assuming a length : width ratio of greater than about 2 . this expression is identical to equation ( [ slabblob_jtrans ] ) , the relation between observables and rest - frame emissivity for a slab - shaped _ blob _ of fixed projected size but unknown @xmath115 . in other words , our lack of knowledge about jet orientation and lorentz factor affects elongated slab - shaped blobs and jets in exactly the same way . however , for a _ sphere _ of fixed observed radius @xmath94 , the inferred fluid - frame emissivity is different by one power of @xmath9 depending on whether we are considering a spherical blob or a spherical jet feature ( compare eqns . [ sphjet_jtrans ] and [ sphblob_jtrans ] ) . this difference between the beaming of an elongated and a spherical blob arises because a spherical blob in effect becomes magnified _ along _ the line of sight , while a slab blob appears magnified _ in the plane of the sky _ , as discussed in [ s : beaming.blobs.restframe ] above ( again compare figure [ f : ray_sph ] ) . the elongation _ along _ the line of sight is unobservable in the projection on the plane of the sky . thus , in the special case of spherical features , there _ is _ a difference between `` jet '' and `` blob '' de - beaming . a synchrotron source , such as the radio jet of an active galactic nucleus , is powered by energy stored either in its magnetic field or in relativistic particles . as first pointed out by @xcite , there is a minimum to the total energy in particles and fields that is required to power a given observed synchrotron luminosity . this minimum can be parameterized in terms of the magnetic field strength at the minimum energy density , which is often used as `` the minimum - energy magnetic field estimate '' . further derivations and ready - to - use formulae for non - relativistic sources are given by @xcite , @xcite and @xcite . @xcite have offered some constructive criticism on this so far well - established formalism , in particular regarding the unknown ratio @xmath116 of total energy in relativistic protons and electrons . @xcite and @xcite have presented detailed derivations of minimum - energy estimates in the rest frame of relativistic jets , but reached slightly different conclusions . armed with the knowledge from the preceding sections , we can now reconsider the question of the correct de - beaming of minimum - energy magnetic field estimates . as noted above , for the case of a `` jet '' geometry , the rest frame of the fluid can not be identified with the rest frame of the `` source '' , as the source volume is _ fixed _ in the observer frame , while the beaming is determined by the lorentz factor at which the fluid is _ moving _ through the observer frame . the same applies for shocks . however , as extragalactic jets are assumed to be close to ideal mhd conditions , the rest frame of the fluid is also the rest frame of the magnetic field , and the frame in which the electron energy distribution is assumed to be isotropic . thus , whether we are considering blobs or jets , the fluid rest frame is the frame in which the minimum - energy field needs to be calculated in order to satisfy the underlying assumptions . as before , the emission is assumed to be optically thin emission with spectral shape @xmath117 and constant emissivity throughout the emission region . in the absence of beaming , the expression for the minimum - energy field in terms of quantities in the rest frame of the fluid ( double primes ) is @xmath118 where the synchrotron spectrum extends over the rest - frame frequency interval @xmath119 ; the first fraction is from the function @xmath120 from ( * ? ? ? 7.8 ) . as pointed out above , neither the rest - frame volume @xmath30 nor the observer - frame volume @xmath121 is fixed by observations when the line - of - sight angle is unknown , but it is the projection of the source on the plane of the sky that needs to be kept fixed . nevertheless , it is instructive to compare the minimum - energy field as function of @xmath77 and @xmath5 inferred for fixed source volume to that inferred for fixed projected appearance . hence both will be rederived here , beginning with the fixed - source case ( [ s : disc.bmin.volfixed ] ) , followed by the fixed - projection case [ s : disc.bmin.obsfixed ] . their direct juxtaposition illustrates why differing opinions have arisen in the literature about how minimum - energy parameters scale with doppler factor . [ [ blob - case ] ] blob case + + + + + + + + + for a blob , the source volume is at rest in the fluid frame , therefore @xmath30 is unambiguous , and the beamed minimum - energy field is obtained from eq . ( [ eq : bmin_dpr ] ) by inserting @xmath122 , @xmath123 , and @xmath124 . hence , @xmath125 and @xmath126 this is identical to equation ( a8 ) of @xcite , which implies that their expression applies to the fixed - volume case considered here , not the fixed - observable case relevant to the interpretation of observations . it is different from equation ( a7 ) of @xcite because those authors explicitly considered the fixed - observable case for a spherical blob whose rest - frame volume @xmath30 is inferred directly from observables , the case that will be discussed in the following subsection . [ [ jet - case ] ] jet case + + + + + + + + from the frequency integral of the definition of monochromatic luminosity ( eq . [ eq : ldef ] ) and the @xmath18-dependence of the emissivity @xmath127 ( eq . [ eq : jtrans ] ; also eqn . 2.5 of * ? ? ? * ) , the bolometric luminosity of a jet section scales with the doppler factor of the fluid as @xmath128 . hence , @xmath129 for a jet scales in exactly the same way as @xmath130 for a blob . the doppler scaling of the frequencies occurring in the @xmath120 expression are identical to the blob case . hence , the expression from eq . ( [ eq : bmin_vfix_blob ] ) is obtained for the jet case also . again , this is the same result obtained by @xcite , again because they were implicitly considering the fixed - volume case . this section describes how to infer the rest - frame minimum - energy field from astronomical observations as function of the unknown lorentz factor and line - of - sight angle ( usually only considered in their combination as doppler factor ) in the case relevant to observations , where it is not the source volume that is kept fixed , but the source s projected appearance . as in [ s : beaming.blobs ] and [ s : beaming.jets ] , the computation of the source volume needs to be done taking into account that changing the lorentz factor and line - of - sight angle not only changes the doppler factor , but also the _ length of the sight line through the object _ due to the relativistic angle aberration ( eq . [ eq : sintrans ] ) . the differences between the formulae presented below and those in the literature arise because the effects of angle aberration are stated here explicitly . to express the minimum - energy field as function of an observed flux density at some frequency @xmath131 , assuming a power - law spectrum , one expresses the rest - frame flux density in terms of the rest - frame luminosity @xmath132 substituting into eq . ( [ eq : bmin_dpr ] ) , one obtains @xmath133 as above ( [ s : beaming.blobs ] ) , the transformation properties for fixed observables depend on the source geometry . before considering this expression for different source geometries , it is important to note that the @xmath131 term in this and the derived expressions does _ not _ scale with the doppler factor , since it is explicitly the fixed observing frequency . the effect of observing the rest - frame spectrum at a doppler - shifted frequency and hence at a different amplitude is already accounted for by the @xmath134 ( @xmath116-correction ) term that appears in the exponent of the doppler factor multiplying the observed flux density . in other words , even when @xmath135 is expressed in terms of @xmath136 , which scales as @xmath89 , the spectral index @xmath88 does not appear in the exponent of @xmath18 in the final expression for the minimum - energy field , or the minimum energy content , because the minimum - energy field depends on _ bolometric _ quantities and all @xmath116-correction terms drop out again . this is also relevant for deriving the boosted version of the yet - lower limit to the energy content of a synchrotron source that ( * ? ? ? * , eqn . 19.29 ) obtains by setting @xmath137 in eq . ( [ eq : lobs_snu ] ) and neglecting the other frequency integration limit . this substitution has to be done in the doppler - boosted expressions for the appropriate geometry , as derived below . hence , it would not be correct to set @xmath138 in eq . ( [ eq : bmin_itosnurest ] ) and apply the boosting afterwards ; in other words , the beamed version of eq . ( 19.29 ) of @xcite _ can not _ be obtained simply by inserting doppler factors into it . once more , the correct boosting transformation needs to be applied not only to the _ integrand _ , but also to the _ integration boundaries _ , and care has to be taken to do both transformations at the same time . [ [ spherical - blobs-2 ] ] spherical blobs + + + + + + + + + + + + + + + since relativistically moving spheres are always observed with a spherical outline , the rest - frame volume of a spherical blob can be inferred directly from the observed radius @xmath94 as @xmath139 . substituting this and the appropriate doppler boosting for flux density ( eq . [ eq : blobfluxdens ] ) and frequency ( eq . [ eq : nutrans ] ) into eq . ( [ eq : bmin_itosnurest ] ) yields @xmath140 this implies that @xmath141 , precisely eq . ( a7 ) of @xcite , and reassuring since they explicitly considered only this special case of a spherical blob . however , hardly any of the currently known jets has knots with morphology that accurately can be described as spherical . [ [ slab - or - cylindrical - blobs ] ] slab or cylindrical blobs + + + + + + + + + + + + + + + + + + + + + + + + + as discussed above , if a slab - shaped or cylindrical blob with intrinsic length : width ratio of greater than 2 is observed to have a projected length ( image extent ) @xmath54 and cross - sectional area @xmath99 , its rest - frame volume is @xmath142 . performing the corresponding substitutions into eq . ( [ eq : bmin_itosnurest ] ) , one obtains @xmath143 the scaling of rest - frame minimum - energy field with the unknown lorentz factor and line - of - sight angle @xmath5 of such a blob is @xmath144 , i.e. , it is not expressible purely as function of the doppler factor @xmath18 because the length of the sight line through the blob scales with @xmath103 . the scaling differs from that of @xcite because they considered only spherical blobs , and from that of @xcite because they did not consider the fixed - observable case . [ [ jet - case-1 ] ] jet case + + + + + + + + in the jet case , the appropriate fluid - frame volume @xmath30 is again the volume of fluid that contributes photons to the supersnapshot , rather than the _ actual _ lorentz - contracted volume inferred from the world map ; hence , the appropriate volume is given by @xmath145 . for a jet with cross - sectional area @xmath99 and projected length @xmath54 , the observer - frame volume is again @xmath146 , and hence @xmath147 identical to the slab / cylindrical blob case . therefore , eq . ( [ eq : bmin_slabblob ] ) applies also to continuous jets , analogous to the identical `` de - beaming '' equations for slab blobs and jets in [ s : beaming.blobs ] and [ s : beaming.jets ] . [ [ an - amusing - special - case ] ] an amusing special case + + + + + + + + + + + + + + + + + + + + + + + usually both @xmath5 and @xmath77 are unknown in jet observations . however , the fact that the jet emission is detectable makes it more likely that the emission is beamed towards the observer than that it is beamed away from the observer . a frequent guesstimate that reduces the number of unknown beaming parameters from 2 to 1 is therefore @xmath148 , equivalent to @xmath72 and @xmath149 . substituting this assumption into eq . ( [ eq : bmin_slabblob ] ) leads to @xmath150 thus , if @xmath148 , the true fluid - frame minimum - energy field is given by @xmath151 where @xmath152 is the minimum - energy field inferred from the observables by neglecting beaming and projection effects and assigning the source a volume @xmath153 . it also turns out that at fixed @xmath5 , @xmath154 . hence , the true minimum - energy field at fixed @xmath5 is always larger than @xmath155 . in [ s : appearance ] , i have illustrated the difference between the _ world map _ of a relativistic jet , what is actually there , and the _ world picture _ , or its special case , the _ supersnapshot _ , that corresponds to what is observable by distant astronomers . for the quantitative interpretation of world pictures in the presence of unknown beaming parameters ( lorentz factor @xmath77 and angle @xmath5 between the fluid motion and the line of sight ) , what matters is that the _ projected appearance _ of the jet is kept fixed and not the _ intrinsic volume_. this gives rise to de - beaming formulae that are slightly different from those in the existing literature . the most important conclusion for the quantitative analysis of jet observations is that the scaling relations relating rest - frame quantities ( volume emissivity , intrinsic source size ) to observables ( projected source size , surface brightness , total flux ) _ can not _ be stated as function of the line - of - sight angle and lorentz factor in a general way , but depend on the details of the source geometry . it is possible to write down explicit scaling relations for certain simple geometries such as spheres and elongated , rotationally symmetric blobs of constant cross - section . for other shapes , such as ellipsoidal blobs or blobs with non - symmetric cross - sections , the projected appearance is affected by edge effects , and additionally by observational effects such as the contrast between the faintest parts of the source and the sky background , as well as the available signal - to - noise level . edge effects are properly taken into account by the ray - tracing in [ s : illust.rays ] , and such ray - tracing modeling of observables is probably the most accurate route to interpreting observations of relativistic objects . indeed , it is part of the prediction of observables from jet simulations such as those by @xcite , e.g. the work of @xcite , which explicitly considers the relation between the jet appearance in a supersnasphot and the underlying physical quantities , taking into account the retardation along the line of sight . most radio , optical and x - ray maps of relativistic jets ( a list of radio jets is given by * ? ? ? * ) show a series of well - separated , distinct features usually referred to as `` knots '' , with diffuse emission linking them . given that relativistic beaming favours the detection of objects with jets at small angles to the line of sight , and the superluminal motions detected in the cores of many such sources , it is plausible that the jet material itself is still relativistic even at large separations ( and indeed , this is required in models accounting for the x - ray emission from powerful radio jets as beamed inverse - compton scattering of cosmic microwave background photons ; see @xcite for the original development of the idea , as well as the recent review by @xcite ) . however , what is not clear is whether the knots themselves are stationary shock features , or themselves moving relativistically . referring to figs . [ f : ray_sph ] and [ f : ray_slab ] , the prevalence of well - separated knots in jet images seems to suggest that the knots are moving at least with mildly relativistic lorentz factors otherwise , there should be _ some _ jets observed at small angles ( favoured by doppler boosting ) where different knots overlap along the line of sight , washing out any individual morphological features . in this case , the knots are subject to retardation magnification and hiding ( illustrated in figs . [ f : magnif ] and [ f : angles ] ) , and we are not seeing _ all _ of the jet features which are actually present between core and hot spot , but just a small fraction ( whose magnitude is given by fig . [ f : visfrac ] ) . an observation of apparent superluminal motion of individual knots would be a direct confirmation that they are moving relativistically , as in the case of parsec - scale knots in vlbi observations . the kiloparsec - scale knots are resolved out in vlbi observations , so that very long - term monitoring programmes at sub - arcsecond spatial resolution are required to make a potential superluminal motion observable . if jet knots are indeed moving relativistically , the retardation magnification and hiding need to be taken into account when interpreting morphological observables such as the ratio between knot separation and jet width , which is important for addressing the question of the origin of the jets morphological features , e.g. , whether they arise from instabilities @xcite or as manifestation of a stable magnetohydrodynamical configuration @xcite . it becomes clear once more that relativistic effects are counter to our non - relativistic intuition , and that familiarity with _ world maps _ , lorentz transformations and the resulting phenomena of length contraction and time dilation is not sufficient for interpreting _ world pictures_. when considering the beaming properties of quantities expressed in terms of integrals , such as a surface brightness , flux or the minimum - energy magnetic field estimate , one needs to consider the transformation properties both of the integrand and the integration volume . apparent differences between the beaming properties of `` blobs '' and `` jets '' disappear when the same source volume is considered . finally , the de - beaming of astronomical observations needs to be done not for a fixed source , but for the fixed _ projection _ of the source . doing so resolves some conflicts ( again only apparent ones ) between different de - beaming formulae in the literature . given that astronomy provides only world pictures , the concepts first laid out by @xcite and @xcite deserve more attention in the interpretation of jet observations . this research has made extensive use of nasa s astrophysics data system bibliographic services . i am grateful to herman marshall for valuable discussions and the impulse to begin this work , and acknowledge fruitful interactions with members of the fermilab experimental astrophysics group , and the astronomy group at the university of southampton . i am particularly grateful to ukasz stawarz for detailed feedback in early stages of this work , to jochen weller , rob fender and dan harris for continued discussions , to arieh knigl for helpful comments on the paper in particular for reminding me of the lorentz invariance of opacity and to w.c . herren for inspiration . i thank the referee , henric , krawczynski , for his constructive criticism which helped me improve the presentation of this material . the portion of this work carried out at fermilab was supported through nasa contract nasgo4 - 5120a and through the u.s . department of energy under contract no . de - ac02 - 76ch03000 . last but not least , i acknowledge support through an otto hahn fellowship from the max - planck - institut fr astronomie that enabled much of this work to be carried out within the astronomy group at the university of southampton , whose hospitality i enjoyed tremendously . the world - map analysis familiar from physics textbooks shows that relativistically moving objects experience lorentz contraction and time dilation . the derivation begins with the lorentz transformations between frames @xmath156 and @xmath157 with aligned @xmath158 and @xmath159 axes and origins coinciding at @xmath160 : @xmath161 to show that there is lorentz contraction of objects at rest in @xmath157 , one considers two observers at rest in @xmath156 which coincide with opposite ends of the moving object at some fixed @xmath162 , e.g. , @xmath163 . solving eq . ( [ eq : lor_t ] ) for @xmath164 and substituting into eq . ( [ eq : lor_x ] ) then yields @xmath165 , i.e. , length contraction . time dilation is obtained by considering a clock at rest in @xmath157 whose time is read by two different observers in @xmath156 , e.g. , one at @xmath166 , the second at @xmath167 . the first observer reads @xmath168 ( by convention ) ; setting @xmath169 in eq . ( [ eq : lor_t ] ) , the second observer reads @xmath170 on the moving clock , i.e. , infers that a shorter time interval has elapsed in the moving frame than in the observers frame and that time is therefore dilated . however , we can ask our observers to do slightly different experiments . imagine that there are a large number of clocks in @xmath157 , and we are told that the clocks are all synchronized in that frame . to investigate the behaviour of time , we ask a single observer to compare the time read on successive clocks that are whizzing past to her own observer - frame clock . in other words , instead of keeping @xmath159 fixed , let us keep @xmath158 fixed , and choose the observer at @xmath171 . the first clock reads @xmath168 . if the clocks are separated in @xmath157 by some distance @xmath172 , the second clock , being at @xmath173 will reach the observer at @xmath174 . at this time , the observer s own clock reads @xmath175 ; thus , the observer s clock has advanced less than the moving ones , and time is accelerated instead of dilated . the reconciliation with time dilation is , of course , that the second moving clock did not read @xmath168 at @xmath163 , but already @xmath176 the clocks that are synchronized in @xmath157 are _ not _ in @xmath156 , simultaneity is relative . indeed , this setup allows the observers at rest in @xmath156 to infer that _ their _ clocks are slowed down relative to observers in @xmath157 : it compares the time interval elapsed on a _ single _ clock in one frame to the interval elapsed between two _ different _ clocks in another frame . the individual clock runs slow compared to two different clocks whizzing past it , or that it whizzes past . we can also derive an alternative length measurement . assume that the observers know about special relativity , and that the observers in @xmath157 have placed a clock at the front and rear end of the object whose length the observers in @xmath156 are trying to measure . these observers decide to measure the length by taking into account the relativity of simultaneity , and they do so by noting the position of each clock in their own frame when it shows a fixed time in the _ moving _ frame , @xmath168 , say . if the rear clock is at @xmath177 , it reads @xmath168 at @xmath163 and will be seen by the observer at @xmath171 . the front clock is at @xmath178 , and thus from eq . ( [ eq : lor_x ] ) it will read @xmath168 when it has reached the observer at @xmath179 . hence , the observers in @xmath156 infer that the moving rod has expanded by a factor @xmath180 compared to what it is in its own rest frame . again , this is just the converse of the observers in @xmath157 measuring the length of an object at rest in @xmath156 and finding that the moving object has contracted . thus , observers in either frame can both observe that moving objects appear contracted to them _ and _ infer that something that is at rest in their own frame will appear contracted to observers in the other frame ; similarly , observers can observe clocks in the other frame run slow _ and _ infer that their own clocks will be observed to run slow by observers in the other frame . the important point is that the outcome of an experiment depends on _ which quantity is held fixed in which frame _ , and the correct interpretation of apparently contradictory results depends on being clear in the description of the experiment . d. e. , krawczynski h. , 2007 , in revista mexicana de astronomia y astrofisica , vol . 27 of revista mexicana de astronomia y astrofisica conference series , constraints on the nature of jets from kpc scale x - ray data . p. 188 c. m. , hughes p. a. , 2008 , in rector t. , de young d. s. , eds , extragalactic jets : theory and observation from radio to gamma - ray vol . 386 of astronomical society of the pacific conference series , understanding the origin of bright features .

making use of the energetics and equations of state of defective uranium dioxide that calculated with first - principles method , we demonstrate a possibility of constraining the formation energy of point defects by measuring the transition pressures of the corresponding pseudo - phase of defects . the mechanically stable range of fluorite structure of uo , which dictates the maximum possible pressure of relevant pseudo - phase transitions , gives rise to defect formation energies that span a wide band and overlap with the existing experimental estimates . we reveal that the knowledge about pseudo - phase boundaries can not only provide important information of energetics that is helpful for reducing the scattering in current estimates , but also be valuable for guiding theoretical assessments , even to validate or disprove a theory . in order to take defect interactions into account and to extrapolate the physical quantities at finite stoichiometry deviations to that near the stoichiometry , we develop a general formalism to describe the thermodynamics of a defective system . we also show that it is possible to include interactions among defects in a simple expression of point defect model ( pdm ) by introducing an auxiliary constant mean - field . this generalization of the simple pdm leads to great versatility that allows one to study nonlinear effects of stoichiometry deviation on materials behavior . it is a powerful tool to extract the defect energetics from finite defect concentrations to the dilute limit . besides these , the full content of the theoretical formalism and some relevant and interesting issues , including reentrant pseudo - transition , multi - defect coexistence , charged defects , and possible consequence of instantaneous defective response in a quantum crystal , are explored and discussed .

1305.6761

defects usually play a prominent role in various properties of a solid . for this reason , the physics and chemistry of defects have been the subjects of much study for several decades.@xcite many of these works focused mainly at the dilute limit , _ i.e. _ , with a small defect concentration . this is the case of interest in doped semiconductors and/or compounds in the immediate vicinity of the stoichiometry . at certain conditions ( usually with high temperature ) deviation from the stoichiometry can span over a wide range of chemical composition . binary oxides ceo@xmath0 and uo@xmath0 represent the paradigms for such kind of non - stoichiometry in the fluorite - related structures , and many others exist as well.@xcite for a comprehensive understanding of these materials , knowledge from just near the stoichiometry is insufficient . this is because many physical quantities depend strongly on non - stoichiometry , and exhibit quite different behavior when at finite deviations . on the other hand , defects at high pressure have received little attention so far , and our knowledge about their general behavior at highly compressional conditions is very limited , in spite of the fact that they are crucial for earth modeling and for planetary evolution description , where plenty of defects presenting in variety of non - stoichiometric minerals in the interior of these celestial bodies . the capability to capture correctly the energetics and other physical properties across the whole stoichiometry range at different temperature - pressure conditions is an essential requirement for the purpose of predicting and controlling the behavior of these complex materials . nevertheless , a general theoretical method for this purpose is still elusive . even at ambient conditions and near the stoichiometry , our understanding about nuclear oxides such as uo@xmath0 and ( u , pu)o@xmath0 is also limited and unsatisfactory.@xcite although there are many papers and reports have been published on various aspects of diffusion in these oxides , a reasonable level of understanding has been reached only in the case of oxygen , from which the formation energy of oxygen frenkel pair can be deduced when the migration energy of the corresponding diffusion process is known.@xcite nevertheless , measuring oxygen diffusion in stoichiometric oxides is difficult because of the need to maintain the stoichiometry , which is almost impossible for a large temperature range.@xcite the difficulty also lies in the interpretation of available experimental data . while theoretical calculations can be applied to individual processes , transport and other data often correspond to a superposition of several entangled processes , and makes extraction of the desired information complicated.@xcite in addition to chemical diffusion and self - diffusion , electrical conductivity@xcite and neutron scattering@xcite have also been performed to measure the defect concentrations and formation energies . in all of these experiments , a poor understanding of the experimental conditions , as well as the inherent difficulties in the measurement and subsequent interpretation of data , have caused dispersed results.@xcite there are also very few experimental data existed for cation defects in nuclear oxides.@xcite for example in uo@xmath0 , only crude estimates of the activation energy for uranium self - diffusion and migration energy of uranium vacancy are available . with the aid of point defect model ( pdm ) , one can extract the schottky defect formation energy from these estimates . but since a great uncertainty remains in the experimental data , the reliability of the derived value is doubtful.@xcite progresses in density functional theory of electrons and in computational algorithms make it possible to calculate the relevant energetics directly from quantum mechanics . such kind of first principles methods have provided better data than previous semi - empirical interatomic potentials , and are comparable to experimental measurements . the electronic structures,@xcite structural phase transformations@xcite and equations of state ( eos),@xcite oxygen diffusion,@xcite and some defect clustering structures@xcite have been modeled . unfortunately , this advance never reached a satisfactory level for defect energetics , even though a lot of efforts have been devoted to it.@xcite it even can not reproduce the experimental fact that oxygen defects dominate the whole stoichiometry range.@xcite also one should note that including oxygen clusters can give rise to the expected predominance of oxygen defects,@xcite but its relevance to the experimentally measured defect formation energy is unclear.@xcite this discrepancy among experimental and theoretical results might be due to the limited accuracy of the theoretical assessments , for example the possible meta - stable electronic states that could be encountered in calculations of strongly correlated materials,@xcite the approximation employed to treat the partially localized 5__f _ _ orbitals,@xcite the variation of the charge state of defects,@xcite and the small size of the simulation cell for defective structure modeling that currently accessible , _ etc . _ all of these might render uncertain error in the final results . but it also can arise from the error lying in the experimental estimates that widely used as the benchmark for theoretical modeling : the data are scattering and not fully self - consistent , and in some cases these estimated data can not reproduce the fact of the predominance of oxygen defects too.@xcite this makes the problem entangled and very hard to treat with . therefore any approach that can constrain the defect energetics and reduce its uncertainty is of decisive help for solving the problem . in this report , we investigate the possibility of a such kind of method by examining the physics that governs the pressure - driven pseudo - transitions@xcite between different defect species . by establishing a theoretical relationship of the pseudo - phase boundary on defect formation energies , we show that strict constraints can be imposed on these energy parameters , which might then be used to refine the experimental estimates . to achieve the final goal , however , one has first to measure the curve of pseudo - transition pressures , and obtain accurate equations of state of defects , then by making use of the theoretical method that we will present below , to get reliable estimate of defect formation energies . in this sense this report is the first step but also the most important step towards this accomplishment . by developing the theoretical basis of this constraining procedure , it not only provides us a new angle to understand the long - standing problem , but also establishes a general method in treating highly defective materials under high pressures . for the clarity of discussion , we will first present a simple theoretical framework using pdm that is intuitive and easy to understand , then a generalization to the general case will be made . in sec.[sec : simple_model ] we discuss the pdm that allows us to calculate the pseudo - phase diagram of point defects , and then establish a relationship between pseudo - phase boundary and defect energetics . using this powerful tool , the influence of intrinsic defect formation energy on pseudo - transition is then investigated in sec.[sec : pdm ] . the obtained information represents a constraint on the possible value of the formation energy of the defects . though our discussions are mainly focused on uranium dioxide in this paper , we also extend the investigation by considering virtual models to explore other interesting phenomena such as reentrant pseudo - transition and multi - defect coexistence in sec.[sec : virtual system ] . it is well known that pdm does not take defect - defect interactions into account and can be applied to only the vicinity of the stoichiometry . in order to deal with highly defective region and extrapolate the energetic and thermodynamic information obtained at finite stoichiometry deviations to the dilute limit , a general formalism that can treat defect interactions is developed in sec.[sec : defect interactions ] . this generalization is necessary for a realistic description of the nonlinear dependence of thermodynamic properties on non - stoichiometry . a brief discussion and remarks on charge state of defects , as well as other relevant issues , are then given in sec.[sec : discussion ] , which is followed by a summary of the main conclusions . at dilute limit where defect concentration is negligibly small , imperfectness in crystal has little impact on thermodynamic and mechanical properties , despite a profound modification on electrical conductivity and/or magnetism might often occur . at large stoichiometry deviation , the defect concentration is governed by the ratio of chemical compositions rather than by thermal excitation , therefore the density of defects could be enormous . in this case , noticeable influence on general thermodynamic quantities can be expected , so as on relevant mechanical properties . to understand the general effects of non - stoichiometry on material s behavior , we require a physical model that expresses the defect density as a function of external conditions usually the hydrostatic pressure and temperature , and how the presentation of defects modifies thermodynamic functions such as enthalpy or gibbs free energy . having such a function that incorporated defect effects , all relevant thermodynamic properties can be derived straightforwardly . in this section we first present the basic picture by considering the simple pdm . a general formalism will be developed in sec.[sec : defect interactions ] . in pdm,@xcite the spatial size of an individual defect is assumed to be of zero dimension , and all interactions among them are neglected . in this simple model , defect concentrations are determined by the corresponding formation energy of isolated defects . considering a structure that contains one defect of type @xmath1 , its gibbs free energy can be written as @xmath2 in which @xmath3 , @xmath4 , and @xmath5 stand for hydrostatic pressure , temperature , and volume of the simulation cell , respectively . the cohesive energy at zero kelvin in static approximation reads @xmath6^{2 } \label{eq : colde}\ ] ] when expressed in birch - murnaghan equation ( other eos model can be used as well ) . here variables with subscript 0 denote the corresponding value in the equilibrium condition of zero pressure , and @xmath7 is the bulk modulus . the contribution of lattice dynamics to the free energy can be approximated in the debye model as @xmath8-k_{b}tf\left(\theta_{d}/t\right)+\frac{9}{8}k_{b}\theta_{d},\ ] ] where @xmath9 is the boltzmann constant and @xmath10 the debye function . the debye temperature can be evaluated approximately by@xcite @xmath11^{1/2}.\ ] ] here the superscript @xmath12 denotes the reference state ( here the defect - free uo@xmath13 ) , @xmath14 is the effective volume per atom , and @xmath15 is the effective atomic weight . the parameters in these equations , namely @xmath16 , @xmath17 , and @xmath18 , can be obtained by fitting to _ ab initio _ results of density functional theory , while @xmath19 can be taken from x - ray diffraction measurement.@xcite the details of determining the value of these parameters have been discussed and presented in ref .. having known @xmath20 , the formation gibbs free energy ( fge ) of intrinsic point defects can be constructed . for example , the fge of a frenkel pair ( fp ) of species @xmath21 is then expressed as @xmath22 and for the schottky defect ( s ) as ( taking uo@xmath0 as the example ) @xmath23 here @xmath24 denotes the number of atoms in a defect - free cell and @xmath25 is the corresponding gibbs free energy , @xmath26 is the gibbs free energy of the cell containing the respective defect . in a closed regime where no particle exchange with the exterior can occur , the defect concentration must satisfy@xcite @xmath27[i_{o}]=\exp\left(\frac{-\delta g_{\mathrm{o{\_}fp}}}{k_{b } t}\right)\label{eq : pdm1},\\ & [ v_{u}][i_{u}]=\exp\left(\frac{-\delta g_{\mathrm{u{\_}fp}}}{k_{b } t}\right)\label{eq : pdm2},\\ & [ v_{o}]^{2}[v_{u}]=\exp\left(\frac{-\delta g_{\mathrm{s}}}{k_{b } t}\right)\label{eq : pdm3}.\end{aligned}\ ] ] the composition equation that expressed in point defect populations is @xmath28-[i_{u}]\right)+[i_{o}]-2[v_{o}]}{1-[v_{u}]+[i_{u } ] } , \label{eq : pdmx}\ ] ] where @xmath29 is the stoichiometry deviation ( for example that in uo@xmath30 ) . notice that eq.([eq : pdmx ] ) is different from the conventional definition of @xmath31-[i_{u}]\right)+[i_{o}]-2[v_{o } ] , \label{eq : pdm4}\ ] ] which is valid only when no cation defect is involved . we thus complete the formalism of pdm , in which the defect concentrations are determined by solving eqs.([eq : pdm1]@xmath32[eq : pdmx ] ) . since no interaction among defects has been taken into account in pdm , the total gibbs free energy of a defective system is a linear superposition of the contribution of each individual defect . that is , @xmath33 where @xmath34 is the gibbs free energy of the defect - free matrix . the defect concentration @xmath35 runs over @xmath36 $ ] , @xmath37 $ ] , @xmath38 $ ] , and @xmath39 $ ] , respectively , with the superscript `` @xmath40 '' indicates the corresponding value in a defective reference system , and @xmath41 . from eq.([eq : gtot ] ) thermodynamic quantities as a function of stoichiometry deviation @xmath29 can be derived . in defective crystals , distinction of the physics mainly originates from the predominant defect species . thanks to the exponential dependence of defect concentrations on the formation energy , most regions in the phase space spanned by temperature , pressure , and chemical composition ( @xmath4-@xmath3-@xmath29 ) are dominated by only one type of defect . one can then use the concept of pseudo phase to simplify the description of defective ( non - stoichiometric ) materials.@xcite in this picture , each pseudo phase corresponds to a region that is governed by a homogeneous distribution of a _ single _ type of defect . here no effects of migration and creation or annihilation of defects are considered , which is justified if we focus mainly on the _ long - time averaged _ properties only . with variation of the thermodynamic conditions of @xmath4 , @xmath3 , and @xmath29 , the predominant defect might change from one type into another , _ i.e. _ , pseudo phase transition ( ppt ) might take place . physical quantities that affected by defects also change rapidly along this transition.@xcite from this perspective , the physics of defects can be greatly simplified to that of each individual pseudo - phases and their respective behavior at the ppt . it is thus important to understand the extent of the control region of each pseudo phase , namely , the pseudo phase diagram ( ppd ) . from eqs.([eq : pdm1]@xmath32[eq : pdmx ] ) , it is evident that such a diagram is completely described by the energetics of each defect . conversely , if we know the ppd , then constraints on defect formation energies can be established . it is natural to define the transition point of a ppt at where defect concentration increasing / decreasing to the half of its saturate value , then the corresponding pseudo - phase boundary ( for example in uo@xmath30 ) is determined by the following equation @xmath42 for a transition between @xmath43 and @xmath44 when @xmath45 , and @xmath46 between @xmath47 and @xmath48 when @xmath49 ( only point defects are considered ) . because all @xmath50 are functions of @xmath4 and @xmath3 ( and also depend on @xmath29 via defect interactions , which we discarded here but will discuss in detail below ) , solutions of eq.([eq : ppbxp ] ) and ( [ eq : ppbxn ] ) provide a set of constraints on formation energy of intrinsic defects . calculated with crocombette ( jpc)@xcite and gengs@xcite formation energies of defects . the constraints on intrinsic defect formation energies when the pseudo transition pressure bounded between 0 and 40gpa are also shown.,width=3 ] this subsection is devoted to a realistic system of uo@xmath0 , where involved parameters are obtained by density functional theory calculations . since early experiments were driven by application of uo@xmath0 as a nuclear fuel , most investigations focused mainly on formation energy of intrinsic defects , which closely relates to the parameter @xmath16 in eq.([eq : colde ] ) . to be consistent with conventional notations , we use @xmath51 , @xmath52 , and @xmath53 to denote the formation energy of schottky defect , oxygen frenkel pair , and uranium frenkel pair at zero pressure and temperature , respectively . based on the defect energetics calculated by geng _ et al . _ for uranium dioxide,@xcite the ppd on the @xmath3-@xmath29 plane are evaluated and shown in fig.[fig : ppd ] . it can be seen that in the region of @xmath49 an increasing of temperature from 300 to 2500k has little influence on the pseudo boundary . however , in the @xmath45 side , such a size of change in @xmath4 leads to an increase of the pseudo transition pressure about 5gpa . except that , the impact of @xmath29 on ppd at the pdm level of approximation is small , and with an opposite trend for the hypo- and hyper - stoichiometry region . there are a few other theoretical assessments of the defect energetics for uo@xmath0 available in literature.@xcite unfortunately in most cases only formation energy at 0gpa and 0k were given , from which , however , one can not determine the pseudo transition pressure because the information about the variation with temperature and pressure is lost . under an assumption that the compression behavior and phonon contribution are the same for all of these calculations , which is a simple but reasonable approximation , we can estimate the corresponding pseudo transition boundary by adjusting the @xmath16 in eq.([eq : colde ] ) accordingly to yield the respective formation energy at 0gpa and 0k using geng s equation of state.@xcite here we choose the data of j. p. crocombette ( jpc ) for the purpose of comparison , since it is a typical one that has considered possible charge states of defects , thus produced a formation energy of oxygen frenkel pair and schottky defect that seems in a good agreement with experimental estimates.@xcite by adjusting @xmath16 to reproduce the @xmath51 and @xmath52 of jpc , we obtained the estimated ppd of jpc s data . note here we have made an assumption that the charge state of each defect is fixed during compression or heating , and the formation energy of uranium interstitial was taken from geng s data because in jpc s work no value for this defect type was given . the ppd calculated with jpc s formation energy is drawn in fig.[fig : ppd ] for comparison . at @xmath49 side , the transition from @xmath48 to @xmath47 takes place at a much higher pressure . this is reasonable since jpc s data has a lower formation energy for @xmath48 than geng s evaluation , which gives rise to a stronger stability of this defect . at the @xmath45 side , uranium vacancy was predicted to be the predominant defect at low pressure , and the transition to @xmath44 occurs at about 8gpa . this result is consistent with previous pdm evaluations at zero hydrostatic pressure , where @xmath43 was predicted to be the major defect component . nevertheless , prevailing @xmath43 is contradictory to the experimental observation that oxygen defect should dominate this region , indicating that the experimental estimate of the defect formation energy might be inconsistent in itself.@xcite this difficulty could stem from the procedure of extraction defect energetics from diffusion measurements . usually the employed physical models were very simple and might lead to inaccurate explanation of the measured data.@xcite on the other hand , as mentioned above , ppd provides valuable information about possible range of the defect formation energy . for uo@xmath0 , the ppt in fluorite structure ( if exists ) should be bounded between a pressure range of 0 and 40gpa , because at higher pressures uo@xmath0 transforms into @xmath54 phase,@xcite which is followed by an iso - structural transition,@xcite and finally converges to isotropic @xmath55 structure@xcite according to recent theoretical predictions . under this restriction , the formation energy of intrinsic defects should satisfy @xmath56 when @xmath49 , and @xmath57 when @xmath45 for a ppt to occur between a pressure of 0 and 40gpa . these constraints are marked in fig.[fig : ppd ] together with dashed lines that indicate the corresponding bounded range of pressure . . , width=3 ] theoretical assessment explicitly suggests that @xmath43 definitely becomes unfavorable under compressional conditions . therefore the experimental observation that @xmath44 prevails at @xmath45 region implies that there should be no ppt from @xmath43 to @xmath44 at any pressures greater than zero . then from eq.([eq : ppd2 ] ) one gets @xmath58 which puts a strong constraint on possible value of defect formation energies . for example the experimental estimates of @xmath52 lie in between @xmath59ev and @xmath51 between @xmath60ev.@xcite if we take @xmath51 as 7.0ev , then @xmath52 must be less than 3.5ev . this value is , however , incompatible with the most reliable experimental assessments.@xcite on the other hand , if we take the neutron scattering measurement@xcite of 4.6ev as a reliable estimate for @xmath52 , then @xmath51 must be greater than 9.3ev . this in turn disqualifies most theoretical estimates with charged defects.@xcite in a word , all of these indicate that we need further scrutiny on these estimates , and any alternative and/or complemental information on defect energetics are decisive to reach the final conclusion . inequalities of eqs.([eq : ppd1]@xmath32[eq : ppd2 ] ) cover most range of the experimental estimates , as shown in fig.[fig : tp ] . thus it can provide new understanding about this issue if we can measure the compression - driven ppt of point defects experimentally . defective behavior of materials at high pressure is determined by the variation of defect formation enthalpy with compression . it is also affected by possible structural transitions of the matrix . above discussion elaborated what might happen in a compressed non - stoichiometric uo@xmath0 . in other materials , however , much more complex phenomena can be expected . from a theoretical perspective , it is helpful to explore all possibilities in order to grasp the general feature of the physics of defects . in the simple pdm approximation , the physics is mainly determined by the parameters appeared in eq.([eq : colde ] ) , _ i.e. _ , the value of @xmath16 , @xmath17 , and @xmath18 of each defective configurations that were employed to derive the formation gibbs free energy ( fge ) . therefore we can arbitrarily alter these parameters to probe other interesting behaviors of defects that are allowed in theory but not in the realistic uo@xmath0 . the first phenomenon we would like to discuss about is reentrant transition . it is a rare type of phase change even for a conventional physical state , where one phase that has been transformed into another re - appears . on the phase diagram the corresponding phase boundary is a reentrant curve . analogous phenomenon can also take place in ppt , where the predominant defect species firstly changes into another type , and then transforms back . the condition of this transition is completely governed by eqs.([eq : pdm1]@xmath32[eq : pdm3 ] ) . put it explicitly , if the derived equation [ eq.([eq : ppbxp ] ) or ( [ eq : ppbxn ] ) ] has multiple solutions , then the corresponding ppt is reentrant . . upper panel : energy variation ; lower panel : defect concentrations . , width=3 ] figure [ fig : reentry ] demonstrates a virtual dioxide compound that has a reentrant ppt . in the upper panel , a graphical solution of eq.([eq : ppbxp ] ) is drawn where the solid line is the term of @xmath61 . another term of @xmath62 is also shown as a dashed line in the figure . the points of intersection a and b correspond to the solutions of eq.([eq : ppbxp ] ) , which also are the locations where ppt taking place . in the lower panel , the change of defect concentrations along compression is illustrated , from which one can clearly see that cation vacancy re - appears at higher pressures . it is necessary to point out that this result was obtained by subtracting 145gpa from the bulk modulus of all defective uo@xmath0 configurations , thus might be an artifact . nevertheless , such a virtual model can help us acquire a profound understanding of material s behavior that having reentrant ppt , if it exists . the resultant modifications on eos and thermodynamic properties across this transition region are interesting . figure [ fig : pv_ppt ] illustrates the compression behavior of the same ppt as in fig.[fig : reentry ] , in which the restoring of the compression curve at high pressure end is evident . there are two volume - collapses in the reentrant ppt at point a and b , respectively . between a and b , the curve is steeper . this is consistent with the requirement for a pressure - driven reentrant transition to occur , namely , the intermediate phase should have a larger bulk modulus and a smaller equilibrium volume . this condition guarantees a double volume - collapse along compression , which is a necessity for a first - order reentrant transition . other thermodynamic properties also show discontinuous or quasi - discontinuous jump at the ppt points . the inset in fig.[fig : pv_ppt ] draws relative variation of thermal expansivity @xmath63 with respect to its initial value along the 300k compression curve . it deviates from the trend of the initial phase ( as the dashed line shows ) at point a and plunge to a new value , but at point b it jumps back to the previous curve , a key feature of reentrant transition . other physical quantities , such as specific heat and compressibility , demonstrate similar characteristics . . inset : the relative variation of thermal expansivity . , width=3 ] the picture of pseudo - phase of defects is only valid when the dominating region of the associated defect type is well defined . for point defects at low temperature , it is usually the case . however , with elevated temperature and/or when complex defect clustering is involved , competition might lead to coexistence of different defect species , where the notation of pseudo phase could lose its physical importance . for realistic uo@xmath0 , pseudo - phase can always be defined , whether oxygen clusters ( _ e.g. _ , cot - o cluster ) involved or not . at high temperatures , the zone of pseudo phase boundary becomes wide , and renders the ppt as a smooth crossover.@xcite in spite of this , the material behavior still can be understood within the framework of pseudo phase . in some conditions , however , a situation that one defect species appears but never gains the dominant role might be possible . this will completely invalidate the picture of pseudo phase , and a detailed analysis of defect concentrations becomes necessary ( in contrast to this , the defect concentrations are determined by stoichiometry deviation if pseudo phase can be applied ) . , see text for details . , width=3 ] an example of this is presented in fig.[fig : coext ] , which was generated with jpc s formation energy data.@xcite in addition to that , the value of cot - o cluster@xcite was also reduced by 4.865ev , namely , we have artificially decreased the stability of oxygen clusters . the result is that oxygen interstitial o@xmath64 gets some promotion , but not enough to become the major one , and persists only within a narrow pressure range . it is evident from fig.[fig : coext ] that in this virtual system u@xmath64 and cot - o ( at least for most pressure range ) can be depicted using pseudo phase , and there is a ppt from u@xmath64 to cot - o at a pressure of 7gpa . on the other hand , o@xmath64 , which appears from 20 to 50gpa , depletes the concentration of cot - o slightly . but it never gains the dominance , and the phenomenon must be taken as a defect coexistence rather than being a ppt . to understand the difference between these is crucial for a correct description of defective materials using the concept of pseudo phase . above discussion is based on the approximation of pdm , which is valid at the dilute limit . with an increase of defect concentrations , however , interaction between defects becomes significant . this could lead to a severe deviation from the prediction of pdm . on the other hand , non - stoichiometry might affect the eos and energetics of defects , thus modify the ppd . to obtain the correct defect behavior near the stoichiometry , a close cooperation between experimental measurements and theoretical analysis is necessary . for that purpose , a thorough and comprehensive understanding of defect interaction is crucial . this is because with such an accurate information , we can deduce the formation energy at dilute limit from the measured ppd at finite stoichiometry deviations . it is worthy of mentioning that the same extraction procedure can be done using the simple pdm , but the resultant error is usually quite large . in this part of the paper , we will first derive a general formalism that includes defect interactions , then we will show that the conventional pdm can be adapted to take these interactions into account effectively with a simple constant mean - field approximation , which greatly expands the applicability of pdm . if one is interested mainly in macroscopic properties of a defective material that averaged over a long enough time scale , the most important contribution comes from defects in static distributions . for a classical crystal at finite temperature , defects have nonzero probability of migrating between available sites . this alters the dynamics and transport property of the system . nevertheless , diffusion does not modify the overall macroscopic properties very much as long as the distribution of defects is still in an equilibrium state . this is because migration is a transient process and in most time defects are trapped in their equilibrium sites . in a migration process , defects induce dynamic deformations in the local lattice , and scatter with phonon . this effect might drive phonon away from its equilibrium distribution . however , considering that this change is very small compared to thermal fluctuations , and our interested time scale is many orders longer than the relaxation time of phonon , it is justified to ignore this effect . the main assumptions in the theory include : ( a ) we work on a lattice model , ( b ) only static distribution of defects is considered , ( c ) the distribution must be homogeneous , ( d ) dynamic effects due to defect migration are ignored , ( e ) no defect creation and annihilation are considered , ( f ) when defect clusters are involved , taking them as single objects , namely , ignore association and/or dissociation effects . with these assumptions , let us consider a grand canonical ensemble on a lattice in which the number of particles fluctuating around the average value . assume there are in total @xmath65 sublattices , which are occupied by @xmath15 species of particles . using @xmath66 denotes the number of @xmath67-th kind of particle that siting on the @xmath1-th sublattice , where @xmath1 can take any value from 1 to @xmath65 and @xmath67 varies from 1 to @xmath15 . it is evident that each system in the ensemble is characterized by the occupation of particles on the lattice , and thus the ensemble can be grouped according to @xmath68 . that is to say , each system can be labeled uniquely by a set of @xmath68 together with an auxiliary index @xmath69 , where @xmath69 runs over all possible configurations that has the same @xmath68 . in this way , if using @xmath70 denotes the probability of configuration @xmath71 in the ensemble , where @xmath71 has been shortened as @xmath72 for brevity , then the gibbs free energy of the ensemble is @xmath73 . \label{eq : gibbs1}\ ] ] here @xmath74 is the gibbs free energy of an individual configuration , and the contribution of configurational entropy has been separated out and presents as the second term in eq.([eq : gibbs1 ] ) . the thermodynamically equilibrium state is achieved when @xmath75 takes a minimum . the normalization condition requires @xmath76 . if @xmath77 denotes the total number of the @xmath67-th particle in the ensemble , then @xmath78 . introducing lagrange multipliers @xmath79 and @xmath80 , the minimization equation of @xmath75 becomes @xmath81 making use of eq.([eq : gibbs1 ] ) , this leads to @xmath82 where the partition function @xmath83 . it can be shown that @xmath79 is just the chemical potential of the @xmath67-th type of particle . the free energy of the @xmath72-th configuration , @xmath84 , which consists of the cold crystal energy at 0k and phonon contributions to the internal energy and entropy ( as well as a term of @xmath85 , if at finite pressure ) , can be calculated with modern first - principles methods . since defects in an individual configuration manifest as an imperfect occupation of the lattice sites , the equilibrium thermodynamics of defects is therefore completely described by eq.([eq : rho ] ) , in which the contribution of defect interactions arises from @xmath84 . when the energy scale of temperature is much smaller than the formation energy of defects , which usually is the case for most applications , the partition function is dominated by the perfect occupation ( _ i.e. _ , the ground - state ) , namely @xmath86 , where the subscript 0 denotes the groundstate . in this situation eq.([eq : rho ] ) reduces to @xmath87 where the fge of configuration @xmath72 is given by @xmath88 . for single point defect @xmath89 . thus the fge of intrinsic defects such as frenkel pairs and schottky defect can be derived straightforwardly from eq.([eq : rhoq ] ) , and have the same form as eqs.([eq : gfp]@xmath32[eq : gs ] ) . furthermore , equations similar to pdm [ eqs.([eq : pdm1]@xmath32[eq : pdm3 ] ) ] also can be constructed easily using eq.([eq : rhoq ] ) , indicating that the simple formalism of pdm is much more flexible than what it was originally proposed . it is necessary to point out that for a system defined on a lattice , there is an orthogonal and complete basis set called correlation functions , which are an alternative but powerful representation of all possible occupation configurations of particles on the lattice using increasingly complex point sets that ranging from single point to nearest pairs and bigger clusters.@xcite any functions defined on the lattice can be expanded with such a basis . for defective crystals , the matrix and interstitial sites define such a lattice naturally . if we further introduce a special species of _ white atom _ , and regard all vacancies as that occupied by white atoms , then such orthogonal and complete correlation functions @xmath90 can be defined . in this way , the probability for a configuration to appear becomes @xmath91 and the gibbs function of eq.([eq : gibbs1 ] ) can be rewritten as @xmath92 with the configurational entropy given by @xmath93 $ ] , and the interaction strength of cluster @xmath94 . it is obvious that eq.([eq : g_alloy ] ) has the same form as the theory for alloys.@xcite namely , both alloying and defects on a lattice can be described by the same unified theoretical framework . within this method , the interaction strength of clusters can be evaluated with cluster expansion method using _ ab initio _ total energy calculations,@xcite and the configurational entropy may be evaluated by cluster variation method.@xcite for a point defect on an infinitely large lattice ( the dilute limit ) , the formalism discussed above naturally leads to the conventional pdm.@xcite when defect concentration is finite but interaction between them is weak and can be ignored safely , the same conclusion holds . this is because for a configuration containing @xmath95 non - interacting point defects , the corresponding fge is just @xmath95 times of that of a single one , and then eq.([eq : rhoq ] ) gives rise to @xmath96 , which is exactly the result of pdm . alternatively , for non - interacting defects , the presence of a defect has no influence on others , we therefore can isolate a defect by cutting it and the associated local lattice out from the matrix , and then extend the surrounding lattice to an infinite range . this operation keeps the defective behavior . it maps non - interacting defects with finite concentrations onto a group of systems at the dilute limit , which justifies the application of pdm . when interactions between defects are substantial , it is almost impossible to isolate a defect from others . at a condition that the distribution of defects is homogeneous , however , we can approximate the interaction by a constant mean - field . in this approximation , the defects and the associated local lattice environment that are cut out from the matrix keep the original size , and subject to a field that takes a role of modeling the interactions with other homogeneously distributed defects that have been removed . to construct accurately an environmental field of such kind is difficult , if not impossible . for practical purpose , we may simplify it by using regularly distributed defects to simulate the field approximately . this sacrifices the rigidity of the theory , but makes the problem more tractable . what one needs to do now is to periodically repeat the piece of defective lattice that has been cut out from the matrix along the three dimensional lattice vectors . in this way the periodical images of the defect take the role of modeling the homogeneous environment produced by other defects . just one such configuration of course can not capture the whole features of the defect interactions . by averaging over all possible regular defective distributions , nevertheless , one eventually can reach a converged result . generally the free energy of a system in the ensemble can be written as @xmath97 where @xmath35 is the concentration of defect @xmath1 in this system . if only up to linear terms are kept , then eqs.([eq : gibbs1 ] ) and ( [ eq : rhoq ] ) reduce back to the conventional pdm . the terms of higher order describe effective defect interactions , and should be important for any real materials with high defect density . on the other hand , if we know the values of parameter @xmath98 , @xmath99 , @xmath7 , and @xmath100 , the free energy ( as a function of defect concentration @xmath35 ) of any defective configuration can be evaluated from eq.([eq : gibbs2 ] ) directly . to determine these parameters , one can solve eq.([eq : gibbs2 ] ) by a least square fit method using _ ab initio _ calculated @xmath84 of a set of configurations . furthermore , since the @xmath50 in eq.([eq : rhoq ] ) can be rewritten as @xmath101 where the first term at the right hand side represents the contribution of non - interacting defects , and the second term arises from defect interactions , eq.([eq : rhoq ] ) then leads to @xmath102 where @xmath103 is the probability predicted by conventional pdm which gives a defect concentration of @xmath104 ( determined by @xmath105 ) . for example , if there are @xmath106 defects of @xmath1-th type appearing in the configuration @xmath72 , where @xmath107 and @xmath108 is the total available sites for that defect , pdm gives @xmath109 . because of the structure of @xmath110 as shown in eqs.([eq : gibbs2 ] ) and ( [ eq : dg ] ) , @xmath70 can be factorized into the same form as @xmath111 . namely , @xmath112 we finally get @xmath113 this explicitly demonstrates that the simple formalism of pdm is still valid even when defect interactions present , as long as the defect distribution is homogeneous . the effect of interaction is to modify the defect concentrations in a constant mean - field way ( here _ constant _ means that the interaction has been averaged over the whole configurational space so that no dependence on the distance between defects presents explicitly ) , and the formation energy of a single point defect @xmath114 has to be changed from its dilute limit value @xmath115 to @xmath116 and the defect concentration equations also become @xmath117 .[tab : cohee ] first - principles results for the energy curve of uo@xmath118 , where @xmath29 is the deviation from the stoichiometric composition of uranium dioxide , @xmath24 is the total number of atoms in the simulation cell . @xmath16 ( in ev ) , @xmath119 ( in @xmath120 ) , and @xmath17 ( in gpa ) are the cohesive energy per atom , the equilibrium lattice parameter of the effective cubic cell , and the zero pressure bulk modulus , respectively . [ cols="<,^,^,^,^,^,^,<",options="header " , ] according to ref . and above discussions , pseudo phases in uo@xmath0 are well defined : at most interested thermodynamic conditions only one type of defect presents , and all other components are suppressed completely . this implies that only the diagonal terms in eq.([eq : gibbs2 ] ) make sense . namely , only interactions between the same kind of defect need to be considered , which greatly reduces the number of _ ab initio _ calculations that are required for extraction of the interaction strengths @xmath7 and @xmath100 . using configurations with different simulation cell size of @xmath121 , @xmath122 , and @xmath123 of the cubic fluorite unit , we extracted the defect interaction strength by solving a set of equations of eq.([eq : gibbs2 ] ) . the employed energy curves for the smallest cell are listed in table [ tab : cohee ] ( in which the defective structures are labeled follwoing the same rule of ref . ) , and others are taken from table i in ref .. in particular , the results in table [ tab : cohee ] were calculated using vasp code , with the same lsda+__u _ _ setting as in ref .. all structures were fully relaxed at a series of fixed volumes . since the supercell size of these structures is relatively small , 36 irreducible @xmath124 points were employed to ensure the total energy convergency . with the effective pdm of eq.([eq : gpdm ] ) , it is not necessary to work on the ensemble average of eq.([eq : gibbs1 ] ) any longer . instead , the problem changes to `` how the effectively independent defects distribute on the lattice '' . for a purpose of investigating the compression behavior of defects , it is helpful to employ a reference supercell , and normalize all involved energetics with respect to it . in doing so , however , the number of defects might no longer be an integer . a re - scaling procedure is thus required when evaluating the formation energy of a _ single _ defect . let @xmath125 be the defect contribution in eq.([eq : gibbs2 ] ) that is evaluated in the reference cell , _ i.e. _ @xmath126 , with a defect concentration equals @xmath127 . the number of unit cells in a supercell which contains _ one _ and _ only one _ of this type of defect is @xmath128 , where @xmath129 is the number of available sites for this defect in a unit cell . then the energy difference for creating a defect is @xmath130 , where @xmath131 is the number of unit cells making up the reference supercell . in this way , the formation energy of a frenkel pair for @xmath132 species becomes @xmath133 and the schottky defect formation energy is @xmath134 here @xmath24 is the total number of atoms in a defect - free reference cell . for a @xmath121 supercell of fluorite uo@xmath0 , @xmath135 and @xmath136 . also for a cubic fluorite unit , @xmath129 takes 8 for o@xmath137 , and 4 for o@xmath64 , u@xmath137 , and u@xmath64 , respectively . these formulations , together with eq.([eq : df_int ] ) , allows for extrapolating the intrinsic defect formation energy to the dilute limit by decrease the defect concentration to an arbitrarily small value . the obtained results are summarized in table [ tab : ef ] . we can see that a @xmath121 cell is not big enough to converge the formation energy to the dilute limit . the deviation from the extrapolated value is less than 1ev . the largest one is oxygen frenkel pair in which @xmath138 reaches a value of 0.6ev . it is at the same level of the finite - size correction of charged defects,@xcite where a value of about 0.6ev was also obtained for o_fp . this good agreement demonstrates that our treatment on defect interaction is at least qualitatively correct . ) and compressibility ( @xmath139 ) with stoichiometry deviation @xmath29 at different level of approximations . notice the sharp tips at @xmath140 due to defect interactions . , width=3 ] inclusion of defect interactions into the pdm makes it possible to study the fine behavior over the whole stoichiometry . figure [ fig : alpha ] shows the relative variation of thermal expansivity @xmath63 and compressibility @xmath139 as a function of stoichiometry deviation @xmath29 at 300k and 0gpa , in which the solid points denote the exact value of the configurations that were employed to extract the interaction strength . it is evident that the linear approximation of eq.([eq : gibbs2 ] ) ( namely the conventional pdm ) fails to reveal the fine behavior of non - stoichiometric uo@xmath0 . far from the point that was used to approximate the defect formation energy , it deviates from the exact value drastically . on the other hand , both quadratic and cubic approximations predict a curved variation of physical quantities correctly . it is interesting to note that the sharp tip appearing at the stoichiometry is very similar to the `` w '' shape anomaly in alloys.@xcite nevertheless , the underlying physics is different . here it mainly originates from two facts : ( a ) the curvature due to defect interactions , and ( b ) the predominant defect type at the hyper- and hypo - stoichiometry sides are different , which gives rise to different variation trends of the physical quantities . from this perspective , measuring the departure of relevant physical quantities from a linear behavior near the stoichiometry would reveal the strength of defect interactions , and constrain the application range of the conventional pdm . above discussions revealed the power of including defect interaction into statistical mechanics model such as pdm . though by comparison with available experimental data one can assess the validity of our approach , due to the uncertainty of these data as mentioned in sec.[sec : intr ] and [ sec : pdm ] , a quantitative validation has to be made with direct _ ab initio _ calculations . to this end , we made an independent calculation on a defective configuration of @xmath141 , using a supercell of @xmath142 of the cubic fluorite unit , in which one oxygen atom has been removed to create one oxygen vacancy . the obtained energetics are listed in table [ tab : cohee ] , and the calculated thermal expansivity and compressibility are compared with that of effective pdm in fig.[fig : alpha ] as the open square points . note that in this separate calculation we employed only the experimental observation that oxygen vacancy prevails in uo@xmath143 , and did not invoke any other approximations . therefore the good agreement between these two results as shown in fig.[fig : alpha ] provides a solid verification of the validity of our constant mean - field treatment of defect interactions for homogeneously distributed defects , as well as the effective pdm that based on it . uo@xmath0 is a semiconductor with a finite energy gap , it is possible that defects in it are charged rather than being neutral . except very few studies,@xcite most _ ab initio _ investigations on defective uo@xmath0 assumed a neutral simulation cell , as we did here . in principle , such `` neutral '' calculations do not correspond to literally neutral defects , since local transfer of electrons might lead to a partial charge of the defects . nevertheless , finite size of the simulation cell imposes a constraint on the charge redistribution , and thus defect might not reach its full charge state . this problem becomes very severe when at the dilute limit or near the stoichiometry , where defects can be fully charged only by exchanging electrons with valence / conduction bands , which is a kind of global charge redistribution . for large stoichiometry deviation , however , defects interact with the matrix strongly , leading to deep defect levels which are either at near the gap middle or hybridization with valence / conduction bands . in both cases a neutral defect could be expected because it is difficult to _ ionize _ the defect at low temperatures in the first case and no charged defect can be supported in the second one . for oxygen clusters , crocombette argued that charged state makes them unfavorable at the stoichiometry.@xcite it is reasonable . actually , `` neutral '' calculations also indicated that oxygen clusters have negligible concentrations when near the stoichiometry.@xcite putting these information together , we ascertain that it should have no defect clustering when @xmath144 . but this does not mean that oxygen clustering is negligible at non - stoichiometry . the experimental evidence for such clustering was in fact observed at large values of @xmath29,@xcite which is compatible with recent neutral calculations that predicted prevailing cot - o clusters at hyper - stoichiometry region . on the other hand , _ ab initio _ electronic structure revealed that the defect levels of cot - o cluster hybridize strongly with the valence band of the matrix,@xcite which implies that the cluster might be `` neutral '' , or at least these `` neutral '' results should partially reflect some physical reality . in these considerations , crocombette s conclusion about charged oxygen defects@xcite might lose the relevance when far from the stoichiometry . but overall the charged state of defects in uo@xmath0 is still an open issue . in the derivation of the general formalism of defects in sec.[sec : defect interactions ] , we did not consider the charged state . to include this is straightforward . one just needs to add an additional index to each type of defect to mark its charge state , and include the chemical potential of free electrons to take the charge contribution at the fermi level into account . eq.([eq : gibbs1 ] ) then becomes @xmath145 , \label{eq : g_charge}\ ] ] where @xmath146 is the total charge of the system , and @xmath147 the fermi level . from this expression , the effective pdm which includes both defect interactions and variable charge state of defects can be derived easily . . , width=3 ] although ppt and the corresponding boundaries can be employed to constrain / extract the dilute limit of defect formation energy an important quantity for understanding the stoichiometric behavior , to measure these boundaries is not a trivial work . at big stoichiometric deviation , the volume change at ppt is prominent , thus it can be detected by measuring the quasi - discontinuous jumps in the eos of defects.@xcite figure [ fig : pv ] plots a compression curve of uo@xmath148 along a hugoniot shocked from 300k and 0gpa . the volume collapse due to ppt from u@xmath137 to o@xmath64 ( here we ignored oxygen clustering ) is evident and therefore detectable . on the other hand , this pseudo transition pressure depends sensitively on the formation energy of @xmath44 : it spans over a wide range of 40gpa when there is a change in @xmath149 about 4.0ev . this property guarantees a good precision for the constraints on defect energetics . at small value of @xmath29 , however , the volume jump would be too weak to be perceptible . this is usually the case when @xmath150.@xcite in these cases , we can not locate the ppt via measuring thermodynamic or mechanical quantities . however , since ppt changes the predominant defect species and thus the position of the defect level within the energy gap , transport properties are also modified . we therefore can detect the occurring of a ppt by measuring the sudden changes in electrical conductivity ( or optical properties).@xcite this method has high sensitivity so that allows us to access the vicinity of the stoichiometry . ) , isothermal bulk sound velocity , and specific heat at constant pressure ( @xmath151 ) along a hugoniot that shocked from 500k and 0gpa in non - stoichiometric uo@xmath0 when defect concentrations have an instantaneous response to thermal fluctuations . the solid circle ( at the bottom left corner ) marks the experimental bulk sound velocity measured at ambient conditions in perfect uo@xmath0 . , width=3 ] in above discussions and also in ref . , we froze the defect concentrations when evaluating the thermodynamic quantities . it is a theoretical requirement for the first derivatives of the gibbs function , such as volume and entropy . but for higher order derivatives of the gibbs function , it has no reason to do so because they are also defined by thermodynamic relations . in practice , however , a justification for this operation can be made . this is because for a classical crystal , the change of defect species can proceed via only atomic diffusion , which is a very slow process , and thus no defect can respond to rapid thermal fluctuations . then an interesting question arises , that is what about if defects can instantaneously respond to any disturbances ? simple analysis shows it might be fantastic . at first the magnitude of anomalies due to ppt would be amplified greatly , thus ease the difficulty in ppt detections . figure [ fig : cpp ] demonstrates this effect on the thermal expansivity @xmath63 , bulk sound velocity , and specific heat at constant pressure @xmath151 . the influence can be fully comprehended by comparing with the fig.2 in ref . , where defect concentrations were fixed when evaluating these quantities . secondly , the compressibility would diverge at zero temperature . considering the relationship between sound velocity and the compressibility , this implies a vanishing sound velocity ( and the bulk modulus ) in the vicinity of a ppt at low temperatures if defects have instantaneous response . on the phonon spectrum , it would manifest as an abnormal softening in acoustic branches at long - wave length ( _ i.e. _ , @xmath152 point in the reciprocal space ) . this observation is tantalizing . but can it be true ? we can not answer it yet . classical atomic diffusion of course can not lead to a rapid response . but what if for a quantum crystal ? in a quantum world particles are described by a wave - function . if the wavefunctions of all defects are in a coherent state , then an instantaneous and simultaneous change of the defect species might be possible.@xcite in this mechanism the requirement for a large scale atomic diffusion has been removed , and a single change in defecton state is enough for a quantum pseudo transition to occur . nevertheless , much more theoretical works are necessary in order to pin down this possibility definitely . a general formalism for the thermodynamics of defects in a crystal was derived based on the statistics of a grand canonical ensemble on a lattice . by introducing idle white atoms for vacancy and extending the sublattice of interstitial sites , this formalism has the same form as the lattice theory for alloys and compounds a reflection of the unified physics underlying these seemingly different systems . with an approximation of constant mean - field , this theory reduces to an effective point defect model in which defect interactions are included by an auxiliary field , whereas each individual defects are treated as independent . in this way , we mapped a _ many - body _ defect system onto a _ single _ defect system by coupling it with an effective external field . if ignoring this field , the conventional pdm is recovered . this generalization greatly expands the applicability of the simple pdm . in order to explore the full content in this theory , we also studied possible reentrant ppt and multi - defect coexistence with virtual systems . using pdm , we investigated the possibility of constraining defect energetics by measuring pseudo phase boundaries . by calculating the possible ppt between interested defects , we showed that the experimental estimates available in literature , as well as variety theoretical assessments , on defective energetics of uo@xmath0 , are not fully consistent . on the other hand , the range of energetics constrained by the ppts overlaps with these estimates largely , therefore has a potential to reduce the inconsistency in these data . by including defect interactions into pdm , we demonstrated that the information obtained at finite stoichiometry deviation can be extrapolated to the dilute limit . finally , we investigated the fine behavior of thermal expansivity and compressibility in the vicinity of the stoichiometry of defective uo@xmath0 , and some relevant issues of charged defects , detection of ppt , and possible instantaneous response of defectons in a quantum crystal are briefly discussed , in which we highlighted the detection of ppt by measuring electrical conductivity when near the stoichiometry , and the complexity arising from possible charged state of defects . through these investigations , we clearly demonstrated that it is valuable to explore the whole non - stoichiometric range in order to acquire a comprehensive understanding about a defective material thoroughly . 99 f. agullo - lopez , c. r. a. catlow , and p. d. townsend , _ point defects in materials _ ( academic press , london , 1988 ) . o. t. srensen , _ non - stoichiometric oxides _ ( academic , new york , 1981 ) matzke , j. chem . faraday trans . 2 * 83 * , 1121 ( 1987 ) . g. e. murch and c. r. a. catlow , j. chem . faraday trans . 2 * 83 * , 1157 ( 1987 ) . t. g. stratton and h. l. tuller , j. chem . faraday trans . 2 * 83 * , 1143 ( 1987 ) . r. tetot and p. gerdanian , j. phys . solids * 46 * , 1131 ( 1985 ) . k. clausen , w. hayes , j. e. macdonald , r. osborn , and m. t. hutchings , phys . lett . * 52 * , 1238 ( 1984 ) . j. p. crocombette , f. jollet , t. n. le , and t. petit , phys . b * 64 * , 104107 ( 2001 ) . l. petit , a. svane , z. szotek , w. m. temmerman , and g. m. stocks , phys . rev . b * 81 * , 045108 ( 2010 ) . s. l. dudarev , g. a. botton , s. y. savrasov , z. szotek , w. m. temmerman , and a. p. sutton , phys . status solidi a * 166 * , 429 ( 1998 ) . i. d. prodan , g. e. scuseria , and r. l. martin , phys . b * 73 * , 045104 ( 2006 ) ; * 76 * , 033101 ( 2007 ) . h. x. song , h. y. geng , and q. wu , phys . rev . 064110 ( 2012 ) . h. y. geng , y. chen , y. kaneta , and m. kinoshita , phys . b * 75 * , 054111 ( 2007 ) . h. y. geng , h. x. song , k. jin , s. k. xiang , and q. wu , phys . rev . b * 84 * , 174115 ( 2011 ) . y. geng , h. x. song , and q. wu , phys . b * 85 * , 144111 ( 2012 ) . f. gupta , a. pasturel , and g. brillant , phys . b * 81 * , 014110 ( 2010 ) . d. a. andersson , t. watanabe , c. deo , and b. p. uberuaga , phys . b * 80 * , 060101(r ) ( 2009 ) . d. a. andersson , f. j. espinosa - faller , b. p. uberuaga , and s. d. conradson , j. chem . phys . * 136 * , 234702 ( 2012 ) . d. a. andersson , j. lezama , b. p. uberuaga , c. deo , and s. d. conradson , phys . b * 79 * , 024110 ( 2009 ) . h. y. geng , y. chen , y. kaneta , m. iwasawa , t. ohnuma , and m. kinoshita , phys . rev . b * 77 * , 104120 ( 2008 ) . h. y. geng , y. chen , y. kaneta , and m. kinoshita , phys . b * 77 * , 180101(r ) ( 2008 ) . h. y. geng , y. chen , y. kaneta , and m. kinoshita , appl . phys . lett . * 93 * , 201903 ( 2008 ) . m. freyss , t. petit , and j. p. crocombette , j. nucl . * 347 * , 44 ( 2005 ) . m. iwasawa , y. chen , y. kaneta , t. ohnuma , h. y. geng , and m. kinoshita , mater . 47 * , 2651 ( 2006 ) . b. dorado , g. jomard , m. freyss , and m. bertolus , phys . b * 82 * , 035114 ( 2010 ) . j. yu , r. devanathan , and w. j. weber , j. phys . : condens . matter * 21 * , 435401 ( 2009 ) . f. gupta , g. brillant , and a. pasturel , philos . mag . * 87 * , 2561 ( 2007 ) . g. brillant and a. pasturel , phys . rev . b * 77 * , 184110 ( 2008 ) . d. a. andersson , b. p. uberuaga , p. v. nerikar , c. unal , and c. r. stanek , phys . b * 84 * , 054105 ( 2011 ) . p. nerikar , t. watanabe , j. s. tulenko , s. r. phillpot , and s. b. sinnott , j. nucl . mater . * 384 * , 61 ( 2009 ) . j. p. crocombette , d. torumba , and a. chartier , phys . b * 83 * , 184107 ( 2011 ) . j. p. crocombette , phys . b * 85 * , 144101 ( 2012 ) . p. larson , w. r. l. lambrecht , a. chantis , and m. van schilfgaarde , phys . b * 75 * , 045114 ( 2007 ) . g. jomard , b. amadon , f. bottin , and m. torrent , phys . b * 78 * , 075125 ( 2008 ) . b. dorado , b. amadon , m. freyss , and m. bertolus , phys . b * 79 * , 235125 ( 2009 ) . h. y. geng , y. chen , y. kaneta , m. kinoshita , and q. wu , phys . b * 82 * , 094106 ( 2010 ) . v. i. anisimov , j. zaanen , and o. k. andersen , phys . b * 44 * , 943 ( 1991 ) . a. i. liechtenstein , v. i. anisimov , and j. zaanen , phys . b * 52 * , r5467 ( 1995 ) . s. l. dudarev , g. a. botton , s. y. savrasov , c. j. humphreys , and a. p. sutton , phys . b * 57 * , 1505 ( 1998 ) . f. zhou and v. ozolins , phys . b * 80 * , 125127 ( 2009 ) . f. zhou and v. ozolins , phys . b * 83 * , 085106 ( 2011 ) . h. y. geng , m. h. f. sluiter , and n. x. chen , phys . b * 72 * , 014204 ( 2005 ) . v. l. moruzzi , j. f. janak , and k. schwarz , phys . b * 37 * , 790 ( 1988 ) . h. serizawa , y. arai , m. takano , and y. suzuki , j. alloys compd . * 282 * , 17 ( 1999 ) . h. serizawa , y. arai , and y. suzuki , j. nucl . mater . * 280 * , 99 ( 2000 ) . h. y. geng , n. x. chen , and m. h. f. sluiter , phys . b * 71 * , 012105 ( 2005 ) . m. sluiter , d. de fontaine , x. q. guo , r. podloucky , and a. j. freeman , phys . b * 42 * , 10460 ( 1990 ) . m. h. f. sluiter , y. watanabe , d. de fontaine , and y. kawazoe , phys . rev . b * 53 * , 6137 ( 1996 ) . j. w. d. connolly and a. r. williams , phys . b * 27 * , 5169 ( 1983 ) . r. kikuchi , phys . rev . * 81 * , 988 ( 1951 ) . b. t. m. willis , proc . . soc . * 1 * , 9 ( 1964 ) . b. t. m. willis , acta crystallogr . , a : cryst . crystallogr . * a34 * , 88 ( 1978 ) . a. d. murray and b. t. m. willis , j. solid state chem . * 84 * , 52 ( 1990 ) .

the low - energy structures of mixed ar xe and kr xe lennard - jones clusters are investigated using a newly developed parallel monte carlo minimization algorithm with specific exchange moves between particles or trajectories . tests on the 13- and 19- atom clusters show a significant improvement over the conventional basin - hopping method , the average search length being reduced by more than one order of magnitude . the method is applied to the more difficult case of the 38-atom cluster , for which the homogeneous clusters have a truncated octahedral shape . it is found that alloys of dissimilar elements ( ar xe ) favor polytetrahedral geometries over octahedra due to the reduced strain penalty . conversely , octahedra are even more stable in kr xe alloys than in kr or xe , and they show a core - surface phase separation behavior . these trends are indeed also observed and further analysed on the 55-atom cluster . finally , we correlate the relative stability of cubic structures in these clusters to the glassforming character of the bulk mixtures .

physics0403048

clusters of heterogeneous materials show a much richer behavior than their homogeneous counterparts . in many bulk compounds , doping can significantly affect some global property , and alloying is a common way to tailor a completely new kind of material . at the mesoscale level , size is another complicating factor , giving rise to further changes with respect to the macroscopic object . to a large extent , most expectations of nanotechnology have been put into the electronic and catalytic properties of small atomic clusters . therefore , it should not be surprising that numerous theoretical studies of mixed clusters were devoted to bimetallic clusters . in particular , there has been a significant amount of work at the level of sophisticated electronic structure calculations,@xcite but these were often limited to small sizes due to the numerical effort involved . on a different scale of chemical complexity , many studies have been carried out using explicit , empirical force fields in order to investigate the segregation properties of these clusters . there are several driving forces toward mixing or segregation in binary systems : * the difference in atomic sizes ; * the difference in surface energies ; * minimization of the overall strain ; * the number of interactions between unlike atoms . these factors can often compete with each other . for instance , minimizing surface energies does usually not increase with the number of interactions between different atoms . also , even though this is not our prime interest here , it should be noted that kinetic factors can be crucial in this problem.@xcite in particular , vach and coworkers have found from experiments and simulations of mixed rare - gas clusters that some anomalous enrichment effects could be observed due to the growth by pick - up of these systems.@xcite very recently , radial segregation and layering have been observed in large ar / xe clusters formed in an adiabatic expansion by tchaplyguine _ et al._@xcite using photoelectron spectroscopy measurements . these data have also been theoretically interpreted by amar and smaby.@xcite fortunately , mixed rare - gas systems can be quite safely described using simple pairwise potentials such as the lennard - jones ( lj ) potential . more accurate potentials are of course also available , even though we will have no need for them in the present , mostly methodological work . hence they are much more convenient to study in a broad size range , not only for their structure but also their dynamics or thermodynamics . it is known from previous studies that the topography of the potential energy surfaces of homogeneous lj clusters can be very peculiar , as for the sizes 38 or 75.@xcite the multiple - funnel structure of these energy landscapes makes it especially hard to locate the most stable structures ( global minima ) or to simulate the finite - temperature behavior of these clusters in an ergodic way . the effects of mixing different rare - gas atoms on cluster structure and thermodynamics have been studied for the specific size 13 by frantz on the examples of ar kr mixtures@xcite as well as ne ar mixtures.@xcite fanourgakis _ have also investigated these latter compounds.@xcite a systematic work of ar xe mixed clusters of 13 and 19 atoms has been carried out by munro and coworkers,@xcite including some global optimization and monte carlo simulations . mixed clusters involving lighter species such as h@xmath1 and d@xmath1 have been investigated using path - integral monte carlo simulations ( pimc ) by chakravarty.@xcite more recently , sabo , doll and freeman reported a rather complete study of the energy landscapes@xciteand melting phase change@xcite in mixed ar ne clusters . in this work quantum delocalization and the effects of impurities on cluster properties were also accounted for using pimc techniques . the main conclusion of these studies is that atomic heterogeneity can be responsible for a drastic increase in complexity of the energy landscapes of rare - gas clusters . this complexity is manifested by numerous new low - lying minima in competitive funnels , characterized by the same overall geometrical arrangement but different permutations of unlike atoms . following jellinek and krissinel,@xcite we will refer to such isomers as `` homotops . '' the presence of several homotops on a given energy landscape often induces solid - solid transitions , which can be detected by some feature in the heat capacity,@xcite even though they can be washed out by quantum effects.@xcite as shown by munro _ et al._,@xcite the various funnels corresponding to different homotops of a same geometry are separated by significant energy barriers . this explains the difficulty or even failure of simulation methods to achieve ergodic sampling of these systems , albeit small.@xcite a similar situation is found in lennard - jones polymers,@xcite where a large number of isomers are based on the same geometrical arrangement , differing only in the path linking the monomers . beyond the actual rare - gases , binary lennard - jones compounds have been investigated in both the cluster and bulk regimes . clarke and coworkers looked at phase separation of small particles with equal compositions.@xcite based on monte carlo simulations , they sketched a phase diagram in the general structure of liquid clusters . bulk binary lennard - jones systems have been seen to provide relatively simple numerical models for glass formation.@xcite most often , the lj interactions in such studies have been tuned in a non - additive way in order to hinder crystallisation . in another related work , lee and coworkers@xcite have investigated the role of atomic size ratio in binary and ternary metallic alloys . interestingly , severaly links between the physics and chemistry of clusters and those of supercooled liquids and glasses have been established since the pioneering work by frank.@xcite the initial suggestion that the local order in simple liquids is not crystalline but icosahedral@xcite ( more generally polytetrahedral ) has since been verified experimentally@xcite and theoretically.@xcite from the clusters viewpoint , the favored finite - size structures of good model glassformers have been shown by doye and coworkers to be polytetrahedral.@xcite the 38-atom homogeneous lennard - jones cluster is known to show some glassy properties , especially slow relaxation to the ground state,@xcite due to the competition between two stable funnels on the energy landscape , corresponding to truncated octahedral and icosahedral shapes , respectively . due to entropic effects,@xcite a solid - solid transition occurs between the two funnels , at temperatures lower than the melting point . the crystal - like configuration of this cluster makes it a good candidate to further investigate the relationship between cluster structure and criteria for glassification . because homogeneous lj@xmath0 constitutes a relatively difficult task for global optimization algorithms , binary clusters of the same size can be expected to be much worse . in this paper , we propose a simple but efficient way to deal with the multiple new minima introduced by unlike atoms within a general monte carlo global minimization scheme . this algorithm will then be applied to the 38- and 55-atom cases , in mixtures of xe with either ar or kr atoms . in the next section , we present the method and test it on the simple cases of the 13- and 19-atom clusters . in sec . [ sec : res ] we give our results obtained at sizes 38 and 55 and we correlate them to the different glassforming abilities of the bulk mixtures . we finally conclude in sec . [ sec : ccl ] . global optimization of cluster structure@xcite is currently best achieved using either genetic algorithms@xcite or the monte carlo+minimization method,@xcite also known as basin - hopping ( bh).@xcite the case of homogeneous lennard - jones clusters is among the most documented of cluster physics , and an up - to - date table of putative global minima can be found in ref . . even though it can never be guaranteed that global minimization has been successful , it is likely that all important structural forms of lj clusters have been found up to more than 100 atoms . these include icosahedral , truncated octahedral , decahedral as well as tetrahedral arrangements . compared to homogeneous clusters , the available data on heterogeneous systems is rather scarce . besides the specific works by frantz on the 13-atom ne ar and kr ar clusters,@xcite munro _ et al . _ used a parallel version of the bh scheme , similar to the replica - exchange monte carlo method,@xcite where several trajectories are run simultaneously at various temperatures.@xcite although these authors looked at moderately large clusters , they reported significant difficulties to locate global minima at specific compositions , as in xe@xmath2ar@xmath3 or xe@xmath4ar@xmath5 , for instance.@xcite a natural problem occuring using the basin - hopping method is that many of the low - lying minima are expected to be related to each other via particle exchange . such a process only occurs via large deformations of the remaining cluster , hence it is quite unprobable . as in condensed matter physics,@xcite allowing exchange moves between particles as a possible monte carlo step may result in notably faster convergence , provided that the interactions are not too dissimilar . actually , optimization of mixed clusters on the lattice formed by the homogeneous system has already been studied by robertson and coworkers.@xcite here we do not wish to restrict to such situations . in the framework of global optimization methods , the local minimization stage removes the possible energetic penalty associated to replacing a small atom by a bigger one . we can thus expect some increased efficiency of the algorithm in case of multiple homotops . now we convert the extra numerical cost of running parallel trajectories at various temperatures into running them at various compositions , at the same fixed temperature @xmath6 for all compositions . for a x@xmath7y@xmath8 compound , each of the @xmath9 trajectories is then labelled with the number @xmath10 of x atoms , running from 0 to @xmath9 . exchange moves between adjacent trajectories ( from @xmath10 to @xmath11 ) thus need to incorporate the transmutation of two atoms ( one for each configuration ) into the other atom type to preserve composition . as in most monte carlo processes , the probability of attempting such moves must be set in advance as a parameter . the global optimization algorithm can thus be summarized into its main steps . keeping the above notations for atom types , and denoting @xmath12 the configuration at step @xmath13 of trajectory @xmath10 , we start the optimization process using fully random configurations , but locally optimized . 1 . with probability @xmath14 , it is decided whether an exchange between adjacent trajectories will be attempted or not . if so , then the two trajectories involved in the exchange are determined randomly . 2 . for each composition @xmath10 not concerned by any exchange , a new configuration @xmath15 is generated from @xmath16 using either several particle exchanges or large atomic moves . the probability to select particle exchanges is denoted @xmath17 , and the number of simultaneous exchanges is allowed to fluctuate randomly between 1 and @xmath18 . if atomic moves are selected , then each atom is displaced randomly around its previous location in the three directions by a random amount of maximum magnitude @xmath19 . in both cases , @xmath20 is obtained after local minimization . 3 . in case of an exchange between adjacent trajectories , the two configurations @xmath12 and @xmath21 corresponding to these trajectories are then swapped , one x atom of @xmath12 being transmuted into y , and one y atom of @xmath21 being transmuted into x. again , the configurations @xmath15 and @xmath22 are obtained after local minimization . each new configuration is accepted with the usual metropolis acceptance probability at temperature @xmath6 . the algorithm has two main parameters , namely @xmath14 and @xmath17 . the maximum number of particle exchange moves , @xmath23 , was set to 4 in this study . we expect that better results could be obtained by adjusting this parameter appropriately , probably taking higher values for larger clusters or for compostitions close to 50% . the amplitude of atomic displacement , @xmath19 , is set to half the equilibrium distance in the x@xmath1 dimer for @xmath24 , half the equilibrium distance in the y@xmath1 dimer for @xmath25 , and is interpolated linearly between these two values for @xmath26 . in the present work , the exchange probabilities were taken as @xmath27 and @xmath28 , hence allowing a rather large probability of sampling among homotops of a same structure . low - energy structures for mixtures of xenon with either argon or krypton atoms have been first investigated for the sizes @xmath29 and @xmath30 , as there are quantitative global optimization data available for ar xe clusters from the jordan group.@xcite we have adjusted the lj values used by leitner _ et al._@xcite to reproduce the clusters energies found by munro and coworkers.@xcite . with respect to argon , the present data for @xmath31 and @xmath32 are thus @xmath33 , @xmath34 , @xmath35 , @xmath36 , @xmath37 , @xmath38 , @xmath39 , and @xmath40 . global optimization of ar xe and kr xe clusters was performed using the parallel algorithm previously described , simultaneously for all compositions , for a maximum number of 10000 minimization steps per trajectory , and at @xmath41 . ten independent runs were carried out to estimate an average search length for each composition . all global minima reported by munro _ _ were always found within the number of mc steps allowed . l|rr|l|rr ar@xmath42xe@xmath43 & global minimum & average & kr@xmath42xe@xmath43 & global minimum & average + cluster & energy & search length & cluster & energy & search length + xe@xmath4 & -82.093 & 3.2 & xe@xmath4 & -82.093 & 3.0 + arxe@xmath44 & -78.698 & 7.9 & krxe@xmath44 & -81.014 & 9.6 + ar@xmath1xe@xmath45&-76.274 & 9.6 & kr@xmath1xe@xmath45 & -79.263 & 4.3 + ar@xmath3xe@xmath2&-74.015 & 5.8 & kr@xmath3xe@xmath2 & -77.550 & 5.7 + ar@xmath46xe@xmath47 & -71.597 & 8.6 & kr@xmath46xe@xmath47 & -75.869 & 26.1 + ar@xmath48xe@xmath49 & -69.017 & 14.0 & kr@xmath48xe@xmath49 & -74.186 & 25.8 + ar@xmath5xe@xmath50 & -66.584 & 37.2 & kr@xmath5xe@xmath50 & -72.498 & 26.4 + ar@xmath50xe@xmath5 & -63.791 & 19.4 & kr@xmath50xe@xmath5 & -70.844 & 45.2 + ar@xmath49xe@xmath48 & -60.733 & 13.9 & kr@xmath49xe@xmath48 & -69.141 & 18.3 + ar@xmath47xe@xmath46 & -57.851 & 22.7 & kr@xmath47xe@xmath46 & -67.473 & 4.7 + ar@xmath2xe@xmath3&-54.594 & 12.0 & kr@xmath2xe@xmath3 & -65.802 & 11.5 + ar@xmath45xe@xmath1&-51.122 & 7.5 & kr@xmath45xe@xmath1 & -64.128 & 4.1 + ar@xmath44xe & -47.698 & 4.1 & kr@xmath44xe & -62.490 & 2.4 + ar@xmath4 & -44.327 & 2.7 & kr@xmath4 & -60.884 & 2.3 + the results for ar@xmath42xe@xmath43 and kr@xmath42xe@xmath43 clusters are given in table [ tab : arxe13 ] . the average search length is generally higher for compositions close to 50% , for which the number of homotops is maximum for a given isomer , regardless of symmetry . the statistics presently obtained for ar xe clusters show that the average search is between 10 and 1000 times faster than using conventional parallel basin - hopping.@xcite kr xe clusters roughly exhibit the same level of difficulty , but we do not see any strong evidence for particularly severe cases : ar@xmath3xe@xmath2 even seems to be one of the easiest . similarly , the results obtained for ar@xmath42xe@xmath51 clusters show a significant improvement over fixed - composition basin - hopping.@xcite they are given in table [ tab : arxe19 ] along with the corresponding data for kr@xmath42xe@xmath51 clusters . this time , the algorithm is about 1100 times faster depending on @xmath9 , the average search length being still longer for equal compositions . for both the 13- and 19-atom clusters , all global minima are homotops of either the single or double icosahedron . this situation is particularly suited for our algorithm , especially the exchange moves . l|rr|l|rr ar@xmath42xe@xmath51 & global minimum & average & kr@xmath42xe@xmath51 & global minimum & average + cluster & energy & search length & cluster & energy & search length + xe@xmath52 & -134.566 & 72.4 & xe@xmath52 & -134.566 & 70.7 + arxe@xmath53 & -131.819 & 64.3 & krxe@xmath53 & -133.651 & 94.0 + ar@xmath1xe@xmath54&-129.116 & 80.3 & kr@xmath1xe@xmath54 & -132.701 & 109.8 + ar@xmath3xe@xmath55&-126.547 & 85.2 & kr@xmath3xe@xmath55 & -130.088 & 84.3 + ar@xmath46xe@xmath56&-123.764 & 238.2 & kr@xmath46xe@xmath56 & -129.067 & 167.2 + ar@xmath48xe@xmath57&-120.786 & 196.6 & kr@xmath48xe@xmath57 & -127.284 & 175.4 + ar@xmath5xe@xmath4&-118.284 & 221.2 & kr@xmath5xe@xmath4 & -125.498 & 265.9 + ar@xmath50xe@xmath44&-115.681 & 391.8 & kr@xmath50xe@xmath44 & -123.709 & 334.6 + ar@xmath49xe@xmath45&-113.075 & 387.9 & kr@xmath49xe@xmath45 & -121.951 & 319.5 + ar@xmath47xe@xmath2&-110.242 & 264.2 & kr@xmath47xe@xmath2 & -120.115 & 287.1 + ar@xmath2xe@xmath47&-107.531 & 295.8 & kr@xmath2xe@xmath47 & -118.304 & 243.6 + ar@xmath45xe@xmath49&-104.576 & 193.8 & kr@xmath45xe@xmath49 & -116.521 & 187.4 + ar@xmath44xe@xmath50&-101.811 & 235.5 & kr@xmath44xe@xmath50 & -114.736 & 214.3 + ar@xmath4xe@xmath5&-98.110 & 158.5 & kr@xmath4xe@xmath5 & -112.947 & 201.3 + ar@xmath57xe@xmath48&-94.396 & 247.3 & kr@xmath57xe@xmath48 & -111.189 & 188.8 + ar@xmath56xe@xmath46&-90.438 & 121.3 & kr@xmath56xe@xmath46 & -108.863 & 176.5 + ar@xmath55xe@xmath3&-86.328 & 131.2 & kr@xmath55xe@xmath3 & -106.609 & 115.1 + ar@xmath54xe@xmath1&-81.907 & 97.3 & kr@xmath54xe@xmath1 & -104.332 & 10.2 + ar@xmath53xe & -77.298 & 86.8 & kr@xmath53xe & -102.036 & 98.1 + ar@xmath52 & -72.660 & 62.2 & kr@xmath52 & -99.801 & 69.8 + initially , the configurations at all compositions are random . the chances to locate the proper structure ( without any consideration of the homotops ) increase linearly with the number of trajectories . as soon as the right structure is found , the algorithm naturally optimizes atom types to find the most stable homotop , hence the global minimum . but it can also communicate the structure to the adjacent trajectories , until all compositions only need to sample among the permutational homotops . when the interactions are not too dissimilar ( as in kr xe clusters ) , it is likely that the mixed clusters share the same isomer as the global minimum of the homogeneous cluster , which justifies the lattice approach of robertson _ et al._@xcite the problem is then reduced to locating the most stable homotops . by setting @xmath17 to one and starting all trajectories from this minimum , the algorithm can be even more successful , and we estimated the average search length to be further reduced by a factor about 3 with respect to the values given in table [ tab : arxe19 ] . however , when the interactions differ significantly among atoms types , or when the energy landscape of the homogeneous cluster does not display a single steep funnel , it becomes much harder to make a guess about structure in these binary clusters . [ fig : bht ] shows the mean first passage time needed to locate the global minima of ar@xmath51xe@xmath42 using the algorithm under different conditions . disabling swap moves between atom types or exchange moves between adjacent trajectories usually attenuates the efficiency . employing a rather high temperature is even worse , because the cluster may easily leave its optimal lattice . this contrasts with optimizing homogeneous clusters , where using a nonzero temperature helps the system to escape from a funnel.@xcite however , if the energy gap between homotops of the same lattice increases and gets close to the gap between different lattices , we expect the zero temperature method not to be the best . but in such cases , even the notion of a lattice should be questionned . in this section we focus on two larger sizes , for which no global optimization result is available . the lj@xmath0 cluster is characterized by its archetypal two - funnel energy landscape.@xcite the high free - energy barrier separating these two funnels and the higher entropy of the less stable minima of the icosahedral funnel make it particularly hard to locate the truncated octahedral lowest - energy minimum using unbiased global optimization algorithms . hence it is not surprising that this peculiar structure was first found by construction.@xcite we have attempted to locate global minima for binary ar xe and kr xe clusters of size 38 , using the parallel basin - hopping algorithm previously described . because of the huge number of homotops at this size , and most importantly because of the structural competition between icosahedra and truncated octahedra , we can not be fully confident that the global optimization was successful . therefore , the energies reported in table [ tab:38 ] for ar xe clusters should be taken with caution , as they could probably be bettered . the same data for 38-atom kr xe clusters is also reported in table [ tab : krxe38 ] . l|lrc|l|lrc @xmath9 & mixing ratio & energy & point group & @xmath9 & mixing ratio & energy & point group + 0 & 0 & -173.928 & @xmath58 & 20 & 0.58 & -268.683 & @xmath59 + 1 & 0.03 & -179.232 & @xmath60 & 21 & 0.59 & -272.465 & @xmath59 + 2 & 0.06 & -186.333 & @xmath60 & 22 & 0.56 & -276.924 & @xmath60 + 3 & 0.15 & -191.890 & @xmath59 & 23 & 0.61 & -280.169 & @xmath60 + 4 & 0.19 & -197.767 & @xmath59 & 24 & 0.60 & -283.955 & @xmath61 + 5 & 0.22 & -203.421 & @xmath60 & 25 & 0.60 & -287.679 & @xmath60 + 6 & 0.25 & -208.709 & @xmath60 & 26 & 0.59 & -290.973 & @xmath59 + 7 & 0.28 & -213.815 & @xmath59 & 27 & 0.58 & -294.157 & @xmath60 + 8 & 0.32 & -218.631 & @xmath60 & 28 & 0.58 & -297.320 & @xmath61 + 9 & 0.34 & -223.491 & @xmath59 & 29 & 0.56 & -300.202 & @xmath61 + 10 & 0.36 & -228.209 & @xmath59 & 30 & 0.50 & -303.484 & @xmath59 + 11 & 0.39 & -232.771 & @xmath59 & 31 & 0.51 & -305.987 & @xmath59 + 12 & 0.40 & -237.337 & @xmath59 & 32 & 0.46 & -308.404 & @xmath59 + 13 & 0.43 & -241.887 & @xmath60 & 33 & 0.43 & -310.521 & @xmath59 + 14 & 0.44 & -246.117 & @xmath59 & 34 & 0.38 & -311.708 & @xmath59 + 15 & 0.45 & -249.677 & @xmath60 & 35 & 0.31 & -313.772 & @xmath59 + 16 & 0.46 & -253.593 & @xmath60 & 36 & 0.25 & -315.988 & @xmath59 + 17 & 0.46 & -257.184 & @xmath60 & 37 & 0.14 & -318.860 & @xmath59 + 18 & 0.46 & -261.079 & @xmath60 & 38 & 0 & -322.115 & @xmath58 + 19 & 0.56 & -264.927 & @xmath59 & & & & + specifically to this cluster size , all minima found during the optimization process were categorized as either icosahedral or cubic - like , depending on the energy of the corresponding homogeneous isomer found by quenching . in cases where the cubic isomer was not found among the isomers , we performed additional optimizations starting from this structure , setting @xmath17 to one . this mainly occured for ar xe clusters . eventually , two series of minima were obtained for each of the icosahedral and octahedral funnels . we did not find any decahedral isomer that could compete with these structural types , even though some evidence for stabilizing decahedra by doping was reported in ref . . we have represented in fig . [ fig : dener ] the relative energy differences @xmath62 between the most stable cubic isomers and the most stable icosahedral isomers , as they were obtained from our optimization scheme , for both the ar xe and kr xe mixtures . besides some strong variations sometimes seen from one composition to the next , and which can be attributed to usual finite - size effects , general trends can be clearly observed . first , kr@xmath63xe@xmath42 clusters are always most stable in the cubic shape . actually , changing the composition most often further stabilizes truncated octahedra , and only rarely enhances the stability of icosahedra , which occurs for @xmath64 and @xmath65 . conversely ar@xmath63xe@xmath42 clusters are preferentially found icosahedral , exceptions being @xmath66 and @xmath67 . this is an example of a composition - induced structural transition between the two funnels of the energy landscape . from a computational point of view , it should be noted that the optimization algorithm was able to locate the truncated octahedral minima for kr xe clusters by itself , starting from disordered minima , and that the extra runs starting from this structure only produced slightly more stable homotops . this is another illustration of the efficiency of the present parallel optimization method . the general degree of disorder is higher in icosahedral structures than in the cubic - like isomer . hence it is more difficult to put up the latter geometry with very unlike interactions , as in ar xe clusters . cubic homotops of argon with xenon are rather distorted , but the strain is much lower with krypton instead of argon . examples of global minima obtained at compositions @xmath68 , 19 , and 29 are represented in fig . [ fig : structex ] . in kr xe compounds , a progressive core - surface phase separation is seen with kr atoms outside , in agreement with energetic arguments : atoms with the larger @xmath32 prefer to occupy interior sites . in icosahedral clusters , the strain increases at such sites , especially in polytetrahedral systems . xe clusters also prefer to have xe atoms at the center , but the increased strain is too high a penalty , which explains that cubic structures are favored over icosahedra . in general , no complete phase separation is found in ar xe clusters , even though ar atoms seem to fit best at the centre of the cluster . in both cases , surface energies thus play an important role . mixing in these clusters can be estimated using radial distribution functions.@xcite here we use the same index as jellinek and krissinel,@xcite namely the overall mixing ratio @xmath69 defined as@xcite @xmath70 where @xmath71 is the binding energy of cluster @xmath72 , @xmath73 the binding energy of subcluster @xmath74 in the homogeneous cluster @xmath75 at the same atomic configuration as @xmath72 , and a similar definition for the last term of eq . ( [ eq : mixing ] ) . as seen from table [ tab:38 ] , the mixing ratio increases notably in ar xe clusters , up to more than 60% for some compositions . kr xe clusters , despite exhibiting some core - surface segregation , show similar variations of the mixing ratio with composition , with only slightly smaller values of @xmath69 . therefore the mixing ratio , as defined in eq . ( [ eq : mixing ] ) , is a rather ambiguous parameter for quantifying the extent of mixing in this small cluster . the optimal structure of an homogeneous cluster described with a pairwise potential results from a competition between maximizing the number of nearest neighbors and minimizing the strain energy , or penalty induced by distorting these bonds.@xcite binary lennard - jones systems exhibit several extra complications due to the various ways of rearranging atom types in a given structure . in these systems , the strain varies notably among the homotops , especially in clusters made of very unlike atoms . however , our results indicate that the same general rules hold for homogeneous and heterogeneous systems . in ar@xmath52xe@xmath52 , the strain is rather high , but the number of contacts is also high . in kr@xmath52xe@xmath52 , both the strain and the number of nearest neighbors are much smaller . to investigate the role of heterogeneity on the strain , we have computed the various contributions to the reduced strain energies in ar@xmath63xe@xmath42 clusters . the strain energies are defined for ar ar , xe xe , and ar xe interactions as follows:@xcite @xmath76 in these equations , @xmath77 is the ( negative ) total binding energy between atoms @xmath78 and @xmath79 , @xmath80 is the number of x y nearest neighbors , and @xmath81 is the lennard - jones well depth corresponding to the interaction between @xmath78 and @xmath79 atoms . reduced strain energies are then defined as @xmath82 , in order to account for the different magnitudes of the interactions among atom types . according to these definitions , the strain energies are always positive quantities . the strain energies in ar@xmath63xe@xmath42 clusters are represented versus composition in fig . [ fig : strain ] . they give us some insight about the possible ways of reducing strain . the pattern exhibited by the reduced strain versus composition shows different behaviors for clusters having mostly argon or xenon atoms . for @xmath83 , most strain is carried by interactions between alike atoms . this case is illustrated by ar@xmath47xe@xmath84 in fig . [ fig : structex ] , where a kind of core / surface phase separation occurs . here surface energies are also important , but the situation is rather different from mixed cubic kr xe clusters . because having the xenon atoms at the inner sites of the icosahedral structure would maximise the strain of these atoms , it is much more favorable to have the smaller atoms inside and the xenon atoms outside . the cubic kr xe structures , on the other hand , are not especially strained , and having the smaller atoms inside would lead to an energetic penalty . when the number of ar atoms increases above about 19 in the 38-atom cluster , interactions between unlike atoms are significantly more strained . the case of ar@xmath84xe@xmath47 depicted in fig . [ fig : structex ] is perticularly informative : xe atoms are located scarcely among the icosahedral cluster , and relieve the structure from too much strain at the expense of only few xe xe interactions . in this case , cluster structure is driven by the number of unlike interactions . it is also worth noting that a few compositions are especially weakly strained ; this occurs when the global minimum is octahedral , but also in the range @xmath85 . for these latter clusters , the core / surface segregation and the number of unlike interactions are both optimal . the cubic to icosahedral transition seen above actually favors polyicosahedral ( or anti - mackay ) structures . the strain reduction produced by size disparity in 38-atom ar xe clusters helps in stabilizing these kinds of structures , which are otherwise replaced by multilayer ( or mackay ) geometries in the homogeneous clusters . since most lj clusters under the size of 38 atoms are most stable as polytetrahedra,@xcite we do not expect that changing composition will affect them to a large extent . as a notable exception , the 6-atom homogeneous lj cluster is more stable in its octahedral isomer . the lowest energy geometries of mixed ar xe clusters containing 6 atoms , represented in fig . [ fig : arxe6 ] , show polytetrahedral transitions for two compositions , namely ar@xmath46xe@xmath1 and ar@xmath3xe@xmath3 . this behavior mimics somewhat what was observed for the larger 38-atom cluster , only at a smaller scale . in particular , and as in fig . [ fig : dener ] , polytetrahedral arrangements are seen to be more convenient for xenon compositions under 50% . possible polytetrahedral structures of mixed ar xe clusters have been investigated for the size 55 , whose most stable isomer is well known as a perfect double layer ( mackay ) icosahedron for the homogeneous system . the global optimization results are summarized in table [ tab:55 ] for all compositions . for this we also conducted complementary calculations on the 2-layer lattice . each putative global minimum was labelled either as mackay icosahedron or , when the lattice structure does not exactly match the multilayer icosahedron , as polytetrahedral . most compositions become increasingly polytetrahedral as the ratio of xenon atoms increases , even though the polytetrahedral character may often be only local . l|lrcc|l|lrcc @xmath9 & mixing ratio & energy & point group & type & @xmath9 & mixing ratio & energy & point group & type + 0 & 0 & -279.248 & @xmath86 & mi & 28 & 0.53 & -417.785 & @xmath59 & pt + 1 & 0.05 & -284.276 & @xmath61 & mi & 29 & 0.53 & -421.225 & @xmath60 & pt + 2 & 0.10 & -289.313 & @xmath60 & mi & 30 & 0.51 & -424.566 & @xmath59 & pt + 3 & 0.15 & -294.360 & @xmath60 & mi & 31 & 0.49 & -427.888 & @xmath59 & pt + 4 & 0.28 & -302.344 & @xmath60 & pt & 32 & 0.47 & -431.565 & @xmath59 & pt + 5 & 0.34 & -310.780 & @xmath60 & pt & 33 & 0.46 & -435.680 & @xmath59 & pt + 6 & 0.38 & -316.331 & @xmath60 & pt & 34 & 0.45 & -439.565 & @xmath59 & pt + 7 & 0.41 & -321.770 & @xmath59 & pt & 35 & 0.42 & -443.335 & @xmath59 & pt + 8 & 0.43 & -327.241 & @xmath60 & pt & 36 & 0.41 & -446.892 & @xmath59 & pt + 9 & 0.44 & -332.356 & @xmath59 & pt & 37 & 0.41 & -449.965 & @xmath59 & pt + 10 & 0.47 & -337.497 & @xmath60 & pt & 38 & 0.36 & -454.356 & @xmath60 & mi + 11 & 0.50 & -342.655 & @xmath60 & pt & 39 & 0.35 & -458.203 & @xmath60 & mi + 12 & 0.55 & -347.608 & @xmath59 & pt & 40 & 0.34 & -462.564 & @xmath61 & mi + 13 & 0.57 & -352.520 & @xmath59 & pt & 41 & 0.30 & -466.709 & @xmath61 & mi + 14 & 0.56 & -357.586 & @xmath60 & pt & 42 & 0.28 & -472.191 & @xmath86 & mi + 15 & 0.58 & -362.483 & @xmath61 & pt & 43 & 0.26 & -475.967 & @xmath87 & mi + 16 & 0.58 & -367.231 & @xmath59 & pt & 44 & 0.24 & -479.739 & @xmath88 & mi + 17 & 0.59 & -372.016 & @xmath60 & pt & 45 & 0.22 & -483.495 & @xmath89 & mi + 18 & 0.59 & -376.740 & @xmath59 & pt & 46 & 0.20 & -487.219 & @xmath61 & mi + 19 & 0.59 & -381.501 & @xmath61 & pt & 47 & 0.18 & -490.910 & @xmath60 & mi + 20 & 0.60 & -386.007 & @xmath60 & pt & 48 & 0.16 & -494.585 & @xmath90 & mi + 21 & 0.60 & -390.135 & @xmath59 & pt & 49 & 0.14 & -498.231 & @xmath60 & mi + 22 & 0.59 & -395.065 & @xmath59 & pt & 50 & 0.12 & -501.860 & @xmath61 & mi + 23 & 0.59 & -399.213 & @xmath59 & pt & 51 & 0.10 & -505.467 & @xmath89 & mi + 24 & 0.60 & -403.205 & @xmath59 & pt & 52 & 0.08 & -509.052 & @xmath88 & mi + 25 & 0.58 & -406.983 & @xmath59 & pt & 53 & 0.06 & -512.616 & @xmath87 & mi + 26 & 0.56 & -410.686 & @xmath59 & pt & 54 & 0.04 & -516.170 & @xmath86 & mi + 27 & 0.56 & -414.427 & @xmath60 & pt & 55 & 0 & -517.168 & @xmath86 & mi + two examples of lowest energy structures are represented in fig . [ fig : arxe55 ] , corresponding to @xmath91 and @xmath92 . the obvious deviations of the geometry of the former from the mackay icosahedron and the various occupation sites of the heavy atoms for both structures illustrate again that there is no simple rule that determine the most stable minima when the atomic sizes do significantly differ from each other . we now go back to the 38-atom clusters of kr and xe atoms , for which the global minimum was always found to be a truncated octahedron . the extremely large number of isomers ( including homotops ) in the energy landscape of binary lennard - jones clusters , added to the expected presence of significant energy barriers between icosahedral and cubic isomers,@xcite prevent finite - temperature simulations from being conducted in a reliably ergodic way with the presently available tools . for example , the particle exchange moves used to accelerate convergence of sampling among homotops will likely have very low acceptance probabilities in mc simulations at low temperatures , especially for ar xe clusters . therefore , even with powerful methods such as parallel tempering or multicanonical monte carlo , reaching convergence in 38-atom lj clusters does not seem currently feasible to us . as an alternative , we have chosen to investigate solid - solid transitions by means of the superposition approach.@xcite for a given cluster , databases of minima in each of the icosahedral ( ico ) and truncated octahedral ( fcc ) funnels were constructed using the optimization algorithm . for each composition and each of the two funnels , no more than 2000 distinct minima were considered . the classical partition function of the y@xmath93xe@xmath7 cluster ( y = ar or kr ) restricted to funnel a = fcc or ico is approximated by a harmonic superposition over all minima of the databases , which belong to this funnel:@xcite @xmath94 where @xmath95 is the inverse temperature , @xmath96 the geometric mean vibrational frequency , @xmath97 with @xmath98 the order of the point group of minimum @xmath13 and @xmath99 . we do not consider quantum effects here , although they may be important at low temperatures,@xcite since delocalization or zero - point effects are not expected to be significant for rare gases as heavy as krypton or xenon . within the harmonic superposition approximation , a solid - solid transition occurs when @xmath100.@xcite this latter equation is solved numerically in @xmath101 or @xmath6 , its solution is denoted @xmath102 . in cases where icosahedra are energetically more stable than octahedra , a solid - solid transition can occur if some cubic structures are entropically favored , which requires lower vibrational frequencies and/or lower symmetries . we did not find such situations in our samples of ar xe clusters , therefore we restrict to kr xe clusters in the following . similar to transitions between funnels , transitions between homotops will happen if their partition functions are equal . the huge number of homotops gives rise to as many values for the corresponding temperatures , and we define the homotop transition temperature @xmath103 such that @xmath104 where @xmath105 is the transition temperature between the global minimum ( homotop 0 ) and its homotop @xmath106 . equating the harmonic partition functions for these two isomers leads to the expression of @xmath105:@xcite @xmath107 since all homotops are characterized by different vibrational and symmetry properties , the transition temperatures @xmath105 are not ordered exactly as the energy differences @xmath108 . this reflects that solid - solid transitions involve crossover in free energy rather than binding energy . the above equation also shows that @xmath105 can take negative values if homotop @xmath106 has a higher symmetry and/or a higher vibrational frequency than the ground state . in this case the global minimum is always the free energy minimum , and no solid - solid transition occurs , hence the form of eq . ( [ eq : th ] ) . finally , a third temperature has a strong consequence on cluster structure , namely the melting temperature . its estimation from either simulations or superpositions approximations is already quite difficult for the homogeneous lj@xmath0 cluster,@xcite and we did not attempt to compute it for binary clusters . however , the previous study by frantz@xcite has shown that the melting point in mixed , 13-atom ar kr clusters varies quite regularly ( approximately quadratically ) with composition . as a simple rule , we will assume that the melting point of kr@xmath63xe@xmath42 , @xmath109 , lies inside some range between the approximate melting points of kr@xmath0 and xe@xmath0 , respectively . from the results obtained by doye and wales@xcite and the monte carlo data of ref . for the lj@xmath0 cluster , we get @xmath110 and @xmath111 in reduced lj units of argon . this provides rough limits to the actual melting points of kr xe clusters , for the price of neglecting finite - size effects . the transition temperatures are represented in fig . [ fig : tss ] for all compositions in kr@xmath63xe@xmath42 clusters . we notice first that the structural transition temperature @xmath102 varies quite regularly with composition in both the ranges @xmath112 and @xmath113 , and that it shows strong size effects between these limits . several situations are predicted to occur depending on the relative values of @xmath114 , @xmath103 and @xmath115 . in most cases , @xmath116 . that melting takes place at temperatures lower than the cubic / icosahedral transition simply nullifies the transition between structural types . however , this extra stability of the octahedral funnel may have a consequence on the melting point itself , which is likely to increase . still , this situation implies that simulations will more easily reach convergence . however , there are notable exceptions for this behavior , at @xmath117 , @xmath65 and @xmath118 . in these clusters , heterogeneity is not sufficient for the thermodynamical behavior of the cluster to deviate too much from those of the homogeneous system . the transition between homotops usually occurs prior to melting . thermal equilibrium within the cubic funnels thus involves several homotops ( and `` restricted '' solid - solid transitions ) , and the corresponding thermodynamical state could be probably simulated using specifically designed exchange moves between outer particles within a monte carlo scheme . a few clusters melt before exhibiting any transition between homotops . this occurs for instance at @xmath119 , 13 or 15 . for these sizes the structural transition also occurs at temperatures higher than the estimated melting point . these cases should pose less problems to conventional simulations than the homogeneous cluster . the previous results have shown that finite - size ar xe compounds show a preferential polytetrahedral order , even for very low doping rates , over octahedral order . on the other hand , kr xe clusters at the same sizes further favor cubic order . since polytetrahedral order is known to be present in liquids and , more generally , in disordered structural glasses , it seems natural to correlate the behavior observed in these clusters to the dynamics of the corresponding bulk materials.@xcite we have simulated the cooling of 108-atom binary rare - gas liquids , using a simple metropolis monte carlo scheme under constant volume and temperature . initially the atoms are placed randomly into a cubic box of side @xmath120 , and periodic boundaries are treated in the minimum image convention . the lj interactions were not truncated , and the simulations consisted of 100 stages of 10@xmath121 mc cycles each , linearly spaced in temperature . three compositions have been selected , following our knowledge of the cluster structure . for each composition , different length sizes @xmath120 and different temperature ranges @xmath122 $ ] were chosen in order to cover both sides of the melting point . in the first mixture , 24 xenon atoms and 84 argon atoms are simulated with @xmath123 lj units of argon , with @xmath124 . in the second mixture , 24 xenon atoms are added to 84 krypton atoms at @xmath125 and @xmath126 . the third mixture consists of 9 argon atoms and 99 xenon atoms at @xmath127 and @xmath128 . even though we did not attempt to locate the most stable crystalline forms for these mixtures , our searches close to the face - centred cubic morphology showed that the most stable configurations for these mixtures always had some cubic order . it is likely that the actual ground states for such systems are indeed crystalline.@xcite the average root mean square fluctuation of the bond distances , also known as the lindemann index @xmath129 , universally characterizes the thermodynamical state of the condensed system as either solid or liquid , depending on its value being lower or higher than about 0.15 . to quantify the extent of crystalline order , we have used the bond order parameter @xmath130 introduced by steinhardt and coworkers.@xcite the two parameters @xmath129 and @xmath130 allow us to follow in monte carlo time the cooling processes for all materials in a simultaneous way , independently of thermodynamical characteristics such as the melting temperature . the correlation between @xmath129 and @xmath130 for ten cooling simulations of each of the three bulk binary compounds is represented in fig . [ fig : glass ] . in all cases , the lindemann parameter covers the whole range @xmath131 , indicating that the melting point was indeed crossed . however , the three compounds display very contrasted cooling behaviors . in the ( ar@xmath132xe@xmath133 ) system , @xmath129 regularly decreases but @xmath130 always remain below 0.05 . therefore crystallization never takes place , and the final state obtained by quenching is significantly higher in energy than some crystalline forms ; this is typical of glass formation . in ( kr@xmath132xe@xmath133 ) , all simulations show some rather sharp transition from a ( high @xmath129 , low @xmath130 ) state to a ( low @xmath129 , high @xmath130 ) state as @xmath129 crosses about 0.1 . the temperatures where crystallisation occurs may vary somewhat among the cooling runs , in the same way as they are expected to depend on the cooling rate . lastly , the case of ( ar@xmath47xe@xmath134 ) is intermediate : while most simulations end up in a nearly fully crystalline phase ( @xmath135 ) , a few of them show a limited degree of cubic ordering in the solid phase , @xmath130 having values close to 0.07 . these results very closely reflect our previous data on binary , 38-atom clusters of the same materials . in terms of composition , the first mixture corresponds to ar@xmath136xe@xmath49 , which clearly favors icosahedral shapes over truncated octahedra . the second mixture reminds of kr@xmath136xe@xmath49 , for which the cubic structure is even more stable than in the homogeneous cluster . the third mixture should be compared to ar@xmath3xe@xmath137 , which favors icosahedra only moderately . this correlation found here between cluster structure and the glassforming ability of the bulk material confirms previous analyses on the icosahedral local order in liquids and glasses,@xcite as well as the recent conclusions obtained by doye _ et al._@xcite that clusters of good glassformers indeed show a polytetrahedral order . as far as structural and dynamical properties are concerned , binary compounds show a significantly richer complexity with respect to homogeneous clusters . the work reported in the present paper was intended to achieve several goals . first , a parallel global optimization algorithm was designed to locate the most stable structures of mixed rare - gas clusters , beyond the lattice approximations of robertson and co - workers.@xcite based on the basin - hopping or monte carlo+minimization algorithm,@xcite this algorithm includes exchange moves between particles at fixed composition as well as exchange moves between configurations at different compositions . tests on simple ar@xmath42xe@xmath43 and ar@xmath42xe@xmath51 clusters show that the method is quite efficient , in addition to being easy to implement . for these systems , we have found that the choice of a very low temperature works best as it allows some significant time to be spent for optimizing the search for homotops on a same common lattice . putative global minima for ar@xmath63xe@xmath42 and kr@xmath63xe@xmath42 clusters have been investigated for all compositions . the structure of ar xe compounds is mainly polytetrahedral , except at very low doping rates . kr xe clusters not only remain as truncated octahedra , but mixing the two rare gases even favors these cubic structures over icosahedra . we see some significant trend toward core / surface phase separation in ar xe clusters with @xmath138 and in all kr xe clusters . however , these demixing behaviors are not due to the same factors , as xe atoms favor outer sites to reduce strain in ar xe icosahedra , while they occupy interior sites to maximize the number of bonds in kr xe truncated octahedra . conversely , ar@xmath63xe@xmath42 clusters with @xmath83 exhibit a higher degree of mixing . analysing the strain in these stable structures confirms the presence of a structural transition near @xmath139 in these systems . polytetrahedral morphologies were also found as the most stable structures of many mixed ar xe clusters with 55 atoms , as soon as the relative number of xe atoms was large enough . the general conclusion thus seems that the extra strain introduced by mixing these different elements penalizes the highly ordered ( cubic or 2-layer icosahedron ) structures . within the harmonic superposition approximation , we have estimated the temperatures required by the 38-atom kr xe clusters to undergo a structural transition toward the icosahedral funnel , or toward other octahedral homotops . for compositions with a doping rate higher than 3/38 , the structural transition temperature was seen to occur at temperatures higher than the extrapolated melting point . this mainly reflects the special stability of the octahedral structures , and has the probable consequence that actual melting points increase somewhat . these predictions could probably be checked with numerical simulations . for most compositions , the transitions between different homotops of the truncated octahedron are seen to be potentially induced by relatively small temperatures . therefore particle exchange moves will be necessary in order that simulations remain close to ergodic . following previous results by other researchers,@xcite we have found some further evidence that criteria for glass formation in bulk materials may also lie in the parameters , which are responsible for stable cluster structures . since the atomistic simulation of the dynamical vitrification process can generally be much harder than obtaining stable configurations of atomic clusters , we expect the approach followed in the present theoretical effort to be also useful in the community of glasses and supercooled liquids . the method is obviously not limited to rare - gases , and its application to other compounds , especially metallic nanoalloys , should be straightforward , except maybe for fine tuning its intrinsic parameters . from a methodological point of view , it could also be applied to materials with more than two components . work on ternary systems is currently in progress . we wish to thank dr . j. p. k. doye and prof . k. j. jordan for very useful discussions . this research was supported by the cnrs - tubtak grant number 15071 . j. l. rousset , a. m. cadrot , f. j. cadete santos aires , a. renouprez , p. mlinon , a. perez , m. pellarin , j. l. vialle , and m. broyer , j. chem . phys . * 102 * , 8574 ( 1995 ) . m. j. lpez , p. a. marcos , and j. a. alonso , j. chem . phys . * 104 * , 1056 ( 1996 ) . g. e. lpez and d. l. freeman , j. chem . phys . * 98 * , 1428 ( 1993 ) . i. l. garzon , x. p. long , r. kawai , and j. h. weare , chem . . lett . * 158 * , 525 ( 1989 ) . p. ballone , w. andreoni , r. car , and m. parrinello , europhys . lett . * 8 * , 73 ( 1989 ) . s. darby , t. v. mortimer - jones , r. l. johnston , and c. roberts , j. chem . phys . * 116 * , 1536 ( 2002 ) . m. s. bailey , n. t. wilson , c. roberts , and r. l. johnston , euro . j. d * 25 * , 41 ( 2003 ) . m. c. vicns and g. e. lpez , phys . a * 62 * , 033203 ( 2000 ) . f. baletto , c. mottet , and r. ferrando , phys . lett . * 90 * , 135504 ( 2003 ) . m. tchaplyguine , r. r. t. marinho , m. gisselbrecht , r. feifel , s. l. sorensen , g. hrwall , m. lundwall , a. naves de brito , j. schulz , n. mrtensson , s. svensson , and o. bjrneholm , j. chem . phys . * 120 * , 345 ( 2004 ) ; m. tchaplyguine , m. lundwall , m. gisselbrecht , g. hrwall , r. feifel , s. sorensen , s. svensson , n. mrtensson , and o. bjrneholm , phys . rev . a * 69 * , 031201 ( 2004 ) . a. s. clarke , r. kapral , b. moore , g. patey , and x .- wu , phys . * 70 * , 3283 ( 1993 ) ; a. s. clarke , r. kapral , and g. n. patey , j. chem . phys . * 101 * , 2432 ( 1994 ) . h. jnsson and h. c. andersen , phys . * 60 * , 2295 ( 1988 ) . w. kob and h. c. andersen , phys . e * 51 * , 4626 ( 1995 ) . b. coluzzi , g. parisi , and p. verocchio , j. chem . phys . * 112 * , 2933 ( 2000 ) . m. utz , p. g. debenedetti , f. h. stillinger , phys . lett . * 84 * , 1471 ( 2000 ) . k. k. bhattacharya , k. broderix , r. kree , and a. zippelius , europhys . lett . * 47 * , 449 ( 1999 ) . r. yamamoto and w. kob , phys . e * 61 * , 5473 ( 2000 ) . lee , t. cagin , w. l. johnson , and w. a. goddard , j. chem . phys . * 119 * , 9858 ( 2003 ) . f. c. frank , proc . london , ser . a * 215 * , 43 ( 1952 ) . t. schenk , d. holland - moritz , v. simonet , r. bellissent , and d. m. herlach , phys . 89 * , 075507 ( 2002 ) . s. mossa and g. tarjus , j. chem . phys . * 119 * , 8069 ( 2003 ) . j. p. k. doye , d. j. wales , f. h. m. zetterling , and m. dzugutov , j. chem . phys . * 118 * , 2792 ( 2003 ) . t. f. middleton , j. hernndez - rojas , p. n. mortenson , and d. j. wales , phys . b * 64 * , 184201 ( 2001 ) ; j. r. fernndez and p. harrowell , phys . e * 67 * , 011403 ( 2003 ) ; j. chem . phys . * 120 * , 9222 ( 2004 ) . p. j. steinhardt , d. r. nelson , and m. ronchetti , phys . b * 28 * , 784 ( 1983 ) . l|lrc|l|lrc @xmath9 & mixing ratio & energy & point group & @xmath9 & mixing ratio & energy & point group + 0 & 0 & -238.897 & @xmath58 & 20 & 0.45 & -286.209 & @xmath60 + 1 & 0.09 & -241.200 & @xmath60 & 21 & 0.44 & -288.240 & @xmath60 + 2 & 0.18 & -243.604 & @xmath61 & 22 & 0.42 & -290.323 & @xmath61 + 3 & 0.25 & -245.962 & @xmath60 & 23 & 0.42 & -292.310 & @xmath60 + 4 & 0.27 & -248.453 & @xmath60 & 24 & 0.41 & -294.347 & @xmath58 + 5 & 0.33 & -250.927 & @xmath60 & 25 & 0.37 & -296.352 & @xmath89 + 6 & 0.35 & -253.489 & @xmath61 & 26 & 0.35 & -298.371 & @xmath61 + 7 & 0.39 & -256.005 & @xmath60 & 27 & 0.33 & -300.382 & @xmath60 + 8 & 0.41 & -258.570 & @xmath140 & 28 & 0.32 & -302.414 & @xmath141 + 9 & 0.43 & -261.156 & @xmath60 & 29 & 0.29 & -304.411 & @xmath141 + 10 & 0.45 & -263.740 & @xmath61 & 30 & 0.25 & -306.457 & @xmath60 + 11 & 0.47 & -266.294 & @xmath60 & 31 & 0.22 & -308.390 & @xmath60 + 12 & 0.48 & -268.867 & @xmath60 & 32 & 0.20 & -310.370 & @xmath60 + 13 & 0.49 & -271.420 & @xmath60 & 33 & 0.16 & -312.315 & @xmath60 + 14 & 0.50 & -273.996 & @xmath61 & 34 & 0.13 & -314.287 & @xmath89 + 15 & 0.50 & -275.996 & @xmath60 & 35 & 0.10 & -316.230 & @xmath60 + 16 & 0.50 & -278.046 & @xmath140 & 36 & 0.07 & -318.200 & @xmath140 + 17 & 0.48 & -280.082 & @xmath60 & 37 & 0.04 & -320.133 & @xmath141 + 18 & 0.47 & -282.129 & @xmath61 & 38 & 0 & -322.115 & @xmath58 + 19 & 0.46 & -284.164 & @xmath60 & & & & +

in this paper , a hub refers to a non - terminal vertex of degree at least three . we study the minimum number of hubs needed in a network to guarantee certain flow demand constraints imposed between multiple pairs of sources and sinks . we prove that under the constraints , regardless of the size or the topology of the network , such minimum number is always upper bounded and we derive tight upper bounds for some special parameters . in particular , for two pairs of sources and sinks , we present a novel path - searching algorithm , the analysis of which is instrumental for the derivations of the tight upper bounds .

1308.2865

consider a network @xmath0 , where @xmath1 denotes the set of vertices in @xmath2 , and @xmath3 denotes the set of edges in @xmath2 . a vertex in @xmath2 is said to be a _ source _ if it is only incident with outgoing edges , and a _ sink _ if it is only incident with incoming edges . often , a source or sink is referred to as a _ terminal _ vertex . a non - terminal vertex is said to be a _ hub _ if its degree is greater than or equal to @xmath4 . in this paper , we are primarily concerned with the minimum number of hubs needed when certain constraints on the flow demand between multiple pairs of sources and sinks are imposed . the flow demand constraints considered in this paper will be in terms of the vertex - cuts between pairs of sources and sinks . this can be justified by a vertex version of the max - flow min - cut theorem @xcite , which states that for a network with infinite edge - capacity and unit vertex - capacity , the maximum flow between one source and one sink is equal to the minimum vertex - cut between them . here , we remark that with appropriately modified setup , our results can be stated in terms of edge - cuts as well . more precisely , for given @xmath5 , let @xmath6 denote the set of all finite networks @xmath2 ( see figure [ illustrator ] for an example ) such that * there are @xmath7 sources @xmath8 and @xmath7 sinks @xmath9 in @xmath2 ; * all edges in @xmath2 , except those incident with a source or sink , are undirected ( alternatively , bi - directional ) ; * for each feasible @xmath10 , the minimum vertex - cut from @xmath11 to @xmath12 is @xmath13 . now , we define @xmath14 where @xmath15 denotes the number of hubs in @xmath16 . the above definition can be roughly interpreted as follows : for a given @xmath2 , we try to find a subgraph @xmath16 that contains the minimum number of hubs required to satisfy the vertex - cut constraints , and @xmath17 gives us the minimum number corresponding to the worst - case scenarios among all possible @xmath2 . with @xmath18 hubs.,scaledwidth=30.0% ] at first glance , @xmath19 can be infinite . one of the main ( and somewhat surprising ) results in this paper , theorem [ finiteness ] , however , states that for any given @xmath20 , @xmath19 is in fact finite . with finiteness confirmed , we are primarily interested in computing the value of @xmath19 . we say a graph @xmath2 is @xmath21 in @xmath22 , if @xmath23 , but for any @xmath24 , @xmath25 . it then follows from the fact that every graph in @xmath6 has at least one minimal subgraph that @xmath26 the vertex - connectivity version of the classical menger s theorem @xcite states that for a network with one pair of source @xmath27 and sink @xmath28 with the minimum vertex - cut between them being @xmath29 , there exist @xmath29 vertex - disjoint paths connecting @xmath27 and @xmath28 , which immediately implies that @xmath30 for any given @xmath29 . theorem [ finiteness ] states that for any given @xmath23 , one can always find a subgraph @xmath16 of @xmath2 such that @xmath15 is upper bounded by a constant , which is independent of the choice of @xmath2 . in some sense , theorem [ finiteness ] can be viewed as a generalization of the vertex - connectivity version of menger s theorem . mathematically , the proposed problem of computing @xmath19 is a natural combinatorial optimization problem . on a more practical side , hubs in networks naturally correspond to more costly vertices . for instance , in a transportation network , as opposed to `` relaying '' vertices with degree @xmath31 , hubs may have to be better equipped for traffic scheduling ; for this reason , when designing the route map , an airline may need to avoid running too many airline hubs to reduce the cost . so , as might be expected , @xmath19 is of significance to cost - minimizing resource allocation in transportation networks . to the best of our knowledge , the proposed problem of computing or estimating @xmath32 has not yet been examined previously and there is little related work in the vast literature of graph theory . on the other hand , to a great extent , this work is motivated by the study of network encoding complexity ( see @xcite and references therein ) , where the number of encoding vertices in directed networks is of primary concern . moreover , our approaches to tackle the problem are influenced by those in network encoding complexity theory , particularly , to a greater extent , those in @xcite . the remainder of the paper is organized as follows . in section [ minimalgraph ] , we give necessary and sufficient conditions ( theorem [ equivalent ] ) for a graph being minimal in @xmath33 . in section [ sectionminimalrepresentation ] , we introduce the notion of a representation of a graph in @xmath33 and we present the structural decomposition theorem ( theorem [ decomposition ] ) for representations of minimal graphs in @xmath33 . we will introduce in section [ xu - algorithm ] a novel path - searching algorithm , the analysis of which will aptly produce an upper bound on @xmath34 for any given @xmath35 . in section [ sectionhmn ] , we derive the value of @xmath36 ( theorem [ nmn2mn ] ) , which is a main result in this paper . another main result is theorem [ finiteness ] , which establishes the finiteness of @xmath19 , @xmath37 , through a recursive bounding argument . the remaining part of section [ more - than - two ] will be devoted to the derivations of the values of @xmath32 for some special parameters ( theorem [ thm11mn ] , [ thmh222 ] ) . we first introduce some notation and terminologies that will be used throughout the paper . a sequence of edges in @xmath2 , @xmath39 , @xmath40 , with @xmath41 , @xmath42 , can be linked to form a _ path _ @xmath43 , denoted by @xmath44 ; furthermore , @xmath43 is called a _ cycle _ if @xmath45 . a path @xmath43 of this form is said to be _ directed _ each @xmath46 is oriented such that all of them concatenate . for a path @xmath47 and two vertices @xmath48 , @xmath49 of @xmath47 , let @xmath50 $ ] denote the subpath of @xmath47 between @xmath48 and @xmath49 . for a directed path @xmath51 , let @xmath52 , @xmath53 denote the head , tail of @xmath47 , respectively ; and we say @xmath46 is @xmath54 than @xmath55 on @xmath47 , denoted by @xmath56 , if @xmath57 is the head of the directed path @xmath58 $ ] ( or alternatively , @xmath55 is _ larger _ than @xmath46 , denoted by @xmath59 ) . for a graph @xmath23 , by the vertex - connectivity version of menger s theorem , for each @xmath10 , one can find a set @xmath60 of @xmath13 vertex - disjoint @xmath61-paths from @xmath11 to @xmath12 . if the choice of @xmath60 is unique , @xmath60 is said to be @xmath62 , otherwise it is said to be _ reroutable_. @xmath2 is said to be _ non - reroutable _ if all @xmath60 s are non - reroutable , _ reroutable _ otherwise . for any index set @xmath63 , let @xmath64 denote the subgraph of @xmath2 induced on the edges of @xmath65-paths . @xmath2 is said to be a _ @xmath66-graph _ if @xmath67 , that is , each edge in @xmath2 belongs to some @xmath60-path . in order to compute @xmath19 , it is enough to only consider all @xmath66-graphs , since @xmath68 is a subgraph of @xmath2 and also in @xmath22 . sections [ minimalgraph ] to [ sectionhmn ] will be devoted to derive the value of @xmath34 . for notational convenience only , we often rewrite @xmath69 , @xmath70 as @xmath61 , @xmath71 , respectively . for a @xmath72-graph @xmath2 , an edge in @xmath2 is said to be _ public _ if it is shared by a @xmath61-path and a @xmath71-path , _ private _ otherwise . evidently , for each @xmath10 , from @xmath11 to @xmath12 , each @xmath61 or @xmath71-path in @xmath2 induces a natural orientation to all its edges . we note that a public edge in @xmath2 may have opposite @xmath61-direction and @xmath71-direction ( such `` inconsistency '' will be dealt with in section [ sectionminimalrepresentation ] ) . we say @xmath2 can be _ naturally orientable _ if each public edge in @xmath2 has the same natural @xmath61 and @xmath71-direction . a cycle @xmath73 in @xmath2 , where @xmath74 , @xmath41 for @xmath75 and @xmath45 , is said to be a _ @xmath61-consistent cycle _ , if it satisfies the following property : for any @xmath76 , if @xmath46 belongs to a @xmath61-path , then its natural @xmath61-direction is from @xmath77 to @xmath78 . and we similarly define a @xmath71-consistent cycle . the following theorem gives necessary and sufficient conditions for a @xmath72-graph being minimal . [ equivalent ] the following three statements are equivalent for a @xmath72-graph @xmath2 : * @xmath2 is minimal ; * @xmath2 is non - reroutable ; * @xmath2 has no @xmath61 or @xmath71-consistent cycle . * 1 . * * ( ii ) @xmath79 ( i ) * : any edge @xmath80 in @xmath2 must belong to some @xmath61 or @xmath71-path . after deleting @xmath80 from @xmath2 , we no longer find @xmath13 vertex - disjoint paths from @xmath11 to @xmath12 for some @xmath81 . so @xmath82 , and therefore @xmath2 is minimal . * 2 . * * ( ii ) @xmath79 ( iii ) * : suppose , by way of contradiction , that @xmath2 has a @xmath61-consistent cycle @xmath83 , which can be written in the following form @xmath84 where each @xmath85 is a subpath of some @xmath61-path , and each @xmath46 is a private @xmath71-edge . furthermore , we assume @xmath83 has the smallest @xmath86 ( the number of private @xmath71-edges ) among all @xmath61-consistent cycles . then , each @xmath85 belongs to a different @xmath61-path ( since otherwise we can always find a @xmath61-consistent cycle with fewer private @xmath71-edges ) , which further implies that @xmath87 . suppose @xmath88 $ ] for @xmath76 , @xmath89 for @xmath90 and @xmath91 . then one can find another group of @xmath92 vertex - disjoint paths @xmath93 from @xmath94 to @xmath95 in @xmath2 , where @xmath96 \circ ( u_1,v_d ) \circ \phi_d[u_d , r_1]&\textrm { for } i=1;\\ \phi_i[s_1,u_i ] \circ ( u_i , v_{i-1 } ) \circ \phi_{i-1}[v_{i-1},r_1]&\textrm { for } 2\le i\le d;\\ \phi_i&\textrm { for } d+1\le i\le c_1,\\ \end{cases}\ ] ] which contradicts the assumption that @xmath2 is non - reroutable . with a parallel argument , we conclude that @xmath71-consistent cycles do not exist either . * 3 . * * ( i ) or ( iii ) @xmath79 ( ii ) * : suppose , by contradiction , that @xmath2 is reroutable , and furthermore , by symmetry , that there exists another group of @xmath92 vertex - disjoint paths @xmath93 from @xmath94 to @xmath95 in @xmath2 with @xmath97 sharing the same outgoing edge from @xmath94 as @xmath98 for every @xmath10 . pick a @xmath99-path , say , @xmath100 , such that @xmath101 , and let @xmath102 denote the smallest vertex on @xmath103 ( under the natural @xmath61-direction ) where they leave each other . assume that , after @xmath102 , @xmath100 first meets some @xmath61-path , say , @xmath104 at the vertex @xmath105 . denote by @xmath106 the smallest vertex where @xmath107 and @xmath104 leave each other . assume that , after @xmath106 , @xmath107 first meets some @xmath61-path , say , @xmath108 at the vertex @xmath109 . continue the procedure in a similar manner to obtain an index sequence @xmath110 , and similarly define @xmath111 s and @xmath112 s . pick the smallest @xmath113 such that @xmath114 for some @xmath115 . notice that @xmath116 is smaller than @xmath112 on @xmath117 , which easily follows from three facts : * @xmath118 first takes apart from @xmath117 at @xmath116 ; * @xmath119 meets @xmath117 at @xmath112 ; * @xmath119 and @xmath118 are vertex - disjoint . hence , we conclude that @xmath120 \circ \phi_{i_{j+1}}[u_{i_j},v_{i_{j+1 } } ] \circ \hat{\phi}_{i_{j+1}}[v_{i_{j+1}},u_{i_{j+1 } } ] \circ \phi_{i_{j+2}}[u_{i_{j+1}},v_{i_{j+2 } } ] \circ \\ \cdots & \circ \hat{\phi}_{i_{k-2}}[v_{i_{k-2}},u_{i_{k-2 } } ] \circ \phi_{i_{k-1}}[u_{i_{k-2}},v_{i_{k-1 } } ] \circ \hat{\phi}_{i_{k-1}}[v_{i_{k-1}},u_{i_{k-1 } } ] \circ \phi_{i_{j}}[u_{i_{k-1}},v_{i_{j}}]\end{aligned}\ ] ] is a @xmath61-consistent cycle in @xmath2 . since all @xmath71-paths are vertex - disjoint and the above cycle does not contain any terminal vertex , at least one edge @xmath80 in this cycle does not belong to any @xmath71-path ( that is , @xmath80 is a private @xmath61-edge ) . notice that each edge of @xmath121 $ ] , @xmath122 , is a private @xmath71-edge , and each edge of @xmath123 $ ] , @xmath122 , does not belong to any @xmath99-path . so , @xmath80 must belong to @xmath123 $ ] for some @xmath124 with @xmath122 , and hence @xmath80 does not belong to any @xmath99-path . in the graph @xmath125 , we can find a set @xmath99 of @xmath92 vertex - disjoint paths from @xmath94 to @xmath95 , and a set @xmath71 of @xmath126 vertex - disjoint paths from @xmath127 to @xmath128 . thus @xmath129 and @xmath2 is not minimal , a contradiction . [ ifreroutable ] the third part of the proof of the theorem has actually proved that for a @xmath130-graph @xmath2 , if @xmath61 ( resp . @xmath71 ) is reroutable , then there exists a private @xmath61 ( resp . @xmath71)-edge @xmath80 such that @xmath131 . this fact will be used later in the paper . in this section , we will transform a minimal @xmath130-graph @xmath2 into @xmath132 , its representation , through the following steps : * step 1 [ remove relays ] * : in this step , we remove all non - terminal vertices in @xmath2 with degree @xmath31 . in more detail , for any non - terminal vertex @xmath49 with @xmath133 and @xmath134 , @xmath135 being the two edges incident with @xmath49 , where @xmath136 , delete edges @xmath137 , @xmath138 and vertex @xmath49 , and then add a new edge @xmath139 . let @xmath140 denote the resulting graph . * step 2 [ stretch crossings ] * : we say a @xmath61-path and a @xmath71-path _ cross _ at vertex @xmath49 if they share @xmath49 , but not any edges incident with @xmath49 . in this step , we convert each crossing into a pair of degree @xmath4 vertices . in more detail , for any vertex @xmath49 in @xmath140 with @xmath141 and @xmath134 , @xmath135 being the two @xmath61-edges incident with @xmath49 , @xmath142 , @xmath143 being the two @xmath71-edges incident with @xmath49 , where all @xmath77 are all distinct , delete edges @xmath137 , @xmath138 , @xmath144 , @xmath145 and vertex @xmath49 , and then add two vertices @xmath146 , @xmath147 and edges @xmath148 , @xmath149 , @xmath150 , @xmath151 and @xmath152 . let @xmath153 denote the resulting graph . * step 3 [ match directions ] * : for any public edge @xmath80 in @xmath153 with inconsistent @xmath61 and @xmath71-direction , we will perform the following operations to obtain consistency : assume edge @xmath154 belongs to both path @xmath155 \circ(w_1,u)\circ ( u , v)\circ(v , w_2)\circ \phi_i[w_2,r_1]$ ] and path @xmath156\circ(w_3,v)\circ(v , u)\circ(u , w_4)\circ\psi_j[w_4,r_2]$ ] . we delete edges @xmath157 , add edges @xmath158 , and then obtain a new @xmath71-path @xmath159\circ ( w_3,u)\circ ( u , v)\circ ( v , w_4)\circ \psi_j[w_4,r_2].\ ] ] let @xmath132 denote the resulting graph , which , evidently , is naturally orientable ; let @xmath160 denote the directed version of @xmath161 , equipped with the consistent natural orientation induced on all its @xmath61 and @xmath71-paths . apparently , @xmath162 with its @xmath61 and @xmath71-paths determined by the original ones , and all the non - terminal vertices in @xmath161 are hubs . the obtained @xmath132 after the above three steps is said to be a _ representation _ of @xmath2 . the following lemma states that @xmath161 must be a minimal @xmath72-graph as well . [ minimalrepresentation ] the representation @xmath161 of a minimal @xmath72-graph @xmath2 is also minimal . first , since @xmath2 is minimal , by theorem [ equivalent ] , @xmath2 is non - reroutable . evidently , @xmath140 is also non - reroutable , and thus minimal . by way of contradiction , suppose that @xmath132 is not minimal . again , by theorem [ equivalent ] , @xmath132 is reroutable . notice that in step 2 and 3 , both crossings or inconsistently oriented public edges are converted into consistently oriented public edges . then , by remark [ ifreroutable ] , one can find a private edge @xmath80 such that @xmath163 , implying @xmath164 also belongs to @xmath38 , which contradicts the minimality of @xmath140 . let @xmath165 denote the subset of @xmath38 consisting of all networks @xmath2 such that @xmath2 is minimal , naturally orientable and all the non - terminal vertices in @xmath2 are of degree 3 . apparently , @xmath166 is the set of the representations of all minimal @xmath130-graphs . the following theorem says that in order to compute @xmath34 , it is enough to only consider all graphs in @xmath165 . @xmath167 the `` @xmath168 '' direction follows from the observation that @xmath169 and the `` @xmath170 '' direction immediately follows from lemma [ minimalrepresentation ] . a path in @xmath171 is said to be an _ alternating _ path if all its edges are privates edges and the terminal pair of this path is one of the following : @xmath172 , @xmath173 , @xmath174 or @xmath175 . an alternating path has the following properties : 1 . each of its @xmath61-edges is only adjacent to @xmath71-edges , and each of its @xmath71-edges is only adjacent to @xmath61-edges . 2 . each of its @xmath61 ( resp . @xmath71)-edge belongs to a different @xmath61 ( resp . @xmath71)-path . this follows from the fact that after step @xmath176 , vertices with degree @xmath31 have been removed and thus no two private @xmath61 ( or @xmath71)-edges are adjacent . we show that in any alternating path , each @xmath61-edge belongs to a different @xmath61-path . suppose , by contradiction , that for an alternating path path @xmath177 , where @xmath74 , two edges @xmath178 , @xmath179 , both belong to @xmath180 . if @xmath181 is smaller than @xmath182 on path @xmath180 in @xmath183 , then @xmath184\circ e_{k+1 } \circ e_{k+2 } \circ \cdots \circ e_{l-1}\ ] ] is a @xmath61-consistent cycle in @xmath132 , which , by theorem [ equivalent ] , gives us a contradiction . if @xmath182 is smaller than @xmath181 on path @xmath180 in @xmath183 , then the subpath @xmath185 $ ] in fact can be expressed as @xmath186 , where @xmath187 are public and @xmath188 are private . then @xmath189 is a @xmath71-consistent cycle in @xmath132 , which , by theorem [ equivalent ] , gives us a contradiction . with a parallel argument , we conclude that each private @xmath71-edge belongs to a different @xmath71-path . we next present the structural decomposition theorem of a representation @xmath190 , which , roughly speaking , states that after deleting public edges in @xmath161 , each connected component in the resulting graph is an alternating path . more precisely , letting @xmath191 denote the subgraph of @xmath132 induced on all private edges in @xmath161 , we have [ decomposition ] @xmath191 consists of @xmath192 alternating paths . consider @xmath193 , the naturally oriented version of @xmath191 . obviously , the degree of any non - terminal vertex in @xmath193 is @xmath31 . now , starting from @xmath94 , traverse along an outgoing @xmath61-edge , say , @xmath137 , and then traverse against the @xmath71-edge adjacent to @xmath137 , say , @xmath138 , and then along a @xmath61-edge adjacent to @xmath138 , say , @xmath144 , and then against a @xmath71-edge adjacent to @xmath144 . continue the procedure in this fashion , we will always reach @xmath95 or @xmath127 , since otherwise the set of edges that we have traversed will contain a cycle , which is both @xmath61-consistent and @xmath71-consistent . evidently , a similar argument can be applied to the case when we start from an incoming edge incident with @xmath128 . it then follows that one can find a set of @xmath192 vertex - disjoint paths in @xmath191 from @xmath194 to @xmath195 ; moreover , it can easily checked that each edge in @xmath191 belong to one of the above - mentioned vertex - disjoint paths . [ head - tail ] let @xmath49 be a non - terminal vertex of an alternating path in @xmath191 and let @xmath196 be the two edges incident with @xmath49 . it then follows from the proof of theorem [ decomposition ] that if @xmath49 is the head ( resp . tail ) of @xmath80 in @xmath193 , then it is also the head ( resp . tail ) of @xmath197 . an alternating path in @xmath191 is said to be an @xmath198 , @xmath199 , @xmath200 , @xmath201-alternating path if its terminal pair is @xmath172 , @xmath173 , @xmath174 or @xmath175 , respectively . a path is said to be an @xmath94-alternating path is it is either an @xmath198-alternating path or @xmath199-alternating path , similarly , a path is also referred to as an @xmath128-alternating path is it is either an @xmath200-alternating path or @xmath201-alternating path . apparently , there are @xmath92 @xmath94-alternating paths and @xmath126 @xmath128-alternating paths . for two vertices @xmath202 of an @xmath94 ( resp . @xmath128)-alternating path @xmath203 , we say @xmath48 is on the right of @xmath49 in @xmath203 if @xmath49 is `` nearer '' to @xmath94 ( resp . @xmath128 ) than @xmath48 in @xmath203 ( more precisely , the number of edges between @xmath48 and @xmath94 ( resp . @xmath128 ) in @xmath203 is more than that between @xmath49 and @xmath94 ( resp . @xmath128 ) ) ( the word `` right '' arises since we will `` position '' the vertices of a path in a plane in section [ xu - algorithm ] for an easy illustration ) . for two edges @xmath204 in @xmath203 , we say @xmath80 is on the right of @xmath197 if one of two vertices incident with @xmath80 is on the right of the two ones incident with @xmath197 in @xmath203 . figure [ illustrator ] shows a naturally oriented minimal @xmath205-graph with @xmath206 hubs . from @xmath94 to @xmath95 , there are two vertex - disjoint @xmath61-paths @xmath207 and @xmath208 ; and from @xmath127 to @xmath128 , there are two vertex - disjoint @xmath71-paths @xmath209 and @xmath210 . the edges @xmath144 , @xmath211 , @xmath212 , @xmath213 are public since each of them is shared by some @xmath61-path and some @xmath71-path . the others are all private . then we can find that @xmath191 consists of four alternating paths : two @xmath94-alternating paths @xmath214 and two @xmath128-alternating paths @xmath215 moreover , @xmath216 is an @xmath217-alternating path , where edge @xmath218 is on the right of edge @xmath219 , vertex @xmath220 is on the right of vertex @xmath29 . in this section , we introduce an algorithm , the analysis of which will be instrumental for deriving the value of @xmath34 . before rigorously describing the algorithm , we roughly illustrate its idea . consider a representation @xmath161 of a minimal @xmath130-graph @xmath2 . imagine on a two - dimensional plane , each alternating path @xmath73 in @xmath191 is `` positioned '' into a `` double deck '' ( see figure [ alternatingpath ] for an example ) , where @xmath221 is on the right of @xmath46 , and all the tails ( in @xmath222 ) are on the upper deck and heads on the lower deck ( see remark [ head - tail ] ) . a vertex on the upper deck is referred to as an _ upper vertex _ if it is not a source , and a vertex on the lower deck is referred to as a @xmath223 if it is not a sink . notice that there is one more lower vertices than upper vertices in an @xmath217-alternating path , while one less in an @xmath224-alternating path , and there are equally many of them in an @xmath225-alternating path or an @xmath226-alternating path . -alternating path , @xmath227 , @xmath228 , @xmath229 are upper vertices and @xmath230 , @xmath231 are lower vertices.,scaledwidth=50.0% ] in each @xmath217-alternating ( resp . @xmath224-alternating ) path , one of the lower ( resp . upper ) vertices will be labeled as a @xmath232 ( which could be a `` bottleneck '' for the path extending procedure in the algorithm ) , which is initially the rightmost lower ( resp . rightmost upper ) vertex of the path and will be updated during the execution of the algorithm . as summarized below , the algorithm will iteratively find a set of the so - called _ interconnecting _ paths in @xmath161 , which will `` link '' the double decks . * in the beginning of each iteration , we initialize the interconnecting path as a public edge , whose tail is an unoccupied lower vertex . see step 2 . * we first traverse along edges in @xmath160 to extend the interconnecting paths to link the double decks . when we reach a choke vertex , we check if it is possible to switch paths ( see figure [ switching ] ) and find new ways to extend the interconnecting paths . see step 3(a ) , 3(b ) . * then we traverse against edges in @xmath160 to extend the interconnecting paths in a parallel manner . see step 5(a ) , 5(b ) . * at the end of @xmath113-th iteration , a set of @xmath113 interconnecting paths is found , whose vertices are all labeled as occupied . see step 6 . as proven later , the algorithm will produce all interconnecting paths and upon termination of the algorithm , all hubs in @xmath161 are occupied . let @xmath233 be the number of @xmath217-alternating paths , then @xmath234 , and moreover , the numbers of @xmath224-alternating paths , @xmath225-alternating paths and @xmath226-alternating paths are @xmath235 , respectively . the following algorithm will find @xmath233 interconnecting paths in @xmath161 . [ algorithm ] input : @xmath161 ; output : a set @xmath236 of @xmath233 interconnecting paths . step 1 : initialize the algorithm . : : orient each edge in the direction of @xmath61 and @xmath71-paths ; + label all hubs as `` unoccupied '' ; + * for * each @xmath217-alternating path * do * + its choke vertex @xmath237 its rightmost lower vertex ; + * for * each @xmath224-alternating path * do * + its choke vertex @xmath237 its rightmost upper vertex ; + @xmath238 ; + @xmath239 . step 2 : initialize the forward extension . : : @xmath240 an arbitrarily picked unoccupied lower vertex and label @xmath49 as `` occupied '' ; + @xmath241 the public edge whose tail is @xmath49 ; + @xmath242 and label @xmath48 as `` occupied '' ; + @xmath243 . step 3(a ) : prepend a private edge . : : @xmath244 the alternating path to which @xmath48 belongs ; + * if * @xmath203 is an @xmath224-alternating path * and * @xmath48 is the choke vertex of @xmath203 * then * * begin * + * if * there are no unoccupied upper vertices in @xmath203 * then * + go to * step 4 * ; + * else * * begin * + @xmath245 the rightmost unoccupied upper vertex of @xmath203 ; + @xmath246 the subpath of @xmath203 between @xmath247 and @xmath48 ; + * if * @xmath248 * then * * begin * + the choke vertex of @xmath249 ; + label @xmath250 as `` occupied '' ; + @xmath251 ; + @xmath252 ; + * end * + * else * ( _ viz . _ @xmath253 ) * begin * + * for * @xmath254 * to * @xmath86 * do * + @xmath255 the interconnecting path in @xmath236 containing @xmath256 ; + @xmath257 ; + the choke vertex of @xmath249 ; + label @xmath250 as `` occupied '' ; + @xmath258 ; + @xmath259 ; + @xmath260 \circ e$ ] ; + * for * @xmath254 * to * @xmath261 * do * + @xmath262 \circ ( x_{i+1},y_i ) \circ p_{i}[y_i , h(p_{i})]$ ] ; + @xmath263 \circ ( u , y_d ) \circ p_{d}[y_d , h(p_{d})]$ ] ; + @xmath264 ; + * end * + * end * + * end * + * else * * begin * + * if * @xmath265 an @xmath94-alternating path * then * + @xmath266 the private @xmath71-edge whose tail is @xmath48 ; + * else * ( _ viz . _ @xmath267 an @xmath128-alternating path ) + @xmath266 the private @xmath61-edge whose tail is @xmath48 ; + label @xmath268 as `` occupied '' ; + @xmath269 . + * end * step 3(b ) : prepend a public edge . : : @xmath241 public edge whose tail is @xmath268 ; + @xmath242 and label @xmath48 as `` occupied '' ; + @xmath270 ; + go to * step 3(a)*. step 4 : initialize the backward extension . : : @xmath271 ; + @xmath272 the interconnecting path in @xmath236 whose tail is @xmath49 ; + @xmath273 ; + @xmath274 . step 5(a ) : append a private edge . : : @xmath275 the alternating path to which @xmath276 belongs ; + * if * @xmath203 is an @xmath217-alternating path * and * @xmath276 is the choke vertex of @xmath203 * then * * begin * + * if * there is no unoccupied lower vertex in @xmath203 * then * + go to * step 6 * ; + * else * * begin * + @xmath277 the rightmost unoccupied lower vertex of @xmath203 ; + @xmath278 the subpath of @xmath203 between @xmath250 and @xmath276 ; + * if * @xmath248 * then * * begin * + the choke vertex of @xmath279 ; + label @xmath247 as `` occupied '' ; + @xmath280 ; + @xmath281 ; + * end * + * else * ( _ viz . _ @xmath253 ) * begin * + * for * @xmath254 * to * @xmath86 * do * + @xmath282 the interconnecting path in @xmath236 containing @xmath256 ; + @xmath257 ; + the choke vertex of @xmath279 ; + label @xmath247 as `` occupied '' ; + @xmath283 ; + @xmath259 ; + @xmath284 $ ] ; + * for * @xmath254 * to * @xmath261 * do * + @xmath285 \circ ( x_i , y_{i+1 } ) \circ p_{{i+1}}[y_{i+1 } , h(p_{{i+1}})]$ ] ; + @xmath286 \circ ( x_d , w ) \circ \widehat{p}[w , h(\widehat{p})]$ ] ; + @xmath264 ; + * end * + * end * + * end * + * else * * begin * + * if * @xmath287 an @xmath128-alternating path * then * + @xmath266 the private @xmath61-edge whose head is @xmath276 ; + * else * ( _ viz . _ @xmath287 an @xmath94-alternating path ) + @xmath266 the private @xmath71-edge whose head is @xmath276 ; + label @xmath288 as `` occupied '' ; + @xmath289 . + * end * step 5(b ) : append a public edge . : : @xmath241 public edge whose head is @xmath288 ; + @xmath290 and label @xmath276 as `` occupied '' ; + @xmath291 ; + go to * step 5(a)*. step 6 : terminate the iteration . : : * if * @xmath292 * then * + terminate the algorithm ; + @xmath271 ; + @xmath293 ; + go to * step 2*. before switching paths 0.5 cm after switching paths * 1 . * for step 2 in each iteration , an unoccupied lower vertex always exists since there is at least one @xmath217-alternating path with its chock vertex left unoccupied . each iteration of algorithm [ algorithm ] consists of forward and backward extensions , and each interconnecting path starts from an @xmath217-alternating path and ends at an @xmath201-alternating path . the algorithm can not be simplified into a `` one - direction extension '' version : for each iteration , the extending procedure will terminate at an @xmath217-alternating path with only one unoccupied lower vertex , but , in the beginning of each iteration , such an alternating path may not exist ( to see this , notice that more than two unoccupied vertices of an alternating path can become `` occupied '' , if we switch paths during one iteration ) . we will need the following two lemmas in the next section . [ distinctaasequence ] for a representation @xmath161 of a minimal @xmath130-graph , upon the termination of algorithm [ algorithm ] , each private edge of any interconnecting path belongs to a different alternating path . by contradiction , suppose that for an interconnecting path @xmath294 , two private edges @xmath295 and @xmath296 , @xmath297 , belong to the same alternating path @xmath203 . let @xmath83 be the cycle @xmath298\circ l[t(e_{2l}),h(e_{2k})].\ ] ] if @xmath296 is on the right of @xmath295 in @xmath203 , then @xmath83 is a @xmath71-consistent cycle in @xmath161 . if @xmath295 is on the right of @xmath296 in @xmath203 , then @xmath83 is a @xmath61-consistent cycle in @xmath161 . by theorem [ equivalent ] , both cases imply that @xmath161 is not minimal , which , by lemma [ minimalrepresentation ] , further implies that @xmath2 is not minimal , a contradiction . [ atmostmconnectingpaths ] for a representation @xmath161 of a minimal @xmath130-graph @xmath2 , upon the termination of algorithm [ algorithm ] , @xmath236 consists of at most @xmath299 interconnecting paths , and each hub in @xmath161 is in exactly one of the interconnecting paths in @xmath236 . each interconnecting path in @xmath236 starts from a lower vertex of a distinct @xmath217-alternating path and ends to an upper vertex of a distinct @xmath224-alternating path . so the number of interconnecting paths is @xmath233 with @xmath234 . now upon the termination of algorithm [ algorithm ] , pick an unoccupied lower vertex , and then execute the algorithm from step 3(b ) ; or pick an unoccupied upper vertex , and then execute the algorithm from step 3(a ) . since all chock vertices have been unoccupied , the algorithm will fail to terminate , violating the fact that @xmath132 is finite . [ nmn2mn ] @xmath300 * the `` @xmath301 '' direction : * for a representation @xmath161 of a minimal @xmath130-graph @xmath2 , apply algorithm [ algorithm ] to obtain a set @xmath236 of @xmath233 interconnecting paths . in the @xmath124-th iteration of the algorithm , the forward extension stops at the choke vertex of some @xmath224-alternating path . let @xmath302 denote this alternating path . note that after @xmath124 iterations , 1 ) all hubs of @xmath302 become occupied ; 2 ) by lemma [ distinctaasequence ] , each of the @xmath124 obtained interconnecting paths contains at most two hubs of @xmath302 ; 3 ) one obtained interconnecting path contains exactly one hub of @xmath302 . hence , we deduce that the number of hubs of @xmath302 is at most @xmath303 , and therefore the total number of hubs of all @xmath224-alternating paths is at most @xmath304 . similarly , the total number of hubs of all @xmath217-alternating paths is at most @xmath305 as well . again , by lemma [ distinctaasequence ] , the total number of hubs of all @xmath225 and @xmath226-alternating paths is at most @xmath306 . by lemma [ atmostmconnectingpaths ] , each hub in @xmath161 belongs to some interconnecting path in @xmath236 . therefore , @xmath307 where the last inequality follows from @xmath234 . * the `` @xmath170 '' direction : * we only need to construct a minimal @xmath130-graph @xmath2 with @xmath308 hubs ( see figure [ picn33 ] for an example ) . the graph @xmath2 can be described as follows : @xmath309 is naturally oriented , and there is a set of @xmath92 vertex - disjoint paths @xmath310 from @xmath94 to @xmath95 and a set of @xmath126 vertex - disjoint paths @xmath311 from @xmath127 to @xmath128 . furthermore , in @xmath312 , the directed version of @xmath2 , paths @xmath98 and @xmath313 meet at vertex @xmath314 and depart at vertex @xmath315 , for @xmath316 , @xmath317 , and * on path @xmath98 , @xmath318 * on path @xmath313 , @xmath319 -graph with @xmath320 hubs.,scaledwidth=40.0% ] in this section , we are concerned with @xmath32 with more than two parameters , which turns out to be much more difficult to compute than @xmath32 with two parameters . we will use an inductive argument on @xmath7 . notice that the case when @xmath323 has been established in theorem [ nmn2mn ] . by way of induction , we assume that the theorem is true for @xmath324 and proceed to prove it for @xmath325 . consider any minimal @xmath326-graph @xmath2 . let @xmath60 denote the set of @xmath13 vertex - disjoint paths from @xmath11 to @xmath12 for @xmath327 . after necessary rerouting of @xmath328 within @xmath329 , we can assume that @xmath329 is minimal and thus @xmath330 let @xmath16 denote the subgraph of @xmath329 induced on all the edges , each of which is simultaneously an @xmath328-edge , and let @xmath331 denote the number of connected components in @xmath16 . obviously , each connected component of @xmath16 is in fact a path , and therefore we have @xmath332 . in this proof , we say a hub in @xmath2 is _ new _ if it is a hub in @xmath2 , however , not one in @xmath329 . and we say a new hub is _ global _ , if this hub belongs to @xmath16 , _ local _ , if this hub is in @xmath333 . for each @xmath338-path , we `` cut '' it at each of its local new hubs and then `` divide '' the path into `` segments '' , each of which , evidently , is a subpath of the original @xmath339-path . let @xmath340 denote the set of all obtained subpaths . then we have @xmath341 . 1 . add sources @xmath345 , sinks @xmath346 . 2 . for any @xmath347 and for any connected component of @xmath16 , whose natural @xmath60-direction is from one end vertex , say @xmath146 , to the other end vertex , say @xmath147 , add two directed edges @xmath348 and @xmath349 ( notice that a connected component may have opposite @xmath350-direction and @xmath351-direction for different @xmath352 , @xmath353 ) . then , for any @xmath354 , we obtain a group @xmath355 of @xmath331 vertex - disjoint paths from @xmath356 to @xmath357 . 3 . for any @xmath340-path , whose natural @xmath339-direction is from one end vertex , say @xmath146 , to the other end vertex , say @xmath147 , add two directed edges @xmath358 and @xmath359 . then , we obtain a group @xmath360 of @xmath361 vertex - disjoint paths from @xmath362 to @xmath363 . obviously , the number of global new hubs in @xmath2 is just @xmath364 . it follows from the minimality of @xmath2 and the observation that for any @xmath365 , any @xmath355-consistent cycle in @xmath344 naturally corresponds an @xmath60-consistent cycle in @xmath2 that @xmath344 is a minimal @xmath343-graph . we then proceed to deduce that @xmath360 is non - reroutable in @xmath344 , since otherwise there exists an edge @xmath80 which is exclusively owned by @xmath360-paths ( this follows from a parallel argument leading to remark [ ifreroutable ] ) such that @xmath366 , violating the fact that @xmath344 is minimal . now , after necessary rerouting of @xmath355 within @xmath344 , we assume that @xmath367 , the subgraph of @xmath344 consisting of all @xmath355-paths and @xmath360-paths , is non - reroutable and thus , by theorem [ equivalent ] , minimal . notice that any hub @xmath49 in @xmath344 must be a hub of some @xmath367 , since otherwise @xmath344 contains an edge incident with @xmath49 that does not belong to any @xmath367 and therefore is not minimal . hence , by theorem [ nmn2mn ] , we obtain that @xmath368 therefore , @xmath369 the proof is then complete . * the `` @xmath168 '' direction : * we will establish this direction using an inductive argument on @xmath7 . suppose the inequality holds when @xmath375 . consider a minimal @xmath376-graph @xmath2 with @xmath92 vertex - disjoint paths @xmath377 from @xmath94 to @xmath95 , @xmath126 vertex - disjoint paths @xmath378 from @xmath127 to @xmath128 , and a path @xmath379 from @xmath11 to @xmath12 for @xmath380 . let @xmath140 be the subgraph of @xmath2 induced on @xmath381 now we split @xmath94 into @xmath92 copies @xmath382 ; @xmath95 into @xmath92 copies @xmath383 ; @xmath127 into @xmath126 copies @xmath384 ; @xmath128 into @xmath126 copies @xmath385 , such that @xmath98 has starting point @xmath386 and ending point @xmath387 for @xmath316 ; @xmath388 has starting point @xmath389 and ending point @xmath390 for @xmath391 . let @xmath153 denote the resulting graph and @xmath331 be the number of weakly connected components ( which means connected components when disregarding the orientation ) in @xmath153 . now for any @xmath392 , we identify @xmath386 and @xmath393 if they belong to the same component and we perform similar operations on @xmath387 s , @xmath389 s and @xmath390 s . let @xmath16 denote the resulting graph , which consists of @xmath331 connected components . note that the @xmath10-th component @xmath394 is in fact a minimal @xmath395-graph , where @xmath396 , @xmath397 and @xmath398 . notice that for any @xmath10 , at least one of @xmath399 , @xmath400 and @xmath401 is nonzero , but some @xmath399 s , @xmath400 s , @xmath401 s can be zero , for which case , a @xmath395-graph can be interpreted as if all zero - valued parameters are simply dropped . for each component @xmath394 , by the induction hypothesis , we have a hub in @xmath2 is said to be _ new _ if it is not a hub in @xmath16 . notice that the path @xmath406 `` meets '' each component at most once , yielding at most two new hubs ( more precisely , it meets one edge in @xmath394 , then departs from one edge in @xmath394 , and thereafter , it will never meet any edge in @xmath394 again ) , otherwise @xmath2 is not minimal . hence , @xmath407\\ \le & \ 2\omega+2(c_1+c_2+t-1-\omega)+2\left[\sum_{i=1}^\omega(c_{1,i}-1)^+\right]\cdot\left[\sum_{i=1}^\omega(c_{2,i}-1)^+\right]\\ \le & \ 2(c_1+c_2+t-1)+2(c_1 - 1)(c_2 - 1)\\ = & \ 2(c_1c_2+t).\end{aligned}\ ] ] it turns out the graph @xmath2 , described below in detail , is such a graph : @xmath2 is naturally orientable ; there is @xmath92 vertex - disjoint paths @xmath377 from @xmath94 to @xmath95 , @xmath126 vertex - disjoint paths @xmath378 from @xmath127 to @xmath128 , and a path @xmath379 from @xmath11 to @xmath12 for @xmath410 ; paths @xmath98 and @xmath313 meet at vertex @xmath314 and depart at vertex @xmath315 for @xmath316 and @xmath317 ; paths @xmath379 and @xmath411 meet at vertex @xmath412 and depart at vertex @xmath413 for @xmath410 . furthermore , in @xmath312 , we have so we only need to prove the other direction . by contradiction , suppose that a minimal @xmath420-graph @xmath2 has at least @xmath422 hubs , and there exists a set of two vertex - disjoint path @xmath423 from @xmath94 to @xmath95 , a set of two vertex - disjoint paths @xmath424 from @xmath127 to @xmath128 , and a set of two vertex - disjoint paths @xmath425 from @xmath426 to @xmath427 . here , we note that when there are three pairs of sources and sinks , the equivalence statements as in theorem [ equivalent ] do not hold any more : it turns out that a minimal graph can be reroutable ( see figure [ n222minimalreroutable ] ) . * we first consider the case @xmath2 is minimal and non - reroutable . * let @xmath428 be the subgraph of @xmath2 induced on the edges of @xmath61-paths and @xmath71-paths ; similarly , we define @xmath429 and @xmath430 . by theorem [ equivalent ] , these three subgraphs are all minimal , since they are all non - reroutable . suppose @xmath428 has the most hubs among them . every hub in @xmath2 belongs to at least one of these three subgraphs , so we have @xmath431 and hence @xmath432 . now , we transform @xmath428 into a corresponding graph @xmath433 by shrinking each public edge into a vertex ( this can be regarded as the `` inverse '' operation of step 2 in section [ sectionminimalrepresentation ] ) . in more detail , for a public edge @xmath148 , say @xmath434 are two private edges incident with @xmath146 , and @xmath435 , @xmath436 are two private edges incident with @xmath147 . then we delete edges @xmath137 , @xmath138 , @xmath144 , @xmath145 and vertices @xmath146 , @xmath147 , and then add a new vertex @xmath49 and edges @xmath437 , @xmath438 , @xmath439 , @xmath440 . if @xmath428 has at least five hubs , then as shown in figure [ n222threecases ] , up to isomorphism , @xmath433 has three possibilities ( note that the first and second are different ) . next , we examine in the ways one can add @xmath441-paths into @xmath442 to form @xmath2 such that @xmath2 is minimal . in the following , a hub in @xmath2 is said to be @xmath443 if it is not a hub in @xmath428 , and an edge is said to be @xmath443 if it does not belong to @xmath428 and not incident with @xmath426 or @xmath427 . * case 1 : * this case is shown in figure [ n222threecases](a ) , where each edge is labeled . since @xmath433 has four hubs , @xmath428 has at most eight hubs . and since @xmath444 , @xmath445 , @xmath446 have to be added to generate at least @xmath447 new hubs . without loss of generality , say , @xmath445 contains at least three new hubs . observe that each of these new hubs is incident with exactly one @xmath445-edge that does not belong to any @xmath61 or @xmath71-path , and @xmath426 and @xmath427 are also incident with one such @xmath445-edge . so at least @xmath448 edges of @xmath445 are exclusively owned by @xmath445 , and thereby at least one of them , say @xmath449 , is a new edge ( not incident with @xmath426 or @xmath427 ) . we next discuss the possible locations of @xmath449 ; here , notice that it is possible that either @xmath146 or @xmath147 is not a new hub . suppose @xmath450 ( this means @xmath146 is one of vertices in @xmath451 , including its two end vertices ) . then , one verifies that if @xmath452 or @xmath453 , then @xmath61 is reroutable ; if @xmath454 or @xmath455 , then @xmath71 is reroutable . for example , if @xmath456 , then we can find another two vertex - disjoint @xmath71-paths @xmath457\circ \phi_2[a , v_1]\circ ( v_1,v_2)\circ \psi_2[v_2,r_2]\end{aligned}\ ] ] from @xmath127 to @xmath128 , and thus @xmath428 is reroutable . and if @xmath458 , then we can find another two vertex - disjoint paths @xmath459\circ \psi_2[b , a]\circ \phi_2[a , r_1],\\ \phi_2'&=\phi_2[s_1,v_1]\circ ( v_1,v_2)\circ \phi_1[v_2,r_1]\end{aligned}\ ] ] from @xmath94 to @xmath95 , and thus @xmath428 is reroutable . hence , we only need to check the last possible case : all new edges ( including @xmath148 ) are incident with public edges in @xmath428 . by symmetry , say , @xmath464 . then , if @xmath465 or @xmath220 , @xmath71 is reroutable ; if @xmath466 , @xmath61 is reroutable ; if @xmath467 , there must be a new edge @xmath468 of @xmath446 with ( @xmath469 and @xmath470 ) or ( @xmath471 and @xmath472 ) ( see figure [ n222case1and2finalsentence](a)(b ) ) , since otherwise @xmath473 is still in @xmath474 , which contradicts the fact that @xmath2 is minimal . however , in both cases , we can still find another two vertex - disjoint paths @xmath475\circ ( v_1',v_2')\circ \phi_2[v_2',r_1],\\ \phi_2'&=\phi_2[s_1,v_1]\circ ( v_1,v_2)\circ \phi_1[v_2,r_1]\end{aligned}\ ] ] from @xmath94 to @xmath95 , and thus @xmath428 is reroutable . hence , we only need to check the last possible case : all new edges ( including @xmath468 ) are incident with public edges in @xmath428 . by symmetry , say @xmath464 or @xmath220 . if @xmath464 and @xmath483 or @xmath484 , then @xmath61 is reroutable ; if @xmath464 and @xmath485 , then @xmath71 is reroutable ; if @xmath486 and @xmath487 or @xmath484 , @xmath71 is reroutable ; if @xmath486 and @xmath466 , there must be a new edge @xmath468 of @xmath446 with @xmath469 and @xmath470 ( see figure [ n222case1and2finalsentence](c ) ) , since otherwise @xmath488 is still in @xmath474 , which contradicts the fact that @xmath2 is minimal . however , we can still find another two vertex - disjoint paths @xmath489\circ ( v_1,v_2)\circ \phi_2[v_2,r_1],\\ \phi_2'&=\phi_2[s_1,v_2']\circ ( v_2',v_1')\circ \phi_1[v_1',r_1]\end{aligned}\ ] ] from @xmath94 to @xmath95 , and thus @xmath428 is reroutable . * case 3 : * this case is shown in figure [ n222threecases](c ) , where each edge is also labeled . since @xmath433 has three hubs , @xmath428 has at most six hubs . and since @xmath444 , @xmath445 , @xmath446 have to be added to generate at least @xmath490 new hubs . so at least @xmath491 edges of @xmath445 and @xmath446 do not belong to any @xmath61 or @xmath71-path , and thereby at least two of them , say , @xmath148 and @xmath468 are two new edges . we next discuss their possible locations . let @xmath504 be the subgraph of @xmath2 induced on the edges of @xmath505 $ ] and @xmath506 $ ] and @xmath507 be the subgraph induced on the edges of @xmath508 $ ] and @xmath509 $ ] . if @xmath510 , either @xmath61 or @xmath71 is reroutable , and hence @xmath511 or @xmath512 . similarly , we deduce that @xmath513 or @xmath514 . then we just need to check the last possibility : @xmath515 and @xmath516 , for which we can also easily rule out the subcases when ( 1 ) @xmath517 and @xmath518 ( 2 ) @xmath519 and @xmath520 ( 3 ) @xmath521 ( 4 ) @xmath522 ( 5 ) @xmath523 and @xmath524 ( 6 ) @xmath525 and @xmath526 . so , in the following , we examine the remaining two subcases : \(2 ) @xmath525 and @xmath524 . by symmetry , we assume @xmath527 on @xmath532 in @xmath529 . if @xmath530 , then @xmath441 is reroutable ; if @xmath531 , then @xmath71 is reroutable . now , we are ready to conclude that any minimal and non - reroutable graph in @xmath474 has at most @xmath206 hubs . * we next consider the case when @xmath2 is a minimal but reroutable @xmath420-graph*. suppose @xmath441 is reroutable in the subgraph @xmath430 . by theorem [ equivalent ] , there exists a @xmath441-consistent cycle @xmath84 where each @xmath85 is a subpath of some @xmath441-path and each @xmath46 is a private @xmath71-edge ( in @xmath533 ) . we further assume that @xmath86 ( the number of private @xmath71-edges ) is the smallest among all possible @xmath441-consistent cycles in @xmath533 . then , each @xmath46 belongs to a different @xmath71-path , which implies @xmath534 . since @xmath2 is minimal , each @xmath441-private edge ( in @xmath533 ) in @xmath535 must belong to @xmath61-paths as well . so , each edge of the @xmath441-consistent cycle in @xmath2 belongs to either some @xmath61-path or @xmath71-path . note that for any edge @xmath80 of this cycle , @xmath536 does not belong to @xmath33 , since otherwise , together with the observation that @xmath441 can be rerouted using edges in @xmath537 , we deduce that @xmath538 , which contradicts the minimality of @xmath2 . then we can find a minimal subgraph @xmath539 of @xmath428 , which is of either case 1 ( when @xmath540 ) or case 2 ( when @xmath541 ) . let @xmath542 be the subgraph induced on the edges of @xmath441-paths . then @xmath543 , by the minimality of @xmath2 . notice that all contradictions in case 1 and case 2 arise from the rerouting of @xmath61 or @xmath71-paths . so we can follow similar arguments and conclude @xmath544 . the proof is then complete .

an exact expression for the spin - spin correlation function is derived for the zero - temperature random - field ising model defined on a bethe lattice of arbitrary coordination number . the correlation length describing dynamic spin - spin correlations and separated from the intrinsic topological length scale of the bethe lattice is shown to diverge as a power law at the critical point . the critical exponents governing the behaviour of the correlation length are consistent with the mean - field values found for a hypercubic lattice with dimension greater than the upper critical dimension .

1106.3424

the zero - temperature random - field ising model ( zt - rfim ) is a prototype model for systems that exhibit avalanche dynamics when slowly driven accross athermal first - order phase transitions @xcite . examples of such transitions include the condensation of fluids in porous media @xcite , the martensitic transformation @xcite , or magnetisation reversal of ferromagnets ( barkhausen effect ) @xcite . for experimentally reasonable time scales , thermal fluctuations do not play an important role and the dynamics in this kind of system proceed along a non - equilibrium path consisting of metastable states @xcite . avalanches are a manifestation of such behaviour and correspond to the driven - induced passage between two metastable states . the zt - rfim predicts that the properties of avalanches ( e.g. their size or duration ) are drastically affected by the degree of quenched disorder in the system . more explicitly , the model predicts three different regimes for avalanche behaviour depending on the degree of disorder . the main feature of the small - disorder regime is the existence of an avalanche that is infinite in extent . in contrast , high degrees of disorder lead to a regime where all avalanches are small . these two regimes are separated by an intermediate situation where the model exhibits critical behaviour ( i.e. a continuous phase transition ) @xcite . for any degree of disorder , avalanche - like dynamics lead to spacial correlations in observable properties averaged over quenched disorder ( e.g. magnetisation , stress or fluid density ) , which extend over a typical length scale , called the correlation length . this quantity is expected to diverge in the critical regime , and this has been shown to be the case in the systems described by mean field @xcite and hypercubic lattices @xcite . concerning other topologies , including a bethe lattice , the situation is less clear . previous studies of the zt - rfim on the bethe lattice have obtained exact results for the magnetisation @xcite , the avalanche size distribution @xcite , the different contributions to the energy @xcite , and the number of metastable states @xcite . in particular , it has been demonstrated that for coordination numbers @xmath0 the system exhibits a discontinuity in its magnetisation hysteresis loop for small amounts of disorder associated with the infinite avalanche . the universality class of the critical point on such a lattice has been suggested to be the same as that of the mean - field system ( complete graph ) @xcite , with the critical exponents for the order parameter , @xmath1 and @xmath2 , being identical to the mean - field values @xcite . the spatial correlations of the model have also been investigated in the past but mostly using numerical simulations @xcite or approximate analytical methods based on mean - field descriptions and/or renormalization group ( rg ) techniques @xcite . in this paper , we derive an exact analytical expression for the spin - spin correlation function corresponding to the zt - rfim with spins placed on a bethe lattice . our results are exact for any coordination number and confirm the validity of the functional form for the correlation function derived independently in @xcite , where it is assumed that the extension from the one - dimensional case ( i.e. with @xmath3 ) is valid . it is well established that in a hypercubic lattice of dimensionality @xmath4 around criticality , the correlation function , @xmath5 , decays exponentially with distance @xmath6 for the zt - rfim , @xmath7 where @xmath8 obeys a power law for large @xmath6 , and @xmath9 is the correlation length . at the critical point , @xmath10 , and so @xmath8 represents the critical behaviour of the correlation function , found to be @xmath11 with anomalous dimension @xmath12 @xcite . in a prototype loopless topology , the cayley tree or bethe lattice , the correlation function exhibits additional exponential behaviour due to the associated hyperbolic topology , @xmath13 where @xmath14 is the coordination number and @xmath15 is a power - law function for large values of chemical distance @xmath6 . the exponential prefactor @xmath16 in gives the number of lattice sites within @xmath6 shells of the bethe lattice , and plays the role of the factor @xmath17 in the equivalent expression for hypercubic lattices . the function @xmath18 with some exponent @xmath19 ( found below to be @xmath20 ) , accounts for all power - law behaviour of the correlation function . there is some inconsistency in the definition of the correlation length for two - state models ( the equilibrium ising model , the zt - rfim and percolation ) on the bethe lattice . in the equilibrium ising model @xcite and the zt - rfim @xcite on the bethe lattice , the prefactor @xmath21 in has been absorbed into the definition of the correlation length @xmath22 with @xmath23 chosen to be the correlation length . in an alternative definition , for both percolation @xcite and the ising model @xcite , the correlation length has instead been identified with @xmath9 in , thus separating the length scale of dynamic correlations from the intrinsic topological length scale of the bethe lattice , @xmath24 . below , we demonstrate that ( i ) the value of @xmath9 diverges at the critical point according to a power law , in contrast to @xmath25 which remains finite at criticality and ( ii ) the critical exponents governing the divergence of @xmath9 around the critical point are consistent with the mean - field values for the divergence of correlation length in hypercubic lattices above the upper critical dimension , @xmath26 . the exponents , describing the divergence of @xmath9 in terms of the external field @xmath27 and degree of disorder @xmath28 , @xmath29 and @xmath30 , are found to be @xmath31 and @xmath32 . as argued below , these values can be related to those in the mean - field model , @xmath33 and @xmath34 , where @xmath35 originates from fractal dimension of a random walk @xcite . therefore these findings motivate us to choose @xmath9 as the correlation length describing the critical behaviour of the zt - rfim on a bethe lattice . such a definition , while in contrast with that used in @xcite , separates the topological contribution of the bethe lattice to the correlation length , and allows its comparison with the correlation length for hypercubic lattices . the zt - rfim involves a set of @xmath36 ising spins , @xmath37 , interacting ferromagnetically with strength @xmath38 , affected by an external field @xmath27 and quenched local disorder , manifested in independent random local fields , @xmath39 , at each site @xmath40 , according to the following hamiltonian , @xmath41 where @xmath42 denotes a sum over all pairs of nearest neighbours . the random fields @xmath39 are assumed to be identically distributed according to the probability density function @xmath43 with , for convenience , zero mean and standard deviation @xmath28 . it has been found that in certain topologies of the network of spins , and for disorder @xmath44 , the zt - rfim exhibits a discontinuity in the hysteresis loop for magnetisation , @xmath45 , at some coersive field @xmath46 , where the magnetisation in a large fraction of the system reverses in a single avalanche . increasing disorder is found to reduce the size of this discontinuity , and to remove it continuously at the critical value of disorder @xmath47 , around which spin - spin correlations become infinite - ranged and the system exhibits scale - free universal properties @xcite . we start with a description of the dynamical rules governing the relaxation of spins within the zt - rfim . the external field is initially set to @xmath48 , forcing all spins to be @xmath49 , and then allowed to increase adiabatically to @xmath50 . when the external field is varied , the system becomes unstable and relaxes into a new metastable state through a series of spin flips . we assume that relaxation takes place according to glauber single - spin flip dynamics for the zero temperature case @xcite , such that a spin flips only if such a flip reduces the overall energy . if a spin at node @xmath40 changes state ( i.e. @xmath51 ) , it induces a change in energy @xmath52 , where @xmath53 is a local field . if @xmath54 then the spin flip will occur , otherwise it remains aligned with the local field . as spins flip one at a time @xcite , the local field at surrounding sites changes , and other spins may become unstable and also flip , therefore relaxation occurs in an avalanche like manner . at a given external field , this process continues until all spins are aligned with their respective local fields at which point the system becomes stable , thus creating clusters of flipped spins around each nucleation site . as the external field increases avalanches continue to progress , and it is known that the set of spins that flip in multiple avalanches when the external field is slowly increased from @xmath55 to @xmath27 coincides with the set of spins that would flip if the field had stepped from @xmath55 to @xmath27 ( see @xcite ) . accordingly , it can be assumed that the correlation function calculated at any external field depends only on the properties of avalanches occurring at that field . 1.5truecm ( 4.5,0.0 ) @xmath56 ( -2.5,1.0 ) @xmath57 ( -2.5,0.0)0.1a1 ( -1.0,1.0 ) @xmath58 ( -1.0,0.0)0.1a2 ( -0.6,0.0)0a2p1 ( 0.6,0.0)0a2p2 ( 1.0,1.0 ) @xmath59 ( 1.0,0.0)0.1a3 ( 2.5,1.0 ) @xmath60 ( 2.5,0.0)0.1a4 ( 5,-1.5 ) @xmath61 ( -4.0,-1.5)0.1b1 ( -2.5,-1.5)0.1b2 ( -1.0,-1.5)0.1b3 ( 1.0,-1.5)0.1b4 ( 2.5,-1.5)0.1b5 ( 4.0,-1.5)0.1b6 ( 5,-2.8 ) @xmath62 ( -4.375,-2.8)0.1c1 ( -3.625,-2.8)0.1c2 ( -2.875,-2.8)0.1c3 ( -2.125,-2.8)0.1c4 ( -1.375,-2.8)0.1c5 ( -0.625,-2.8)0.1c6 ( 0.625,-2.8)0.1c7 ( 1.375,-2.8)0.1c8 ( 2.125,-2.8)0.1c9 ( 2.875,-2.8)0.1c10 ( 3.625,-2.8)0.1c11 ( 4.375,-2.8)0.1c12 ( -4.275,-3.1)0.0c1p1 ( -3.525,-3.1)0.0c2p1 ( -2.775,-3.1)0.0c3p1 ( -2.025,-3.1)0.0c4p1 ( -1.275,-3.1)0.0c5p1 ( -0.525,-3.1)0.0c6p1 ( 0.525,-3.1)0.0c7p1 ( 1.275,-3.1)0.0c8p1 ( 2.025,-3.1)0.0c9p1 ( 2.775,-3.1)0.0c10p1 ( 3.525,-3.1)0.0c11p1 ( 4.275,-3.1)0.0c12p1 ( -4.475,-3.1)0.0c1p2 ( -3.725,-3.1)0.0c2p2 ( -2.975,-3.1)0.0c3p2 ( -2.225,-3.1)0.0c4p2 ( -1.475,-3.1)0.0c5p2 ( -0.725,-3.1)0.0c6p2 ( 0.725,-3.1)0.0c7p2 ( 1.475,-3.1)0.0c8p2 ( 2.225,-3.1)0.0c9p2 ( 2.975,-3.1)0.0c10p2 ( 3.725,-3.1)0.0c11p2 ( 4.475,-3.1)0.0c12p2 ( -4.0,-4.2)(-4.0,-4.8 ) ( 0.0,-4.2)(0.0,-4.8 ) ( 4.0,-4.2)(4.0,-4.8 ) ( 4.5,-5.5 ) @xmath63 ( -2.5,-5.5)0.1d1 ( 2.5,-5.5)0.1d2 ( 4.5,-7.0 ) @xmath64 ( -3.0,-7.0)0.1e1 ( -2.0,-7.0)0.1e2 ( 2.0,-7.0)0.1e3 ( 3.0,-7.0)0.1e4 ( 4.5,-8.2 ) @xmath65 ( -3.25,-8.2)0.1f1 ( -2.75,-8.2)0.1f2 ( -2.25,-8.2)0.1f3 ( -1.75,-8.2)0.1f4 ( 1.75,-8.2)0.1f5 ( 2.25,-8.2)0.1f6 ( 2.75,-8.2)0.1f7 ( 3.25,-8.2)0.1f8 ( -2.0,-7.0)(2.0,-7.0 ) 10truecm the spin - spin correlation function , @xmath5 , between two spins labelled @xmath57 and @xmath60 , separated by the shortest chemical distance @xmath66 , is given by , @xmath67 where @xmath68 denotes an average over quenched disorder . in order to evaluate the correlation function , it is convenient to define a cayley tree of height @xmath69 by dividing it into a set of @xmath69 generations @xmath70 of spins ( see ) . the first generation , @xmath56 , forms a central chain consisting of @xmath71 connected spins , in contrast to the standard definition with a single central ( root ) site ( presented in [ sec : dhar ] ) . the two boundary spins of the central chain , @xmath57 and @xmath60 , are each connected to one spin within the central chain , and to @xmath72 spins in generation @xmath73 of the cayley tree . the interior spins of the central chain @xmath74 interact with @xmath61 other neighbouring spins ( see thick horizontal lines in ) in the central chain and @xmath75 second generation spins . each spin in generation @xmath76 then interacts with a single spin in generation @xmath77 , and @xmath72 spins in generation @xmath78 . spins in the last ( boundary ) generation , @xmath65 , interact only with a single spin in generation @xmath64 . in the limit of large number of generations , a cayley tree defined in this way tends to the same limit , i.e. a bethe lattice , as the more standard cayley tree defined in @xcite . for finite values of external magnetic field , spins in all generations are relaxed . first , we relax the spins in generations @xmath76 , so that the spins in generation @xmath73 are up with probability @xmath79 , which is the solution to the self - consistent equation , @xmath80 the function @xmath81 is given by @xcite , @xmath82}^m{\left[1-p^*\right]}^{q-1-m}\nonumber\\ i(-h+(q-2m)j,\infty)~ , \label{eq : fm}\end{aligned}\ ] ] with , @xmath83 then , knowing the state of spins in generation @xmath73 , we relax the spins in the central chain @xmath56 and calculate the correlation function ( see appendix a ) . the relaxed central chain can contain both flipped and unflipped spins . in order to know the state of spin @xmath84 $ ] we need to know the local field , @xmath85 , which depends on the neighbourhood of this spin and the random field , @xmath39 , @xmath86 here , the state of the neighbourhood is represented by the number @xmath87 of flipped neighbours in the central chain and the number @xmath88 of flipped neighbours in generation @xmath61 prior to the relaxation of the central chain , so that the total number of flipped neighbours @xmath89 . according to the values of the local fields , @xmath90 , where the variables @xmath39 and @xmath88 are fixed during the relaxation of the central chain , spins in nodes @xmath91 can be divided into three categories : ( @xmath92 ) those that experience a positive local field when all neighbours within the central chain are down @xmath93 , i.e. spins in the set @xmath94 ; ( @xmath61 ) those that only experience a positive local field after one of their two neighbours in the central chain has flipped i.e. @xmath95 ; ( @xmath62 ) all other spins @xmath96 . avalanches within the central chain are nucleated at all sites of category @xmath92 , and propagate through sites of category @xmath61 , terminating either at the end of the chain , or at a site of category @xmath62 . therefore , for a given configuration of categories @xmath97 of spins in nodes @xmath74 it is possible to determine the state of the spins at the end of the chain , @xmath57 and @xmath60 , by finding the first site from each end which is not of category @xmath61 . if this first site not of category @xmath61 is of category @xmath92 then the respective end spin will be up and if such first site is of category @xmath62 the end spin will be down ( see appendix b ) . summing over all combinations of such systems it is possible to calculate @xmath98 . similarly , it is possible to consider just one spin @xmath112 . therefore the correlation function can be found by the formula , and is given by , @xmath113 + \frac{4q_1q_3{q_2^\prime}^2}{(1-q_2)q_2 ^ 2}r\right\}\left(q_2\right)^r~.\label{eq : corrfuncfinal}\end{aligned}\ ] ] is the main result of our derivation , and represents the exact expression for the correlation function in a bethe lattice . for @xmath114 , the correlation function should follow the behaviour @xcite , @xmath115 where , by comparing with in the limit of large @xmath6 , it is found that @xmath116 and the correlation length is given by , @xmath117^{-1}~. \label{eq : xi}\ ] ] the exact expression for correlation function given by can be supported numerically by evaluation of the correlation function on random @xmath14-regular graphs . these are random graphs with a fixed coordination number @xmath14 whose topology at local scales is similar to that of a cayley tree with the same coordination number @xcite . the absence of boundaries in @xmath14-regular graphs make them more suitable for numerical simulations than the bethe lattice used in our analytical derivation ( see appendix c ) . demonstrates that the results of numerical calculations closely match the analytical formula , . on @xmath6 for @xmath118 according to ( lines ) and numerical simulations ( symbols ) for several values of external field : @xmath119 ( circles , solid line ) , @xmath120 ( squares , dashed line ) , and @xmath121 ( diamonds , dot - dashed line ) . all results were obtained for systems characterised by @xmath122 . numerical simulations consider a @xmath14-regular graph ( @xmath118 ) of @xmath123 spins , and calculate the mean of the correlation function between all pairs of sites in the system for @xmath124 realisations of quenched disorder . [ fig : numerics],width=226 ] near the critical point the correlation length diverges according to standard scaling relations , @xmath125 with @xmath126 and @xmath127 being the critical exponents . in order to find these exponents , it is convenient to rewrite using the following property of @xmath128 , @xmath129 introducing this expression in , one obtains , @xmath130^{-1}~,\label{eq : xiing}\end{aligned}\ ] ] where @xmath131 . can be expanded in terms of @xmath132 which is equal to @xmath133 at criticality and small nearby , resulting in , @xmath134 in the case of bethe lattices with coordination numbers @xmath3 or @xmath135 , there is no transition at any non - zero disorder . however , for such coordination numbers , it is possible to show by our method that the correlation length diverges exponentially as disorder goes to zero . this result is in agreement with the calculation given in @xcite for the case @xmath3 . for higher coordination numbers @xmath136 , where a transition exists , expansion of around criticality can be done in terms of @xmath137 , @xmath138 and @xmath139 . using the known mean - field exponents @xmath2 and @xmath1 @xcite describing the behaviour of @xmath140 in terms of @xmath137 and @xmath138 , respectively , around criticality and noting that both the first and second derivatives @xmath141 one obtains the values @xmath142 and @xmath143 for the correlation length critical exponents ( see appendix d ) . the values of these exponents can be explained in terms of the mean - field values using the argument for percolation of @xcite . in this argument , it is demonstrated that the critical percolation cluster in high dimensions ( @xmath144 ) is sparse enough as for the loops to be unimportant . in this case , the critical percolation cluster is essentially the same as that on a bethe lattice , i.e. its backbone consists of chains of links ( which form random walks ) , with occasional branch points ( nodes ) . the mean chain length between branching nodes is defined as the correlation length which scales as @xmath145 ( where @xmath146 and @xmath147 are bond probability and the critical bond probability ) with the power @xmath148 for chemical distance and @xmath149 for euclidean distance . the same arguments can be applied to the critical avalanche clusters in the zt - rfim on a hypercubic lattice and bethe lattice , so that the correlation length should scale with the exponents @xmath31 , @xmath32 for chemical distance through the avalanche cluster and @xmath150 , @xmath151 for euclidean distance across the hypercubic lattice . to conclude , we derived an exact analytical formula for the spin - spin correlation function for the zt - rfim defined on a bethe lattice and investigated the scaling of the correlation length near criticality . we demonstrate that if the correlation length is defined not to include the intrinsic topological length scale , then it diverges around criticality with exponents consistent with the mean - field description . in fact , choosing the dimensionality of the bethe lattice to be a function of the length scale @xmath6 @xcite , that is , replacing the term @xmath152 in the standard definition of the correlation length with @xmath153 , allows for analysis consistent with that performed for hypercubic lattices . we thank m. l. rosinberg for fruitful discusions . t.p.h . would like to thank the uk epsrc for financial support . 1.5truecm ( 4.5,0.0 ) @xmath56 ( -2.5,1.0 ) @xmath57 ( -2.5,0.0)0.1a1 ( -1.0,1.0 ) @xmath58 ( -1.0,0.0)0.1a2 ( -0.6,0.0)0a2p1 ( 0.6,0.0)0a2p2 ( 1.0,1.0 ) @xmath59 ( 1.0,0.0)0.1a3 ( 2.5,1.0 ) @xmath60 ( 2.5,0.0)0.1a4 ( 5,-1.5 ) @xmath61 ( -4.0,-1.5)0.1b1 ( -2.5,-1.5)0.1b2 ( -1.0,-1.5)0.1b3 ( 1.0,-1.5)0.1b4 ( 2.5,-1.5)0.1b5 ( 4.0,-1.5)0.1b6 ( 5,-2.8 ) @xmath62 ( -4.375,-2.8)0.1c1 ( -3.625,-2.8)0.1c2 ( -2.875,-2.8)0.1c3 ( -2.125,-2.8)0.1c4 ( -1.375,-2.8)0.1c5 ( -0.625,-2.8)0.1c6 ( 0.625,-2.8)0.1c7 ( 1.375,-2.8)0.1c8 ( 2.125,-2.8)0.1c9 ( 2.875,-2.8)0.1c10 ( 3.625,-2.8)0.1c11 ( 4.375,-2.8)0.1c12 ( -4.275,-3.1)0.0c1p1 ( -3.525,-3.1)0.0c2p1 ( -2.775,-3.1)0.0c3p1 ( -2.025,-3.1)0.0c4p1 ( -1.275,-3.1)0.0c5p1 ( -0.525,-3.1)0.0c6p1 ( 0.525,-3.1)0.0c7p1 ( 1.275,-3.1)0.0c8p1 ( 2.025,-3.1)0.0c9p1 ( 2.775,-3.1)0.0c10p1 ( 3.525,-3.1)0.0c11p1 ( 4.275,-3.1)0.0c12p1 ( -4.475,-3.1)0.0c1p2 ( -3.725,-3.1)0.0c2p2 ( -2.975,-3.1)0.0c3p2 ( -2.225,-3.1)0.0c4p2 ( -1.475,-3.1)0.0c5p2 ( -0.725,-3.1)0.0c6p2 ( 0.725,-3.1)0.0c7p2 ( 1.475,-3.1)0.0c8p2 ( 2.225,-3.1)0.0c9p2 ( 2.975,-3.1)0.0c10p2 ( 3.725,-3.1)0.0c11p2 ( 4.475,-3.1)0.0c12p2 ( -4.0,-4.2)(-4.0,-4.8 ) ( 0.0,-4.2)(0.0,-4.8 ) ( 4.0,-4.2)(4.0,-4.8 ) ( 6.5,-4.1 ) @xmath63 ( -5.0,-4.4)0.2d1 ( -5.0,-4.4 ) @xmath154 ( 5.0,-4.4)0.2d2 ( 5.0,-4.4 ) @xmath154 ( -4.86,-4.0)0.0d1s ( 4.86,-4.0)0.0d2s ( 6.5,-5.6 ) @xmath64 ( -5.5,-5.6)0.2e1 ( -5.5,-5.6 ) @xmath155 ( -4.5,-5.6)0.2e2 ( -4.5,-5.6 ) @xmath155 ( 4.5,-5.6)0.2e3 ( 4.5,-5.6 ) @xmath154 ( 5.5,-5.6)0.2e4 ( 5.5,-5.6 ) @xmath155 ( 6.5,-6.8 ) @xmath65 ( -5.75,-6.8)0.2f1 ( -5.75,-6.8 ) @xmath155 ( -5.25,-6.8)0.2f2 ( -5.25,-6.8 ) @xmath156 ( -4.75,-6.8)0.2f3 ( -4.75,-6.8 ) @xmath154 ( -4.25,-6.8)0.2f4 ( -4.25,-6.8 ) @xmath154 ( 4.25,-6.8)0.2f5 ( 4.25,-6.8 ) @xmath155 ( 4.75,-6.8)0.2f6 ( 4.75,-6.8 ) @xmath154 ( 5.25,-6.8)0.2f7 ( 5.25,-6.8 ) @xmath154 ( 5.75,-6.8)0.2f8 ( 5.75,-6.8 ) @xmath156 ( -1.7,-5.5)(1.7,-5.5 ) 8truecm in this section , we describe the relaxation of spins in all generations with @xmath76 using a procedure similar to that described by @xcite for obtaining the magnetisation in the zero - temperature random field ising model ( zt - rfim ) . the system of spins within the ferromagnetic zt - rfim , relaxing according to zero - temperature glauber dynamics , obeys the abelian property @xcite , i.e. the final metastable state of the spin system after relaxation is independent of the order in which spins relax . we therefore choose a relaxation order convenient to analysis . we start relaxation with boundary generation @xmath65 at the bottom of the tree ( see ) , at the first step , the spins in all generations except the boundary generation @xmath69 are held in the their initial down ( @xmath157 , @xmath158 ) state , while allowing any spin in generation @xmath69 to flip ( change state , @xmath159 ) if the local field acting on that spin is positive , and thus the flip reduces the overall energy ( see the up spins , @xmath160 , in in generation @xmath65 ) . at the next step , spins in generation @xmath64 are similarly allowed to flip if doing so reduces the overall energy ( see the up spins , @xmath160 , in in generation @xmath161 ) , while the sites in generations @xmath162 are again kept fixed down . spins flipping in generation @xmath64 will cause a change in the local field acting at their non - flipped neighbours in generation @xmath69 , which may become positive and thus cause these spins to flip as well ( see the up spins , @xmath163 , in in generation @xmath65 ) . such secondary flips in generation @xmath69 can not influence the spin in the above generation , which is already flipped and thus in its final state ( at zero temperature ) . because of the loopless structure of the cayley tree , these secondary flips are isolated from the above generations . therefore , they can not affect those generations and , in particular , the central chain ( @xmath56 ) in which we are interested for calculating the correlation function . at subsequent steps , relaxation progresses up the generations in a similar way until the spins in generation @xmath73 are relaxed . the above relaxation procedure is a random process , due to the contribution of quenched independent random fields @xmath164 to the local fields @xmath165 at generation @xmath70 . at the first step , all other contributions to the local fields @xmath166 are identical for all sites , meaning that the local fields are independent identically distributed random variables . as a consequence any spin @xmath167 in generation @xmath69 may flip independently of all others in this generation with probability @xmath168 , the probability that @xmath169 . at the second step , randomness in the local fields at sites in generation @xmath64 is introduced by the random fields @xmath170 at those sites in combination with the field produced by a random number of flipped neighbours in generation @xmath69 , both of which are independent random variables . therefore , all the flips in generation @xmath64 at the second step are independent events , occurring with probability @xmath171 . similarly , according to this relaxation procedure , at the @xmath172-th step , the flips in generation @xmath173 are all independent events . in order to calculate the values of the probabilities @xmath174 , consider first an arbitrary spin @xmath167 in generation @xmath175 and calculate the conditional probability @xmath176 of it flipping given that it is surrounded by @xmath177 spins in the up state . the condition for the spin to flip is that the local field , @xmath178 is greater than zero , i.e. @xmath179 here , @xmath180 gives the number of neighbouring spins @xmath181 ( in generations @xmath182 ) of the spin @xmath40 ( in generation @xmath70 ) which are in the up state , @xmath183 . the probability of the random field @xmath39 meeting condition is , @xmath184 where @xmath185 stands for the heaviside step function , and @xmath186 is given by . the variable @xmath177 in is a random variable depending on the orientation of neighbouring spins , which are located in generations @xmath187 . for @xmath188 , when a spin in generation @xmath70 is first relaxed ( at relaxation step @xmath189 ) , the neighbouring spin in generation @xmath190 is in the down - state , while the @xmath72 neighbouring spins in generation @xmath191 may each independently ( as stated above ) be in the up - state with probability @xmath192 . therefore , the total number of neighbouring spins which are in the up - state is binomially distributed according to @xmath193 , i.e. the probability of @xmath177 neighbours of a spin at generation @xmath70 being up is given by @xmath194}^n{\left[1-p^{(g+1)}\right]}^{q-1-n}~.\ ] ] the probability @xmath174 can then be found as a sum over all possible configurations of @xmath177 neighbouring spins in the up - state ( @xmath195 ) , @xmath196}^n{\left[1-p^{(g+1)}\right]}^{q-1-n}p_n~ , \label{eq : pg}\ ] ] which is a recursion relation for @xmath174 valid for @xmath188 . far from the boundary of very large cayley trees ( @xmath197 , @xmath198 ) , @xmath174 tends to some limiting value , @xmath199 , identical for all interior generations . this leads to the self - consistent equation for @xmath79 ( see and ) , @xmath200}^m{\left[1-p^*\right]}^{q-1-m}p_m~,\label{eq : f } \\ p^*&=&f(p^*)~,\label{eq : selfconsistant}\end{aligned}\ ] ] so that @xmath201 for all @xmath76 , i.e. after the relaxation of all spins except those in the central chain , spins in generation @xmath61 are in the up - state with probability @xmath79 . 2truecm ( 7.0,0.0 ) @xmath56 ( 0.0,0.0)0.2a1 ( 7.0,-2.0 ) @xmath73 ( -4.0,-2.0)0.2b1 ( 0.0,-2.0)0.2b2 ( 4.0,-2.0)0.2b3 ( 7.0,-3.5 ) @xmath202 ( -5.0,-3.5)0.2c1 ( -3.0,-3.5)0.2c2 ( -1.0,-3.5)0.2c3 ( 1.0,-3.5)0.2c4 ( 3.0,-3.5)0.2c5 ( 5.0,-3.5)0.2c6 ( -4.0,-4.2)(-4.0,-4.8 ) ( 0.0,-4.2)(0.0,-4.8 ) ( 4.0,-4.2)(4.0,-4.8 ) ( 7.0,-5.5 ) @xmath203 ( -4.0,-5.5)0.2d1 ( 4.0,-5.5)0.2d2 ( 7.0,-7.0 ) @xmath161 ( -5.0,-7.0)0.2e1 ( -3.0,-7.0)0.2e2 ( 3.0,-7.0)0.2e3 ( 5.0,-7.0)0.2e4 ( 7.0,-8.5 ) @xmath65 ( -5.5,-8.5)0.2f1 ( -4.5,-8.5)0.2f2 ( -3.5,-8.5)0.2f3 ( -2.5,-8.5)0.2f4 ( 2.5,-8.5)0.2f5 ( 3.5,-8.5)0.2f6 ( 4.5,-8.5)0.2f7 ( 5.5,-8.5)0.2f8 ( -2.0,-7.0)(2.0,-7.0 ) 7truecm in the case of the standard cayley tree ( see ) , the relaxation procedure described above allows the magnetisation of the central spin ( generation @xmath56 in ) to be calculated in the following way . as mentioned above , and are only valid for @xmath188 , and the central spin at generation @xmath56 should be treated separately . the central spin has @xmath14 neighbours in generation @xmath73 , which , on the relaxation of that spin , are in the up - state with probability @xmath204 . the number of neighbours of the central site that are in the up - state is binomially distributed , @xmath205 and the expression for @xmath206 is given by with @xmath72 replaced by @xmath14 and @xmath192 by @xmath79 , i.e. @xmath207}^n{\left[1-p^*\right]}^{q - n}p_n~. \label{eq : pcentral}\ ] ] the value of @xmath206 calculated by solving the self - consistent equations and and substituting into allows the mean magnetisation of the lattice to be evaluated , @xmath208 . the result of this calculation for magnetisation vs external field is given in , showing the known spinodal transition at low disorder @xcite . mean magnetisation , @xmath209 , vs external field , @xmath27 , scaled by interaction strength , @xmath210 , for zt - rfim on a bethe lattice of coordination number @xmath118 , as magnetic field is swept upwards from @xmath55 to @xmath211 . the solid , dot - dashed and dashed lines correspond to different degrees of disorder around critically : @xmath212 , @xmath213 and @xmath214 , respectively . the critical field ( @xmath215 ) is marked by the dotted vertical line . in the region where the magnetisation curve is multi - valued ( marked by circles ) , the system follows the lower branch , so that a jump in magnetisation ( represented by the vertical arrow ) occurs when the external field passes the coercive field strength.,width=226 ] 2truecm ( -7.0,0.0 ) ( a ) ( -5.0,0)0.3sa0 ( -4.0,0)0.3sa1 ( -3.0,0)0.3sa2 ( -2.0,0)0.3sa3 ( -1.0,0)0.3sa4 ( 0.0,0)0.3sa5 ( 1.0,0)0.3sa6 ( 2.0,0)0.3sa7 ( 3.0,0)0.3sa8 ( 4.0,0)0.3sa9 ( 5.0,0)0.3sa10 ( -5.3,-0.4)0.0sa0ll ( -5.0,-0.4)0.0sa0l ( -4.0,-0.4)0.0sa1l ( -3.0,-0.4)0.0sa2l ( -2.0,-0.4)0.0sa3l ( -1.0,-0.4)0.0sa4l ( 0.0,-0.4)0.0sa5l ( 1.0,-0.4)0.0sa6l ( 2.0,-0.4)0.0sa7l ( 3.0,-0.4)0.0sa8l ( 4.0,-0.4)0.0sa9l ( 5.0,-0.4)0.0sa10l ( 5.3,-0.4)0.0sa10lr ( -5.0,-1.0 ) @xmath216 ( 5.0,-1.0 ) @xmath217 ( 7.3,-0.0 ) @xmath218 , @xmath219 ( -7.0,-2.5 ) ( b / d ) ( -5.0,-2.5)0.3sb0 ( -4.0,-2.5)0.3sb1 ( -3.0,-2.5)0.3sb2 ( -2.0,-2.5)0.3sb3 ( -1.0,-2.5)0.3sb4 ( 0.0,-2.5)0.3sb5 ( 1.0,-2.5)0.3sb6 ( 2.0,-2.5)0.3sb7 ( 3.0,-2.5)0.3sb8 ( 4.0,-2.5)0.3sb9 ( 5.0,-2.5)0.3sb10 ( -5.3,-2.9)0.0sb0ll ( -5.0,-2.9)0.0sb0l ( -4.0,-2.9)0.0sb1l ( -3.0,-2.9)0.0sb2l ( -2.0,-2.9)0.0sb3l ( -1.0,-2.9)0.0sb4l ( 0.0,-2.9)0.0sb5l ( 1.0,-2.9)0.0sb6l ( 2.0,-2.9)0.0sb7l ( 3.0,-2.9)0.0sb8l ( 4.0,-2.9)0.0sb9l ( 5.0,-2.9)0.0sb10l ( 5.3,-2.9)0.0sb10lr ( -5.0,-3.5 ) @xmath216 ( -4.0,-3.5 ) ( -3.0,-3.5 ) ( -2.0,-3.5 ) ( -1.0,-3.5 ) ( 0.0,-3.5 ) @xmath220 ( 1.0,-3.5 ) @xmath61 ( 2.0,-3.5 ) @xmath61 ( 3.0,-3.5 ) @xmath61 ( 4.0,-3.5 ) @xmath61 ( 5.0,-3.5 ) @xmath221 [ ur](0.9,-4.3)@xmath222 ( 7.3,-2.5 ) @xmath218 , @xmath223 ( -7.0,-5.0 ) ( c / e ) ( -5.0,-5.0)0.3sc0 ( -4.0,-5.0)0.3sc1 ( -3.0,-5.0)0.3sc2 ( -2.0,-5.0)0.3sc3 ( -1.0,-5.0)0.3sc4 ( 0.0,-5.0)0.3sc5 ( 1.0,-5.0)0.3sc6 ( 2.0,-5.0)0.3sc7 ( 3.0,-5.0)0.3sc8 ( 4.0,-5.0)0.3sc9 ( 5.0,-5.0)0.3sc10 ( -5.3,-5.4)0.0sc0ll ( -5.0,-5.4)0.0sc0l ( -4.0,-5.4)0.0sc1l ( -3.0,-5.4)0.0sc2l ( -2.0,-5.4)0.0sc3l ( -1.0,-5.4)0.0sc4l ( 0.0,-5.4)0.0sc5l ( 1.0,-5.4)0.0sc6l ( 2.0,-5.4)0.0sc7l ( 3.0,-5.4)0.0sc8l ( 4.0,-5.4)0.0sc9l ( 5.0,-5.4)0.0sc10l ( 5.3,-5.4)0.0sc10lr ( -5.0,-6.0 ) @xmath216 ( -5.0,-6.5 ) @xmath224 ( -4.0,-6.0 ) @xmath61 ( -3.0,-6.0 ) @xmath61 ( -2.0,-6.0 ) @xmath61 ( -1.0,-6.0 ) @xmath61 ( 0.0,-6.0 ) @xmath61 ( 1.0,-6.0 ) @xmath61 ( 2.0,-6.0 ) @xmath61 ( 3.0,-6.0 ) @xmath61 ( 4.0,-6.0 ) @xmath61 ( 5.0,-6.0 ) @xmath221 [ ur](-4.1,-6.8)@xmath225 ( 7.3,-5.0 ) @xmath218 , @xmath226 ( -7.0,-7.5 ) ( f ) ( -5.0,-7.5)0.3sf0 ( -4.0,-7.5)0.3sf1 ( -3.0,-7.5)0.3sf2 ( -2.0,-7.5)0.3sf3 ( -1.0,-7.5)0.3sf4 ( 0.0,-7.5)0.3sf5 ( 1.0,-7.5)0.3sf6 ( 2.0,-7.5)0.3sf7 ( 3.0,-7.5)0.3sf8 ( 4.0,-7.5)0.3sf9 ( 5.0,-7.5)0.3sf10 ( -5.3,-7.9)0.0sf0ll ( -5.0,-7.9)0.0sf0l ( -4.0,-7.9)0.0sf1l ( -3.0,-7.9)0.0sf2l ( -2.0,-7.9)0.0sf3l ( -1.0,-7.9)0.0sf4l ( 0.0,-7.9)0.0sf5l ( 1.0,-7.9)0.0sf6l ( 2.0,-7.9)0.0sf7l ( 3.0,-7.9)0.0sf8l ( 4.0,-7.9)0.0sf9l ( 5.0,-7.9)0.0sf10l ( 5.3,-7.9)0.0sf10lr ( -5.0,-8.5 ) @xmath227 ( -4.0,-8.5 ) @xmath61 ( -3.0,-8.5 ) @xmath61 ( -2.0,-8.5 ) @xmath228 [ ur](-5.5,-9.3)@xmath229(-1.0,-8.5 ) ( 0.0,-8.5 ) ( 1.0,-8.5 ) @xmath220 ( 2.0,-8.5 ) @xmath61 ( 3.0,-8.5 ) @xmath61 ( 4.0,-8.5 ) @xmath61 ( 5.0,-8.5 ) @xmath221 ( 7.3,-7.5 ) @xmath230 , @xmath231 ( 7.3,-7.9 ) @xmath232 [ ur](1.9,-9.3)@xmath233(-7.0,-10.0 ) ( g ) ( -5.0,-10.0)0.3sg0 ( -4.0,-10.0)0.3sg1 ( -3.0,-10.0)0.3sg2 ( -2.0,-10.0)0.3sg3 ( -1.0,-10.0)0.3sg4 ( 0.0,-10.0)0.3sg5 ( 1.0,-10.0)0.3sg6 ( 2.0,-10.0)0.3sg7 ( 3.0,-10.0)0.3sg8 ( 4.0,-10.0)0.3sg9 ( 5.0,-10.0)0.3sg10 ( -5.3,-10.4)0.0sg0ll ( -5.0,-10.4)0.0sg0l ( -4.0,-10.4)0.0sg1l ( -3.0,-10.4)0.0sg2l ( -2.0,-10.4)0.0sg3l ( -1.0,-10.4)0.0sg4l ( 0.0,-10.4)0.0sg5l ( 1.0,-10.4)0.0sg6l ( 2.0,-10.4)0.0sg7l ( 3.0,-10.4)0.0sg8l ( 4.0,-10.4)0.0sg9l ( 5.0,-10.4)0.0sg10l ( 5.3,-10.4)0.0sg10lr ( -5.0,-11.0 ) @xmath227 ( -4.0,-11.0 ) @xmath61 ( -3.0,-11.0 ) @xmath61 ( -2.0,-11.0 ) @xmath61 [ ur](-5.5,-11.8)@xmath234(-1.0,-11.0 ) @xmath235 ( 0.0,-11.0 ) @xmath61 ( 1.0,-11.0 ) @xmath61 ( 2.0,-11.0 ) @xmath61 ( 3.0,-11.0 ) @xmath61 ( 4.0,-11.0 ) @xmath61 ( 5.0,-11.0 ) @xmath221 [ ur](-0.1,-11.8)@xmath236(7.3,-10 ) @xmath230 , @xmath231 ( 7.3,-10.4 ) @xmath237 11truecm in order to calculate the correlation function , we need to know the states of the chain - boundary spins @xmath57 and @xmath60 which are determined entirely by the categories @xmath97 of all the spins in the central chain . however if the categories of the boundary spins are either @xmath92 or @xmath62 ( see for definition ) , their state depends only on their own category . in particular , if the category of a boundary spin is @xmath92 then it will have a positive local field and flip . conversely , if a boundary spin has a category @xmath62 then it will have a negative local field and will never flip . in the case that a boundary spin is of category @xmath61 , its state depends on the categories of the other spins in the central chain . specifically , a boundary spin ( @xmath57 or @xmath60 ) of category @xmath61 will flip only if its neighbour in the chain ( either spin @xmath58 or @xmath59 ) flips first . the state of the spins @xmath58 and @xmath59 , given that their respective neighbours at the boundary are of category @xmath61 , can be found in the same way . therefore , the state of spin @xmath57 depends on and coincides with the state of the first spin , @xmath238 ( @xmath239 ) , counting from the @xmath57 boundary , that is not of category @xmath61 . the state of spin @xmath238 is , itself , determined only by its category @xmath240 , with @xmath241 or @xmath242 but , by definition , @xmath243 . similarly , the state of spin @xmath6 is determined by the state of spin @xmath244 ( @xmath245 ) , the first spin counting from the @xmath60 boundary which is not of category @xmath61 , being instead of category @xmath246 . in the special case of all the spins in the central chain being of category @xmath61 , the values of @xmath228 , @xmath220 , @xmath238 and @xmath247 can not be defined . however , this situation is simpler because all spins in the central chain after relaxation are necessarily in the down - state , @xmath49 . the probability , @xmath248 , that all spins @xmath40 ( @xmath249 ) are of category @xmath61 with spin @xmath238 being of category @xmath228 , and simultaneously all spins @xmath181 ( @xmath250 ) are of category @xmath61 with spin @xmath244 being of category @xmath220 , can be calculated explicitly ( see for details ) . in we show the central chain of length @xmath6 with white circles representing spin of either category @xmath92 or @xmath62 , and grey circles are used for spins of category @xmath61 . circles with crosses represent spins for which the category is irrelevant to the state of the boundary spins . all distinct configurations of spin categories ( excluding the special case when all spins are of category @xmath61 ) are represented by rows ( a)-(g ) . in configuration ( a ) , the boundary spin are not of category @xmath61 , so that the categories of all other spins are irrelevant . in configurations ( b ) and ( d ) , one of the boundary spins is of category @xmath61 , and the first spin which is not of category @xmath61 linked to this spin does not coincide with the other boundary spin . in configurations ( c ) and ( e ) , all the spins are of category @xmath61 , except one of the boundary spins , meaning @xmath235 . in configurations ( f)/(g ) , both boundary spins are of category @xmath61 , and the first spins not of category @xmath61 , counting from spin @xmath57 and from @xmath60 do not coincide / do coincide . there are two important observations to be taken into account in order to evaluate @xmath248 . the first observation is that @xmath228 and @xmath220 are independent variables only if @xmath251 ( see configurations ( a ) , ( b ) , ( d ) and ( f ) in ) , while otherwise , i.e. when @xmath252 , they are equal , @xmath235 ( see configurations ( c ) , ( e ) and ( g ) ) . the second point to consider is that the configurations where one or both of the boundary spins are not of category @xmath61 should be treated separately ( i.e. in configurations ( a ) , ( b ) , ( c ) , ( d ) and ( e ) ) , because in such configurations the probability distribution of @xmath228 and @xmath220 are described by @xmath253 and @xmath254 rather than @xmath255 and @xmath256 . bearing these two points in mind , the probability @xmath248 for each of the configurations presented in is given by the following expressions , @xmath257 the probability of each configuration is given by a product of probabilities that each relevant spin is of a specific category . for example , for configuration ( g ) the quantity @xmath258 refers to the probability of both boundary spins being of category @xmath61 , the power @xmath259 refers to the probability of @xmath260 interior spins also being of category @xmath61 and the term @xmath255 describes the single interior spin not of category @xmath61 . the probability of the special case that all spins are of category @xmath61 is @xmath261 . the states of spins @xmath133 and @xmath6 are determined by @xmath228 and @xmath220 in , so that @xmath262 and @xmath263 where , @xmath264 then the expectation value can be calculated by the sum , @xmath265 substitution of the probabilities given by into gives , @xmath266 the value of @xmath267 in can be found similarly by considering a central chain of infinite length ( @xmath268 ) and evaluating the probabilities , @xmath269 , that the sites @xmath40 , @xmath249 ( @xmath270 ) are of category @xmath61 and that spin @xmath238 is of category @xmath228 , @xmath271 therefore the mean magnetisation can be evaluated by a sum , @xmath272 using the definition of the correlation function ( see ) it can be found that , @xmath273q_2^{r-2}}\over{(1-q_2)^2}}\nonumber\\ & + & { q_2^\prime}^2(r-1)q_2^{r-2}(1-q_2)+2q_2^\prime q_2^{r-1}(1-q_2^\prime)+{q_2^\prime}^2q_2^{r-1}\nonumber\\ & + & \left[(q_1^\prime - q_3^\prime)+{{q_1-q_3 } \over { 1-q_2}}q_2^\prime\right]^2-{\langle s_0\rangle}^2~,\end{aligned}\ ] ] the final two ( constant ) terms cancel , and the above formula can be rewritten as , @xmath274\right.\nonumber\\ & + & \left.\frac{4q_1q_3{q_2^\prime}^2}{(1-q_2)q_2 ^ 2}r\right\}\left(q_2\right)^r~,\label{eq : corrfuncfinalr}\end{aligned}\ ] ] which completes the derivation of the correlation function given by . the exact expression for correlation function can be supported by means of numerical simulations . since the number of sites in a cayley tree grows exponentially with the number of generations it contains , a useful numerical description of a system is computationally expensive on this system . instead , our numerical calculations have been undertaken on a @xmath14-regular graph ( thin random graph @xcite ) , i.e. a graph of @xmath36 vertices , in which bonds are placed randomly in such a way that each node has a coordination number exactly equal to @xmath14 . due to the absence of a boundary , finite size effects are much less pronounced in such a system than in a cayley tree of the same size . this graph contains loops , however these loops are small in number , and are irrelevant for large enough systems @xcite . therefore , a @xmath14-regular graph gives a good approximation to a bethe lattice with the same coordination number . a set of spins is placed on the nodes of the @xmath14-regular graph . these spins interact according to the hamiltonian given by and relax by glauber dynamics as described in . a similar comparison of numerical data for magnetisation versus external field obtained for a @xmath14-regular graph with analytical results derived for a bethe lattice has been performed in @xcite . in our numerical model , each spin in the @xmath14-regular graph is assigned a random field @xmath39 according to a normal probability distribution , @xmath275 with zero mean and variance @xmath276 . all the spins are initally set in the down state , and the system is allowed to evolve according to the dynamics described in until it is in a stable state . in order to calculate the correlation function @xmath5 given by , a mean @xmath277 is calculated for all pairs of spins @xmath40 and @xmath181 separated by a chemical distance @xmath6 . the correlation function is then averaged over several ( @xmath124 ) realisations of disorder . the @xmath6-dependence of the correlation function @xmath5 presented in fig . ( 2 ) is in good agreement with the exact analytical expression derived in this paper . locus of points in parameter space @xmath278 for various number of real solutions of the self - consistent equation , @xmath279 , with the shaded and white regions corresponding to three and one solutions , respectively . at the boundaries ( continuous and dotted lines ) , the equation @xmath279 has two roots , one of which is a multiple root . the values of @xmath278 marked by the continuous line correspond to the occurrence of the infinite avalanche for increasing external field . these loci were obtained for the zt - rfim defined on a bethe lattice with @xmath118 with random fields distributed according to a normal distribution , @xmath280 , in the presence of an external field @xmath27 . the critical point at @xmath281 is marked by @xmath282 . , width=226 ] the behaviour of the zt - rfim can be determined by the number of solutions of the self - consistent equation @xmath279 ( where @xmath131 ; see ) . in the presence of external field @xmath27 , and with random fields distributed according to a pdf with zero mean and variance @xmath276 , the number of solutions of the self - consistent equation depends on the location in parameter space @xmath283 @xcite . for a bethe lattice with @xmath118 and a normal distribution of random fields , @xmath280 , the loci of points corresponding to different numbers of real solutions of @xmath279 are presented in . the shaded area ( excluding the boundaries ) corresponds to the range of parameters where the self - consistent equation has three real solutions corresponding to different branches of the multivalued function , @xmath284 , for mean magnetisation . ( see the part of the solid curve in marked by circles ) . the boundaries correspond to the case of two solutions , with the boundary marked by a solid line corresponding to the locus of points where an infinite avalanche occurs as external field increases from @xmath55 to @xmath285 ( see arrow in ) . outside the shaded region , the self - consistent equation has a single solution ( see dashed line , dot - dashed line and the part of the solid line not marked by circles in ) . the point at which all three solutions converge is a critical point marked by c in , and represents the boundary between low and high disorder , i.e. with and without an infinite avalanche , respectively ( see dotted line in ) . in this section , we analyse the behaviour of the correlation length in the zt - rfim around the critical point @xmath286 . in order to do this , we expand in terms of the differences @xmath140 , @xmath137 and @xmath138 measuring the distance to the critical point , @xmath287 . the derivative @xmath132 in this equation can be expanded in terms of the values @xmath79 , @xmath27 and @xmath28 near @xmath287 in the following way , @xmath288 where terms of higher order have been neglected because they do not play an essential role in the vicinity of the critical point . in the above expansion , we have used the fact that the condition for the merging of the three solutions of the self - consistent equation is @xmath141 . the observed critical point is known to be a saddle - node transition @xcite , meaning that the solution , @xmath79 , of exhibits the known mean - field exponents around this point . explicitly , the behaviour of @xmath140 around criticality with fixed @xmath289 can be obtained by following the standard method for saddle - point transitions ( see e.g. @xcite ) , i.e. expanding @xmath279 along the dot - dashed line in in terms of @xmath140 and @xmath137 . keeping terms only to lowest order in @xmath137 and @xmath140 , the equation reduces to the following expression , @xmath290 where @xmath291 refers to the sign of @xmath137 and the exponent takes its mean - field value , @xmath2 . similarly , expanding @xmath279 along the dashed line in and keeping lowest order terms in @xmath138 and @xmath140 , the behaviour is found to be , @xmath292 where the exponent also takes its mean - field value , @xmath1 . the substitution of and into results in , @xmath293{6}}{3}\left(-\frac{\partial^3g}{\partial { p^*}^3}\right)^{-\frac{1}{3}}\left({\frac{\partial g}{\partial h}}\right)^{-\frac{2}{3}}\left|h - h_c\right|^{-\mu}~,\label{eq : dgh } \\ \xi=-\left(\left.\frac{\partial g}{\partial p^*}\right\vert_{p^*,h_c,\delta}\right)^{-1}&= & \cases { \left|{\left.{\frac{\partial^2g}{\partial\delta\partial p^*}}\right.}\right|^{-1}\left|\delta-\delta_c\right|^{-\nu},\delta>\delta_c\\ \frac{1}{2}\left|{\left.{\frac{\partial^2g}{\partial\delta\partial p^*}}\right.}\right|^{-1}\left|\delta-\delta_c\right|^{-\nu},\delta<\delta_c~ , } \label{eq : dgdelta}\end{aligned}\ ] ] where the critical exponents for the correlation length take the values , @xmath32 and @xmath31 . as remarked in , these values are different from the mean - field exponents , @xmath294 and @xmath295 . in order to confirm the values of @xmath126 and @xmath127 , we have calculated the spin - spin correlation length for zt - rfim in the case of normally distributed random fields , @xmath296 by solving and numerically and using and . the results for the correlation length _ vs _ @xmath137 ( for fixed @xmath297 ) and @xmath138 ( for fixed @xmath298 ) are compared with and in ( a ) and ( b ) , respectively . a similar check has been performed up to @xmath299 , finding no deviation of the exponents from that for @xmath118 . spin - spin correlation length for the zt - rfim in the case of normally distributed random fields , @xmath296 _ vs _ ( a ) external field , @xmath300 , at critical value of disorder ( @xmath297 ) and ( b ) degree of disorder @xmath301 at critical value of external field ( @xmath298 ) . in panel ( a ) , the behaviour on both sides of criticality is identical ( it depends only on @xmath300 ) , and is shown by the circles . in panel ( b ) , the behaviour is different above and below criticality , and both behaviours are shown ( squares for @xmath302 and circles for @xmath44 ) . the solid line in panel ( a ) and the solid and dashed lines in panel ( b ) are plots of and , showing convergence of the exact correlation length to these equations near criticality . , width=415 ] 10 xavier illa , jordi ortn , and eduard vives . exact calculation of the energy contributions to the @xmath304 random - field ising model with metastable dynamics on the bethe lattice . , 71(18):184435 , may 2005 . xavier illa , prabodh shukla , and eduard vives . zero - temperature hysteresis in a random - field ising model on a bethe lattice : approach to mean - field behavior with increasing coordination number @xmath305 . , 73:092414 , mar 2006 .

in this paper for any dimension we give a complete description of lattice convex polytopes in that are regular with respect to the group of affine transformations preserving the lattice .

math0602193

consider an @xmath0-dimensional real vector space . let us fix a full - rank lattice in it . convex polytope _ is a convex hull of a finite number of points . a hyperplane @xmath2 is said to be _ supporting _ for a ( closed ) convex polytope @xmath3 , if the intersections of @xmath3 and @xmath2 is not empty , and the whole polytope @xmath3 is contained in one of the closed half - spaces bounded by @xmath2 . an intersection of any polytope @xmath3 with any its supporting hyperplane is called a _ face _ of the polytope . zero- and one - dimensional faces are called _ vertices _ and _ faces_. consider an arbitrary @xmath0-dimensional convex polytope @xmath3 . an arbitrary unordered @xmath4-tuple of faces containing the whole polytope @xmath3 , some its hyperface , some hyperface of this hyperface , and so on ( up to a vertex of @xmath3 ) is called a _ face - flag _ for the polytope @xmath3 . a convex polytope is said to be lattice if all its vertices are lattice points . an affine transformation is called _ lattice - affine _ if it preserves the lattice . two convex lattice polytopes are said to be _ lattice - congruent _ if there exist a lattice - affine transformation taking one polytope to the other . a lattice polytope is called _ lattice - regular _ if for any two its face - flags there exist a lattice - affine transformation preserving the polytope and taking one face - flag to the other . in this paper we give a complete description of lattice - regular convex lattice polytopes in @xmath1 for an arbitrary @xmath0 ( theorem [ main_theorem ] in section [ formulation ] ) . the study of convex lattice polytopes is actual in different branches of mathematics , such as lattice geometry(see , for example @xcite , @xcite , @xcite , @xcite ) , geometry of toric varieties ( see @xcite , @xcite , @xcite ) and multidimensional continued fractions ( see @xcite , @xcite , @xcite , @xcite , @xcite ) . mostly , it is naturally to study such polytopes with respect to the lattice - congruence equivalence relation . now we formulate two classical examples of unsolved problems on convex lattice polytopes . the first one comes from the geometry of toric varieties . find a complete invariant of lattice - congruence classes of convex lattice ( two - dimensional ) polygons . only some estimates are known at this moment ( see for example @xcite and @xcite ) . the second problem comes from lattice geometry and theory of multi - dimensional continued fractions . a lattice symplex is called _ empty _ if the intersection of this ( solid ) symplex with the lattice coincides with the set of its vertices . find a description of lattice - congruence classes of empty symplices . the answer to the second problem in the two - dimensional case is simple . all empty triangles are lattice - congruent . tree - dimensional case is much more complicated . the key to the description gives white s theorem ( 1964 ) shown in @xcite ( for more information see @xcite , @xcite , and @xcite ) . the problems similar to the shown above are complicated and seem not to be solved in the nearest future . nevertheless , specialists of algebraic geometry or theory of multidimensional continued fractions usually do not need the complete classifications but just some special examples . in the present paper we make the first steps in the study of the lattice polytopes with non - trivial group of _ lattice - symmetries _ ( i.e. the group of lattice - affine transformations , preserving the polytope ) . we describe the `` maximally '' possible lattice - symmetric polytopes : the lattice - regular polytopes . let us formulate statement for the second step in the study of the lattice polytopes with non - trivial group of lattice - symmetries . a convex lattice pyramid @xmath3 with the base @xmath5 is said to be _ lattice - regular _ if @xmath5 is a lattice - regular polytope , and the group of lattice - symmetries of the base @xmath5 ( in the hyperplane containing @xmath5 ) is expandable to the group of lattice - symmetries of the whole pyramid @xmath3 . find a description of lattice - regular convex lattice pyramids . this paper is organized as follows . we give a well - known classical description of euclidean and abstract regular polytopes in terms of schlfli symbols in section 1 . in section 2 we give necessary definitions of lattice geometry and formulate a new theorem on lattice - affine classification of lattice - regular convex polytopes . further in section 2 we prove this theorem for the two - dimensional case . we study the cases ( in any dimension ) of lattice - regular symplices , cubes , and generalized octahedra in sections 4 , 5 , and 6 respectively . finally in section 7 we investigate the remaining cases of low - dimensional polytopes and conclude the proof of the main theorem . * acknowledgement . * the author is grateful to professors v. i. arnold and a. g. khovanskii for useful remarks and discussions , and universit paris - dauphine ceremade for the hospitality and excellent working conditions . for the proof of the main theorem on lattice - regular polytopes we use the classification of abstract convex polytopes . we start this section with the description of euclidean regular polytopes and schlfli symbols for them , and then continue with the case of abstract regular polytopes . consider an arbitrary @xmath0-dimensional euclidean regular polytope @xmath3 . let @xmath6 be one of its flags . denote by @xmath7 the mass center of the face @xmath8 considered as a homogeneous solid body ( for @xmath9 ) . the @xmath0-dimensional tetrahedron @xmath10 is called the _ chamber _ of a regular polytope @xmath3 corresponding to the given flag . denote by @xmath11 ( for @xmath12 ) the reflection about the @xmath13-dimensional plane spanning the points @xmath14 , @xmath15 . see figure [ schlsymb ] . these reflections are sometimes called _ basic_. the following classical statement holds . the reflections @xmath16 generate the group of euclidean symmetries of the euclidean regular polytope @xmath3 . + for @xmath17 the angle between the fixed hyperplanes of the symmetries @xmath18 and @xmath11 equals @xmath19 , where @xmath20 is an integer greater than or equivalent to @xmath21 . the symbol @xmath22 is said to be the _ schlfli symbol _ for the polygon @xmath3 . traditionally , the string @xmath23 of the length @xmath24 in schlfli symbol is replaced by the symbol @xmath25 . since all face - flags of any regular polytope are congruent , the schlfli symbol is well - defined . @xmath26 * theorem a. on classification of regular euclidean polytopes . * _ any regular convex euclidean polytope is homothetic to some polytope of the following list . _ * list of regular euclidean polytopes . * + * dimension 1 : * a segment with schlfli symbol @xmath27 . + * dimension 2 : * a regular polygon with @xmath28 vertices @xmath29for any @xmath30@xmath31 with schlfli symbol @xmath32 . + * dimension 3 : * a regular tetrahedron @xmath33 , a regular octahedron @xmath34 , a regular cube @xmath35 , a regular icosahedron @xmath36 , a regular dodecahedron @xmath37 . + * dimension 4 : * a regular symplex @xmath38 , a regular cube @xmath39 , a regular generalized octahedron @xmath29or cross polytope , or hyperoctahedron ; with schlfli symbol @xmath40@xmath31 , a regular 24-cell @xmath29or hyperdiamond , or icositetrachoron ; with schlfli symbol @xmath41@xmath31 , a regular 600-cell @xmath29or hypericosahedron , or hexacosichoron ; with schlfli symbol @xmath42@xmath31 , a regular 120-cell @xmath29or hyperdodecahedron , or hecatonicosachoronor ; with schlfli symbol @xmath43@xmath31 . * dimension n ( n@xmath444 ) : * a regular symplex @xmath45 , a regular cube @xmath46 , a regular generalized octahedra @xmath47 . the cases of dimension one , two , and three were already known to the ancient mathematicians . the cases of higher dimensions were studied by schlfli ( see in @xcite ) . in this subsection we consider arbitrary convex polytopes . consider two @xmath0-dimensional polytopes . a homeomorphism of two @xmath0-dimensional polytopes is said to be _ combinatorical _ if it takes any face of one polytope to some face of the same dimension of the other polytope . two polytopes are called _ combinatorically isomorphic _ if there exist a combinatorical homeomorphism between them . a convex polytope is called _ combinatorical regular _ if for any two its face - flags there exist a combinatorical homeomorphism taking the polytope to itself and one face - flag to the other . * theorem b. ( mcmullen @xcite . ) * _ a polytope is combinatorical regular iff it is combinatorically isomorphic to a regular polytope . _ the proof of this statement essentially uses the work of coxeter @xcite . theorem b implies the classification of real affine and projective polytopes ( see @xcite ) . both classifications coincide with the classification of euclidean regular polytopes . for further investigations of abstract polytopes see for example the work of l. danzer and e. schulte @xcite and the book on abstract regular polytopes by p. mcmullen and e. schulte @xcite . let us fix some basis of lattice vectors @xmath48 for @xmath49 generating the lattice in @xmath1 . denote by @xmath50 the origin in @xmath1 . consider arbitrary non - zero integers @xmath51 for @xmath52 . by @xmath53 we denote the greater common divisor of the integers @xmath54 , where @xmath55 . we write that @xmath56 if the reminders of @xmath57 and @xmath58 modulo @xmath59 coincide . let @xmath60 be an arbitrary lattice polytope with the vertices @xmath61 ( where @xmath62 lattice vectors ) for @xmath63 , and @xmath64 be an arbitrary positive integer . the polygon @xmath3 with the vertices @xmath65 for @xmath66 is said to be the _ @xmath64-multiple _ of the polygon @xmath60 . a lattice polytope @xmath3 is said to be _ elementary _ if for any integer @xmath67 and any lattice polytope @xmath60 the polytope @xmath3 is not lattice - congruent to the @xmath64-multiple of the lattice polytope @xmath60 . we will use the following notation . * symplices . * for any @xmath68 we denote by @xmath69 the @xmath0-dimensional symplex with the vertices : @xmath70 * cubes . * any lattice cube is generated by some lattice point @xmath3 and a @xmath0-tuple of linearly independent lattice vectors @xmath62 : @xmath71 we denote by @xmath72 for any @xmath73 the lattice cube with a vertex at the origin and generated by all basis vectors . + by @xmath74 for any @xmath73 we denote the lattice cube with a vertex at the origin and generated by the first @xmath75 basis vectors and the vector @xmath76 . + by @xmath77 for any @xmath78 we denote the lattice cube with a vertex at the origin and generated by the vectors : @xmath79 , and @xmath80 for @xmath81 . * generalized octahedra . * we denote by @xmath82 for any @xmath73 the lattice generalized octahedron with the vertices @xmath83 for @xmath49 . + by @xmath84 for any positive @xmath0 we denote the lattice generalized octahedron with the vertices @xmath83 for @xmath85 , and @xmath86 . + by @xmath87 for any positive @xmath0 we denote the lattice generalized octahedron with the vertices @xmath50 , @xmath88 , @xmath89 for @xmath81 , and @xmath90 for @xmath81 . * a segment , octagons , and 24-sells . * denote by @xmath91 the lattice segment with the vertices @xmath50 and @xmath92 . + by @xmath93 we denote the hexagon with the vertices @xmath94 , @xmath95 , @xmath96 . + by @xmath97 we denote the hexagon with the vertices @xmath98 , @xmath99 , @xmath96 . + by @xmath100 we denote the 24-sell with 8 vertices of the form @xmath101 and 16 vertices of the form @xmath102 by @xmath103 we denote the 24-sell with 8 vertices of the form @xmath104 and 16 vertices of the form @xmath105 now we formulate the main statement of the work . [ main_theorem ] any elementary lattice - regular convex lattice polytope is lattice - congruent to some polytope of the following list . * list of the polygons . * + * dimension 1 : * the segment @xmath91 . + * dimension 2 : * the triangles @xmath106 and @xmath107 ; + the squares @xmath108 and @xmath109 ; + the octagons @xmath93 and @xmath97 . + * dimension 3 : * the tetrahedra @xmath110 , for @xmath111 ; + the octahedra @xmath112 , for @xmath113 ; + the cubes @xmath114 , for @xmath113 . + * dimension 4 : * the symplices @xmath115 and @xmath116 ; + the generalized octahedra @xmath117 , for @xmath113 ; + the 24-sells @xmath100 and @xmath103 ; + the cubes @xmath118 , for @xmath113 . + * dimension n ( n@xmath444 ) : * the symplices @xmath119 where positive integers @xmath120 are divisors of @xmath121 ; + the generalized octahedra @xmath122 , for @xmath113 ; + the cubes @xmath123 , for @xmath113 . all polytopes of this list are lattice - regular . any two polytopes of the list are not lattice - congruent to each other . @xmath124 on figure [ adjac ] we show the adjacency diagram for the elementary lattice - regular convex lattice polygons of dimension not exceeding 7 . lattice - regular lattice polygons of different ( six ) types are shown on figure [ pic_reg ] in the next section . lattice - regular lattice three - dimensional polygons of different ( nine ) types are shown on figures [ 3dsymp ] , [ 3dcube ] , and [ 3doct ] further in sections [ symplex ] , [ cube ] , and [ octahedron ] respectively . further in the proofs we will use the following definition . consider a @xmath125-dimensional lattice polytope @xmath3 . let its euclidean volume equal @xmath126 . denote the euclidean volume of the minimal @xmath125-dimensional symplex in the @xmath125-dimensional plane of the polytope by @xmath127 . the ratio @xmath128 is said to be the _ lattice volume _ of the given polytope ( if @xmath129 , or @xmath130 the _ lattice length _ of the segment , or the _ lattice area _ of the polygon respectively ) . in this section we prove theorem [ main_theorem ] for the two - dimensional case . [ regular_poligons ] any elementary lattice - regular @xmath29two - dimensional@xmath31 lattice convex polygon is lattice - congruent to one of the following polygons @xmath29see figure [ pic_reg]@xmath31 : + * 1 ) * @xmath131 ; * 2 ) * @xmath132 ; * 3 ) * @xmath133 ; * 4 ) * @xmath134 ; * 5 ) * @xmath135 ; * 6 ) * @xmath136 . @xmath137 suppose that the lattice polygon @xmath138 , where @xmath139 , is primitive and lattice - regular . let us prove that all edges of @xmath138 are of unite lattice lengths . since the polygon is lattice - regular , all its edges are lattice - congruent , and hence they are of the same lattice length . suppose that the lattice lengths of all edges equal some positive integer @xmath125 . then our polygon is lattice - congruent to the @xmath125-tuple of the polygon @xmath140 , where @xmath141 , and @xmath142 for @xmath143 . therefore , @xmath129 . denote by @xmath144 the midpoint of the edge @xmath145 for @xmath146 , and by @xmath147 the midpoint of the edge @xmath148 . suppose that @xmath0 is even ( @xmath149 ) . denote by @xmath150 the midpoint of the segment @xmath151 . note that the point @xmath150 is the common intersection of the segments @xmath152 for @xmath153 . suppose that @xmath0 is odd ( @xmath154 ) . denote by @xmath150 the intersection point of the segments @xmath155 and @xmath156 . note that the point @xmath150 is the common intersection of the segments @xmath157 , @xmath158 for @xmath159 , and the segment @xmath160 . for any integer @xmath120 such that @xmath161 the following holds . the transformation that preserves the points @xmath150 and @xmath144 , and taking the point @xmath162 to the point @xmath163 ( or @xmath164 to @xmath165 in the case of @xmath166 ) is lattice - affine and preserve the polygon @xmath138 . suppose that the polygon @xmath138 contains some lattice point not contained in the union of its vertices and segments @xmath167 for @xmath168 . then by symmetry reasons the triangle @xmath169 contains at least one lattice point , that is not contained in the edges @xmath170 and @xmath171 . denote one of such points by @xmath3 . let @xmath60 be the point symmetric to the point @xmath3 about the line @xmath171 . the segment @xmath172 is parallel to the segment , and hence the lattice point @xmath173 is contained in the interior of the segment @xmath174 . then the lattice length of the edge @xmath174 is not unit . we come to the contradiction with the above . therefore , all inner lattice points of the polygon @xmath175 are contained in the union of the segments @xmath167 for @xmath168 and vertices . now we study all different cases of configurations of lattice points on the segment @xmath171 . _ suppose that @xmath171 does not contain lattice points . then by symmetry reasons the polygon @xmath138 does not contain lattice points different to its vertices . hence the vectors @xmath176 and @xmath177 generate the lattice . consider the linear system of coordinates such that the points @xmath165 , @xmath178 , and @xmath179 have the coordinates @xmath180 , @xmath181 , and @xmath182 in it respectively . if @xmath183 , then the triangle @xmath184 is lattice - congruent to the triangle @xmath185 . let @xmath186 . since the vectors @xmath187 and @xmath177 generate the lattice , and the vectors @xmath177 and @xmath188 generate the lattice , the point @xmath189 has the coordinates @xmath190 for some integer @xmath57 . since the segment @xmath191 does not contain lattice points distinct to the endpoints , @xmath192 . by the same reasons @xmath193 . therefore , @xmath194 , and the lattice polygon @xmath195 is lattice - congruent to the lattice - regular quadrangle @xmath196 . _ _ suppose that the point @xmath150 is lattice and the segment @xmath171 does not contain lattice points distinct to @xmath150 . then the vectors @xmath197 and @xmath198 generate the lattice . consider the linear system of coordinates such that the points @xmath165 , @xmath150 , and @xmath178 have the coordinates @xmath180 , @xmath181 , and @xmath182 in it . since the vectors @xmath199 and @xmath198 generate the lattice , and the vectors @xmath200 and @xmath201 generate the lattice , the point @xmath179 has the coordinates @xmath202 for some integer @xmath57 . if @xmath203 , then the polygon is not convex or it contains straight angles . if @xmath204 , then the vectors @xmath187 and @xmath177 generate the lattice . since the vectors @xmath205 and @xmath206 generate the lattice , and the vectors @xmath177 and @xmath188 generate the lattice , the new coordinates of the point @xmath189 are @xmath207 . since @xmath208 , we have @xmath209 therefore , the lattice - regular polygon @xmath210 is lattice - congruent to the lattice - regular hexagon @xmath211 . if @xmath212 , then @xmath213 . hence @xmath214 , and @xmath194 . therefore , the polygon @xmath195 is lattice - congruent to the lattice - regular quadrangle @xmath215 . if @xmath216 , then @xmath179 is contained in the line @xmath171 . hence @xmath183 . therefore , the lattice triangle @xmath217 is lattice - congruent to the lattice - regular triangle @xmath218 . since @xmath219 , the edges @xmath148 and @xmath174 intersect for the case of @xmath220 . _ suppose that the segment @xmath171 contains the unique lattice point @xmath3 distinct from the endpoints of the segment @xmath171 . then the vectors @xmath221 and @xmath222 generate the lattice . consider the linear system of coordinates such that the points @xmath165 , @xmath3 , and @xmath178 have the coordinates @xmath180 , @xmath181 , and @xmath182 in it . since the polygon is lattice - regular , the point @xmath223 is also a lattice point , and hence @xmath224 . denote the point ( -1,-2 ) by @xmath225 . ( note that the point @xmath150 is the midpoint of the segment @xmath226 ) . the vectors @xmath227 and @xmath228 generate a sublattice of index @xmath21 . the vectors @xmath229 and @xmath230 generate a sublattice of index @xmath21 . the segment @xmath231 is of unit lattice length . therefore , the point @xmath179 has the coordinates @xmath232 for some integer @xmath57 . if @xmath233 , then the polygon is not convex , or it contains straight angles . since @xmath234 , the edges @xmath148 and @xmath174 intersect for the case of @xmath235 . _ suppose that the point @xmath150 is lattice and the segment @xmath171 contains the unique interior lattice point @xmath3 . then the vectors @xmath221 and @xmath222 generate the lattice . consider the linear system of coordinates such that the points @xmath165 , @xmath3 , and @xmath178 have the coordinates @xmath180 , @xmath181 , and @xmath182 in it . since the polygon is lattice - regular , the point @xmath236 is also lattice , and hence @xmath237 . the vectors @xmath227 and @xmath228 generate a sublattice of index @xmath21 . the vectors @xmath229 and @xmath230 generate a sublattice of index @xmath21 . the segment @xmath231 is of unit lattice length . therefore , the point @xmath179 has the coordinates @xmath238 for some integer @xmath57 , such that @xmath239 . if @xmath233 , then the polygon is not convex , or it contains straight angles , but this is impossible . if @xmath240 , then the vectors @xmath187 and @xmath177 generate a sublattice of index @xmath21 . since the vectors @xmath241 and @xmath242 generate a sublattice of index @xmath21 , and the vectors @xmath177 and @xmath188 generate a sublattice of index @xmath21 , the point @xmath243 . since @xmath208 , we have @xmath244 therefore , the lattice polygon @xmath210 is lattice - congruent to the lattice - regular hexagon @xmath245 . since @xmath246 , the edges @xmath148 and @xmath174 are intersecting for the case of @xmath247 . _ the remaining cases . _ suppose that the segment @xmath171 contains at least two interior lattice points . let @xmath248 and @xmath249 be two distinct interior lattice points of the segment @xmath171 . let also the segment @xmath250 contains the point @xmath248 . consider an lattice - affine transformation @xmath251 taking the point @xmath150 to itself , and the segment @xmath174 to the segment @xmath231 . the points @xmath252 and @xmath253 are contained in the segment @xmath254 . since the lines @xmath255 and @xmath256 are parallel and the triangle @xmath257 contains the segment @xmath255 , the lattice point @xmath258 is contained in the interior of the segment @xmath255 . hence the lattice point @xmath259 of the polygon @xmath138 is not contained in the union of segments @xmath167 for @xmath168 . we come to the contradiction . we have studied all possible cases of configurations of lattice points contained inside lattice polygons . the proof of proposition [ regular_poligons ] is completed . in this section we study all lattice - regular lattice symplices for all integer dimensions . let us fix some basis of lattice vectors @xmath48 , for @xmath49 generating the lattice in @xmath1 and the corresponding coordinate system . denote by @xmath50 the origin in @xmath1 . [ regular_symplices ] * sym@xmath260 . * all elementary lattice - regular one - dimensional lattice symplices are lattice segments of unit lattice length . * sym@xmath261 ( for @xmath68 ) . * _ i@xmath31 _ the symplex @xmath262 where @xmath263 is a positive divisor of @xmath121 is elementary and lattice - regular ; + _ ii@xmath31 _ any two symplices listed in _ @xmath29i@xmath31 _ are not lattice - congruent to each other ; + _ iii@xmath31 _ any elementary lattice - regular @xmath0-dimensional lattice symplex is lattice - congruent to one of the symplices listed in _ @xmath29i@xmath31_. the three - dimensional tetrahedra @xmath264 , @xmath265 , and @xmath266 are shown on figure [ 3dsymp ] . @xmath267 we start the proof of studying of some low - dimensional cases . the one - dimensional case is trivial and is omitted here . the two - dimensional case was described in proposition [ regular_poligons ] . let us study the three - dimensional case . _ three - dimensional case . _ consider an arbitrary elementary lattice - regular three - dimensional lattice tetrahedron @xmath259 . since its faces are lattice - regular , by proposition [ regular_poligons ] the faces are lattice - congruent either to @xmath185 or to @xmath268 . suppose that the faces of @xmath259 are lattice - congruent to @xmath185 . then there exist a positive integer @xmath58 , nonnegative integers @xmath269 , @xmath270 less than @xmath58 , and a lattice - affine transformation taking the the tetrahedron @xmath259 to the tetrahedron @xmath271 with the vertices @xmath272 since @xmath271 is also a lattice - regular tetrahedron , the group of its symmetries is isomorphic to the group of permutations of order 4 . this group is generated by the following transpositions of vertices : @xmath273 and @xmath274 , @xmath274 and @xmath275 , and @xmath127 and @xmath274 . the first two transpositions are linear , and the third one is linear after shifting by the vector @xmath276 . direct calculations shows , that the matrices of the corresponding linear transformations are the following : @xmath277 since the listed transformations are lattice - linear we have only the following possibilities , all these matrices are integer . therefore , @xmath278 . since , positives @xmath269 and @xmath270 was chosen to be smaller than @xmath58 , we have @xmath279 . since the matrix @xmath179 is integer , the coefficient @xmath280 is also integer . so we have to check only the following cases for @xmath269 , @xmath270 , and @xmath58 : @xmath281 , and @xmath282 ; @xmath283 , and @xmath284 ; @xmath285 , and @xmath286 . these cases corresponds to the tetrahedra @xmath264 , @xmath265 , and @xmath266 respectively . since the lattice volume of @xmath287 equals @xmath263 , the above tetrahedra are not lattice - congruent . let us prove that the faces of @xmath259 are not lattice - congruent to @xmath268 by reductio ad absurdum . suppose it is so . let @xmath127 , @xmath273 , @xmath274 , and @xmath275 be the vertices of @xmath259 . since the faces @xmath288 and @xmath289 are congruent to @xmath268 , the face @xmath288 contains a unique lattice point in its interior ( we denote it by @xmath248 ) , and the face @xmath289 contains a unique lattice point in its interior ( we denote it by @xmath249 ) . consider a lattice - symmetry of @xmath259 permuting @xmath127 and @xmath273 and preserving @xmath274 and @xmath275 . this symmetry takes the face @xmath288 to the face @xmath289 , and hence it maps the point @xmath248 to @xmath249 . thus , the lattice vector @xmath290 is parallel to the vector @xmath291 . hence , the point @xmath292 is interior lattice point of the segment @xmath291 . therefore , the segment @xmath291 is not of unit lattice length and the face @xmath293 is not lattice - congruent to @xmath268 . we come to the contradiction . this completes the proof of proposition [ regular_symplices ] for the three - dimensional case . in the higher dimensional case we first study two certain families of symplices . _ the first type of @xmath0-dimensional symplices . _ consider any symplex ( for @xmath186 ) with the vertices @xmath294 we denote it by @xmath295 . we suppose that all @xmath20 are nonnegative integers satisfying @xmath296 for @xmath297 . let us find the conditions on @xmath20 for the symplex to be lattice - regular . if @xmath298 is a lattice - regular symplex then the group of its symmetries is isomorphic to the group of permutations of order @xmath121 . this group is generated by the following transpositions of vertices : the transposition exchanging @xmath299 and @xmath300 for @xmath301 , and the transposition exchanging @xmath127 and @xmath274 . the first @xmath75 transpositions are linear ( let their matrices be @xmath302 for @xmath301 ) , and the last one is linear after shifting by the vector @xmath276 ( denote the corresponding matrix by @xmath164 ) . let us describe the matrices of these transformations explicitly . the matrix @xmath302 for @xmath303 coincides with the matrix transposing the vectors @xmath304 and @xmath305 , except the last column . the @xmath0-th column contains the coordinates of the vector @xmath306 the matrix @xmath307 coincides with the matrix of identity transformation except the last two columns . these columns contain the coefficients of the following two vectors respectively : @xmath308 the matrix @xmath309 coincides with the unit matrix except the second row . this row is as follows : @xmath310 the determinants of all such matrices equal @xmath311 . so the corresponding affine transformations are lattice iff all the coefficients of all the matrices are lattice . the matrices @xmath302 for @xmath312 are lattice iff @xmath313 since @xmath296 , we have the equalities . suppose @xmath314 for some positive integer @xmath263 . the matrix @xmath307 is integer , iff @xmath315 . therefore , @xmath316 is divisible by @xmath317 , and hence @xmath318 . the matrix @xmath164 is integer , iff @xmath319 is divisible by @xmath263 , or equivalently @xmath121 is divisible by @xmath263 . so , we have already obtained that the symplex @xmath320 where @xmath186 and @xmath321 for @xmath301 is lattice - regular iff it is coincides with some @xmath322 for some @xmath263 dividing @xmath0 . since the lattice volume of @xmath323 equals @xmath263 , the above symplices are not lattice - congruent . _ the second type of @xmath0-dimensional symplices . _ here we study symplices ( for @xmath186 ) with the vertices @xmath324 denote such symplices by @xmath298 . we also suppose , that all @xmath20 are nonnegative integers satisfying @xmath296 for @xmath301 , and @xmath325 . let us show that all these symplices are not lattice - regular . consider an arbitrary symplex @xmath298 , satisfying the above conditions . consider the symmetry exchanging @xmath326 and @xmath327 . this transformation is linear . its matrix coincide with the matrix of identity transformation , except for the last two columns . these columns contains the coefficients of the following two vectors respectively : @xmath328 if this transformation is lattice - linear , then @xmath329 is divisible by @xmath263 . consider the symmetry exchanging @xmath330 and @xmath327 . this transformation is linear . its matrix coincide with the matrix of identity transformation , except for the first column and the last two columns . these columns contains the coefficients of the following three vectors respectively : @xmath331 if this transformation is lattice - linear then @xmath270 is divisible by @xmath332 . since @xmath270 and @xmath329 are divisible by @xmath332 and @xmath333 , the symplex @xmath298 is not lattice - regular . _ conclusion of the proof . _ statement _ @xmath29i@xmath31 _ is already proven . since @xmath263 is a lattice volume of @xmath323 , statement _ @xmath29ii@xmath31 _ holds . we prove statement _ @xmath29iii@xmath31 _ of the proposition by the induction on the dimension @xmath0 . for @xmath334 the statement is already proven . suppose that it is true for an arbitrary @xmath335 . let us prove the statement for @xmath121 . consider any lattice - regular @xmath4-dimensional lattice symplex @xmath259 . since it is lattice - regular , all its faces are lattice - regular . by the induction assumption there exist a positive integer @xmath263 dividing @xmath121 such that the faces of @xmath259 are lattice - congruent to @xmath322 . therefore , @xmath259 is lattice - affine equivalent to the symplex @xmath336 , where @xmath337 , and @xmath296 for @xmath301 . by the above cases the lattice - regularity implies , that @xmath338 , and that there exist a positive integer @xmath339 dividing @xmath340 such that the symplex @xmath341 coincides with @xmath342 . this concludes the proof of the statement _ @xmath29iii@xmath31 _ for the arbitrary dimension . proposition [ regular_symplices ] is proven . in this section we describe all lattice - regular lattice cubes for all integer dimensions . [ regular_cubes ] * cube@xmath260 . * all elementary lattice - regular one - dimensional lattice cubes are lattice segments of unit lattice length . * cube@xmath343 . * all elementary lattice - regular two - dimensional lattice cubes are lattice - congruent to @xmath196 , or to @xmath215 . the cubes @xmath196 and @xmath215 are not lattice - congruent . * cube@xmath261 ( for @xmath344 ) . * all elementary lattice - regular @xmath0-dimensional lattice cubes are lattice - congruent to @xmath345 , @xmath346 , or @xmath347 . the cubes @xmath345 , @xmath346 , and @xmath347 are not lattice - congruent to each other . the three - dimensional cubes @xmath345 , @xmath346 , and @xmath347 are shown on figure [ 3dcube ] . @xmath348 we use the following two facts . [ interior_points_cube ] any elementary lattice - regular @xmath0-dimensional lattice cube contains at most one lattice point in its interior . if the cube contains an interior lattice point , this point coincides with the intersection point of the diagonals of the cube . suppose the converse is true . there exist a primitive @xmath0-dimensional cube with an interior lattice point @xmath349 distinct to the intersections of the diagonals . then there exist a lattice - linear reflection @xmath251 of the cube above some @xmath75 dimensional plane , that do not preserve @xmath349 . so the line of vector @xmath350 coincide with the line containing one of the generating vectors of the cube . hence , the one - dimensional faces of the cube are not elementary . therefore , the cube is not elementary . we come to the contradiction . let us give the following important definition . consider some hyperplane containing a sublattice of the lattice of corank 1 and a lattice point in the complement to this hyperplane . let the euclidean distance from the given point to the given hyperplane equals @xmath351 . the minimal value of nonzero euclidean distances from the points of the lattice to the hyperplane is denoted by @xmath352 . the ratio @xmath353 is said to be the _ lattice distance _ from the given lattice point to the given lattice hyperplane . [ ind_dis_cube ] the lattice distances from any vertex of any elementary lattice - regular @xmath0-dimensional lattice cube to any its @xmath13-dimensional face @xmath29that does not containing the given vertex@xmath31 equals either @xmath354 , or @xmath130 . _ proof of proposition [ regular_cubes ] . _ the one - dimensional case is trivial . the two - dimensional case was described in proposition [ regular_poligons ] . let us study higher - dimensional cases . _ the lattice cubes @xmath355 , @xmath356 , and @xmath357 . _ first , let us study the cases of the polytopes @xmath355 , @xmath356 , and @xmath357 for any @xmath78 . since lattice volumes of @xmath355 , @xmath356 , and @xmath357 are @xmath358 , @xmath359 , and @xmath360 respectively , the listed cubes are not lattice - congruent to each other . let us prove that these polytopes are lattice - regular for any @xmath78 . since the vectors of @xmath355 generate lattice , it is lattice - regular . now we study the case of @xmath356 . denote by @xmath62 the vector @xmath48 for @xmath17 and by @xmath361 the vector @xmath362 . the group of lattice symmetries of the cube @xmath356 is generated by the linear operators @xmath302 transposing the vectors @xmath62 and @xmath363 for @xmath17 , and the last one : the composition of the symmetry @xmath164 sending @xmath364 to @xmath365 and preserving @xmath62 for @xmath366 and the lattice shift on the vector @xmath364 . let us check that all the linear transformations @xmath302 are lattice - linear . we show explicitly the matrices of @xmath302 in the basis @xmath48 for @xmath168 . the matrix of @xmath302 for @xmath367 coincides with the the matrix of the transposition of the vectors @xmath304 and @xmath305 . the matrix of @xmath307 coincides with the matrix of identity transformation except the last two columns . these columns contain the coefficients of the vectors @xmath361 and @xmath368 respectively . the matrix of @xmath164 coincides with the matrix of identity transformation except the first row , which is @xmath369 . since all these matrices are in @xmath370 , the cube @xmath356 is lattice - regular . let us consider now the case of @xmath357 . put by definition @xmath371 , and @xmath372 for @xmath81 . the group of lattice symmetries of the cube @xmath357 is generated by the linear operators @xmath302 transposing the vectors @xmath62 and @xmath363 for @xmath17 , and the last one : the composition of the symmetry @xmath164 sending @xmath364 to @xmath365 and preserving @xmath62 for @xmath373 , and the lattice shift on the vector @xmath364 . let us check that all the linear transformations @xmath302 are lattice - linear . the matrix of @xmath165 coincides with the matrix of identity transformation except the second row , which is @xmath374 . the matrix of @xmath302 for @xmath375 coincides with the matrix transposing the vectors @xmath304 and @xmath305 . the matrix of @xmath164 coincides with the matrix of identity transformation except the first row , which is @xmath376 . since all these matrices are in @xmath370 , the cube @xmath357 is lattice - regular . _ conclusion of the proof of proposition [ regular_cubes ] . _ now we prove , that any elementary lattice - regular @xmath0-dimensional lattice cube for @xmath335 is lattice - congruent to @xmath355 , @xmath356 , or @xmath357 . by the induction on @xmath0 . _ the base of induction . _ any face of any three - dimensional lattice - regular cube is lattice - regular . suppose , that the faces of three - dimensional lattice - regular cube @xmath377 are lattice - congruent to @xmath196 . then @xmath377 is lattice - congruent to the cube generated by the origin and the vectors @xmath79 , @xmath378 , and @xmath379 . by corollary [ ind_dis_cube ] we can choose @xmath380 equals either @xmath354 or @xmath130 . if @xmath381 then @xmath377 is lattice - congruent to @xmath382 . if @xmath383 , then we can choose @xmath269 and @xmath270 being @xmath384 , or @xmath354 . direct calculations show , that the only possible case is @xmath385 , and @xmath386 , i.e. the case of @xmath387 . suppose now , that the faces of three - dimensional lattice - regular cube @xmath377 are lattice - congruent to @xmath215 . then @xmath377 is lattice - congruent to the cube generated by the origin and the vectors @xmath79 , @xmath388 , and @xmath379 . by corollary [ ind_dis_cube ] we can choose @xmath380 equals either @xmath354 or @xmath130 . then we can also choose @xmath269 and @xmath270 being @xmath384 , or @xmath354 . direct calculations show , that the only possible case @xmath385 , @xmath386 , and @xmath383 corresponds to @xmath389 . _ the step of induction . _ suppose that any elementary lattice - regular @xmath13-dimensional lattice cubes ( @xmath390 ) are lattice - congruent to @xmath391 , @xmath392 , or @xmath393 . let us prove that any elementary lattice - regular @xmath0-dimensional lattice cubes are lattice - congruent to @xmath355 , @xmath356 , or @xmath357 . any face of any lattice - regular cube is lattice - regular . suppose , that the faces of @xmath13-dimensional lattice - regular cube @xmath377 are lattice - congruent to @xmath391 . then @xmath377 is lattice - congruent to the cube @xmath394 generated by the origin and the vectors @xmath395 for @xmath396 and the vector @xmath397 . by corollary [ ind_dis_cube ] we can choose @xmath317 equals either @xmath354 or @xmath130 . if @xmath398 , then the lattice volume of @xmath377 is @xmath358 and it is lattice - congruent to @xmath355 . if @xmath399 , then we can choose @xmath20 being @xmath384 , or @xmath354 for @xmath17 . consider a symmetry of @xmath394 transposing the vectors @xmath400 and @xmath401 for @xmath367 . this transformation is linear and its matrix coincides with the matrix of the transposition of the vectors @xmath304 and @xmath305 , except the last column . the @xmath0-th column contains the coordinates of the vector @xmath402 since the transformation is lattice , @xmath403 since any @xmath20 is either zero or unit , the above imply @xmath404 . if @xmath405 , then the vector @xmath361 is not of the unit lattice length , but the vector @xmath364 is of the unit length , so @xmath394 is not lattice - regular . if @xmath385 then @xmath394 coincides with @xmath356 . suppose , that the faces of @xmath13-dimensional lattice - regular cube @xmath377 are lattice - congruent to @xmath392 . then @xmath377 is lattice - congruent to the cube @xmath394 generated by the origin and the vectors @xmath395 , for @xmath406 , @xmath407 , and @xmath408 . consider a symmetry of @xmath394 transposing the vectors @xmath409 and @xmath410 . this transformation is linear and its matrix coincides with the matrix of identity transformation except the last two columns . these columns contain the coefficients of the following two vectors respectively : @xmath411 since the described transformation is lattice , the integers @xmath412 and @xmath317 are even , and @xmath413 is odd . consider now a symmetry of @xmath394 transposing the vectors @xmath414 and @xmath409 . this transformation is linear and its matrix coincides with the matrix of identity transformation except the last three columns . these columns contain the coefficients of the following three vectors respectively : @xmath415 since the described transformation is lattice , the integer @xmath416 is even , and thus @xmath413 is even . we come to the contradiction with the divisibility of @xmath413 by 2 . so @xmath377 is not lattice - regular . suppose , that the faces of @xmath13-dimensional lattice - regular cube @xmath377 are lattice - congruent to @xmath393 . then @xmath377 is lattice - congruent to the cube @xmath394 generated by the origin and the vectors @xmath417 , @xmath372 for @xmath418 , and @xmath408 . by corollary [ ind_dis_cube ] we can choose @xmath317 equals either @xmath354 or @xmath130 . then we choose @xmath20 being @xmath384 , or @xmath354 for @xmath419 . consider a symmetry of @xmath394 transposing the vectors @xmath409 and @xmath410 . this transformation is linear and its matrix coincides with the matrix of identity transformation except the last two columns . these columns contain the coefficients of the following two vectors respectively : @xmath420 since the described transformation is lattice , the integer @xmath269 is odd , and the integers @xmath20 for @xmath421 are even . thus @xmath422 therefore @xmath394 coincides with @xmath357 . we have already studied all possible @xmath0-dimensional cases . this proves the statement for the dimension @xmath0 . all statements of proposition [ regular_cubes ] are proven . in this section we describe all lattice - regular lattice generalized octahedra for all integer dimensions greater than 2 . [ regular_oct ] all elementary lattice - regular @xmath0-dimensional lattice generalized octahedra for @xmath78 are lattice - congruent to @xmath423 , @xmath424 , or @xmath425 . the generalized octahedra @xmath423 , @xmath424 , and @xmath425 are not lattice - congruent to each other . we show the ( three - dimensional ) octahedra @xmath426 , @xmath427 , and @xmath428 on figure [ 3doct ] . @xmath429 consider an arbitrary elementary @xmath0-dimensional lattice - regular generalized octahedron @xmath3 . let the vertices @xmath430 of @xmath3 be enumerated in such a way that for any positive integer @xmath431 there exist a lattice symmetry exchanging @xmath432 and @xmath433 and preserving any other vertex . so , @xmath434 is a lattice - regular square for @xmath435 . therefore , the midpoints of the segments @xmath436 coincide for @xmath437 . denote the common midpoint of the segments @xmath436 by @xmath349 . suppose the point @xmath349 is lattice . consider the lattice cube with the vertices @xmath438 , we denote it by @xmath439 . the cube @xmath439 is also lattice - regular . note that the lattice - regular generalized octahedra @xmath440 and @xmath441 have lattice - congruent cubes @xmath442 and @xmath443 iff @xmath440 and @xmath441 are lattice - congruent . since @xmath3 is elementary , the segment @xmath444 is of unit lattice length . therefore , the cube @xmath439 is lattice - congruent to the 2-multiple of some @xmath445 , for @xmath446 . if @xmath439 is lattice - congruent to the 2-multiple of @xmath355 , or to the 2-multiple of @xmath356 , then @xmath3 is lattice - congruent to @xmath423 , or to @xmath424 , respectively . if @xmath439 is lattice - congruent to the 2-multiple of @xmath357 , then @xmath3 is not elementary . suppose now , the common midpoint @xmath349 of the diagonals is not lattice . if the lattice length of @xmath447 equals @xmath448 for some positive @xmath125 , then the generalized octahedron @xmath3 is not elementary . suppose the segment @xmath447 is of unite lattice length . consider a 2-multiple to the polygon @xmath3 and denote in by @xmath449 . since the segment @xmath447 is of unit lattice length , the cube @xmath450 is the 2-multiple of some @xmath445 , for @xmath446 . if @xmath450 is lattice - congruent to the 2-multiple of @xmath355 , or of @xmath356 , then @xmath3 is not a lattice polytope . if @xmath450 is lattice - congruent to the 2-multiple of @xmath357 , then @xmath3 is lattice - congruent to @xmath425 . the generalized octahedra @xmath423 , @xmath424 , and @xmath425 are lattice - regular , since so are the cubes @xmath451 , @xmath452 , and @xmath453 . the generalized octahedra @xmath423 , @xmath424 , and @xmath425 are not lattice - congruent to each - other , since the corresponding elementary cubes @xmath451 , @xmath452 , and @xmath453 are not lattice - congruent . in this section we obtain proof of theorem [ main_theorem ] by combining the results of propositions from the previous sections and describing the remaining low - dimensional cases . consider any convex lattice - regular lattice polytope . since it is lattice - regular and convex it is combinatorically regular . therefore , by theorem b it is combinatorically isomorphic to one of the euclidean polytopes of theorem a. in section [ 2d ] we gave the description of the two - dimensional case . in sections [ symplex ] , [ cube ] , and [ octahedron ] we studied the cases of lattice - regular polytopes combinatorically isomorphic to regular symplices @xmath45 , regular cubes @xmath46 , and regular generalized octahedra @xmath47 respectively . we have already classified all lattice - regular elementary tetrahedra , cubes , and octahedra . there is no lattice - regular dodecahedron , since there is no lattice - regular pentagon . there is no lattice - regular icosahedron , since there is no lattice - affine transformation with a fixed point of order 5 . _ the case of 24-sell . _ suppose that @xmath3 is a lattice - regular 24-sell . it contains 16 vertices such that the subgroup of the group of all lattice - symmetries of @xmath3 preserving these 16 vertices is isomorphic to the group of the symmetries of the four - dimensional cube . so we can naturally define a combinatorial - regular cube associated with this 16 vertices . any two - dimensional face of this cube is an euclidean parallelogram , since such face is the diagonal section containing four lattice points of some lattice - regular octahedron . if all two - dimensional faces are parallelograms , then all three - dimensional faces are parallelepipeds and these 16 vertices are vertices of a four - dimensional parallelepiped . denote this parallelepiped by @xmath377 . since all transformations of @xmath377 are lattice - affine , the polytope @xmath377 is a lattice - regular cube . ( note that for any @xmath454-sell there exist exactly three such ( distinct ) cubes ) . suppose that @xmath377 is a lattice - regular cube generated by the origin and some vectors @xmath62 for @xmath455 . consider also the coordinates @xmath456 corresponding to this basis . let the point @xmath457 of @xmath3 connected by edges with the vertices of @xmath377 is in the plane with the unit last coordinate . then , the point @xmath458 is also a vertex . thus , the point @xmath459 is also a vertex . note that the points @xmath457 and @xmath460 are symmetric about the center of the cube : @xmath461 . so @xmath462 . thus , @xmath463 . similar calculations show that @xmath464 . therefore , the eight points do not contained in @xmath377 coincide with the following points : @xmath465 consider a lattice - regular primitive 24-sell @xmath3 and @xmath377 one of the corresponding cubes . since the edges of @xmath377 are the edges of @xmath3 , the cube @xmath377 is also primitive . let us study all three possible cases of lattice - affine types of @xmath377 . in the case of @xmath468 and @xmath469 all the vertices are lattice and in our notation coincide with @xmath470 and @xmath471 respectively . straightforward calculations shows , that both resulting 24-sells are lattice - regular . _ the case of 120-sell . _ the 120-sell is non - realizable as an lattice - regular lattice polytope , since its two - dimensional faces should be lattice - regular lattice pentagons . by the above , lattice - regular lattice pentagons are not realizable . _ the case of 600-sell . _ consider an arbitrary polytope with topological structure of 600-sell having one vertex at the origin @xmath50 . let @xmath472 be some edge of this 600-sell . the group of symmetries of an abstract 600-sell with fixed vertex @xmath50 is isomorphic to the group of symmetries of an abstract icosahedron . the group of symmetries of an abstract 600-sell with fixed vertices @xmath50 and @xmath273 is isomorphic to the group of symmetries of an abstract pentagon . so , there exist a symmetry @xmath349 of the 600-sell of order 5 preserving the vertices @xmath50 and @xmath273 . if the polytope @xmath3 is lattice - regular , then this symmetry is lattice - linear . since @xmath473 is the identity transformation and the space is four - dimensional , the characteristic polynomial of @xmath349 in the variable @xmath474 is either @xmath475 or @xmath476 . since @xmath477 , the characteristic polynomial of @xmath349 is divisible by @xmath475 . if the characteristic polynomial is @xmath475 , then the operator @xmath349 is the identity operator of order 1 and not of order 5 . therefore , there is no lattice - regular lattice polytope with the combinatorial structure of the 600-sell . 99 v. i. arnold , _ continued fractions _ m. : moscow center of continuous mathematical education , 2002 . v. i. arnold , _ statistics of integer convex polygons _ , func . an . and appl . , v. 14(1980 ) , n. 2 , pp . 13 . i. brny j .- kantor , _ on the number of lattice free polytopes _ , european j. combin . v. 21(1 ) , 2000 , pp . 103110 . i. brny a. m. vershik , _ on the number of convex lattice polytopes _ , geom v. 2(4 ) , 1992 , pp . coxeter , _ the complete enumeration of finite groups of the form @xmath478 _ , j. london math . soc , v. 10 ( ser . 1 ) , no 37 , 1935 , pp . 2125 . l. danzer , e. schulte , _ regulre inzidenzkomplexe i _ , geom . dedicata , v. 13(1982 ) , pp . w. fulton , _ introduction to toric varieties _ , annals of mathematics studies ; princeton university press , v. 131(1993 ) , j .- kantor , k. s. sarakaria _ on primitive subdivisions of an elementary tetrahedron _ , pacific j. of math . v. 211(1 ) , 2003 , pp . . o. n. karpenkov , _ two - dimensional faces of multidimensional continued fractions and pyramids on integer lattices _ , preprint n 2005 - 26 , cahiers du ceremade , unite mixte de recherche du c.n.r.s . n7534 ( 2005 ) , http://www.ceremade.dauphine.fr/preprint/cmd/2005-26.pdf . a. g. khovanskii , a. pukhlikov , _ finitely additive measures of virtual polytopes _ , algebra and analysis , v. 4(2 ) , 1992 , pp . 161185 . e. i. korkina , _ two - dimensional continued fractions . the simplest examples _ , proceedings of v. a. steklov math . ins . , v. 209(1995 ) , pp . 243166 . g. lachaud , _ voiles et polyhedres de klein _ , act . , 176 pp . , hermann , 2002 . p. mcmullen , _ combinatorially regular polytopes _ , mathematika , v. 14(1967 ) , pp 142150 . p. mcmullen , _ affinely and projectively regular polytopes _ , j. london math . soc , v. 43(1968 ) , pp 755757 . p. mcmullen , e. schulte , _ abstract regular polytopes _ , series : encyclopedia of mathematics and its applications , no . 92(2002 ) , 566 p. moussafir , _ voiles et polydres de klein : geometrie , algorithmes et statistiques _ , docteur en sciences thse , universit paris ix - dauphine , ( 2000 ) , + see also at http://www.ceremade.dauphine.fr/~msfr/. t. oda , _ convex bodies and algebraic geometry , an introduction to the theory of toric varieties _ , springer - verlag , survey in mathemayics , 15(1988 ) . b. reznick , _ lattice point symplices _ , discrete math . , v. 60(1986 ) , pp . 219242 . l. schlfli , _ theorie der vielfachen kontinuitt _ , written 1850 - 1852 ; zrcher und furrer , zrich 1901 ; denkschriften der schweizerischen naturforschenden gesellschaft v. 38(1901 ) , pp . 1237 ; reprinted in _ ludwig schlfli 1814 - 1895 , gesammelte mathematische abhandlungen _ , i , birkhuser , basel 1950 , pp . . g. k. white , _ lattice tetrahedra _ , canadian j. of math . 16(1964 ) , pp . 389396 .

the definition of the grafting operation for quasifuchsian groups is extended by bromberg to all-groups . although the grafting maps are not necessarily continuous at boundary groups , in this paper , we show that the grafting maps take every `` standard '' convergent sequence to a convergent sequence . as a consequence of this result , we extend goldman s grafting theorem for quasifuchsian groups to all boundary-groups .

math0411133

in this paper we are interested in the behavior of the holonomy map from projective structures to representations . especially , we study the continuity and the non - continuity of the local inverse of the holonomy map at the boundary of the space of discrete faithful representations . let @xmath1 be an oriented closed surface of genus @xmath2 . a projective structure on @xmath1 is a @xmath3-structure where @xmath4 is a riemann sphere @xmath5 and @xmath6 is the group of projective automorphism of @xmath5 . let @xmath7 denote the space of projective structures on @xmath1 and @xmath8 the space of conjugacy classes of representations @xmath9 . the holonomy map @xmath10 takes a projective structure to its holonomy representation . we are interested in the quasifuchsian space @xmath11 of faithful representations with quasifuchsian images . although @xmath12 is a connected , contractible , open submanifold of @xmath8 , the topology of its closure @xmath13 in @xmath8 is very complicated ; for instance mcmullen @xcite showed that @xmath14 self - bumps , i.e. , there exists @xmath15 such that @xmath16 is disconnected for any sufficiently small neighborhood @xmath17 of @xmath18 , by using projective structures and ideas of anderson and canary @xcite . in fact , since the holonomy map @xmath10 is a local homeomorphism , some kind of complexity of the topology of @xmath19 can be observed by studying the inverse image @xmath20 of @xmath14 in @xmath7 . by goldman s grafting theorem @xcite , the set of components of @xmath21 is in one - to - one correspondence with the set @xmath22 of integral points of measured laminations . for every @xmath23 , the associated component @xmath24 of @xmath21 is taken by the holonomy map biholomorphically onto @xmath14 . we let @xmath25 denote the univalent local branch of @xmath26 and call it the _ grafting map _ for @xmath23 . the _ standard component _ @xmath27 is the component of @xmath21 of projective structures with injective developing map . then mcmullen @xcite actually showed that there exists a sequence in @xmath28 converging to @xmath29 , which implies that @xmath14 self - bumps . moreover , we showed in @xcite that for any @xmath30 , there exists @xmath31-components of @xmath21 which bump simultaneously ( i.e. , having intersecting closures ) , that any component of @xmath21 except for @xmath27 self - bumps , and that any two components of @xmath21 bump . all these phenomena are obtained by using exotic convergent sequence in @xmath14 ( see definition below ) constructed by anderson and canary @xcite . let @xmath32 be the parameterization of @xmath14 of bers simultaneous uniformization and let @xmath33 be a sequence converging to @xmath15 . then we say that the convergence @xmath34 is _ standard _ if there exists a compact subset @xmath35 of @xmath36 which contains all @xmath37 or all @xmath38 ; otherwise it is _ exotic_. we let @xmath39 denote the set of limits of standard convergence sequences , or the set of all boundary @xmath0-groups ( see 2.1 ) . then by using bromberg s observation in @xcite , the grafting map @xmath40 is naturally extended to @xmath41 for every @xmath23 . here @xmath42 denotes the one - point compactification of @xmath7 . as described above , anderson and canary @xcite showed that there exist sequences @xmath43 converging exotically to @xmath44 , which course the non - continuity of the map @xmath45 at @xmath44 for all @xmath23 . on the contrary , we study the continuity of the grafting maps in this paper . the following theorem ( theorem 5.1 ) is the main result of the paper , which states that each grafting map @xmath40 behaves as a continuous map for every standard convergent sequence . let @xmath46 be a sequence converging standardly to @xmath44 . then the sequence @xmath47 converges to @xmath48 in @xmath49 for every @xmath23 . let @xmath50 denote the closure of @xmath27 in @xmath7 , _ not _ in @xmath49 . as we see in 3.3 , a sequence @xmath51 converges to @xmath52 if and only if the sequence @xmath53 converges standardly to @xmath54 . hence we have the next corollary ( theorem 6.10 ) of theorem 1.1 . the map @xmath55 is extended continuously to @xmath56 for each @xmath23 , where @xmath50 is the closure of @xmath27 in @xmath7 . an important remark is that , since we do not know whether @xmath27 self - bumps or not , the statement of theorem 1.1 ( or corollary 1.2 ) is not trivial and worth considering . the following theorem plays an important roll in the proof of theorem 1.1 , which have some interest in its own light ( see 4 ) . a bers slice @xmath57 is a slice of @xmath14 of the type @xmath58 or @xmath59 . in addition , we let @xmath60 denote the projection which takes a projective structure to its underling conformal structure . let @xmath57 be a bers slice and @xmath61 a sequence of distinct elements of @xmath22 . then the sequence @xmath62 eventually escapes any compact subset @xmath35 of @xmath36 ; that is , @xmath63 for all large enough @xmath31 . as a consequence of theorem 1.1 , we extend goldman s grafting theorem as follows ( see theorem 6.1 ) : for every boundary @xmath0-groups @xmath64 , all projective structures with holonomy @xmath18 are obtained by grafting of @xmath18 ; that is , @xmath65 as mentioned above , any two components of @xmath20 bump . on the contrary , we obtain the following theorem ( theorem 6.11 ) , which reveals that only exotic convergent sequences cause the bumping of distinct components of @xmath21 . set @xmath66 for a compact subset @xmath67 . then the inverse image @xmath68 of @xmath69 in @xmath7 is discrete ; that is , every connected component @xmath70 of @xmath68 has an open neighborhood in @xmath7 which is disjoint from any other one . it is conjectured by bers , sullivan and thurston that the closure of @xmath14 is equal to the space @xmath71 of discrete , faithful representations . this conjecture is closely related to thurston s ending lamination conjecture , which was recently announced by minsky to be solved affirmatively . but we do not make use of these deep results in this paper . the author would like to thank katsuhiko matsuzaki for useful discussions on the topics of 4 . a kleinian group @xmath72 is a discrete subgroup of @xmath73 , which acts on hyperbolic @xmath74-space @xmath75 as isometries and on the sphere at infinity @xmath5 as conformal automorphisms . the union @xmath76 is naturally topologized as a closed 3-ball so that @xmath73 acts continuously on it . for a kleinian group @xmath72 , we let @xmath77 denote the region of discontinuity and @xmath78 the limit set . we associate to a kleinian group @xmath72 the following orbit spaces : @xmath79 where @xmath80 is called the _ conformal boundary _ of @xmath81 . in general if @xmath82 is an oriented maniford with boundary @xmath83 , we orient @xmath83 by requiring that the frame @xmath84 has positive orientation whenever @xmath85 is a positively oriented frame on @xmath83 and @xmath31 is an inward - pointing vector . let @xmath1 be an oriented closed surface of genus @xmath2 . let @xmath8 be the space of conjugacy classes @xmath86 $ ] of representations @xmath9 whose images @xmath87 are non - abelian . the space @xmath8 is equipped with the algebraic topology , the topology of convergence on generators up to conjugation . ( by abuse of notation , we also denote @xmath86 $ ] by @xmath18 if there is no confusion . ) it is known that @xmath8 is a @xmath88-dimensional complex manifold ( see theorem 4.21 in @xcite ) . let @xmath89 be the subset of @xmath8 of discrete , faithful representations , which is known to be a closed subset of @xmath8 by jrgensen @xcite . let @xmath34 be an algebraic convergence sequence in @xmath89 . then it is known by jrgensen and marden @xcite that , by passing to a subsequence if necessary , the sequence @xmath90 of kleinian groups converges geometrically to some kleinian group @xmath91 , which contains the algebraic limit @xmath92 . here the convergence @xmath93 is _ geometric _ if and only if for any @xmath94 , there exist @xmath95 such that @xmath96 and if for every convergent sequence @xmath97 , the limit is contained in @xmath91 . a convergence @xmath34 in @xmath89 is said to be _ strong _ if @xmath90 converges geometrically to the algebraic limit @xmath92 . for every @xmath98 with image @xmath99 , bonahon s theorem @xcite guarantees that there exists an orientation preserving homeomorphism @xmath100 , where @xmath101 is an open interval . the conformal boundary @xmath102 ( possibly empty ) of @xmath103 decomposes into two parts @xmath104 , where @xmath105 ( resp . @xmath106 ) is the limit of @xmath107 in @xmath108 as @xmath109 ( resp . @xmath110 . associated to this decomposition , @xmath77 decomposes into @xmath111 for which @xmath112 and @xmath113 . let @xmath99 be the image of some @xmath98 . then @xmath72 is called a _ @xmath0-group _ if exactly one of @xmath114 or @xmath115 is simply connected and @xmath72-invariant . for a @xmath0-group @xmath18 , we denote by @xmath116 its _ parabolic locus _ , a collection of homotopy classes of disjoint simple closed curves @xmath117 on @xmath1 such that @xmath118 is parabolic . similarly @xmath72 is called a _ quasifuchsian group _ if both @xmath114 and @xmath115 are simply connected and @xmath72-invariant . quasifuchsian space @xmath12 is the subset of @xmath8 of faithful representations with quasifuchsian images . it is known by marden @xcite and sullivan @xcite that @xmath14 equals the interior of @xmath89 . hence @xmath14 is a @xmath88-dimensional complex manifold in @xmath8 . on the other hand , it is trivial that @xmath119 and is conjectured that @xmath120 , which is so called the bers - sullivan - thurston conjecture . we let @xmath121 denote the relative boundary of @xmath14 in @xmath8 , whose element is called a _ boundary group_. now let @xmath122 . then there exists an orientation preserving homeomorphism @xmath123 \to \overline{m_\rho}$ ] which induces the representation @xmath124 . moreover , it provides orientation preserving homeomorphisms @xmath125 and @xmath126 , where @xmath127 denotes @xmath1 with its orientation reversed . hence @xmath122 determines a pair of marked reimann surfaces @xmath128 in the product of the teichmller spaces . on the other hand , bers @xcite showed that each pair @xmath129 has the unique simultaneous uniformization @xmath130 . therefore the map @xmath131 gives us a global parameterization of @xmath14 . we define _ vertical _ and _ horizontal bers slices _ in @xmath14 by @xmath132 and @xmath133 , respectively . it is known by bers that both @xmath134 and @xmath135 are precompact in @xmath8 , whose frontiers are denoted by @xmath136 and @xmath137 . a representation @xmath98 is called a _ bers boundary group _ if it is contained in the frontier of some bers slice . we introduce the notion of `` standard '' and `` exotic '' convergence for a sequence @xmath46 tending to a limit @xmath138 . one of our main purpose of this paper is to show that every grafting map behaves as a continuous map for every standard convergent sequence ( see theorem 5.1 ) . for a given subset @xmath35 of @xmath36 , we set @xmath139 , where @xmath140 denote the complex conjugation of @xmath141 . suppose that a sequence @xmath142 in @xmath14 converges to @xmath143 . then the sequence @xmath144 is said to converge _ standardly _ to @xmath145 if there exists a compact subset @xmath67 such that ( i ) @xmath146 for all @xmath31 , or ( ii ) @xmath147 for all @xmath31 . otherwise , we say that @xmath144 converges _ exotically _ to @xmath148 . we let @xmath149 and @xmath150 denote the subsets of @xmath121 consisting of limits of standard convergent sequences of type ( i ) and ( ii ) , respectively , and set @xmath151 . it is easily seen that every bers boundary group is an element of @xmath152 and that every element in @xmath152 is a _ boundary @xmath0-group _ , a @xmath0-group in @xmath121 . on the contrary , it follows from the arguments in the paper of brock , bromberg , evans and souto @xcite that every boundary @xmath0-group is a bers boundary group ( see theorem 2.3 below ) . hence we see that the set @xmath152 equals the set of all boundary @xmath0-groups and that the following hold : @xmath153 moreover , bromberg @xcite obtained the following result by developing the grafting operation for elements in @xmath150 ( see 3 ) and by combining with minsky s result @xcite . every @xmath0-group @xmath98 without parabolics is a bers boundary group . we now outline the proof of theorem 2.3 by following the arguments in @xcite . in the argument , we also obtain corollary 2.4 below , which is required in 5 . every boundary @xmath0-group is a bers boundary group . let @xmath154 with parabolic locus @xmath116 . if @xmath155 ( resp . @xmath156 ) , there exists a sequence @xmath157 which converges standardly to @xmath18 and which satisfies @xmath158 ( resp . @xmath159 ) as @xmath160 . here @xmath161 denotes the total sum of hyperbolic lengths of components of @xmath116 on @xmath162 . let @xmath15 with @xmath116 , which is possibly empty . we may assume that the positive part @xmath163 of the conformal boundary of @xmath164 is homeomorphic to @xmath1 , and hence determine a point @xmath141 . by theorem 3.1 in @xcite , the drilling theorem plays an important role in its proof , there exists a strong convergent sequence @xmath165 in @xmath13 for which each @xmath144 is a geometrically finite representation whose parabolic locus @xmath166 equals @xmath116 . since the convergence @xmath165 is strong , @xmath167 converges to @xmath4 in @xmath36 . moreover , for each @xmath31 , there exists a strong convergent sequence @xmath168 of quasifuchsian representations all of whose positive parts of the conformal boundaries equal @xmath169 ( see theorem 3.4 in @xcite ) . therefore @xmath170 is a sequence in the bers slice @xmath171 for each @xmath31 . since the convergence @xmath172 is strong and since @xmath144 is geometrically finite , the hyperbolic lengths @xmath173 of @xmath116 on @xmath174 tend to @xmath175 as @xmath176 . by a diagonal argument , we can choose a sequence @xmath177 from @xmath178 so that @xmath179 and that @xmath158 as @xmath160 , which satisfies the desired property in corollary 2.4 . now let us consider new sequence @xmath180 in @xmath134 . then this sequence @xmath181 also converges to @xmath18 because maximal dilatations of quasiconformal automorphism of @xmath5 conjugating @xmath182 to @xmath183 tend to @xmath184 as @xmath160 . this implies that @xmath185 . we remark that the set @xmath186 is not empty ; for instance , it contains limits of sequences which appear in thurston s double limit theorem . on the other hand , anderson and canary @xcite showed that there exists a sequence in @xmath14 which converges exotically to some point in @xmath152 . all the known such sequences are basically obtained by their technique and here is a typical example : let @xmath117 be a simple closed curve on @xmath1 and let @xmath187 be the dehn twist along @xmath117 . then for a fixed pair @xmath129 , the sequence @xmath188 in @xmath14 converges exotically to some @xmath189 as @xmath190 . we will make use of this sequence in 3.3 to explain how the grafting maps fail to be extended to a continuous map . we only give a brief summary of projective structures and refer to @xcite and elsewhere for more details . a projective structure on @xmath1 is a @xmath3-structure where @xmath4 is a riemann sphere @xmath5 and @xmath6 is the group of projective automorphism of @xmath5 . let @xmath7 be the space of marked projective structures on @xmath1 . a projective structure @xmath191 determine the underlying conformal structure @xmath192 . it is known that @xmath7 is the holomorphic affine bundle over @xmath36 with the projection @xmath60 , and that @xmath7 is a @xmath88-dimensional complex manifold . a projective structure @xmath193 on @xmath1 can be lifted to that @xmath194 on @xmath195 , where @xmath196 is the universal cover on which @xmath197 acts as a covering group . since @xmath194 is simply connected , we obtain a developing map @xmath198 by continuing charts of @xmath194 analytically , which induce a holonomy representation @xmath199 satisfying @xmath200 for every @xmath201 . we remark that the pair @xmath202 is determined uniquely up to @xmath73 . we now define the _ holonomy map _ @xmath203 by @xmath204 $ ] . hejhal @xcite showed that the map @xmath205 is a local homeomorphism and earle @xcite and hubbard @xcite independently showed that the map is holomorphic . the holonomy map @xmath10 is a holomorphic local homeomorphism . we denote by @xmath20 the set of projective structures with quasifuchsian holonomy . an element of @xmath21 is said to be _ standard _ if its developing map is injective ; otherwise it is _ exotic_. we denote by @xmath206 the subset of standard projective structures . let @xmath207 with image @xmath99 . then the quotient surface @xmath208 can be regarded as a standard projective structure on @xmath1 with bijective developing map @xmath209 , with holonomy representation @xmath210 , and with underlying conformal structure @xmath141 . let @xmath211 be the map defined by the correspondence @xmath212 as described above . then the map @xmath213 turns out to be a univalent local branch of @xmath26 onto the connected component @xmath27 of @xmath21 , which is called the _ standard component_. it is known by bers that every bers slice @xmath214 is embedded by the map @xmath213 into a bounded domain @xmath215 of a fiber @xmath216 . observe that @xmath217 is bijective , where @xmath50 is the closure of @xmath27 in @xmath7 . we let @xmath218 denote the set of integral points of measured laminations on @xmath1 . in other words , each element of @xmath23 is a isotopy class of disjoint union @xmath219 of homotopically distinct simple closed curves @xmath220 on @xmath1 with positive integer @xmath221 weights . we do not distinguish the isotopy class @xmath222 and its representative if there is no confusion . the `` zero - lamination '' @xmath175 is also contained in @xmath22 . in what follows , the parabolic locus @xmath116 of a @xmath0-group @xmath18 is also regarded as an element of @xmath22 . for each non - zero @xmath23 , we will explain haw to obtain the grafting map @xmath223 which satisfies @xmath224 on @xmath27 . we give here two equivalent definitions of grafting operation ; the first one is as usual , and the second one is introduced by bromberg in @xcite so that it also makes sense for elements of @xmath150 . for a while , we assume that @xmath222 is a simple closed curve @xmath117 of weight one for simplicity . in addition , we fix our notation as follows : for a given @xmath122 , let @xmath72 be the quasifuchsian image of @xmath18 and @xmath225 the region of discontinuity of @xmath72 . we regard quotient surfaces @xmath226 and @xmath227 as projective structures on @xmath1 and @xmath127 , respectively . let @xmath228 and @xmath229 be simple closed curves corresponding to @xmath230 and suppose that @xmath231 is a representative of the homotopy class of @xmath117 . let @xmath232 and @xmath233 be the @xmath234-invariant lifts of @xmath228 and @xmath229 , respectively . let @xmath235 be open subsets of a topological space @xmath236 such that @xmath237 and that @xmath238 . then @xmath239 denotes the interior of @xmath240 . in the situation as described above , the _ grafting _ @xmath241 of the standard projective structure @xmath242 along @xmath117 is a projective structure obtained by the following ( equivalent ) procedures : i : : let @xmath243 be a cylinder @xmath244 equipped with a projective structure induced from that of @xmath5 . we obtain @xmath245 by cutting @xmath193 along @xmath117 and inserting @xmath243 at the cut locus without twisting ; that is , @xmath246 ii : : here we further assume that @xmath117 separates @xmath1 into two surfaces @xmath247 and @xmath248 with boundaries . ( the non - separating case is described precisely in @xcite . ) accordingly , @xmath193 and @xmath249 decompose into @xmath250 and @xmath251 , respectively . let @xmath252 denotes @xmath184 or @xmath253 . let @xmath254 be the connected component of the inverse image of @xmath255 in @xmath256 which contains @xmath257 in its closure @xmath258 . then the stabilizer subgroup @xmath259 of @xmath260 is identified with @xmath261 . since @xmath262 is a purely loxodromic free group with non - empty region of discontinuity , maskit s result @xcite implies that @xmath262 is a schottky group . note that the conformal boundary @xmath263 of @xmath264 with natural projective structure containing both projective surfaces @xmath265 and @xmath266 . then @xmath267 is obtained from projective surfaces @xmath268 and @xmath269 by gluing their boundaries without twisting ( see figure 1 ) ; that is , @xmath270 of @xmath193 along @xmath117 . ] observe that the definitions i and ii are equivalent . the grafting @xmath271 of @xmath193 along general @xmath272 is similarly obtained ; for instance , @xmath273 is obtained by inserting @xmath274-copies of @xmath243 in definition i. an important fact is that the grafting operation does not change the holonomy representation , that is , @xmath275 is always satisfied . since the grafting map @xmath276 satisfies @xmath224 on @xmath27 and since @xmath277 is a biholomorphic map with its inverse @xmath213 , the map @xmath278 takes @xmath27 biholomorphically onto the image @xmath279 , which is denoted by @xmath24 . hence we obtain a univalent local branch @xmath280 of @xmath26 which is defined by @xmath281 and is also called the _ grafting map _ for @xmath222 ( see the commutative diagram below ) . by goldman s grafting theorem @xcite below , we obtain the decomposition @xmath282 of @xmath21 into its connected components . for every @xmath122 , all projective structures with holonomy @xmath18 are obtained by grafting of @xmath18 ; that is , we have @xmath283 . let @xmath49 denotes the one - point compactification @xmath284 of @xmath7 . we now extend the grafting map @xmath285 to @xmath286 . ( in fact @xmath287 is defined for all @xmath0-groups @xmath98 by the same manner . ) observe that definition i works well even for @xmath155 whenever @xmath288 is loxodromic , because there still exists a @xmath234-invariant simple arc @xmath289 in non - degenerate component @xmath290 , for which @xmath244 is still an annulus . on the other hand , definition ii works well for @xmath291 whenever every connected component of @xmath116 intersects @xmath117 essentially . in fact , on this assumption , @xmath292 and @xmath293 in definition ii are still schottky groups , @xmath288 is still loxodromic , and there still exists a @xmath234-invariant simple arc @xmath257 in non - degenerate component @xmath294 . for general @xmath23 , we can also define the grafting @xmath295 of @xmath18 along @xmath222 if the pair @xmath296 satisfies the following condition : the pair @xmath296 of @xmath23 and @xmath154 is said to be _ admissible _ if * @xmath155 and @xmath116 and @xmath222 have no parallel component in common , or * @xmath156 and every component of @xmath116 intersects @xmath222 essentially . if the pair @xmath296 is not admissible , we set @xmath297 . the extended grafting map @xmath41 is also denoted by the same symbol @xmath40 . we collect in table 1 below the equivalent conditions with standard / exotic convergence of quasifuchsian representations ( see ( * ? ? ? * appendix a ) and @xcite ) . the situation in which we consider is as follows : let @xmath46 be a sequence converging to @xmath298 . we may assume that the sequence @xmath90 converges geometrically to a kleinian group @xmath91 , which contains the algebraic limit @xmath92 . let @xmath299 be a unique projective structure in @xmath29 such that @xmath300 and let @xmath301 be a univalent local branch of @xmath26 which is defined on a neighborhood @xmath17 of @xmath148 and takes @xmath148 to @xmath299 . then @xmath302 converges to @xmath303 . in this situation , all conditions in the same line in table 1 are equivalent . .equivalent conditions with standard / exotic convergence . [ cols="^,^",options="header " , ] in the same setting as above , let @xmath304 be the sequence introduced at the end of 2.1 which converges exotically to @xmath145 . then @xmath305 are exotic projective structure for all large enough @xmath306 . since @xmath307 , the sequence @xmath308 of standard projective structure can not accumulates on a point in @xmath29 and hence diverges . this implies that the map @xmath309 is not continuous at @xmath189 . in addition , we have shown in @xcite that @xmath310 and that @xmath311 . on the other hand , we have shown in @xcite that @xmath312 for every simple closed curve @xmath313 on @xmath1 which intersects @xmath117 essentially . this implies that the grafting map @xmath314 is not continuous at @xmath315 . more general , we have the following for any @xmath23 , the grafting map @xmath41 is not continuous . let @xmath316 with @xmath317 and let @xmath4 be a subset of @xmath5 which is @xmath72-invariant . then we naturally obtain a subset @xmath318 of @xmath193 since the inverse image @xmath319 of @xmath4 is also invariant under the action of the covering group @xmath197 of the covering map @xmath320 . we call @xmath321 the _ pull - back _ of @xmath4 in @xmath193 . for @xmath322 , a _ realization _ @xmath323 of @xmath222 is a disjoint union of simple closed curves on @xmath1 which realize each weighted simple closed curve @xmath324 by @xmath221 parallel disjoint simple closed curves which are homotopic to @xmath220 . let @xmath325 denotes a regular neighborhood of @xmath323 in @xmath1 . let @xmath122 with @xmath99 , let @xmath326 be the grafting of @xmath18 along @xmath23 , and let @xmath327 denotes either @xmath328 or @xmath329 . then we denote by @xmath330 the pull - back @xmath331 of an invariant component @xmath332 of @xmath77 . from the definition of grafting , we have the following lemma ( see also the proof of theorem 4.2 ) . let @xmath333 . then there is a automorphism of @xmath193 which is homotopic to the identity and which takes @xmath334 onto @xmath335 and @xmath325 onto @xmath336 . therefore the topological type of the pull - backs @xmath335 and @xmath336 are constant for every elements @xmath333 . the following lemma states that the similar result holds for a grafting @xmath326 of @xmath154 and for the pull - back of the unique invariant component of @xmath77 ( see figure 2 ) . in fact , we use this fact in 5 to show that @xmath326 for @xmath154 is actually a limit of a sequence in @xmath24 . similar arguments can be found in @xcite and @xcite . let @xmath296 be an admissible pair of @xmath23 and @xmath154 , and set @xmath326 . * if @xmath155 , there is a one - to - one correspondence between the set of connected components of @xmath337 and that of @xmath338 . moreover , for each connected component of @xmath339 , there is a homeomorphism from @xmath340 onto the corresponding component of @xmath334 which is isotopic to the identity . * if @xmath156 , there is a one - to - one correspondence between the set of connected components of @xmath341 and that of @xmath342 . moreover , for each connected component of @xmath343 , there is a homeomorphism from @xmath340 onto the corresponding component of @xmath325 which is isotopic to the identity . we discuss in the same setting of definition 3.1 and 3.2 , except for one change ; here @xmath193 denotes the grafting @xmath287 but not a standard projective structure . we first suppose that @xmath344 . let us consider the quotient torus @xmath345 , in which the arc @xmath346 descends to an essential curve @xmath347 in @xmath348 . then the grafting annulus @xmath349 is identified with the subset @xmath350 of @xmath348 . on the other hand , since the unique invariant component @xmath290 of @xmath77 is simply connected , @xmath351 is also an annulus with core curve @xmath347 . hence the intersection @xmath352 equals @xmath353 in @xmath348 , which is a disjoint union of two essential annuli @xmath354 and @xmath355 . we now regard that @xmath356 . then the pull - back @xmath337 of @xmath114 coincides with the union @xmath357 , which is obtained by gluing those pieces along copies of @xmath347 . thus the result follows in this case . next we suppose that @xmath358 . by the same argument as above , we see that the quotient @xmath359 of the unique invariant component @xmath115 of @xmath77 is an annulus with core curve @xmath360 , and that the disjoint union @xmath361 descends to that of essential annuli @xmath362 in @xmath359 , where @xmath363 for @xmath364 . we may assume that @xmath365 for @xmath364 . recall that @xmath366 is a schottky group . since @xmath367 , the image of @xmath368 via the covering map @xmath369 is @xmath370 , which is obtained from @xmath371 and @xmath372 by gluing along @xmath373 . similarly , the image of @xmath368 via the covering map @xmath374 is @xmath375 . since @xmath193 is obtained from @xmath268 and @xmath269 by gluing along @xmath373 , we see that the pull - back @xmath376 of @xmath115 equals the annulus @xmath377 with core curve @xmath373 . thus we have obtained the result also in this case . for @xmath155 and @xmath376 for @xmath156 . ] let @xmath378 , @xmath379 and @xmath154 . if both pairs @xmath296 and @xmath380 are admissible then @xmath381 . the aim of this section is to prove theorem 4.3 , which guarantees some finiteness : suppose that a sequence @xmath46 converges standardly to @xmath382 and that @xmath301 is a univalent local branch of @xmath26 defined on a neighborhood @xmath17 of @xmath18 . then the sequence @xmath383 is actually contained in some finite union of components of @xmath21 . this finiteness theorem plays an important role in the proof of our main theorem ( theorem 5.1 ) , which guarantees some continuity of the grafting maps and which can be seen also as the uniqueness theorem . recall that the _ modulus _ @xmath384 of a conformal annulus @xmath385 is defined uniquely as the ratio of the height and the circumference of an euclidian annulus which is conformally equivalent to @xmath385 . the following inequality is a direct consequence of the geometric and analytic definitions of the extremal length . let @xmath386 be a complete hyperbolic surface of finite area @xmath387 and let @xmath385 be an essential annular domain in @xmath386 . then we have @xmath388 where @xmath389 is the hyperbolic length of the homotopy class @xmath117 of a core curve of @xmath385 . let @xmath390 denotes the extremal length of the homotopy class @xmath117 in @xmath386 . from the analytical definition of @xmath390 , we have @xmath391 where the supremum is taken over all metrics @xmath18 consistent with the conformal structure of @xmath386 and the infimum is taken over all closed curves @xmath313 in the homotopy class @xmath117 . on the other hand , from the geometrical definition of @xmath390 , we have @xmath392 where the supremum is taken over all annuli @xmath393 whose core curve is in the homotopy class @xmath117 . from the above two inequality , the desired inequality follows . we introduce the notion of a quasiconformal deformation of a projective structure with quasifuchsian holonomy , which was developed by shiga and tanigawa in @xcite . let @xmath394 and let @xmath395 be its holonomy . suppose that @xmath396 is a quasiconformal deformation of @xmath397 induced by a quasiconformal automorphism @xmath398 , whose beltrami differential is denoted by @xmath399 . then we obtain a new projective structures @xmath400 with holonomy @xmath401 which is characterized as follows : 1 . there is a quasiconformal map @xmath402 whose beltrami differential is equal to the pull - back @xmath403 of @xmath399 via the developing map @xmath404 . moreover , the map @xmath402 descends to a quasiconformal map @xmath405 , which is consistent with their markings . the developing map of @xmath400 is defined by @xmath406 . here we say that a map between projective structures is _ quasiconformal _ if it is a quasiconformal map between their underling conformal structures . we call @xmath400 as the _ quasiconformal deformation _ of @xmath193 . we remark that every grafting map @xmath285 is obtained by quasiconformal deformations of some fixed @xmath333 and its holonomy @xmath395 . fix @xmath141 arbitrarily . suppose that @xmath23 contains a weighted simple closed curve @xmath407 of weight @xmath408 . let @xmath409 and let @xmath410 denotes the underlying conformal structure of @xmath287 . then we have @xmath411 , where @xmath412 denotes the genus of @xmath1 . we make use of the inequality above mostly in the following form : @xmath413 which ensures that if @xmath414 is large enough then the teichmller distance @xmath415 between @xmath4 and @xmath416 is also large . we first treat the case where @xmath222 is a simple closed curve @xmath117 of weight one . let @xmath417 be the fuchsian representation with image @xmath418 . we normalize @xmath419 so that @xmath420 equals the upper half plane @xmath421 , @xmath422 equals the lower half plane @xmath423 , and that the hyperbolic element @xmath424 fixes the positive imaginary axis @xmath425 . we let @xmath426 denote the standard projective structure @xmath427 . recall that the grafting @xmath428 of @xmath429 , which is denoted by @xmath430 , is obtained from @xmath426 by cutting along @xmath117 and inserting the annulus @xmath431 . we set @xmath432 . then @xmath433 and then @xmath434 , where @xmath435 , @xmath436 and @xmath437 . here @xmath438 and @xmath439 . from now on , we regard @xmath434 as a subset of @xmath430 . by definition of grafting , it hold that @xmath440 and that @xmath441 on @xmath430 , where @xmath442 ( resp . @xmath443 ) is the pull - back of the upper half plane @xmath444 ( resp . the lower half plane @xmath445 ) . let @xmath409 be a quasiconformal deformation of @xmath446 which is induced by a quasiconformal map @xmath398 . if @xmath18 is contained in @xmath134 ( resp . @xmath447 ) we may assume that the support of the beltrami differential of @xmath448 is contained in the lower half plane @xmath445 ( resp . the upper half plane @xmath444 ) . associated to the quasiconformal deformation @xmath18 of @xmath419 , we have a quasiconformal deformation @xmath449 of @xmath450 and a quasiconformal map @xmath451 . it is important to notice that the map @xmath452 takes @xmath453 ( resp . @xmath454 ) conformally into @xmath455 if @xmath456 ( resp . @xmath457 ) . here we claim that the annulus @xmath458 always satisfies @xmath459 . in fact , if @xmath456 then the the map @xmath452 takes @xmath460 conformally into @xmath458 and thus @xmath461 . similarly , if @xmath457 then @xmath452 takes @xmath462 conformally into @xmath458 and thus @xmath463 . recall that @xmath416 denote the underlying conformal structure of @xmath464 . by lemma 4.1 , we have @xmath465 . by combining this inequality with the inequality @xmath459 , the result follows in the case where @xmath466 . next suppose that @xmath23 contains a weighted simple closed curve @xmath407 with @xmath408 . then @xmath467 contains an annulus @xmath468 which is a union of succeeding @xmath274 parallel annuli each of whose modulus @xmath469 . thus @xmath470 and hence we have obtained the desired inequality in general . from theorem 4.2 , we see that the both @xmath471 and @xmath472 are proper subsets of @xmath36 for every non - zero @xmath23 . in fact if @xmath473 , they are contained in the proper subset @xmath474 of @xmath36 . compare this fact with a result which was independently obtained by gallo @xcite and tanigawa @xcite ; they showed that the map @xmath475 takes the fuchsian space @xmath476 bijectively onto @xmath36 for every @xmath23 . in addition , we remark that both @xmath471 and @xmath472 are non - precompact in @xmath36 for every non - zero @xmath23 . in fact if @xmath477 , we can make the length @xmath478 arbitrarily small by varying @xmath479 in @xmath134 so that @xmath480 or @xmath481 in @xmath447 so that @xmath482 . as a consequence of theorem 4.2 , we obtain the following suppose that a sequence @xmath483 of quasifuchsian projective structures converges to @xmath484 and that the sequence @xmath485 of their holonomies converges standardly to @xmath486 . then there is a finite union @xmath487 of components of @xmath21 in which @xmath305 are contained for all @xmath31 . instead proving this theorem directly , we prove the following theorem , which is a contraposition of theorem 4.3 . suppose that a sequence @xmath46 converges standardly to @xmath488 and @xmath61 is a sequence of distinct element of @xmath22 . then the sequence @xmath489 converges to @xmath490 in @xmath49 as @xmath160 , that is , the set @xmath491 has no accumulation point in @xmath7 . set @xmath492 . we first assume that @xmath493 . then @xmath494 for some @xmath141 and @xmath495 in @xmath36 . one can choose a sequence of weighted simple closed curves @xmath496 which satisfies @xmath497 as @xmath160 . let @xmath498 denote the underlying conformal structure of @xmath489 . then it follows from theorem 4.2 that @xmath499 and that the teichmller distance @xmath500 between @xmath37 and @xmath501 tends to @xmath490 as @xmath160 . since @xmath495 in @xmath36 , @xmath502 also tends to @xmath490 , which implies that @xmath503 in @xmath49 . the proof for @xmath358 is completely parallel . the following is our main theorem in this paper . let @xmath46 be a sequence which converges standardly to @xmath504 . then the sequence @xmath47 converges to @xmath48 in @xmath49 for every @xmath23 . we devote this section to the proof of the above theorem , which is divided into two cases ; the case where @xmath505 and that where @xmath506 . throughout this proof , we let @xmath327 denote either @xmath328 or @xmath329 for which @xmath507 . let @xmath508 be a univalent local branch of @xmath26 from a neighborhood @xmath17 of @xmath145 to that @xmath509 of @xmath510 and set @xmath511 and @xmath512 . then the sequence @xmath305 converges to @xmath299 in @xmath7 . we claim in lemma 5.2 below that @xmath513 for all large enough @xmath31 . assuming this lemma , we have @xmath514 and thus obtain the desired convergence @xmath515 . @xmath513 for all large enough @xmath31 . the rest of this subsection is devoted to the proof of lemma 5.2 . the idea of the proof is as follows : by lemma 3.6 , we have already known the topological type of the set @xmath516 in relation to @xmath222 . as we will see in lemma 5.5 , the set @xmath517 coincides with the essential part of the carathodory limit of the sequence @xmath518 as @xmath160 . it is essential that the sequence @xmath305 is contained in a finite union of components of @xmath21 by theorem 4.3 . then we will show that for any subsequence @xmath519 of @xmath520 contained in a component of @xmath21 , the topological types of @xmath521 is the same to that of @xmath517 for all @xmath522 , which implies that @xmath513 for all @xmath31 . now we fill in the details . we may assume that the sequence @xmath90 of quasifuchsian groups converges geometrically to some kleinian group @xmath91 , which contains the algebraic limit @xmath92 . then kerckhoff and thurston ( * ? ? ? * corollary 2.2 ) showed that the sequence @xmath523 of the regions of discontinuity of @xmath524 converges to the region of discontinuity @xmath525 of @xmath91 in the sense of carathodory . here @xmath526 in this sense if ( i ) for every compact @xmath527 , we have @xmath528 for all large enough @xmath31 and if ( ii ) for any open set @xmath529 , we have @xmath530 . moreover @xmath525 decomposes into two @xmath91-invariant regions @xmath531 , @xmath532 , each of which are the carathodory limits of @xmath533 , @xmath534 , respectively . since the convergence @xmath535 is standard , the argument in the proof of ( * ? ? ? * proposition 2.3 ) reveals that @xmath536 is a component of @xmath537 , that @xmath538 , and that @xmath539 fixes @xmath536 if and only if @xmath540 . we put them together in the following we have @xmath541 . recall that @xmath542 denotes the pull - back of @xmath543 in @xmath305 . similarly @xmath544 is the pull - back of @xmath536 in @xmath299 . in addition , we let @xmath545 denotes the pull - back of @xmath546 in @xmath299 . by the same argument of ( * ? ? ? * lemma 3.3 ) , we obtain the following the sequence @xmath518 converges to @xmath547 in the sense of carathodory . more precisely , there exist @xmath548-bi - lipshiz diffeomorphisms @xmath549 between hyperbolic surfaces ( consistent with their markings ) such that @xmath550 as @xmath160 and that the sequence @xmath551 converges to @xmath552 in the sense of carathodory . we claim in the following lemma that @xmath517 carries the full fundamental group of @xmath552 . each connected component of @xmath553 is homotopically trivial in @xmath299 . suppose for contradiction that there exists a connected component @xmath340 of @xmath553 which contains an essential closed curve @xmath554 . let @xmath555 be a lift ( i.e. , a connected component of the inverse image ) of @xmath556 and @xmath557 a lift of the curve @xmath558 which is contained in @xmath559 . let @xmath560 be the connected component of @xmath561 which contains the developing image @xmath562 . let @xmath563 be an element of @xmath564 which fixes @xmath565 , and hence @xmath560 . now we see that @xmath288 is loxodromic from definition 3.3 and lemma 3.5 . in fact if @xmath189 , then each connected component of @xmath566 is an annulus whose core curve is homotopic to a component of @xmath222 . hence the curve @xmath554 in @xmath567 is homotopic into @xmath222 and hence is not homotopic into @xmath568 . on the other hand , if @xmath358 then each component of @xmath568 must traverse @xmath222 , and thus @xmath569 , in @xmath299 . therefore the curve @xmath554 in @xmath570 is not homotopic into @xmath568 . from lemma 5.3 , there exists @xmath571 such that @xmath572 . since @xmath288 fixes the component @xmath560 , @xmath573 fixes the component @xmath536 , which implies that @xmath574 . but this contradicts to the following lemma ( lemma 2.4 in @xcite ) . let @xmath575 be a finitely generated group . let @xmath576 be a sequence of discrete faithful representations converging algebraically to @xmath577 . moreover , assume that @xmath524 converges to @xmath91 geometrically . then for any @xmath571 , @xmath578 is @xmath579 or rank - one parabolic subgroup . let us combine the above observations to complete the proof of lemma 5.2 , which states that @xmath580 for all large enough @xmath31 . note that the sequence @xmath305 is contained in a finite union of components of @xmath21 by theorem 4.3 . therefore to show lemma 5.2 , we only need to show that @xmath581 by assuming that the sequence @xmath305 is contained in a component @xmath582 of @xmath21 . let @xmath583 be a fixed element and set @xmath584 . then every @xmath305 is a quasiconformal deformation of @xmath429 in @xmath582 , which is induced by the quasiconformal deformation @xmath144 of @xmath585 , and which induces the quasiconformal map @xmath586 . note that the map @xmath587 is consistent with their markings and takes @xmath588 onto @xmath589 . since the convergence @xmath535 is standard , the dilatation of @xmath590 tends to @xmath184 , and hence we assume that @xmath591 is conformal for every @xmath31 for simplicity . now let @xmath592 , @xmath593 denote the set of connected components of @xmath588 , @xmath594 , respectively . then lemma 3.6 tells us that @xmath595 if and only if there exists a bijection @xmath596 which satisfies @xmath597 for every @xmath598 . here we identify @xmath599 and @xmath600 with @xmath197 via their markings . note that @xmath601 converge to @xmath552 is the sense of lemma 5.4 and that @xmath602 hold for all @xmath598 and @xmath31 . we first define a map @xmath603 , which will turn out to be the inverse of @xmath412 . let @xmath604 and let @xmath605 be a compact subset such that @xmath606 . then there exists @xmath598 such that @xmath607 for all large enough @xmath31 , and hence that @xmath608 . now we let @xmath609 . next we define the map @xmath610 as follows : let @xmath598 . by passing to a subsequence if necessary , the sequence @xmath611 converges to an open subset @xmath612 of @xmath613 . we claim that @xmath612 contains exactly one essential component @xmath614 of @xmath552 , which satisfies @xmath615 . we postpone the proof of this claim and continue the argument . we have @xmath604 by lemma 5.5 and let @xmath616 . then it follows immediately that @xmath412 is a bijective map with @xmath617 and hence that @xmath595 . we have completed the proof of lemma 5.2 , and hence that of theorem 5.1 in the case where @xmath618 . we only need to show the above claim . suppose for contradiction that @xmath619 is trivial or contains more that two essential connected components . since @xmath620 converge to @xmath612 , @xmath621 then become more and more constricted , that is , there exists an arc @xmath622 joining two distinct components of @xmath623 whose hyperbolic length tends to @xmath175 as @xmath160 . but since the map @xmath624 is conformal , this contradicts to lemma 5.7 below . thus we have proved the claim . let @xmath386 be a hyperbolic surface , @xmath625 a simple closed curve and @xmath626 is an annulus whose core curve is homotopic to @xmath117 . then there is a positive constant @xmath627 , depending only on @xmath389 and @xmath384 , such that @xmath628 holds for every arcs @xmath629 joining points in distinct components of @xmath630 . we first prepare the following let @xmath23 and let @xmath632 be a sequence which converges standardly to @xmath44 and which satisfies the property as in corollary 2.4 . suppose that @xmath631 . then the sequence @xmath633 in @xmath49 . we first suppose that @xmath189 . since @xmath634 is not admissible , @xmath222 and @xmath568 have parallel components in common and let @xmath117 be one of such components . since @xmath635 , there are annular neighborhoods @xmath636 of @xmath117 in @xmath162 such that @xmath637 as @xmath160 . we set @xmath514 . then observe that the annulus @xmath636 is conformally embedded in @xmath638 for every @xmath31 . hence the underlying conformal structures @xmath639 of @xmath47 diverge in @xmath36 , and hence @xmath633 in @xmath49 . next we suppose that @xmath358 . then there is a component @xmath117 of @xmath568 which do not intersect @xmath222 essentially . since @xmath640 , there are annular neighborhoods @xmath636 of @xmath117 in @xmath37 such that @xmath637 as @xmath160 . then observe that the annulus @xmath636 is conformally embedded in @xmath641 for every @xmath31 . by the same argument as above , we see that @xmath633 in @xmath49 . now we back to the proof of theorem 5.1 in the case where @xmath642 ; we intend to show that @xmath643 for every standard convergent sequence @xmath644 . set @xmath645 . to obtain a contradiction , we suppose that the sequence @xmath305 has a subsequence ( which is denoted by the same symbols ) converging to some @xmath484 . in addition , let @xmath646 be a sequence converging to @xmath145 as in lemma 5.8 and let @xmath647 be a univalent local branch of @xmath26 . then we have @xmath302 for all large enough @xmath31 . note that both @xmath302 and @xmath648 converge to @xmath299 . from theorem 4.3 , the sequence @xmath649 has a subsequence ( also denoted by the same symbols ) which is contained in a component @xmath650 of @xmath21 for some @xmath651 . then we have @xmath652 , and thus @xmath653 , which implies that @xmath654 by lemma 5.8 and that @xmath379 . since both @xmath655 converge standardly to @xmath145 , theorem 5.1 for admissible pairs guarantees that both @xmath656 convege to @xmath657 . on the other hand , since @xmath658 , we have @xmath659 . now we obtain two sequences @xmath660 both of which converge to @xmath299 . but since @xmath379 , this contradicts to the fact that the map @xmath205 is a local homeomorphism . now we have completed the proof of theorem 5.1 . we collect in this section some results which are obtained as consequences of theorem 5.1 . let @xmath154 and @xmath663 . we only need to show that @xmath664 for some @xmath23 . let @xmath46 be a sequence converging standardly to @xmath18 . since the map @xmath205 is a local homeomorphism , there exists a convergent sequence @xmath665 of projective structures with @xmath666 for all @xmath31 . from theorem 4.3 , there exists a subsequence @xmath519 of the sequence @xmath520 which is contained in a component @xmath667 of @xmath21 . then we have @xmath668 for all @xmath522 and thus @xmath669 as @xmath670 . on the other hand , since the convergence @xmath165 is standard , theorem 5.1 implies that @xmath671 in @xmath49 . hence @xmath664 and the result follows . suppose that @xmath673 for @xmath23 and @xmath154 . then there exists a path @xmath674 \to r(s)$ ] with @xmath675 and @xmath676 along which the grafting map @xmath45 is continued analytically to a univalent local branch @xmath677 of @xmath26 which is defined on some neighborhood @xmath17 of @xmath18 . we may assume that @xmath155 , and hence that @xmath678 for some @xmath141 . let @xmath679 de a bounded domain with @xmath680 and set @xmath681 . we claim that there exists a neighborhood @xmath17 of @xmath18 such that @xmath682 holds on @xmath683 . for otherwise there is a sequence @xmath684 converging to @xmath18 for which @xmath685 hold for all @xmath31 . on the other hand , theorem 5.1 implies that both sequences @xmath686 converge to @xmath687 , which contradicts to the fact that @xmath688 is univalent . now let @xmath674 \to r(s)$ ] be an arc such that @xmath675 , @xmath689 and that @xmath690 ) \subset b_d \cup u$ ] . then @xmath45 is continued analytically to @xmath301 along @xmath554 . let @xmath674 \to r(s)$ ] be a path as in theorem 6.3 . then the above theorem implies that there exists the lift @xmath691 \to p(s)$ ] of @xmath554 via the holonomy map with the starting point @xmath692 . on the other hand , we have the following we may assume that @xmath155 , and hence that @xmath678 for some @xmath141 . let take a path @xmath674 \to r(s)$ ] with @xmath675 and @xmath689 for which there exists a sequence @xmath694 such that @xmath695 for some bounded domain @xmath679 with @xmath680 . then the sequence @xmath696 converges standardly to @xmath18 , and thus @xmath697 by theorem 5.1 . hence there is no lift of @xmath554 to @xmath7 with the starting point @xmath692 . let @xmath698 be topological spaces . a continuous map @xmath699 is said to be a _ covering map _ or a _ weak - covering map _ if for any @xmath700 there is a neighborhood @xmath17 of @xmath701 such that @xmath702 is a homeomorphism we explain this fact in our context . let @xmath154 with @xmath707 . then there are @xmath708 such that @xmath673 and @xmath709 . one can find a path @xmath674 \to r(s)$ ] with @xmath675 and @xmath689 which satisfies both conditions in the proofs of theorems 6.3 and 6.4 . then there is the lift of the path @xmath554 with the starting point @xmath692 but there is no lift with the starting point @xmath710 , which implies that the map @xmath205 is not a covering map . we remark that the idea used in the argument above is the same to that of hejhal @xcite , but he made use of a path @xmath674 \to r(s)$ ] of shottky representations . let @xmath154 . for every @xmath23 with @xmath673 , suppose that @xmath711 is a univalent local branch of @xmath26 defined on a neighborhood @xmath712 of @xmath18 . then @xmath18 can not be an interior point of @xmath713 , where the intersection is taken over all @xmath23 with @xmath673 . we may assume that @xmath678 for some @xmath141 . then there exists a sequence @xmath144 of maximal cusps converging to @xmath18 in @xmath136 for which the parabolic locus @xmath714 has no parallel component in common with @xmath116 ( which is possibly empty ) for every @xmath31 ( see ( * ? ? ? * theorem 5.5 ) ) . since @xmath715 , there exists a univalent local branch @xmath716 for every @xmath31 . on the other hand , since @xmath717 , @xmath718 for every @xmath31 . therefore we have obtained the result . we extend our definition of standard convergence to sequences in @xmath13 . let @xmath35 be a compact subset of @xmath36 and @xmath722 its complex conjugation . we set @xmath723 and @xmath724 . recall that a sequence @xmath46 converges standardly to @xmath138 if there is a compact subset @xmath67 such that @xmath725 for all @xmath31 . similarly we say that a sequence @xmath726 converges _ standardly _ to @xmath138 if there is a compact subset @xmath67 such that all @xmath144 are contained in the _ closure _ of @xmath727 in @xmath8 . note that the limit of a standard convergent sequence in @xmath13 is contained in @xmath152 and that if a sequence @xmath728 converges standardly to @xmath729 then @xmath730 for all large enough @xmath31 , where @xmath327 denotes @xmath328 or @xmath329 . we only have to consider the case where @xmath728 for all @xmath31 . then @xmath34 in @xmath152 . if @xmath618 then the result follows from theorem 6.3 . hence we suppose that @xmath642 . ( we remark that the argument below also works in the case where @xmath618 . ) to obtain a contradiction , we assume that the sequence @xmath731 converges to some @xmath732 . then for each @xmath31 there exists a sequence @xmath733 converging standardly to @xmath734 as @xmath176 , which is taken by @xmath40 to a convergent sequence @xmath735 from theorem 5.1 . then by a diagonal argument , we can choose a sequence @xmath182 in @xmath14 which converges standardly to @xmath145 and which satisfies @xmath736 . on the other hand , @xmath737 by theorem 5.1 , which is a contradiction . now we have obtained the result . let @xmath50 denote the closure of @xmath27 in @xmath7 , _ not _ in @xmath49 . from the observation in 3.3 , we see that a sequence @xmath738 converges to @xmath52 in @xmath7 if and only if the sequence @xmath739 converges standardly to @xmath740 . hence , as a direct consequence of the above theorem , we obtain the following let @xmath35 be a given compact subset of @xmath36 and set @xmath742 . then the inverse image @xmath743 of @xmath69 in @xmath7 is discrete ; that is , every connected component of @xmath68 has open neighborhood which is disjoint from any other one . suppose for contradiction that there exist @xmath744 and @xmath23 such that @xmath745 is a limit of a sequence @xmath305 in @xmath746 . by theorem 4.3 , we may assume that the sequence @xmath305 is contained in a component @xmath747 of @xmath21 with @xmath379 . since the sequence @xmath748 converges standardly to @xmath18 , we have @xmath749 by theorem 5.1 . but this contradicts the fact that @xmath381 ( corollary 3.7 ) .

there are now about fifty known radio pulsars in binary systems , including at least five in double neutron star binaries . in some cases , the stellar masses can be directly determined from measurements of relativistic orbital effects . in others , only an indirect or statistical estimate of the masses is possible . we review the general problem of mass measurement in radio pulsar binaries , and critically discuss all current estimates of the masses of radio pulsars and their companions . we find that significant constraints exist on the masses of twenty - one radio pulsars , and on five neutron star companions of radio pulsars . all the measurements are consistent with a remarkably narrow underlying gaussian mass distribution , . there is no evidence that extensive mass accretion ( ) has occurred in these systems . we also show that the observed inclinations of millisecond pulsar binaries are consistent with a random distribution , and thus find no evidence for either alignment or counteralignment of millisecond pulsar magnetic fields . # 1to 0pt#1

astro-ph9803260

neutron stars have been the subject of considerable theoretical investigation since long before they were discovered as astronomical sources of radio and x - ray emission ( @xcite ) . their properties are determined by the interplay of all four known fundamental forces electromagnetism , gravitation , and the strong and weak nuclear forces but neutron stars remain sufficiently simple in their internal structure that realistic stellar modeling can be done . measurements of their masses and radii ( as well as detailed study of their cooling histories and rotational instabilities ) provide a unique window on the behavior of matter at densities well above that found in atomic nuclei ( @xmath2 ) . observations of neutron stars also provide our only current probe of general relativity ( gr ) in the `` strong - field '' regime , where gravitational self - energy contributes significantly to the stellar mass . the most precisely measured physical parameter of any pulsar is its spin frequency . the frequencies of the fastest observed pulsars ( psr b1937 + 21 at 641.9 hz and b1957 + 20 at 622.1 hz ) have already been used to set constraints on the nuclear equation of state at high densities ( e.g. , @xcite ) under the assumption that these pulsars are near their maximum ( breakup ) spin frequency . however , the fastest observed spin frequencies may be limited by complex accretion physics rather than fundamental nuclear and gravitational physics . a quantity more directly useful for comparison with physical theories is the neutron star mass . the basis of most neutron star mass estimates is the analysis of binary motion . soon after the discovery of the first binary radio pulsar ( @xcite ) , it became clear that the measurement of relativistic orbital effects allowed extremely precise mass estimates . indeed , the measurement uncertainties in several cases now exceed in precision our knowledge of newton s constant @xmath3 , requiring masses to be quoted in solar units @xmath4 rather than kilograms if full accuracy is to be retained . after several recent pulsar surveys , there are now about fifty known binary radio pulsar systems , of which five or six are thought to contain two neutron stars . it is thus possible for the first time to consider compiling a statistically significant sample of neutron star masses . it is our purpose here to provide a general , critical review of all current estimates of stellar masses in radio pulsar binaries . the resulting catalog , with a careful , uniform approach to measurement and systematic uncertainties , should be of value both to those who wish to apply mass measurements to studies of nuclear physics , gr , and stellar evolution , and as a guide to the critical observations for observational pulsar astronomers . we begin with a discussion of known methods for pulsar mass determination ( [ sec : methods ] ) , including a new statistical technique for estimating the masses of millisecond pulsars in non - relativistic systems . in [ sec : estimates ] we review all known mass estimates , including new data and analysis where possible . statistical analysis of the available pulsar mass measurements is presented in [ sec : stat ] . we summarize in [ sec : summ ] . a second paper will consider mass estimates for neutron stars in x - ray binary systems ( ( * ? ? ? * paper ii ) ) . a detailed discussion of the implications of the combined results of this work and paper ii for studies of supernovae and neutron star formation , mass transfer in binary evolution , the nuclear equation of state , and gr will occur elsewhere ( paper iii ) . it is a familiar circumstance that estimates of astronomical masses are available only for bodies in gravitationally bound binary systems or clusters . compact stars introduce the additional possibility of directly measuring the surface gravitational potential , and hence @xmath5 , through the study of redshifted spectral features . although this technique has been used with considerable success in the case of white dwarfs , and attempts have been made to fit redshifted x - ray spectra from neutron stars ( paper ii ) , no lines have been identified in radio pulsar spectra and other gravitational effects on the observed emission from pulsars are sufficiently complex and theory dependent that no useful limits on the neutron star properties have yet been possible . in the following , we thus limit ourselves only to the determination of stellar masses in binary systems . in any binary pulsar system , five keplerian parameters can be very precisely measured by pulse timing techniques ( @xcite ) : the binary period @xmath6 , the projection of the pulsar s semimajor axis on the line of sight @xmath7 ( where the binary inclination @xmath8 is the angle between the line of sight and the orbital angular momentum vector , defined to lie in the first quadrant ) , the eccentricity @xmath9 , and the time and longitude of periastron , @xmath10 and @xmath11 . it is frequently more convenient to use the orbital angular frequency in place of the orbital period : @xmath12 . these observational parameters are related to the pulsar and companion masses , @xmath13 and @xmath14 , through the mass function @xmath15 where @xmath16 , the masses are measured in solar units , and we introduce the constant @xmath17s . relativistic corrections to the binary equations of motion are most often parameterized in terms of one or more post - keplerian ( pk ) parameters ( @xcite ) . in gr , the most significant pk parameters have familiar interpretations as the advance of periastron of the orbit @xmath18 , the combined effect of variations in the transverse doppler shift and gravitational redshift around an elliptical orbit @xmath19 , the orbital decay due to emission of quadrupole gravitational radiation @xmath20 , and the `` range '' and `` shape '' parameters @xmath21 and @xmath22 that characterize the shapiro time delay of the pulsar signal as it propagates through the gravitational field of its companion . in terms of measured quantities and the pulsar and companion masses ( in solar units ) , these pk parameters are given by ( @xcite ) : @xmath23 note , by combining equations ( [ eqn : massfunc ] ) and ( [ eqn : pks ] ) , that @xmath24 for gr . ) the measurement of the mass function @xmath25 ( equation [ eqn : massfunc ] ) together with any two pk parameters ( equations [ eqn : pkomdot][eqn : pks ] ) is sufficient in the context of gr to uniquely determine the component masses @xmath13 and @xmath14 . with additional assumptions , such as a uniform prior likelihood for orbital orientations with respect to the observer ( [ sec : uniform ] ) , strong statements about the posterior distribution of the masses is often possible if even a single pk parameter is measured . as an alternative to the pk formalism , which is designed for testing gravitation theory , it is sometimes advantageous to fit the timing data to a model that _ assumes _ the correctness of gr . for example , the ddgr model of damour , deruelle , and taylor ( @xcite ) describes the pulsar phase as a function of the five keplerian parameters , the companion mass @xmath14 , and the total mass @xmath26 . a least - squares fit of the timing data to the ddgr model thus gives direct estimates of the uncertainties in @xmath14 and @xmath26 , as well as the covariances between the mass estimates and estimates of other parameters . it is of interest to note that mass measurements from pulsar timing observations depend on the unknown relative motion of the solar system and the binary barycenters . damour and deruelle ( 1986 ) have shown that neglecting this velocity is equivalent to changing units of mass and time . in particular , the rest frame mass @xmath27 and the barycenter frame mass @xmath28 are related by @xmath29 , with doppler factor @xmath30 where @xmath31 is the line of sight unit vector and @xmath32 is the barycentric velocity of the pulsar . although the transverse velocity of the binary system can be estimated from proper motion measurements , the radial component @xmath33 is unknown . for a typical velocity of 100 km s@xmath34 , the systematic mass error is about @xmath35 ; small , but in some cases much larger than other uncertainties . while timing measurements of relativistic corrections to the keplerian orbital equations provide the most accurate and theory - independent estimates of neutron star masses , they are possible only for close , eccentric binary orbits or when the orbit is observed nearly edge - on . in the great majority of observed binaries , the mass function provides the only timing information about the component masses , and the pulsar mass can only be determined if additional constraints are found on the companion mass and binary inclination through other techniques . this section describes several ways to use observations or theoretical considerations to limit @xmath14 ; alternate limits on @xmath36 are the topic of [ sec : sini ] . in recent years , over a dozen companion stars in radio pulsar binaries have been optically detected ( e.g. , @xcite ) . in most cases , the companions are white dwarfs ; the two main sequence exceptions are discussed in the next section . many of the white dwarf companions are extremely faint ( @xmath37 ) , allowing detection only with the _ hubble space telescope _ and the largest ground - based telescopes . there are a number of ways to determine white dwarf masses ( e.g. , @xcite and references therein ) . given a theoretical relation between the white dwarf mass and radius , the measurement of any combination of the mass and radius is sufficient to determine the mass . for example , the radius can be estimated directly from estimates of the optical flux , effective temperature , and distance . alternately , the surface gravity , @xmath38 , can be found by fitting a model atmosphere to the observed spectrum ( @xcite ) , and combining the result with a temperature or luminosity estimate . in practice , difficulties arise from several sources . first , the white dwarf companions of millisecond pulsars , with typical masses @xmath39 , are usually believed to be helium white dwarfs which were insufficiently massive to burn to carbon ; the luminosity and temperature evolution of such stars has received much less study than for more massive white dwarfs ( e.g. , @xcite ) , leading to uncertainties in the finite temperature contributions to the mass - radius relationship . hansen and phinney ( 1998a , b ) have recently calculated cooling curves for helium core dwarfs , using their own calculations of h and he opacities at temperatures below 6000 k , and applied their results to a number of radio pulsar companions . we discuss their results in more detail in [ sec : estimates ] . the measurement of surface gravity in cool stars is also potentially problematic . a higher helium abundance , as may occur through enhanced convective mixing in cool stars , will produce higher pressure and broader lines and hence mimic a higher white dwarf mass ( @xcite ) ; indeed , there is some evidence that surface gravity measurements overestimate masses below about 12,000 k ( @xcite ) . two pulsars have been found in binaries with main sequence companions : psr b1259@xmath4063 , with a be companion ( @xcite ) , and psr j0045@xmath407319 , with a b star companion ( @xcite ) . although in both cases the optical companion is quite bright and easily observed , knowledge of the companion mass @xmath14 and pulsar mass function @xmath25 is of little use in limiting @xmath13 when @xmath41 . if the mass function of the companion can also be measured , then the mass ratio can be determined . this has been done in the case of j0045@xmath407319 ( [ sec:0045 ] ) . the binary millisecond pulsars are believed to be spun up to high spin frequencies through mass transfer from a companion star . in many cases , these pulsars are observed in wide , low - eccentricity binary systems with white dwarf secondaries . these characteristics indicate that the secondary must have passed through a red giant phase after the formation of the neutron star primary , during which tidal torques circularized the orbit and the giant probably filled its roche lobe , causing the mass transfer which spun up the pulsar to millisecond periods . at the end of mass transfer , the envelope of the giant is exhausted or ejected , leaving the degenerate core as a white dwarf secondary . there is a close relation between the core mass and the radius of low mass giants ( @xcite ) . combined with the assumption that the giant filled its roche lobe during the mass transfer , this yields a relation between the binary period at the end of mass transfer and the remnant white dwarf mass ( @xcite ) . following rappaport _ et al . _ ( 1995 ) and references therein , the relation between the effective radius of the roche lobe @xmath42 and the binary separation @xmath43 can be written @xmath44 where @xmath45 is the total mass of the giant ( core and envelope ) . near the end of mass transfer , the envelope mass can be neglected , so that @xmath46 ( where @xmath14 is the final white dwarf mass ) , and the giant radius @xmath47 . using kepler s third law , @xmath48 where @xmath49 and @xmath14 are in solar units . we note that equation ( [ eqn : pb ] ) is independent of the pulsar mass @xmath13 , and relates the orbital period at the end of mass transfer to the final white dwarf mass and the giant radius @xmath49 . the radius depends , in general , on the composition and history of the giant as well as on @xmath14 ; the utility of equation ( [ eqn : pb ] ) derives from the relatively narrow distribution of @xmath49 for a given @xmath14 . et al . _ ( 1995 ) have examined stellar models with a wide range of chemical compositions , varying between population i and population ii values , as well as a variety of envelope masses and convective mixing lengths . over a wide range of core masses greater than @xmath50 , they find the data are well described by the equation @xmath51 where @xmath52 is the best fit to the stellar models . in all models studied by rappaport _ et al . _ , equation ( [ eqn : rapp1 ] ) was correct within a factor of 1.8 . over the limited range @xmath53 , rappaport and joss ( 1997 ) found that the alternate expression @xmath54 could be used , with the smaller uncertainty of a factor @xmath55 . we regard these to be approximately 95% confidence regions for @xmath49 . it is important to determine the range of applicability of equations ( [ eqn : rapp1 ] ) and ( [ eqn : rapp2 ] ) . at orbital periods less than a few days , heating of the surface of the companion by x - rays produced during accretion may cause significant bloating and modification of the core mass radius relations ( @xcite ) . following rappaport _ , _ we trust the modeling only for binary orbits longer than three days . further , the @xmath6@xmath14 relation can not be applied to systems like psr j2145@xmath400750 or j1022 + 10both 16 ms pulsars that were most likely recycled in common envelopes that occurred when their companions overflowed their roche surfaces while on the asymptotic giant branch , leaving carbon - oxygen rather than helium white dwarfs ( @xcite)or to pulsars like b0820 + 02 , a slow ( 0.864 s ) , high - field ( @xmath56 g ) pulsar without evidence of significant recycling . to avoid both classes of pulsar without introducing biases by cutting on apparent companion mass , we apply the @xmath6@xmath14 relation only to pulsars with @xmath57 ms , for which the assumption of an extended period of mass transfer during a low - mass x - ray binary phase seems secure . using equations ( [ eqn : pb])([eqn : rapp2 ] ) , the observed orbital period in a millisecond pulsar binary can be used to limit the range of @xmath14 . when combined with the mass function ( equation [ eqn : massfunc ] ) and the restriction @xmath58 , an upper limit can be placed on the pulsar mass . the resulting limits are discussed below , in [ sec : nswd ] . if a random distribution of orbits is assumed for the set of observed millisecond pulsars , then statistical arguments lead to limits on the masses of milliseconds pulsars as a class . these arguments are described in [ sec : stat ] . if the companion mass can be estimated and bounds can be found on the orbital inclination , then the mass of the pulsar can be found using the keplerian mass function . in this section , we discuss measurements that may allow direct estimates of @xmath36 . in the standard model of millisecond pulsar formation , the pulsar is spun - up by mass transfer from a companion star . after spin - up , the pulsar spin axis will be aligned with the orbital angular momentum , so a measurement of the angle @xmath59 between the pulsar spin axis and the line of sight also determines the orbital inclination : @xmath60 , or @xmath61 for @xmath59 in the second quadrant . ( although this is also true at an early stage of the formation of double neutron star binaries , an asymmetry in the second supernova explosion may leave the spin and orbit misaligned . ) in the rotating vector model of pulsar emission , the radio signal is elliptically polarized with the major axis of the polarization ellipse aligned with the plane of curvature of the dipolar magnetic field lines . the position angle of the observed linear polarization @xmath62 can be expressed as a function of the rotational phase @xmath63 of the pulsar : @xmath64 where @xmath65 is the angle between the pulsar spin axis and the magnetic pole . in practice , the difference @xmath66 can often be estimated quite well from equation ( [ eqn : rvm ] ) , but accurate estimates of @xmath59 or @xmath65 are possible only if polarized emission can be detected over a broad range of pulse phase @xmath63 . application of equation ( [ eqn : rvm ] ) to millisecond pulsars , in which the light cylinder bounds the magnetosphere at only a few stellar radii , may be complicated by multipolar field geometries , aberration , and other deviations from the simple rotating vector model . indeed , attempts to fit equation ( [ eqn : rvm ] ) to millisecond pulsar data have met with few unqualified successes ( @xcite ) . observed pulsar signal strengths vary in both frequency and time because of scintillation in the interstellar medium . a phase - changing screen along the line of sight produces an interference pattern across which the pulsar moves ; the characteristic scintillation decorrelation timescale @xmath67 is inversely proportional to the transverse component of the pulsar velocity . observations of scintillation rates have been widely used to estimate the proper motions of isolated pulsars ( e.g. , @xcite ) . in binary systems , the transverse velocity is modulated by the orbital motion , with a small amplitude of modulation for orbits viewed face - on , and a large amplitude for orbits viewed on edge . scintillation measurements can thus provide an alternate path to estimating the inclination angle @xmath36 ( @xcite ) . the real physical situation is more complex . the measured scintillation velocity @xmath68 of a pulsar depends not only on its proper motion @xmath69 and orbital motion @xmath70 , but also on the earth s motion @xmath71 , the mean motion of the scattering medium @xmath72 and the ratio of the effective distance to the scattering screen to the pulsar distance , @xmath25 : @xmath73 if the proper motion is known , from timing measurements or interferometry , then the annual modulation of @xmath74 and the orbital modulation of @xmath75 can in principle be used to find not only the orbital inclination @xmath36 but also @xmath25 , the pulsar distance @xmath76 , and the position angle on the sky of the orbital ascending node @xmath77 . only a few attempts to determine orbital inclinations from scintillation observations have been published . in no case was @xmath69 known , nor has the annual modulation due to @xmath74 been measured ( although the latter has been observed in the isolated millisecond pulsar b1937 + 21 by ryba ( 1991 ) ) . lyne ( 1984 ) found that the inclination of psr b0655 + 64 is either @xmath78 or @xmath79 ( the discrete ambiguity could be broken through observations at another time of year , using variations in @xmath74 ) . jones and lyne ( 1988 ) later expressed reservations about this measurement . in the most convincing success of the scintillation technique , dewey _ et al . _ ( 1988 ) found a limit on the the inclination of psr b1855 + 09 : @xmath80 . shapiro time delay measurements have since shown that @xmath81 ( [ sec:1855 ] ) . the problem of estimating scintillation parameters from data is considered by cordes ( 1986 ) . the dominant error source in estimating scintillation parameters is the finite number @xmath82 of scintillation features sampled : @xmath83 . scintillation intensity fluctuations are exponentially distributed ; for observing bandwidth @xmath84 and time @xmath85 , we have roughly @xmath86 . considering a case relevant to many recently discovered pulsar binaries , a pulsar at a dispersion measure dm@xmath87 will have typical scintillation parameters at 430 mhz @xmath88 minutes and @xmath89 khz ( @xcite ) . with @xmath90 mhz and @xmath91 hr , @xmath92 , so a study of orbital dependence of scintillation parameters requires a substantial amount of observing time . furthermore , to use a bandwidth of 10 mhz requires a spectrometer with @xmath93 frequency channels . such observations will be more easily done with the new generation of flexible , all - digital pulsar data recorders ( @xcite ) . it is possible that long term variability in @xmath25 or @xmath68 will limit the use of the annual variation in @xmath74 to estimate @xmath25 . proper motion of the binary system across the sky leads to a secular change in the projected semimajor axis @xmath95 . if @xmath77 is the position angle of the ascending node and @xmath96 and @xmath97 are the components of the proper motion in right ascension and declination , then @xmath98 the angle @xmath77 is generally unknown ( though it is accessible to scintillation measurements , [ sec : scint ] ) , but eqn . [ eqn : xdot ] can be rewritten as the limit @xmath99 a similar expression can be found relating proper motion to a purely geometric advance of periastron @xmath18in principle , measurement of both @xmath100 and @xmath18 would determine @xmath8but the effect is too small to have yet been seen in any pulsar binary . when no other information is available on the orbital inclination , it is sometimes useful in statistical calculations to assume that binary orbits are randomly oriented with respect to the line of sight . the differential distribution of inclinations is then proportional to @xmath36 ( i.e. , the most likely orbital viewing angle is edge - on ) . values of @xmath101 should then be uniformly distributed between 0 and 1 . in binary systems where the directions of the pulsar spin axis and the orbital angular momentum vector are expected to be correlated , such as millisecond pulsar systems in which significant mass transfer has occurred since the last supernova , non - random orientations of the pulsar magnetic field axis with repect to the spin axis may lead to a pulsar discovery bias that skews the observed orbital orientation distribution . some models of pulsar evolution predict magnetic field alignment , counteralignment , or both ( e.g. , @xcite ) . backer ( 1998 ) has claimed , based on the distribution of observed mass functions in 21 pulsar white - dwarf binaries , that millisecond pulsars have a non - random distribution of observed binary inclinations . he suggests that an apparent preference for high inclination orbits may arise because the pulsars magnetic fields are preferentially oriented perpendicular to the spin axis . however , his analysis makes several assumptions and approximations that prove unwarranted . in particular , he compares the observed distribution of minimum companion masses @xmath102 with the predicted distribution given the observed mass functions , random inclinations and a single fixed value for the true companion masses @xmath14 ( i.e. , a delta function distribution in @xmath14 ) , finding poor agreement . however , the predicted distribution for @xmath102 actually depends quite sensitively upon the assumed distribution for @xmath14 ; the delta function distribution is an unreasonable assumption . indeed , the @xmath6@xmath14 relation suggests that the observed @xmath14 values should vary by a factor of three . knowing the precise limits and distribution of these masses is crucial for analyzing the distribution of @xmath102 . we repeated backer s analysis , with different assumptions about the distribution of @xmath14 values . if we assume the true values @xmath14 fall uniformly in the range 0.150.35 @xmath103 , then the predicted and observed distributions for @xmath102 agree at an 81% confidence level according to a kolmogorov - smirnov ( ks ) test ( e.g. , @xcite ) . however , assuming a 0.10.4 @xmath103 range reduces the agreement to the 33% confidence level . the discrepancy is primarily due to a deficit of observed systems with small values of @xmath102 , as noted by backer . unfortunately , our knowledge of the true underlying distribution of @xmath14 in these binaries is sufficiently uncertain that no conclusion regarding the distribution of inclinations is possible from this line of analysis . however , we argue below ( [ sec : stat ] ) that for systems with orbital periods longer than three days , where the @xmath6@xmath14 relation can be applied to estimate the @xmath14 distribution , the observed mass functions are consistent with a random distribution of inclinations . the parameters of 47 radio pulsar binary systems are listed in table [ tab : allbins ] . at least five of these pulsars have neutron star companions , so a total of 52 neutron stars are known in radio pulsar binaries . useful mass constraints can be placed on half of these stars . psr j1518 + 4904 is in a moderately relativistic binary system ; one pk parameter has been measured , the relativistic advance of periastron @xmath104 , yielding a total system mass of @xmath105 ( @xcite ) . given the mass function @xmath106 , the lower limit on the companion mass ( using @xmath58 ) is @xmath107 , and the lower limit on the inclination ( given @xmath108 ) is @xmath109 ( or @xmath110 ) . using a uniform prior distribution in @xmath36 over the interval @xmath111 , we find the central 68% confidence intervals @xmath112 and @xmath113 and central 95% confidence intervals @xmath114 and @xmath115 . the identification of the companion as a neutron star is compelling , although it is not possible to completely rule out a low - mass black hole . optical observations ( van kerkwijk , quoted in sayer 1996 ) show no source at the pulsar position to @xmath116 , excluding a main sequence companion , and a white dwarf in an eccentric orbit is not expected on evolutionary grounds . psr b1534 + 12 is in a highly relativistic binary system , presumably with a second neutron star companion . all five of the pk parameters defined by equations ( [ eqn : pkomdot])([eqn : pks ] ) have been measured , three with better than 1% precision ( @xcite ) . using the ddgr timing model , the total system mass is found to be @xmath117 , while the individual component masses @xmath13 and @xmath14 are both @xmath118 . ( the quoted errors are 68% confidence regions ; the 95% confidence regions are about twice as large . ) it is remarkable that the pulsar and companion masses agree to better than @xmath119 ; the assumption that the companion is a second neutron star seems secure . the uncertainty on the individual masses is expected to decrease significantly in the next year , when observations are again possible with the arecibo telescope . psr b1913 + 16 was the first binary pulsar discovered and , with observations stretching over two decades , it remains one of the best studied . psr b1913 is a highly relativistic system : the pk parameters @xmath18 , @xmath19 , and @xmath20 have all been measured precisely . a fit to the ddgr timing model yields the total system mass @xmath120 and component masses @xmath121 and @xmath122 ( @xcite ) . ( the quoted errors are 68% confidence regions ; the 95% confidence regions are about twice as large . ) the uncertainties are approaching the level where they will be dominated by kinematic effects ( equation [ eqn : doppler ] ) . psr b2127 + 11c is a relativistic binary in the globular cluster m15 . precise measurements have been made of the pk parameters @xmath18 and @xmath19 , resulting in mass estimates for the pulsar and companion of @xmath123 and @xmath124 , respectively , and a total system mass @xmath125 ( @xcite ) . ( the quoted errors are 68% confidence regions ; the 95% confidence regions are approximately twice as large . ) psr b2303 + 46 is a pulsar in a moderately relativistic orbit with , most likely , a neutron star companion . one pk parameter , the relativistic advance of periastron , has been measured . a timing analysis was published by thorsett _ et al . _ ( 1993 ) , and updated by arzoumanian ( 1995 ) . using data from early 1985 to late 1994 , we find an improved value @xmath126 , yielding a total system mass @xmath127 and the constraints @xmath128 and @xmath129 . using a uniform prior distribution in @xmath36 , we find the central 68% confidence intervals @xmath130 and @xmath131 and central 95% confidence intervals @xmath132 and @xmath133 . psr j0437@xmath404715 is the brightest and closest known millisecond pulsar , and therefore one of the best studied . it has a mass function @xmath134 , and a white dwarf companion that has been detected optically ( @xcite ) ; and both thermal and nonthermal x - ray emission from the neutron star have been observed ( @xcite ) . the distance is known , @xmath135 pc , from the effects of parallax on the pulsar timing signal ( @xcite ) . using the optical data of danziger _ _ hansen and phinney ( 1998b ) found an effective temperature of the companion @xmath136 k. combining the @xmath137 and distance estimates with their cooling curves , they find consistent companion models for all masses @xmath138 . this range encompasses the mass limits derived from the @xmath6@xmath14 relation ( [ sec : cmrr ] ) , @xmath139 . the binary orbit is nearly circular , and despite the very high timing precision achieved , no pk timing parameters have been measured . however , the proximity of the pulsar leads to a high proper motion ( @xmath140 mas / yr ) , changing the projected orbital size ( [ sec : xdot ] ) at a rate @xmath141 ( @xcite ) . the implied limit on the inclination angle is @xmath142 , or @xmath143 . attempts to measure the system geometry using the polarization of the radio beam ( [ sec : polar ] ) have proven difficult , because of the very complex emission pattern . well - calibrated intensity and polarization data have been reported by manchester and johnston ( 1995 ) , who have modeled the sweep of the linear polarization position angle across the profile in terms of the rotating vector model . they find that an impact parameter of the line of sight on the magnetic pole @xmath144 and an angle between the line of sight and the spin axis @xmath145 produced reasonable agreement with the data , but there are strong systematic deviations from the simple rotating vector model . using the same data , gil and krawczyk ( 1997 ) model the multicomponent profile and find a comparable impact parameter , @xmath146 , but a very different @xmath147 . they show that this geometry also explains the observed polarization sweep over much of the pulse period . with the assumption that the pulsar spin axis is aligned with the orbital axis , the implied orbital inclination is @xmath148 ( manchester and johnston model ) or @xmath149 ( gil and krawczyk model ) . either is consistent with the timing data . the timing limit on @xmath36 , together with the mass function and the upper limit on @xmath14 from the core mass orbital period relation , gives an upper limit to the pulsar mass of @xmath150 , while the optical upper limit on @xmath14 gives only the much weaker limit @xmath151 . the manchester and johnston inclination together with the @xmath6@xmath14 mass range gives @xmath152 , while the gil and krawczyk inclination gives @xmath153 , a physically implausible result . in any case , the complexity of the pulse shape modeling leads us to adopt the weaker one - sided limit @xmath150 . improvements of the parallax measurement and the optical photometry and spectroscopy are needed to further constrain the companion mass . an independent mass determination would test the core mass orbital period relation as well as improve the limit on @xmath13 . measurement of the position angle of the ascending node ( perhaps through scintillation studies ) , combined with the measurement of @xmath154 , would give the inclination of the orbit . psr j1012 + 5307 has a hot , bright white dwarf companion that has been extensively studied . van kerkwijk , bergeron , and kulkarni ( 1996 ) have used the models of bergeron , wesemael , and fontaine ( 1991 ) to determine the effective temperature @xmath155 k and surface gravity @xmath156 of the companion . the latter value is in disagreement with an unpublished value of callanan and koester , @xmath157 , quoted by hansen and phinney ( 1998b ) . using the optical observations and their cooling models for low mass helium white dwarfs , hansen and phinney find a companion mass @xmath158 for the van kerkwijk _ et al . _ gravity measurement and @xmath159 for the callanan and koester value . the radial velocity of the companion has been measured , making j1012 + 5307 a double - line spectroscopic pulsar binary . van kerkwijk _ et al . _ have found @xmath160 ( van kerkwijk , private communication ; their earlier published value @xmath161 was corrupted by a calibration problem ) . depending on the gravity measurement , the resulting ( @xmath162 ) pulsar mass limit is @xmath163 or @xmath164 . clearly resolution of the discrepant gravity measurements must be a high priority , the radial velocity measurements now contribute negligibly to the total error on @xmath14 . for now , we adopt hansen and phinney s conservative conclusion that @xmath165 , and we regard the error range as roughly a 68% confidence interval . psr j1045@xmath404509 is in a 4.08 day orbit with mass function @xmath166 . the @xmath6@xmath14 relation gives an upper limit to the companion mass of @xmath167 , leading to a limit on the pulsar mass of @xmath168 . psr j1713 + 0747 is a bright pulsar in a nearly circular , 68 day orbit with a white dwarf companion . at the timing precision reached ( about 500 ns for 1.4 ghz observations ) , camilo ( 1995 ) found that the shapiro pk parameters @xmath21 and @xmath22 were required to adequately model the data , but strong covariances between the range parameter @xmath21 and other orbital parameters ( especially @xmath95 ) prevented him from setting interesting limits on the pulsar or companion mass . the @xmath6@xmath14 relation predicts @xmath169 . companion mass estimates are also possible by combining unpublished optical observations of the white dwarf by lundgren _ et al . _ with the parallax from pulsar timing ( @xcite ) . the resulting limits are @xmath170 if the white dwarf has a thick h envelope , and @xmath171 if it has a thin h envelope ( @xcite ) . if the observed mass function @xmath172 is combined with the @xmath6@xmath14 limit on @xmath14 and the restriction @xmath58 , an upper limit to the pulsar mass is found : @xmath173 . the optical observations yield a tighter limit , @xmath174 , as was noted by hansen and phinney . more interesting limits can be obtained by combining the shapiro measurements with the limits on @xmath14 ( recalling that @xmath175 ) . for example , if @xmath176 , the timing data of camilo is consistent with only the restricted range @xmath177 . by letting @xmath14 vary over the range allowed by the @xmath6@xmath14 relation , we find the allowed pulsar mass @xmath178 , where the uncertainties are dominated by the systematic uncertainties in the @xmath6@xmath14 relation . if the companion mass is limited to the intersection of the regions allowed by @xmath6@xmath14 and the optical measurements , @xmath179 , then we find @xmath180 . psr b1802@xmath4007 is in the globular cluster ngc 6539 . its companion is most likely a white dwarf ; the system s large eccentricity can be understood as the result of gravitational perturbations of the system by close stellar encounters in the dense cluster . the relativistic advance of periastron @xmath18 was measured by thorsett _ et al . _ ( 1993 ) , and an improved measurement was published by arzoumanian ( 1995 ) . using the same analysis techniques , with data extending through october 1997 , we find a slightly improved value @xmath181 , implying a total system mass @xmath182 . the lower bound on the companion mass from the requirement that @xmath58 is @xmath183 , and the lower bound on the inclination from the requirement @xmath108 is @xmath184 ( or @xmath185 ) . using a uniform prior for @xmath36 in the range @xmath186 , we find a 68% confidence bound on the pulsar mass @xmath187 and a 95% confidence bound @xmath188 . psr j1804@xmath402718 is in an 11.1 day orbit . the @xmath6@xmath14 relation limits its companion mass to @xmath189 ; the resulting upper limit on the pulsar mass is @xmath190 . psr b1855 + 09 is a 5.4 ms pulsar in a 12.3 day circular orbit ( @xmath191 ) with a white dwarf companion . because it is a bright pulsar for which high precision timing measurements can be made and because the orbital inclination is high , measurements have been made of the pk parameters @xmath21 and @xmath22 . we have reanalyzed all of the available data , extending from january 1986 to january 1994 . ( for the data collected after mid-1989 , taken with the princeton mark iii timing system ( @xcite ) , a new algorithm was used for identifying times and frequencies where the signal strength was enhanced by interstellar scintillation and weighting this data in subsequent analysis . ) the details and complete timing solution will be published elsewhere . the shapiro parameters are @xmath192 and @xmath193 . the resulting limits on the pulsar mass are @xmath194 ( 68% confidence ; the 95% confidence region will be about twice as large ) . the new measurement is in good agreement with previous values : @xmath195 ( @xcite ) and @xmath196 ( @xcite ) . by calculating the _ joint _ 68% confidence bound on @xmath197 and @xmath24 and then interpreting this as a 68% confidence bound on each of @xmath14 and @xmath36 . ] the @xmath6@xmath14 relation predicts @xmath198 , in good agreement with the timing measurement . the corresponding upper limit to the pulsar mass is @xmath150 . as noted above , scintillation measurements have also been used to limits @xmath199 ( @xcite ) . psr j2019 + 2425 is in a 76.5 day orbit . the @xmath6@xmath14 relation gives a limit on the companion mass @xmath200 ; the resulting upper limit on the pulsar mass is @xmath201 . the pulsar j0045@xmath407319 is the only known pulsar in the small magellanic cloud . it is in a binary orbit , with mass function @xmath202 ( @xcite ) . the companion has been identified as a b1 v star . the radial velocity of the companion has been measured , giving a mass ratio @xmath203 ( @xcite ) we have compared the observed optical luminosity @xmath204 and temperature @xmath205 k to the grids of stellar models calculated by schaller _ et al . _ ( 1992 ) for the low metallicity ( @xmath206 ) appropriate for the smc , and estimate the companion mass to be @xmath207 . the pulsar mass is then @xmath208 ( 68% confidence ) , where the uncertainty is dominated by the uncertainty in the amplitude of the companion s radial velocity curve . for a dozen neutron stars , useful mass constraints are available with no assumptions beyond the applicability of the general relativistic equations of orbital motion to binary pulsar systems . ten of these stars are members of double neutron star binaries . with the possible exception of psr b2127 + 11c , in the globular cluster m15 , the pulsar in each system is believed to have undergone a short period of mass accretion during a high - mass x - ray binary phase ( @xmath209 , taam and van den heuvel 1986 ) . the companion stars have not undergone accretion ; their masses most directly preserve information about the initial mass function of neutron stars . only two `` millisecond '' pulsars , the end products of extended mass transfer in low - mass x - ray binaries , have interesting mass estimates based on gr alone : psrs b1802@xmath4007 and b1855 + 09 . because such pulsars must accrete @xmath210 to reach millisecond periods ( @xcite ) , and much more ( @xmath211 ) in some field decay models ( e.g. , @xcite ) , obtaining additional mass measurements of millisecond pulsars is of particular interest in testing evolutionary models and in locating the maximum neutron star mass . as noted in [ sec : cmrr ] , the @xmath6@xmath14 relation can be used to estimate the companion mass in recycled binary systems with circular orbits and orbital periods @xmath212 d. there are now thirteen such millisecond ( @xmath213 ms ) pulsars known , excluding those in globular clusters ( where gravitational interactions may have significantly perturbed the orbital parameters since spin - up ) . in each case , the measured mass function and the inferred companion mass , together with the requirement that @xmath58 , then yields an upper limit on the mass of the pulsar itself . a number of systems in which this upper limit is particularly constraining have been mentioned in [ sec : nswd ] . additional constraints on the neutron star mass in these systems can be derived using statistical arguments , given a prior assumption about the distribution of binary inclinations . the simplest such assumption is that the binaries are randomly oriented on the sky , though biases toward high or low inclinations are possible in some models ( [ sec : uniform ] ) . however , as discussed below , we believe there is currently no evidence for such a bias , so for the remainder of this discussion we assume random orbital orientations . for an individual system , we are interested in the probability distribution is the ( marginal ) probability density for the random variable @xmath95 , where @xmath95 depends upon the parameter @xmath214 . also , @xmath215 is the conditional probability density for the random variable @xmath95 for a given value of the random variable @xmath216 and parameter @xmath214 . ] @xmath217 for the neutron star mass @xmath13 given the measured mass function @xmath25 and binary period @xmath6 . we can neglect the measurement uncertainty in @xmath25 and @xmath6 . then , the probability distribution for @xmath13 can be written schematically as @xmath218 where @xmath219 is restricted to positive values . we have evaluated this numerically for each system , assuming that @xmath220 is uniform between zero and unity and that @xmath221 is uniformly distributed within the appropriate factor ( see [ sec : cmrr ] ) of the @xmath14 implied by equations ( 9)(11 ) . not surprisingly , the width of the distribution @xmath217 is dominated by the range of allowed @xmath101 rather than the uncertainty in @xmath14 for a given @xmath6 . for each of the 13 binaries , we have plotted the cumulative distribution cdf@xmath222 in figure [ fig : cumprob ] . the median and 68% and 95% confidence regions for each pulsar mass is given in table [ tab : pbm2 ] . although several of the pulsars have , under the assumptions made , most likely masses well above @xmath223 , some such results are expected even if all the masses are quite low . in fact , in only one case of the 13 pulsars does @xmath224 lie outside the 95% central confidence region ( j1045@xmath404509 ) , and in 6 cases of 13 is @xmath224 excluded at 68% confidence , consistent with chance . it is interesting to ask whether a single , simple distribution of neutron star masses is consistent with all of our observational constraints . we considered two models for this question : a gaussian distribution of masses with mean @xmath225 and standard deviation @xmath226 , and a uniform distribution of masses between @xmath227 and @xmath228 ( _ cf . _ finn 1994 ) . a maximum likelihood analysis was used to estimate the parameters @xmath225 , @xmath226 , @xmath227 , and @xmath228 ( assuming a uniform prior distribution for all four parameters ) . the resulting 68% and 95% joint confidence limits on @xmath225 and @xmath226 are shown in figure [ fig : all26 ] , and on @xmath227 and @xmath229 in figure [ fig : all26uni ] . in each model , the distribution of neutron star masses is remarkably narrow : the maximum likelihood solutions are @xmath230 and @xmath231 , and @xmath232 and @xmath233 . of course , _ any _ model ( even a poor one ) will yield maximum likelihood parameters for a given data set . however , it is obvious by inspection that both the gaussian and uniform distributions for the neutron star mass are good fits to the extremely narrow observed range of neutron star masses in the double neutron star binaries . while it is difficult to quantify the goodness - of - fit for the entire data set , because of the diverse assumptions made in the various mass estimates and the sometimes highly non - gaussian error estimates , we can easily test some neutron star subsamples against the maximum likelihood gaussian model @xmath234 . for the thirteen neutron - star white - dwarf binaries , we used a monte carlo technique to evaluate the fit quality . for each binary ( with its measured @xmath6 and @xmath25 ) , we simulated a large number of monte carlo trials where the neutron star mass @xmath13 was drawn from the maximum likelihood model , @xmath14 was drawn from the appropriate uniform distribution implied by @xmath6 , and @xmath101 was drawn from a uniform distribution . the monte carlo trials were then used to construct the probability distribution for the mass function , and this distribution was used to compute the cumulative probability for the measured mass function , @xmath235 . if the model and the associated assumptions are correct , then the cumulative probabilities for the 13 measured mass functions should be consistent with a uniform distribution betwen zero and unity ( _ cf . _ a classical @xmath236 test , schmidt 1968 ) . a ks test of the distribution shows consistency with a uniform distribution at the improbably good 99% level ( figure [ fig : ks ] ) . for the ten stars for which gaussian error estimates @xmath237 are available ( both stars in the relativistic binaries b1534 + 12 , b1913 + 16 , and b2127 + 11c , as well as j1012 + 5307 , j1713 + 0747 , b1855 + 09 , and j0045@xmath407319 ) , we can calculate a @xmath238 statistic , @xmath239 , consistent with expectations for a chi - square distribution with @xmath240 degrees of freedom . we conclude , therefore , that at least in the radio pulsar systems , there is no evidence for neutron star masses above about @xmath241 . indeed , the data appear very well modeled by very narrow distributions centered around @xmath224 . there are now 26 neutron stars in binary radio pulsar systems for which useful mass constraints can be derived . of these , about half are neutron star - white dwarf binaries in which the mass determination depends on the validity of the @xmath6@xmath14 relation and the isotropy of the binary orbits with respect to the line - of - sight , as discussed in the previous section . all other mass constraints are listed in table [ tab : masssum ] and shown in figure [ fig : massrev ] . although we defer a full discussion of the underlying neutron star mass distribution and it implications for neutron star formation and evolution until after analysis of the x - ray binary systems ( paper ii ) , we note here a few points of particular interest about the radio pulsar binaries . figure [ fig : massrev ] is striking primarily for the very small variations in the masses of well measured stars . this has , of course , been noted before ( e.g. , thorsett _ et al . _ 1993 ) , but it remains surprising that no new mass measurements differ greatly from @xmath242 . in the five double neutron star binaries for which a relativistic periastron advance yields an accurate total ( and hence average ) mass , the average neutron star masses vary by less than 7% . the most surprising implication of the current results is that there is little evidence for mass tranfer of @xmath243 or more in the millisecond pulsar systems . it is important , then , to reiterate the assumptions upon which this conclusion rests : ( 1 ) the @xmath6@xmath14 is correct , at least within the ( modest ) claimed precision , and ( 2 ) binary orbits are randomly oriented with respect to the line of sight . there are good prospects for testing both assumptions . the reliability of the @xmath6@xmath14 relation depends principally upon the core - mass radius relation for red giants ( eqns . [ eqn : rapp1 ] and [ eqn : rapp2 ] ) . the latter relation can be tested observationally by careful study of nearby red giants . precise measurements of bolometric flux and angular size ( through optical / infrared photometry and interferometry ; see , for example , dyck _ et al . _ 1996 , perrin _ et al . _ 1998 ) together with accurate distance measurements ( e.g. , using _ hipparcos _ ) can be used to probe the relationship between luminosity and radius . since the core - mass luminosity relation for red giants should be inherently more precise than the core - mass radius relation ( rappaport _ et al . _ 1995 and references therein ) , such observations would provide an effective test of the core - mass radius ( and hence @xmath6@xmath14 ) relation . further , we hope that the power of the @xmath6@xmath14 relation as a statistical tool will encourage more detailed theoretical investigations of , in particular , the extension of the relation to orbital periods below about three days , where x - ray heating and bloating of the companion star become important . although any technique that limits @xmath36 ( [ sec : sini ] ; e.g. , polarization or scintillation studies ) can be used to test the hypothesis that orbits are randomly inclined , the most precise measurements will come from pulsar timing and shapiro time delay measurements . in table [ tab : pbm2 ] we list the mean predicted amplitude of the shapiro delay signal @xmath244 for each of the thirteen pulsars discussed in [ sec : stat ] , under the assumption that the neutron stars are all @xmath242 and the companion masses are the central value predicted by the @xmath6@xmath14 relation . of these systems , only b1855 + 09 has a well measured signal ( @xcite ) ; it and the less studied j2019 + 2425 have the largest predicted signals . ( it is also interesting to note that of the pulsars in table [ tab : pbm2 ] , the only two that have been clearly shown to have `` classical '' interpulse emission separated from the main radio emission peak by @xmath245 are b1855 + 09 and j1804@xmath402718 . these are two of the four pulsars that the @xmath6@xmath14 relation suggests are observed on lines of sight that are most nearly equatorial , as would be expected for systems in which both magnetic poles are seen . ) an assumed pulsar mass distribution and the @xmath6@xmath14 relation allow us to test our assumption that the orbital inclinations are randomly distributed : i.e. , @xmath101 is uniformly distributed . in figure [ fig : ks ] we show the cumulative distribution of @xmath101 for the thirteen pulsar white - dwarf binaries discussed above , using @xmath36 values from table [ tab : pbm2 ] except for b1855 + 09 , for which a better estimate is available from timing . we find that the measured values are consistent with uniform at the 81% level . because the pulsar spin and orbital angular momenta are most likely aligned , we further conclude that there is no evidence for either alignment or counteralignment of pulsar magnetic fields in millisecond pulsars , contrary to some predictions ( e.g. , @xcite ) . the observations on which this paper is based were carried out by many people . set particularly thanks his collaborators z. arzoumanian , d. j. nice , and j. h. taylor . v. m. kaspi , l. rawley , and m. ryba all contributed significantly to the psr b1855 + 09 dataset , as did a. vasquez and other staff members of the arecibo observatory , operated by cornell university for the u. s. national science foundation . observations were also made with the facilities of the national radio astronomy observatory . we thank m. h. van kerkwijk for sharing results in advance of publication , and d. c. backer for valuable discussions . the research of set is supported by the nsf . dc is supported by a nasa compton gro postdoctoral fellowship , under grant nag 5 - 3109 . dewey , r. j. , cordes , j. m. , wolszczan , a. , and weisberg , j. m. 1988 , in _ radio wave scattering in the interstellar medium , aip conference proceedings no . 174 _ , ed . cordes , b. j. rickett , and d. c. backer , ( new york : american institute of physics ) , 217 . rickett , b. j. 1988 , in _ radio wave scattering in the interstellar medium , aip conference proceedings no . 174 _ , ed . cordes , b. j. rickett , and d. c. backer , ( new york : american institute of physics ) , 2 . llllc j1518 + 4904&0.040935&8.634&0.24948&b1534 + 12&0.037904&0.421&0.27368&b1913 + 16&0.05903&0.323&0.61713&b2127 + 11c&0.030529&0.335&0.68141&1b2303 + 46&1.066371&12.34&0.65837 & b0021@xmath4072e&0.003536&2.257&0.000&1b0021@xmath4072i&0.003485&0.226&0.00&1b0021@xmath4072j&0.002101&0.121&0.00&1j0034@xmath400534&0.001877&1.589&0.0000&j0218 + 4232&0.002323&2.029&0.00000&j0437@xmath404715&0.005757&5.741&0.00000&j0613@xmath400200&0.003062&1.199&0.00000&b0655 + 64&0.195671&1.029&0.00000&j0751 + 1807&0.003479&0.263&0.0000&b0820 + 02&0.864873 & 1232.47&0.01187&j1012 + 5307&0.005256&0.605&0.00000&j1022 + 10&0.016453&7.805&0.0001&j1045@xmath404509&0.007474&4.084&0.00000&b1310 + 18&0.033163&255.8&0.002&1j1455@xmath403330&0.007987&76.175&0.00017&j1603@xmath407202 & 0.014842 & 6.309 & 0.0000 & 2b1620@xmath4026&0.011076&191.443&0.02531&1,4j1640 + 2224&0.003163&175.461&0.0008&b1639 + 36b&0.003528&1.259&0.005&1j1643@xmath401224&0.004622&147.017&0.00051&j1713 + 0747&0.004570&67.825&0.00007&b1718@xmath4019&1.004037&0.258&0.000&1b1744@xmath4024a&0.011563&0.076&0.0000&1b1800@xmath4027&0.334415&406.781&0.00051&b1802@xmath4007&0.023101&2.617&0.212&1j1804@xmath402717 & 0.009343 & 11.129 & 0.00004 & 2b1820@xmath4011&0.279828&357.762&0.79462&3b1831@xmath4000&0.520954&1.811&0.000&b1855 + 09&0.005362&12.327&0.00002&j1910 + 0004&0.003619&0.141&0.00&1j1911@xmath401114 & 0.003626 & 2.717 & 0.0000 & 2b1953 + 29&0.006133&117.349&0.00033&b1957 + 20&0.001607&0.382&0.00000&j2019 + 2425&0.003935&76.512&0.00011&j2033 + 17&0.005949&56.2&0.00&j2129@xmath405721 & 0.003726 & 6.625 & 0.0000 & 2j2145@xmath400750&0.016052&6.839&0.00002&j2229 + 2643&0.002978&93.016&0.00026&j2317 + 1439&0.003445&2.459&0.00000 & j0045@xmath407319&0.926276&51.169&0.808&b1259@xmath4063&0.047762 & 1236.724&0.86993 & b1257 + 12&0.006219&66.536&0.0182 & & 2nd planet & 98.223&0.0264 & lllclllllll j0437@xmath404715 & 5.741 & 1.243 & 0.164 & 0.106 & 0.472 & 1.044 & 1.573 & 1.971 & 0.87 & 3.5 j1045@xmath404509 & 4.083 & 1.765 & 0.132 & 0.047 & 0.234 & 0.557 & 0.965 & 1.273 & 0.97 & 6.2 j1455@xmath403330 & 76.174 & 6.272 & 0.305 & 0.098 & 0.491 & 1.144 & 1.681 & 2.058 & 0.85 & 5.9 j1640 + 2224 & 175.460 & 5.907 & 0.351 & 0.135 & 0.651 & 1.495 & 2.190 & 2.681 & 0.73 & 4.6 j1643@xmath401224 & 147.017 & 0.783 & 0.341 & 0.553 & 2.106 & 4.439 & 6.335 & 7.674 & 0.38 & 1.6 j1713 + 0747 & 67.825 & 7.896 & 0.299 & 0.078 & 0.405 & 0.960 & 1.419 & 1.741 & 0.91 & 7.6 j1804@xmath402718 & 11.128 & 4.137 & 0.212 & 0.073 & 0.369 & 0.856 & 1.265 & 1.558 & 0.94 & 6.7 b1855 + 09 & 12.327 & 5.557 & 0.219 & 0.060 & 0.315 & 0.745 & 1.101 & 1.351 & 0.97 & 9.0 b1953 + 29 & 117.349 & 2.417 & 0.328 & 0.236 & 1.024 & 2.251 & 3.251 & 3.955 & 0.58 & 2.8 j2019 + 2425 & 76.511 & 10.686 & 0.305 & 0.062 & 0.338 & 0.818 & 1.218 & 1.500 & 0.95 & 10.0j2033 + 17 & 56.2 & 2.75 & 0.290 & 0.172 & 0.772 & 1.720 & 2.489 & 3.029 & 0.67 & 3.2 j2129@xmath405721 & 6.625 & 1.049 & 0.176 & 0.138 & 0.591 & 1.293 & 1.900 & 2.348 & 0.80 & 2.7 j2229 + 2643 & 93.015 & 0.839 & 0.315 & 0.463 & 1.789 & 3.787 & 5.409 & 6.552 & 0.42 & 1.7 lllll pulsar & 1.56 & @xmath246 & @xmath247 & gr , rocompanion & 1.05&@xmath248&@xmath249&gr , roaverage & 1.31 & @xmath250 & @xmath251 & gr pulsar & 1.339 & @xmath252 & @xmath253 & grcompanion & 1.339 & @xmath252 & @xmath253 & gr pulsar & 1.4411 & @xmath254 & @xmath255 & gr companion & 1.3874 & @xmath254 & @xmath255 & gr pulsar & 1.349 & @xmath256 & @xmath257 & grcompanion & 1.363 & @xmath256 & @xmath257 & graverage & 1.3561 & @xmath258 & @xmath259 & gr pulsar & 1.30 & @xmath260 & @xmath261 & gr , rocompanion & 1.34 & @xmath262 & @xmath263&gr , roaverage & 1.32 & @xmath264 & @xmath265 & gr j0437@xmath404715 & & & @xmath266 & @xmath100 , @xmath267 j1012 + 5307 & 1.7 & @xmath268 & @xmath269 & opt j1045@xmath404509 & & & @xmath270 & @xmath267 j1713 + 0747 & 1.45 & @xmath271 & @xmath272 & @xmath267 , gr & 1.34 & @xmath273 & @xmath274 & @xmath267 , gr , optb1802@xmath4007 & & @xmath275 & @xmath276 & gr & 1.26 & @xmath277 & @xmath278 & gr , @xmath267j1804@xmath402718 & & & @xmath279 & @xmath267b1855 + 09 & 1.41 & @xmath280 & @xmath273 & gr j2019 + 2425 & & & @xmath281 & @xmath267 j0045@xmath407319 & 1.58 & @xmath282 & @xmath283 & opt , ms

we explore high - redshift gamma - ray bursts ( grbs ) as promising tools to probe pre - galactic metal enrichment . we utilize the bright afterglow of a population iii ( pop iii ) grb exploding in a primordial dwarf galaxy as a luminous background source , and calculate the strength of metal absorption lines that are imprinted by the first heavy elements in the intergalactic medium ( igm ) . to derive the grb absorption line diagnostics , we use an existing highly - resolved simulation of the formation of a first galaxy which is characterized by the onset of atomic hydrogen cooling in a halo with virial temperaturek . we explore the unusual circumburst environment inside the systems that hosted pop iii stars , modeling the density evolution with the self - similar solution for a champagne flow . for minihalos close to the cooling threshold , the circumburst density is roughly proportional to with values of about a few . in more massive halos , corresponding to the first galaxies , the density may be larger , . the resulting afterglow fluxes are weakly dependent on redshift at a fixed observed time , and may be detectable with the _ james webb space telescope ( jwst ) _ and very large array ( vla ) in the near - ir and radio wavebands , respectively , out to redshift . we predict that the maximum of the afterglow emission shifts from near - ir to millimeter bands with peak fluxes from mjy to jy at different observed times . the metal absorption line signature is expected to be detectable in the near future . grbs are ideal tools for probing the metal enrichment in the early igm , due to their high luminosities and featureless power - law spectra . the metals in the first galaxies produced by the first supernova ( sn ) explosions are likely to reside in low - ionization stages ( , , and ) . we show that if the afterglow can be observed sufficiently early , analysis of the metal lines may distinguish whether the first heavy elements were produced in a pair - instability supernova ( pisn ) , or a core - collapse ( type ii ) sn , thus constraining the initial mass function of the first stars .

1207.3879

one of the key goals in modern cosmology is to study the formation of the first stars and galaxies at the end of the cosmic dark ages , and how they shaped the subsequent evolution of the universe ( barkana & loeb 2001 ; bromm & larson 2004 ; ciardi & ferrara 2005 ; bromm et al . 2009 ; loeb 2010 ) . the first , so - called population iii ( pop iii ) stars are predicted to have formed at @xmath5 in minihalos with virial mass @xmath6 and temperatures @xmath7k ( haiman et al . 1996 ; tegmark et al . 1997 ; yoshida et al . the pop iii initial mass function ( imf ) is thought to be top - heavy ( bromm et al . 1999 , 2002 ; abel et al . 2002 ) , possibly extending to @xmath8 , but recent simulations indicate that the pop iii imf may be quite broad , also including a fraction of lower - mass stars ( stacy et al . 2010 ; clark et al . 2011b ; greif et al . 2011 , 2012 ) . the first bona - fide galaxies are expected to have formed at a later stage in hierarchical structure formation ( bromm & yoshida 2011 ) , when @xmath9 halos assembled at @xmath10 via the merging of progenitor minihalos ( wise & abel 2007 ; greif et al . these systems are often termed ` atomic cooling halos ' , because their virial temperature , @xmath11k , exceeds the threshold to enable efficient cooling via lines of atomic hydrogen ( oh & haiman 2002 ) . direct observations of the first galaxies at redshifts @xmath12 have so far been out of reach . in the coming decade , the _ james webb space telescope ( jwst ) _ promises to directly probe this critical period ( gardner et al . the detection of metal absorption lines in the afterglow spectrum of high - redshift gamma - ray bursts ( grbs ) , imprinted by enriched gas in the first galaxies , offers an unusual opportunity to study the physical conditions inside them . we may thus be able to derive constraints on the temperature , metallicity , ionization state , and kinematics in the interstellar medium ( ism ) of high - redshift galaxies , and in the surrounding intergalactic medium ( igm ) . it is encouraging that such diagnostics can already be obtained for bursts at somewhat lower redshifts . an example is grb 081008 , where high - resolution spectroscopy with the vlt has probed the ism of a host galaxy at @xmath13 ( delia et al . 2011 ) . long - duration grbs have been shown to be associated with the death of massive stars ( stanek et al . 2003 ; hjorth et al . 2003 ; woosley & bloom 2006 ) . their high luminosities make them detectable out to the edge of the visible universe ( ciardi & loeb 2000 ; lamb & reichart 2000 ; bromm & loeb 2002 , 2006 ; gou et al . 2004 ; inoue et al . 2007 ; mszros & rees 2010 ) , with the current record held by grb 090429b at @xmath14 ( cucchiara et al . 2011 ) . grbs provide ideal probes of the high - redshift universe , including the star formation rate ( totani 1997 ; wijers et al . 1998 ; porciani & madau 2001 ; chary et al . 2007 ; yksel et al . 2008 ; wang & dai 2009 , 2011 ; elliott et al . 2012 ) , reionization ( gallerani et al . 2008 ) , dark energy ( dai et al . 2004 ; wang et al . 2011 ) , and the igm metal enrichment ( barkana & loeb 2004 ; totani et al . 2006 ; toma et al . 2011 ; bromm & loeb 2012 ) . the leading contender for the central engine of long - duration grbs is the collapsar model ( woosley 1993 ; macfadyen et al . 2001 ) . because of their predicted high characteristic mass , a significant fraction of pop iii stars might end their lives as a black hole , potentially leading to a large number of high - redshift grbs . thus , pop iii stars are viable progenitors of long - duration grbs , triggered by the collapsar mechanism , as long as they can lose their outer envelope and retain sufficient angular momentum in their center ( bromm & loeb 2006 ; belczynski et al . 2007 ; komissarov & barkov 2010 ; stacy et al . 2011 ) . it might even be possible for pop iii collapsars to occur if the extended outer envelope were not lost ( suwa & ioka 2011 ) . the history of pre - galactic metal enrichment has several important consequences for structure formation ( madau et al . 2001 ; karlsson et al . 2012 ) . an early phase of metal injection may qualitatively change the character of star formation , from a predominantly high - mass ( pop iii ) mode to a normal , low - mass dominated ( pop i / ii ) one , once the enrichment has exceeded a ` critical metallicity ' of @xmath15 ( bromm et al . 2001a ; schneider 2002 , 2006 ; bromm & loeb 2003 ; mackey et al . the transition between these two modes has crucial implications , e.g. , for the expected redshift distribution of grbs ( bromm & loeb 2006 ; campisi et al . 2011 ; de souza et al . 2011 ) , for reionization ( cen 2003 ; wyithe & loeb 2003 ; furlanetto & loeb 2005 ) , and for the chemical abundance patterns of low - metallicity stars ( qian & wasserburg 2001 ; frebel et al . 2007 , 2009 ; tumlinson 2010 ) . it is therefore important to explore the topology of early metal enrichment , and to determine when particular regions in the universe become supercritical ( tornatore et al . 2007 ; maio et al . 2010 ) . absorption lines imprinted on the spectra of bright background sources , such as grbs or quasars , are one of the main sources of information about the physical and chemical properties of high - redshift systems ( oh 2002 ; furlanetto & loeb 2003 ; oppenheimer et al . these lines are due mainly to absorption by neutral hydrogen present in the low column - density ly@xmath16 forest , and by metals in low - ionization stages ( e.g. , , , , , ) which arise in the higher column - density gas associated with damped ly@xmath16 absorbers ( dlas ) . the analysis of the spectrum of distant grb 050904 ( totani et al . 2006 ) has resulted in a wealth of detailed insight into the physical conditions within the host galaxy at @xmath17 , and salvaterra et al . ( 2009 ) claimed that they identified two absorption lines ( and ) , although at poor signal - to - noise , in the spectrum of grb 090423 , the most distant spectroscopically confirmed burst at @xmath18 ( salvaterra et al . 2009 ; tanvir et al . grbs as background sources offer a number of advantages compared to traditional lighthouses such as quasars ( bromm & loeb 2012 ) . their number density drops much less precipitously than quasars at @xmath19 ( fan et al . 2006 ) , and the absence of a strong proximity effect , together with the near power - law character of their spectra , renders them ideal probes of the early igm . in this paper , we discuss the observational signatures of pop iii grbs and study pre - galactic metal enrichment utilizing absorption lines in the spectra of high-@xmath20 grbs which were imprinted by the first galaxies . recently , it has become feasible to study the formation of the first galaxies , including the metal enrichment from pop iii supernovae ( sne ) , with highly - resolved cosmological simulations ( wise & abel 2008 ; greif et al . we here place a bright grb into the simulation box of greif et al . ( 2010 ) , and derive the spectral signature as the afterglow light escapes from the first galaxy , thereby probing the partially enriched igm in its vicinity . the fluxes in the near - ir and radio bands may be detectable by the _ jwst _ and the very large array ( vla ) out to @xmath5 . the structure of this paper is as follows . in section 2 , we derive the circumburst density profile of pop iii grbs , followed by a brief description of the underlying first galaxy simulation ( section 3 ) . we discuss the properties of the grb afterglow in section 4 , and the metal absorption line diagnostics in section 5 , followed by our conclusions . two different environments are currently discussed as pop iii star formation sites , minihalos and atomic cooling halos ( bromm et al . the latter mode is often termed pop iii.2 to indicate physical conditions , such as a higher degree of ionization , that differs from the canonical minihalo case , possibly resulting in somewhat lower masses ( mckee & tan 2008 ) . for normal grbs , either a constant number - density profile , or a power - law dependence , as might be expected in the stellar wind from the progenitor , is discussed in the literature ( bloom 2011 ) . the grb afterglow emission sensitively depends on the circumburst density ( ciardi & loeb 2000 ; gou et al . 2004 ; inoue et al . 2007 ) , and this is one of the crucial uncertainties in making predictions for pop iii bursts . we next derive the likely circumburst conditions in the two possible pop iii host systems . for simplicity , we assume that pop iii grbs are triggered in minihalos close to the cooling threshold for collapse , which is only weekly dependent on redshift ( yoshida et al . 2003 ) . for definiteness , we choose @xmath21 . to determine the pre - burst density , we need to consider the build - up of the region around the central pop iii star , and how the density structure is modified through the strong photoionization - heating inside of it ( whalen et al . the production of ionizing photons strongly depends on the stellar mass , which in turn is determined by how the accretion flow onto the growing protostar proceeds under the influence of this radiation field ( e.g. , mckee & tan 2008 ; hosokawa et al . 2011 ; smith et al . 2011 ; stacy et al . thus , the assembly of pop iii stars and the development of an region around them proceed simultaneously and affect each other . the shallow potential wells in the host minihalos , with corresponding circular velocities of a few kms@xmath22 , are unable to retain photo - ionized gas , so that the gas is effectively blown out of the minihalo . the resulting photo - evaporation has been studied with 3d radiative transfer calculations ( alvarez et al . 2006 ; abel et al . 2007 ; greif et al . 2009 ) , where one massive pop iii star at the center of the minihalo acts as an embedded point source , that also take into account the hydrodynamic response of the photo - heated gas . it is possible to understand the key physics of the photoevaporation from minihalos with the self - similar solution for a champagne flow ( shu et al . 2002 ) . assuming a @xmath23 density profile , which describes the typical situation in minihalos outside a nearly flat inner core , one can reformulate the spherically symmetric continuity and euler equations for isothermal gas as follows : @xmath24{1 \over \alpha } { d \alpha \over d x } = \left[\alpha - { 2 \over x}(x - v ) \right](x - v ) , \label{alpha}\ ] ] @xmath24{d v \over d x } = \left[(x- v)\alpha - { 2 \over x } \right](x - v ) , \label{vel}\ ] ] where @xmath25 is the similarity variable , and @xmath26 and @xmath27 introduce the reduced density and velocity , respectively . @xmath28 is the sound speed of the ionized gas and @xmath29 the hydrogen mass fraction . the density globally decreases with time , after the ionizing source has turned on . to derive the immediate pre - explosion value , we will therefore set @xmath30 yr , the typical lifetime of a massive pop iii star ( e.g. , bromm et al . 2001b ) . different solutions are obtained depending on the ratio @xmath31 , where @xmath32 and @xmath28 are the initial and ionized isothermal sound speeds . we take @xmath33 k and @xmath34 , where the virial temperature is defined as @xmath35 the intuition behind the latter identification is that @xmath36 , the specific gravitational potential energy of the gas inside a halo , is a measure of how strongly the fluid is bottled up , before a champagne flow can occur . using table 1 in shu et al . ( 2002 ) , we can obtain the density profiles at different redshifts as follow . first , the value of @xmath37 can be calculated at the given redshift . we then derive the corresponding reduced central density , @xmath38 , via interpolation . for the given @xmath38 we can numerically solve equations ( [ alpha ] ) and ( [ vel ] ) subject to the inner boundary conditions : @xmath39 to avoid the singularity near @xmath40 , shu et al . ( 2002 ) provide the series expansion @xmath41 @xmath42 once the pop iii star has turned on at @xmath43 , the photo - heating commences , putting in place a pressurized inner bubble which leads to an outward - flowing shock at @xmath44 . the shock obeys the usual isothermal jump conditions , written in terms of the reduced variables as @xmath45 where the subscript @xmath46(@xmath47 ) indicates the value of the reduced velocity and density upstream ( downstream ) of the shock . in figure [ shu ] , we show the resulting density profiles at the end of the pop iii progenitor s life , and therefore at the completion of the photoionization - heating as well . as can be seen , the circumstellar densities have dropped significantly from the high values , @xmath0@xmath4 , present prior to pop iii star formation , and are nearly uniform at small radii . such a flat density profile is markedly different from that created by stellar winds . the absence of any wind around a pop iii star is indeed predicted , as a consequence of very low metallicity ( baraffe et al . 2001 ; kudritzki 2002 ) , and could serve as an indicator of a low-@xmath48 grb progenitor . the post - photoheating density evolves with redshift , approximately according to @xmath49 , normalized such that @xmath50@xmath4 at @xmath51 . this law differs from earlier estimates of the pop iii circumburst evolution , where no @xmath20-dependence was assumed , or a scaling with the density of the background universe , @xmath52 ( e.g. , gou et al . 2004 ) . this relation can be understood as follows . prior to the onset of protostellar collapse , the baryonic density profile in the minihalo can be described by the lane - emden equation for an isothermal sphere @xmath53 where @xmath54 is the density at which the gravitational potential @xmath55 is set to zero . a simple analytical solution to this equation is the well - known singular isothermal sphere ( sis ) : @xmath56 , where @xmath57 . this law approximately describes the profile found in realistic simulations of first star formation ( e.g. , yoshida et al . 2003 ) , outside an inner core of radius @xmath58pc , the jeans length of the primordial gas ( bromm et al . the baryonic mass inside the shock radius , @xmath59pc , is about @xmath60 at the end of photoionization - heating , the mass in the flat inner core is about @xmath61 if we neglect any outflows at large radii , we have approximate mass conservation , such that @xmath62 , resulting in @xmath63 , as before . our post - photoheating densities are consistent with simulation results ( kitayama et al . 2004 ; alvarez et al . 2006 ; abel et al . 2007 ; greif et al . 2009 , 2010 ) . we also agree with whalen et al . ( 2004 ) , if we take into account that their halo had a somewhat lower mass , @xmath64 . empirical support is provided in chandra et al . ( 2010 ) who found that the circumburst density of grb 090423 at @xmath18 was about @xmath65@xmath4 by fitting the radio , x - ray and infrared afterglow , although this burst almost certainly did not originate in a minihalo environment . star formation and radiative feedback inside atomic cooling halos is significantly less well - understood than the minihalo case ( e.g. , johnson et al . 2009 ; safranek - shrader et al . 2010 , 2012 ) , and there are no comparable high - resolution simulations yet . what is the character of star formation , in terms of imf and stellar multiplicity ( clark et al . 2011a ) ? given current uncertainty , we here again use the formalism of the shu solution as developed above . we further assume for simplicity that either a single pop iii star forms , or a small stellar cluster , with properties that would again heat the surrounding region to @xmath66k . we stress that the numbers derived here only provide us with rough guesses at best , and the hope is that simulations will eventually become available to firm them up . if a cluster of pop iii stars forms , the time where photoheating stops will be prolonged . we thus need to evaluate @xmath67 at @xmath68 , where @xmath69 is the timescale over which star formation is ongoing , and @xmath70 again is the lifetime of a single pop iii star . we can estimate @xmath71 , because disruptive feedback effects will tend to terminate star formation when the first member stars die . the resulting densities at the end of photoheating are shown in figure [ atomic ] . similar to the minihalo case , densities are nearly constant at small radii , but overall values are much higher . the reason is that here the potential wells are deeper , so that photoheated gas can more easily be retained . typical circumburst densities are @xmath72@xmath4 . because such large densities would correspond to high afterglow fluxes , pop iii grbs originating in atomic cooling halos may be extremely bright , rendering them visible out to very high redshifts . for some of the most distant bursts known , such large circumburst densities have indeed been inferred . e.g. , gou et al . ( 2007 ) have argued that fitting the grb 050904 afterglow at @xmath73 requires a circumburst density of a few @xmath74@xmath4 . similar values have been suggested for grb 080913 at @xmath75 ( greiner et al . 2009 ; zhang et al . in order to predict afterglow spectra from pop iii grbs , we consider explosions that are embedded in a realistic cosmological setting . specifically , we use the first galaxy simulation carried out by greif et al . ( 2010 ) , where the assembly of an atomic cooling halo was tracked , resolving all prior pop iii star formation in the progenitor minihalos , together with the radiative feedback from regions around those stars . in addition , the simulation allowed one of the pop iii progenitor stars to explode as an energetic sn , explicitly following the transport and mixing of the ejected metals into the igm . we here summarize a few key aspects of the simulation , and refer the reader to greif et al . ( 2010 ) for further details . the simulation is carried out in a cosmological box of size @xmath76 ( comoving ) , and is initialized at @xmath77 according to a concordance @xmath78 cold dark matter ( @xmath78cdm ) model with matter density @xmath79 , baryon density @xmath80 , hubble parameter @xmath81 , spectral index @xmath82 , and normalization @xmath83 ( spergel et al . density and velocity perturbations are imprinted at recombination with a gaussian distribution , and they are propagated to @xmath77 , when the simulation is started , by applying the zeldovich approximation . within the highest resolution region of our nested , zoomed - in initial conditions , the dm and gas particle masses are @xmath84 and @xmath85 , respectively , the baryonic mass resolution , roughly given by the mass contained inside an sph kernel , is @xmath86 , close to the relevant jeans mass in the primordial star forming gas . to capture the chemical evolution of the gas , the simulation follows the abundances of h , h@xmath87 , h@xmath88 , h@xmath89 , h@xmath90 , he , he@xmath87 , he@xmath91 , and e@xmath88 , as well as the five deuterium species d , d@xmath87 , d@xmath88 , hd and hd@xmath87 . all relevant cooling mechanisms are included , i.e. , h and he atomic line cooling , bremsstrahlung , inverse compton scattering , and collisional excitation cooling via h@xmath89 and hd ( johnson & bromm 2006 ) . in figure [ galsim ] , we show the hydrogen number density and metallicity averaged along the line of sight within the central @xmath92 kpc ( comoving ) at two different output times , from @xmath93 , briefly after the first star - forming minihalo has collapsed , to @xmath94 , closer to the virialization of the first galaxy . the distribution of metals produced by the first sn explosion is highly inhomogeneous , and the metallicity can reach up to @xmath95 , which is already larger than any of the values discussed for the critical metallicity , @xmath96 ( e.g. , bromm et al . 2001a ; bromm & loeb 2003 ; frebel et al . 2007 ) , including those predicted for a dust - driven pop iii / pop ii transition , where @xmath97 ( omukai et al . 2005 ; schneider et al . 2006 ) . the implication is that both pop iii and pop i / ii stars will form during the assembly of the first galaxies ( johnson et al . 2008 ; maio et al . 2010 ) , in turn giving rise to the simultaneous occurrence of pop iii and normal grbs at a given redshift ( bromm & loeb 2006 ; de souza et al . 2011 ) . here , we will focus on the case of a pop iii burst exploding in one of the ( still metal - free ) first galaxy progenitor minihalos at @xmath98 ( see section 5 ) . in figure [ tem ] , we show the temperature distribution in the vicinity of the explosion site , allowing us to calculate ( thermal ) line broadening in a realistic fashion . our modeling of the grb broad - band afterglow emission follows standard prescriptions , and we here only briefly summarize the basic physics involved ( for general reviews , see mszros 2006 ; zhang 2007 ) . we specifically consider a relativistic shell ejected from the progenitor of a grb ( e.g. , kobayashi 2000 ) . the shell has a rest mass of @xmath99 , an observed thickness of @xmath100 , and exhibits an initial lorentz factor of @xmath101 , where @xmath102 is the isotropic - equivalent explosion energy . the interaction between the shell and the surrounding medium with a density of @xmath103 produces a forward and a reverse shock . the shocks , in turn , accelerate electrons to high energies , distributed according to a power - law : @xmath104 for @xmath105 , where @xmath106 is the internal lorentz factor for the electrons . the minimum lorentz factor is set by assuming that a fraction @xmath107 of the post - shock energy is transferred to the electrons , resulting in : @xmath108 , where @xmath109 is the lorentz factor of the shocked fluid ( e.g. , blandford & mckee 1976 ; sari et al . if one also assumes that a fraction @xmath110 of the post - shock energy is deposited into random magnetic fields , the gyration of the shock - accelerated electrons around the magnetic field will generate synchrotron radiation , thus giving rise to the afterglow emission . there is then a second characteristic lorentz factor , @xmath111 , determining the threshold below which electrons do not lose a significant fraction of their energy to radiation . equating the synchrotron cooling time with the expansion time of the shell , one finds : @xmath112 , where @xmath113 is the synchrotron radiation power ( rybicki & lightman 1979 ) , and @xmath114 the expansion time , measured in the comoving fluid frame ( sari et al . 1998 ; panaitescu & kumar 2000 ) . therefore , the cooling lorentz factor can be written as : @xmath115 . hz ( m band ) as a function of observed time . we show the emission from the forward shock ( _ dashed line _ ) , the reverse shock ( _ solid red line _ ) , and their combination ( _ solid black line _ ) . we use the parameters given in the text , for a grb exploding in a minihalo with circumburst density @xmath116@xmath4 at @xmath117 . it can be seen that the reverse shock dominates during the first few hours , reaching a brightness in excess of @xmath118mjy.,scaledwidth=50.0% ] for the resulting spectrum , there are now two cases : in the fast cooling case ( @xmath119 ) , the flux can be calculated from @xmath120 whereas in the opposite , slow - cooling ( @xmath121 ) , case , one has @xmath122 the break frequencies correspond to the electron lorentz factors introduced above . the frequency @xmath123 describes the onset of synchrotron self - absorption , which is important at low ( radio ) frequencies . the afterglow emission exhibits a different character before and after the reverse shock ( rs ) has crossed the shell . this happens at a time @xmath124 cm , as measured in the observer frame . before rs crossing ( @xmath125 ) , we have for the forward shock ( `` @xmath126 '' ) : @xmath127 where @xmath128 is the luminosity distance . here and in the following we equate the pre - shock density with that established by photoionization - heating , at the time of the pop iii star s death : @xmath129 ( see section 2 ) . for the reverse shock ( `` @xmath130 '' ) , we have @xmath131 after rs crossing , we similarly have @xmath132 for the forward shock . for the reverse shock , we have @xmath133 the exact expressions , and further details , can be found in kobayashi ( 2000 ) and wu et al . note that the latter study assumes the presence of a circumburst wind profile , whereas we here consider the constant - density ( ism ) case , similar to kobayashi ( 2000 ) . the basic parameters of the afterglow emission region are : @xmath134 , @xmath135 , @xmath136 , @xmath137 , @xmath138 , and @xmath139 . we have verified that our afterglow predictions are consistent with those in gou et al . ( 2004 ) , and inoue et al . ( 2007 ) , as these papers focus on pop iii bursts at high redshifts as well . as an example , in figure [ lcm ] , we show the m - band light curve , corresponding to @xmath140hz . it can be seen that the reverse shock emission dominates during the first few hours , whereas the forward shock component takes over at later times . we here briefly summarize the key aspects of afterglows triggered by pop iii stars . figure [ lck ] shows the observed flux at @xmath141hz as a function of redshift for pop iii grbs in a minihalo . according to section 2.1 , circumburst densities can then be written as @xmath142 . the lines with filled dots , black triangles and open dots correspond to an observed time of @xmath143 minutes , 1 hour and 1 day respectively . the straight line marks the k - band sensitivity for the near - infrared spectrograph ( nirspec ) on _ jwst _ , estimated for a resolution of @xmath144 , a signal - to - noise ratio of s / n@xmath145 , and an integration time of 1 hour ( gardner et al . once the observed frequency corresponds to an emitted frequency above the ly@xmath16 resonance , @xmath146hz , which occurs at @xmath147 for the k - band , all flux will be completely absorbed by the intervening , still largely neutral igm . as can be seen , in the @xmath148 band the _ jwst _ will be able to detect grbs , and to conduct spectroscopy on their afterglows , out to @xmath149 even after 1 day . in the @xmath150 band , the redshift horizon is extended further still , to @xmath151 . at these frequencies , the afterglow flux is only weakly dependent on redshift . the main reason is that cosmic time dilation implies that a given observed time after the trigger corresponds to successively earlier emission times , where intrinsic afterglow luminosities are much brighter ( ciardi & loeb 2000 ; bromm & loeb 2012 ) . secondly , circumburst densities modestly increase with redshift ( see section 2 ) . in figure [ radio ] , we show the flux at @xmath152ghz , again after @xmath153 minutes , 1 hour and 1 day . we calculate the sensitivity of the vla , using the following expression ( ioka & mszros 2005 ) @xmath154 where @xmath155 is the integration time , @xmath156 the bandwith , and @xmath157@xmath158k@xmath22 . the bandwidth of the expanded vla ( evla ) can be up to 8ghz . for a 2h integration in the 8ghz band , the evla can thus reach a sensitivity of up to 6.4 @xmath159jy , which is shown as a horizontal line in figure [ radio ] . we can see that the radio afterglows are easily within reach for the evla , up to @xmath160 ( e.g. , chandra et al . 2010 ) . according to hierarchical cosmic structure formation , the characteristic halo mass varies as a function of redshift . thus , the typical observed peak flux and wavelength of pop iii grbs are also a function of redshift . in the following , we derive ballpark estimates for these key quantities by considering @xmath161 overdensities ( e.g. , figure 4 in clarke & bromm 2003 ) . this provides us with the typical halo mass , @xmath162 , at a given redshift . in the @xmath161 case , halo masses become less than @xmath163 at redshift @xmath164 . since this has dropped below the threshold mass for h@xmath165 cooling ( e.g. , yoshida et al . 2003 ) , pop iii stars could not form in such low - mass halos . for higher - sigma peaks , however , super - critical halos still exist at earlier times . therefore , for simplicity we adopt @xmath166 at @xmath164 . after the characteristic halo mass is thus determined , the virial temperature is fixed as well , and the ratio @xmath167 is known . the shu solution then gives the circumburst densities at a given redshift ( see section 2 ) , and grb afterglow spectra can finally be calculated . figure [ fluxz ] shows the observed peak flux at observed time @xmath153 minutes , 1 hour and 1 day at @xmath168 . the peak flux varies from milli - jansky to sub - jansky . figure [ lambdaz ] gives the observational peak wavelength at the same observed times for the same redshift range . the observed peak wavelengths vary from millimeter to mid - infrared bands . early afterglows are best observed with facilities such as the _ jwst _ , or possibly adaptive - optics supported next - generation , extremely - large telescopes on the ground . at late times , radio facilities provide the optimal chance for detection . among them is alma , which will allow to probe the afterglow in the mm and sub - mm regime , where the peak of the emission is located about 1 day after the explosion for @xmath169 sources . as we have seen above , pop iii grbs can in principle be detected out to very high redshifts . their afterglow spectra provide us with a unique probe into the state of the early igm , including its degree of ionization ( e.g. , totani et al . 2006 ) and metal enrichment . the latter will be encoded in absorption lines imprinted on the smooth afterglow emission with a signature that depends on the nature of the pop iii sne . here , we will consider for simplicity that prior to the grb only one nearby sn exploded beforehand , dispersing its complement of heavy elements into the otherwise pristine igm . such a scenario is consistent with the clustered nature of pop iii star formation , as is expected for the high-@xmath170 peak host systems ( e.g. , greif & bromm 2006 ) . specifically , we here explore nucleosynthetic yields for type ii core - collapse sne , and for pair - instability supernovae ( pisne ) . in table 1 , we summarize the average yields per sn , @xmath171 , for select elements that are expected to dominate the spectral signature . for type ii sne , we use the yields of woosley & weaver ( 1995 ) . these depend on the explosion energy , which introduces uncertainties of @xmath172 . for the pisn scenario , we utilize the yields of heger & woosley ( 2002 ; 2010 ) . table 1 also lists the properties of several key transitions , including their rest - frame wavelength , @xmath173 , and their oscillator strength , @xmath174 . we employ a simple model for the ionization structure of the metal - enriched gas . if the hydrogen in this region is substantially neutral , metals will reside in states typical of regions in the milky way ( , , , and for the elements of interest ) , because photons able to further ionize these elements will be absorbed by ( furlanetto & loeb 2003 ) . the ionization potential of is @xmath175 ev , and it remains locked in charge exchange equilibrium with ( oh 2002 ) . a modest extragalactic radiation background at frequencies below the lyman limit could easily maintain these metals in a singly - ionized state ( e.g. , furlanetto & loeb 2003 ) . we thus assume that c , o , si , and fe are predominantly in the ionization states typical for galactic clouds . employing comoving coordinates , the equation of cosmological radiative transfer is ( e.g. , abel et al . 1999 ) : @xmath176 where @xmath177 is the intensity of the radiation field , @xmath178 a unit vector along the direction of the ray , @xmath179 the hubble constant , and @xmath180 the ratio of cosmic scale factors separated by @xmath181 . here @xmath182 and @xmath183 denote the emission and absorption coefficient , respectively . equation ( [ equ : crt ] ) can be simplified if the effects of cosmic expansion are negligible . this will be the case if the light - crossing time , @xmath184 , over the region of interest , here the ( physical ) size of our simulation box , @xmath185 , is small compared to the hubble time , @xmath186yr@xmath187^{-3/2}$ ] . since @xmath188 , the transfer equation reverts to its non - cosmological form : @xmath189 we further neglect any diffuse ( re-)emission of radiation , such that @xmath190 , and assume near steady - state conditions ( @xmath191 ) , resulting in the static transfer equation : @xmath192 if @xmath193 is the intensity of the central afterglow , approximately treated as a point source , we have the standard solution : @xmath194 , where @xmath195 is the optical depth along the line - of - sight ( los ) . we can obtain the optical depth as follows . for a ( physical ) differential distance element , @xmath196 , along the given los , we have : @xmath197 , where @xmath198 is the frequency dependent absorption cross - section , and @xmath199 the number density of species x ( e.g. , , , , , ) . in the above equation , @xmath200 is the doppler width with parameter @xmath201 . the voigt function , @xmath202 , is given by @xmath203 here , the relative strength of natural ( damping ) to doppler broadening is described by the parameter @xmath204 where @xmath205 is the damping constant for transition @xmath206 ( morton & smith 1973 ) . the variables @xmath207 and @xmath208 are the frequency difference relative to the line center in doppler units , and the normalized particle velocity , with respect to the grb location , respectively . the total los value of the optical depth is then : @xmath209 . we note that we can neglect any wavelength differences between the grb source and the metal absorber , because of the very small expansion - generated redshift on the scale of our simulation box . specifically , we estimate that @xmath210yr)@xmath211 , where we assume @xmath169 , and @xmath212yr . we extract the optical depths along random lines of sight with the ray - tracing algorithm of greif et al . ( 2009 ) . we choose @xmath213 and @xmath214 in the spherical coordinates , which corresponds to 51200 rays . for each ray , we use 200 logarithmically - spaced radial bins , and approximate the optical depth integral with a sum along the los . the pop iii grb will likely explode in a still substantially neutral igm . any flux shortward of the ly@xmath16 resonance is therefore completely absorbed , and because of the corresponding very large optical depth , there will be absorption in the red damping wing as well . we model this with the analytical formula given by miralda - escud ( 1998 ) , using @xmath215 and standard cosmological parameters ( komatsu et al . the line shape is not very sensitive to the choice of @xmath216 ( bromm et al . 2001b ) . cccccc c & 0.1 m@xmath217 & 4.1 m@xmath217 & & 1334.5 & 0.1278 + & & & & 1548.2 & 0.1908 + & & & & 1550.8 & 0.09522 + o & 0.5 & 44 & & 1302.2 & 0.04887 + si & 0.06 & 16 & & 1304.4 & 0.094 + & & & & 1393.8 & 0.514 + & & & & 1402.8 & 0.2553 + fe & 0.07 & 6.4 & & 1608.5 & 0.058 + & & & & 2344.2 & 0.114 + & & & & 2382.8 & 0.300 + [ cols=">,^,^,^,^ " , ] s. we show the spectral region redshifted into the near ir , accessible to the nircam and nirspec instruments on board the _ jwst_. metal absorption lines are imprinted according to the pop iii sn event , pisn vs. core - collapse ( type ii ) . the former originates from a very massive star ( vms ) progenitor , whereas the latter from a less massive one . in each case , the cutoff at short wavelengths is due to lyman-@xmath16 scattering in the neutral igm . , scaledwidth=50.0% ] s. we use the metal yields in the vms case . the observed equivalent widths ( ews ) are also shown in the figure . if multiple lines are present , the given ews belong to the strongest lines in each panel . , scaledwidth=50.0% ] , but displaying metal absorption lines for a normal ( scalo ) pop iii imf . individual ews are again given for the strongest lines in each panel.,scaledwidth=50.0% ] in figure [ totalspetc ] , we show the part of the resulting afterglow spectrum that has been redshifted into the near - ir , where the _ jwst _ is most sensitive . for the two assumptions regarding the pop iii imf , top - heavy ( vms ) and more normal ( scalo ) , the spectra are depicted at the reverse shock crossing time . the metal absorption lines are imprinted on the smooth underlying grb afterglow spectrum , rendering their identification straightforward . the cutoff at short wavelenghts is due to lyman-@xmath16 scattering in the igm which is expected to be still completely neutral at @xmath12 . for detailed predictions , it is advantageous to quantify the strength of absorption lines through equivalent widths ( ews ) . the observed ews can be calculated from @xmath218d\lambda \mbox{\ .}\ ] ] in figures [ ewvmstc ] and [ ewsnetc ] , we zoom in on the vicinity of the strongest lines , again at the reverse shock crossing time for both imf cases . in addition , we explicitly give the derived ews , which are about a few ten for a top - heavy imf , and an order of magnitude smaller for a normal ( scalo ) imf . the ew values do not change significantly , if evaluated at 1day after the explosion . such line strengths are well within reach of the nirspec instrument on board the _ jwst _ , which has a spectral resolution up to @xmath219 , and a line sensitivity of @xmath220ergs@xmath22@xmath221 ( gardner et al . 2006 ) . in general , the limiting equivalent width of an unresolved line at observed wavelength @xmath222 , at @xmath223 significance , is ( tumlinson et al . 2002 ) : @xmath224 where s / n is the signal - to - noise ratio per resolution element . within our assumptions , therefore , spectroscopy of pop iii grb afterglows will be able to distinguish between different enrichment events , pisn vs. core - collapse . there is , however , a possible complication . the total enrichment level would be rather similar to that from one pisn for the case of 10 type ii sne . such a situation is not implausible , given the number of pop iii star formation sites , i.e. , minihalos , in the vicinity of the pop iii grb ( see sec . however , we could then still discriminate between the cases using absorption line flux ratios , which remain quite different . in table 2 , we list ranges for select flux ratios , as encountered along different loss . we consider the ratios /x , where x denotes another species , at both the reverse shock crossing time and at 1 day after the explosion . in the former case , there is an order of magnitude difference between the two imfs , allowing a robust identification . at later times , however , this clear - cut signature disappears again . in figure [ linesta ] , we show the cumulative ew distribution for 100 randomly selected sightlines at redshift @xmath94 . again , the resulting distributions are quite different for the two imf cases . it is , however , important to note that the statistical significance of this analysis is limited by the fact that it is based on only one simulation , and only one grb afterglow . in reality , observations will sample multiple explosion sites with respective los distributions that sample many clouds at various evolutionary stages along the way . our analysis should still provide a representative view of the typical situation , whereas a more complete study has to await the further increase in available computational power . in this paper , we develop a diagnostic to study pre - galactic metal enrichment in the vicinity of the first galaxies . specifically , we utilize the bright afterglow of a pop iii grb as a featureless background source , and calculate the strength of metal absorption lines that are imprinted by the first heavy elements produced by pop iii sne . to approximately derive the metal absorption line statistics , we use an existing highly - resolved simulation of the formation of a first galaxy which is characterized by the onset of atomic hydrogen cooling in a halo with virial temperature @xmath0k . the reverse shock initially dominates the afterglow flux a few hours after explosion , followed by the forward shock emission later on . the fluxes in the near - ir and radio bands may be detectable with the _ jwst _ and the vla out to @xmath225 . we predict that the afterglow emission peaks from near - ir to millimeter bands with peak fluxes from mjy to jy at different observed times . metal absorption lines in the grb afterglow spectrum , giving rise to ews of a few tens of , may allow us to distinguish whether the first heavy elements were produced in a pop iii star that died as a pisn , or a core - collapse sn . to this extent , the spectrum needs to be obtained sufficiently early , within the first few hours after the trigger . the absorption signature of the first sn events might allow us to constrain the underlying mass scale of pop iii stars . this is the key input parameter to predict their evolution , nucleosynthetic yields , and modes of death , which in turn determine how the first stars impact subsequent cosmic history . the strength of the associated stellar feedback governs the assembly process of the first galaxies , in the sense that the stronger feedback from more massive stars shifts their formation to later stages in the hierarchical build - up of structure ( e.g. , ricotti et al . 2002 ; greif et al . 2010 ; frebel & bromm 2012 ; ritter et al . 2012 ; wise et al . 2012 ) . it will , however , be very challenging to directly probe the initial epoch of cosmic star formation . the reason is that even the _ jwst _ will not be able to detect individual first stars , but instead only more massive stellar systems or clusters that form later on in more massive host systems ( e.g. , pawlik et al . pop iii grbs may thus afford us one of the few direct windows into the crucial epoch of first light , another one being the extremely energetic pair - instability sn or hypernova explosions that are predicted to end the lives of the most massive pop iii stars ( e.g. , pan et al . 2012 ; hummel et al . the promise provided by the combination of a wide field grb trigger mission , such as janus or lobster , with the next - generation of highly - sensitive near - ir telescopes , such as the _ jwst _ or the planned ground - based extremely large facilities ( the e - elt , gmt , and tmt ) , is huge . grbs are likely to play a key role in finally opening up the high - redshift frontier , all the way back to the very beginning of star and black hole formation . we are indebted to anna frebel and steve finkelstein for helpful discussions . this work is supported by the national natural science foundation of china ( grants 11103007 and 11033002 ) . v. b. acknowledges support from nsf grant ast-1009928 and nasa atfp grant nnx09aj33 g . k. s. c. is supported by the grf grants of the government of the hong kong sar under hku 7011/09p . the simulations presented here were carried out at the texas advanced computing center ( tacc ) . abel , t. , norman , m. l. , & madau , p. 1999 , apj , 523 , 66 abel , t. , bryan , g. l. , & norman , m. l. 2002 , science , 295 , 93 abel , t. , wise , j. h. , & bryan , g. l. 2007 , apj , 659 , l87 alvarez , m. a. , bromm , v. , & shapiro , p. r. 2006 , apj , 639 , 621 baraffe , i. , heger , a. , & woosley , s. e. 2001 , apj , 550 , 890 barkana , r. & loeb , a. 2001 , phys . , 349 , 125 barkana , r. & loeb , a. 2004 , apj , 601 , 64 belczynski , k. , bulik , t. , heger , a. , & fryer , c. 2007 , apj , 664 , 986 blandford r. d. , & mckee c. f. 1976 , phys . fluid , 19 , 1130 bloom , j. s. 2011 , what are gamma - ray bursts ? ( princeton : princeton univ . press ) bromm , v. , coppi , p. s. , & larson , r. b. 1999 , apj , 527 , l5 bromm , v. , coppi , p. s. , & larson , r. b. 2002 , apj , 564 , 23 bromm , v. , ferrara , a. , coppi , p. s. , & larson , r. b. 2001a , mnras , 328 , 969 bromm , v. , kudritzki , r. p. , & loeb , a. 2001b , apj , 552 , 464 bromm , v. , & larson , r. b. 2004 , ara&a , 42 , 79 bromm , v. , & loeb , a. 2002 , apj , 575 , 111 bromm , v. , & loeb , a. 2003 , nature , 425 , 812 bromm , v. , & loeb , a. 2006 , apj , 642 , 382 bromm , v. , & loeb , a. 2012 , in gamma - ray bursts , ed . c. kouveliotou , s. e. woosley & r. a. m. j. wijers ( cambridge : cambridge univ . press ) , arxiv:0706.2445v2 bromm , v. , & yoshida , n. 2011 , ara&a , 49 , 373 bromm , v. , yoshida , n. , hernquist , l. & mckee , c. f. 2009 , nature , 459 , 49 campisi , m. a. , maio , u. , salvaterra , r. , & ciardi , b. 2011 , mnras , 416 , 2760 cen , r. 2003 , apj , 591 , 12 chandra , p. , et al . 2010 , apj , 712 , l31 chary , r. , berger e. & cowie l. 2007 , apj 671 , 272 ciardi , b. , & loeb , a. 2000 , apj , 540 , 687 ciardi , b. , & ferrara , a. 2005 , space sci . , 116 , 625 clark , p. c. , glover , s. c. o. , klessen , r. s. , & bromm , v. 2011a , apj , 727 , 110 clark , p. c. , glover , s. c. o. , smith , r. j. , greif , t. h. , klessen , r. s. , & bromm , v. 2011b , science , 331 , 1040 clarke , c. j. , & bromm , v. 2003 , mnras , 343 , 1224 cucchiara , a. , et al . 2011 , apj , 736 , 7 dai , z. g. , liang , e. w. , & xu , d. 2004 , apj , 612 , l101 delia , v. , campana , s. , covino , s. , davanzo , p. , piramonte , s. , & tagliaferri , g. 2011 , mnras , 418 , 680 de souza , r. s. , yoshida , n. , & ioka , k. 2011 , a&a , 533 , a32 elliott , j. , greiner , j. , khochfar , s. , schady , p. , johnson , j. l. , & rau , a. 2012 , a&a , 539 , a113 fan , x. , carilli , c. l. , & keating , b. 2006 , ara&a , 44 , 415 frebel , a. , johnson , j. l. , & bromm , v. 2007 , mnras , 380 , l40 frebel , a. , johnson , j. l. , & bromm , v. 2009 , mnras , 392 , l50 frebel , a. , & bromm , v. 2012 , apj , in press ( arxiv:1010.1261 ) furlanetto , s. r. , & loeb , a. 2003 , apj , 588 , 18 furlanetto , s. r. , & loeb , a. 2005 , apj , 634 , 1 gallerani , s. , salvaterra , r. , ferrara , a. , & choudhury , t. r. 2008 , mnras , 388 , l84 gardner , j. p. , et al . 2006 , space sci . , 123 , 485 gou , l. j. , mszros , p. , abel , t. & zhang , b. 2004 , apj , 604 , 508 gou , l. j. , fox , d. b. & mszros , p. 2007 , apj , 668 , 1083 greif , t. h. , & bromm , v. 2006 , mnras , 373 , 128 greif , t. h. , johnson , j. l. , klessen , r. s. & bromm , v. 2008 , mnras , 387 , 1021 greif , t. h. , johnson , j. l. , klessen , r. s. & bromm , v. 2009 , mnras , 399 , 639 greif , t. h. , glover , s. c. o. , bromm , v. , & klessen , r. s. 2010 , apj , 716 , 510 greif , t. h. , springel , v. , white , s. d. m. , glover , s. c. o. , clark , p. c. , smith , r. j. , klessen , r. s. , & bromm , v. 2011 , apj , 737 , 75 greif , t. h. , bromm , v. , clark , p. c. , glover , s. c. o. , smith , r. j. , klessen , r. s. , yoshida , n. , & springel , v. 2012 , mnras , 424 , 399 greiner , j. , et al . 2009 , apj , 693 , 1610 haiman , z. , thoul , a. a. , & loeb , a. 1996 , apj , 464 , 523 heger , a. , & woosley , s. e. 2002 , apj , 567 , 532 heger , a. , & woosley , s. e. 2010 , apj , 724 , 341 hjorth , j. , et al . 2003 , nature , 423 , 847 hosokawa , t. , omukai , k. , yoshida , n. , & yorke , h. w. 2011 , science , 334 , 1250 hummel , j. a. , pawlik , a. h. , milosavljevi , m. , & bromm , v. 2012 , apj , in press ( arxiv:1112.5207 ) inoue , s. , omukai , k. , & ciardi , b. 2007 , mnras , 380 , 1715 ioka , k. , & mszros , p. 2005 , apj , 619 , 684 johnson , j. l. , & bromm , v. 2006 , mnras , 366 , 247 johnson , j. l. , greif , t. h. , & bromm , v. 2008 , mnras , 388 , 26 johnson , j. l. , greif , t. h. , bromm , v. , klessen , r. s. , & ippolito , j. 2009 , mnras , 399 , 37 karlsson , t. , bromm , v. , & bland - hawthorn , j. 2012 , rev . , in press ( arxiv:1101.4024 ) kitayama , t. , yoshida , n. , susa , h. , & umemura , m. 2004 , apj , 613 , 631 kobayashi , s. 2000 , apj , 545 , 807 komatsu , e. , et al . 2011 , apjs , 192 , 18 kudritzki , r. p. 2002 , apj , 577 , 389 komissarov , s. s. , & barkov , m. v. 2010 , mnras , 402 , l25 lamb , d. q. , & reichart , d. e. 2000 , apj , 536 , 1 loeb , a. 2010 , how did the first stars and galaxies form ? ( princeton : princeton univ . press ) macfadyen , a. i. , woosley , s. e. , & heger , a. 2001 , apj , 550 , 410 mackey , j. , bromm , v. , & hernquist , l. 2003 , apj , 586 , 1 madau , p. , ferrara , a. , & rees , m. j. 2001 . apj , 555 , 92 maio , u. , ciardi , b. , dolag , k. , tornatore , l. , & khochfar , s. 2010 , mnras , 407 , 1003 mckee , c. f. , & tan , j. c. 2008 , apj , 681 , 771 mszros , p. 2006 phys . , 69 , 2259 mszros , p. & ress , m. j. 2010 , apj , 715 , 967 miralda - escud . , j. 1998 , apj , 501 , 15 morton , d. c. , & smith , w. h. 1973 , apjs , 26 , 333 oh , s. p. 2002 , mnras , 336 , 1021 oh , s. p. , & haiman , z. 2002 , apj , 569 , 558 omukai , k. , tsuribe , t. , schneider , r. , & ferrara , a. 2005 , apj , 626 , 627 oppenheimer , b. d. , dav , r. , & finlator , k. 2009 , mnras , 396 , 729 pan , t. , kasen , d. , & loeb , a. 2012 , mnras , 422 , 2701 panaitescu , a. , & kumar , k. 2000 , apj , 543 , 66 pawlik , a. h. , milosavljevi , m. , & bromm , v. 2011 , apj , 731 , 54 porciani , c. , & madau , p. 2001 , apj , 548 , 522 qian , y. z. , & wasserburg , g. j. 2001 , apj , 559 , 925 ricotti , m. , gnedin , n. y. , & shull , j. m. 2002 , apj , 575 , 33 ritter , j. s. , safranek - shrader , c. , gnat , o. , milosavljevi , m. , & bromm , v. 2012 , apj , submitted ( arxiv:1203.2957 ) rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : wiley ) safranek - shrader , c. , bromm , v. , & milosavljevi , m. 2010 , apj , 723 , 1568 safranek - shrader , c. , agarwal , m. , federrath , c. , dubey , a. , milosavljevi , m. , & bromm , v. 2012 , mnras , submitted ( arxiv:1205.3835 ) salvaterra , r. , et al . 2009 , nature , 461 , 1258 sari r. , piran t. , & narayan r. 1998 , apj , 497 , l17 schneider , r. , ferrara , a. , natarajan , p. , & omukai , k. 2002 , apj , 571 , 30 schneider , r. , omukai , k. , inoue , a. k. , & ferrara , a. 2006 , mnras , 369 , 1437 shu , f. h. , lizano , s. , galli , d. , cant , j. , & laughlin , g. 2002 , apj , 580 , 969 smith , r. j. , glover , s. c. o. , clark , p. c. , greif , t. h. , & klessen , r. s. 2011 , mnras , 414 , 3633 spergel , d. n. , et al . 2003 , apjs , 148 , 175 stacy , a. , greif , t. h. , & bromm , v. 2010 , mnras , 403 , 45 stacy , a. , bromm , v. , & loeb , a. 2011 , mnras , 413 , 543 stacy , a. , greif , t. h. , & bromm , v. 2012 , mnras , 422 , 290 stanek , k. z. , et al . 2003 , apj , 591 , l17 suwa , y. , & ioka , k. 2011 , apj , 726 , 107 tanvir , n. r. , et al . 2009 , nature , 461 , 1254 tegmark , m. , silk , j. , rees , m. j. , blanchard , a. , abel , t. , & palla , f. 1997 , apj , 474 , 1 toma , k. , sakamoto , t. & mszros , p. 2011 , apj , 731 , 127 tornatore , l. , ferrara , a. & schneider , r. 2007 , mnras , 382 , 945 totani , t. 1997 , apj , 486 , l71 totani , t. , et al . 2006 , pasj , 58 , 485 tumlinson , j. , et al . 2002 , apj , 566 , 857 tumlinson , j. 2010 , apj , 708 , 1398 wang , f. y. & dai , z. g. 2009 , mnras , 400 , l10 wang , f. y. & dai , z. g. 2011 , apj , 727 , l34 wang , f. y. , qi , s. & dai , z. g. 2011 , mnras , 415 , 3423 whalen , d. , abel , t. , & norman , m. l. 2004 , apj , 610 , 14 wijers , r. a. m. j. , bloom , j. s. , bagla , j. s. , & natarajan , p. 1998 , mnras , 294 , l13 wise , j. h. , & abel , t. 2007 , apj , 665 , 899 wise , j. h. , & abel , t. 2008 , apj , 685 , 40 wise , j. h. , turk , m. j. , norman , m. l. , & abel , t. 2012 , apj , 745 , 50 woosley , s. e. 1993 , apj , 405 , 273 woosley , s. e. & bloom , j. s. 2006 , ara&a , 44 , 507 woosley , s. e. , & weaver , t. a. 1995 , apjs , 101 , 181 wu , x. f. , dai , z. g. , huang , y. f. , & lu , t. 2003 , mnras , 342 , 1131 wyithe , j. s. b. , & loeb , a. 2003 , apj , 586 , 693 yoshida , n. , abel , t. , hernquist , l. , sugiyama , n. 2003 , apj , 592 , 645 yksel , h. et al . 2008 , apj , 683 , l5 zhang , b. 2007 , chin . j. astron . astrophys . , 7 , 1 zhang , b. , et al . 2009 , apj , 703 , 1696

new ccd photometric observations of the eclipsing system ar boo were obtained from february 2006 to april 2008 . the star s photometric properties are derived from detailed studies of the period variability and of all available light curves . we find that over about 56 years the orbital period of the system has varied due to a combination of an upward parabola and a sinusoid rather than in a monotonic fashion . mass transfer from the less massive primary to the more massive secondary component is likely responsible for at least a significant part of the secular period change . the cyclical variation with a period of 7.57 yrs and a semi - amplitude of 0.0015 d can be produced either by a light - travel - time effect due to an unseen companion with a scaled mass of=0.081 or by a magnetic period modulation in the secondary star . historical light curves of ar boo , as well as our own , display season - to - season light variability , which are best modeled by including both a cool spot and a hot one on the secondary star . we think that the spots express magnetic dynamo - related activity and offer limited support for preferring the magnetic interpretation of the 7.57-year cycle over the third - body understanding . our solutions confirm that ar boo belongs to the w - subtype contact binary class , consisting of a hotter , less massive primary star with a spectral type of g9 and a companion of spectral type k1 .

0905.4094

in the vicinity of the globular cluster m3 , ar boo ( gsc 1999 - 0011 ; @xmath2=@xmath312.76 , @xmath4@xmath2=@xmath30.89 ) was announced by kurochkin ( 1960 ) to be an eclipsing binary with an asserted period of 0.416718 d. houck & pollock ( 1986 ) obtained a photographic light curve and determined a new period of 0.3447 d. kurochkin s period turned out to be incorrect probably because there were only 97 magnitude estimates over a ten - year interval . since then , ccd light curves have been made by wolf et al . ( 1998 ) and by samec et al . ( 2006 ) in the @xmath5 and @xmath6 bandpasses , respectively . the former authors reported that the orbital period history of the system can be represented either by an abrupt period increase about 1983 or by a positive quadratic term indicating a continuous period increase . samec et al . analyzed their light curves using the wilson - devinney synthesis code ( wilson & devinney 1971 , hereafter wd ) and classified the binary as a member of the w - subtype cateory defined empirically by binnendijk ( 1970 ) . such binaries show light curves in which the deeper minimum corresponds to an occultation of the smaller , less massive component by the larger , more massive one . in addition , these latter authors suggested that a second - order light variation is due to a single hot spot on the more massive component and that the orbital period is increasing . in this article , we present improved photometric properties of ar boo from detailed studies of the period and light variations , based on our new ccd observations as well as on historical data . we performed time - series ccd photometry of ar boo on 27 nights between february 2006 and april 2008 , using an fli img4301e ccd camera and a @xmath7 filter set attached to the 1.0-m reflector at the mt . lemmon optical astronomy observatory ( loao ) in arizona , usa . the detecting system had been installed by the korea astronomy and space science institute ( kasi ) and was operated by remote control from daejeon , korea via an internet connection ( han et al . the ccd chip has 2084@xmath82084 pixels and a pixel size of 24 @xmath9 m . the image field - of - view is 22@xmath10.2@xmath822.2 at the f/7.5 cassegrain focus of the telescope . with the conventional iraf package , we processed the ccd frames to correct for bias level , dark noise , and pixel - to - pixel inhomogeneities of quantum efficiency and applied simple aperture photometry to obtain instrumental magnitudes for about a hundred stars near to the variable . the nearby stars gsc 1999 - 0009 and gsc 1999 - 0067 ( on the same ccd frame as the variable ) were selected as comparison ( c ) and check ( k ) stars , respectively . for these stars no light variability had been reported in the previous observations by samec et al . and we also detected no variations in the brightness difference between them during our observing runs . the observations of our first season were dedicated to obtaining new times of minimum light . from the 2007 and 2008 seasons , a total of 1838 individual observations was obtained in the three bandpasses ( 617 in @xmath4 , 619 in @xmath2 , and 602 in @xmath11 ) and a sample of them is listed in table 1 . the light curves are displayed in the upper panel of figure 1 as magnitude differences _ versus _ orbital phase , where the open circle and plus symbols show the individual observations of the 2007 and 2008 seasons , respectively . the differences ( 07-08 ) between the two seasons are plotted in the middle panel . to obtain these seasonal differences , we interpolated in phase and magnitude difference between phase - consecutive measures of the 2008 season to the phase of an appropriate 2007 observation and performed the subtraction . all ( k c ) magnitude differences appear in the lower panel . whereas the reference stars were evidently constant , ar boo has clearly varied seasonally . from our ccd observations , times of minimum light in each filter were determined by the method of kwee & van woerden ( 1956 ) . twenty - two weighted mean timings from these determinations and 84 epochs ( 12 photographic plate , 13 visual , 3 photographic , and 56 ccd ) from the literature were used for a period analysis of ar boo . most of the eclipse timings were extracted from the data base referenced by kreiner et al . all available ccd timings are listed in table 2 , where the second column gives the standard deviation of each timing . because the other timings were published with no errors , the following standard deviations were assigned to timing residuals based on observational method : @xmath120.0130 d for photographic plate , @xmath120.0096 d for visual , and @xmath120.0012 d for photographic minima . for ephemeris computations , relative weights were then calculated as the inverse squares of these values consistent with errors and weights for the ccd timings . as previously mentioned , the period variation of ar boo has been interpreted to be due to either a sudden period jump or a continuous period increase . after testing several ephemeris forms , we found that the history of the period shows a short - term oscillation superposed on a upward parabolic variation , rather than varying in monotonic or bi - linear fashions . this type of period variation appears in contact and near - contact binaries such as er ori ( kim et al . 2003 ) , bx peg ( lee et al . 2004 ) , tu boo ( lee et al . 2007 ) , and v432 per ( lee et al . 2008 ) , for which orbital period changes of the systems can be satisfactorily explained as combinations of secular variations and light - travel time ( hereafter ltt ) effects . in general , there also exists the possibility that a binary orbit has shrunk from an initially wider dimension by secularly transferring the angular momentum of an inner , close pair to a distant companion of a triple system ( cf . pribulla & rucinski 2006 ) . a recent study by these authors shows that about 42% of 151 known contact binaries brighter than @xmath13 + 10 exist in multiple systems . we , therefore , initially assumed the oscillation to be due to an ltt effect and all eclipse timings were fitted to a quadratic _ plus _ ltt ephemeris as follows : @xmath14 where @xmath15 is the ltt due to a hypothetical distant companion to the eclipsing close binary ( irwin 1952 , 1959 ) and includes five parameters ( @xmath16 , @xmath17 , @xmath18 , @xmath19 and @xmath20 ) . here , @xmath16 , @xmath17 and @xmath18 are the orbital parameters of the eclipsing pair around the mass center of the triple system . the parameters @xmath19 and @xmath20 denote keplerian mean motion of the mass center of the eclipsing pair and the epoch of its periastron passage , respectively . the levenberg - marquart algorithm ( press et al . 1992 ) was used to evaluate the eight parameters of the ephemeris and the final results are summarized in table 3 , together with other derived quantities . the @xmath21@xmath22 residuals calculated from the linear terms in equation ( 1 ) are plotted in the upper panel of figure 2 . the middle and lower panels display the ccd residuals from the linear _ plus _ quadratic term and from the complete ephemeris , respectively . the @xmath21@xmath23 residuals are given in the fourth column of table 2 . as displayed in figure 2 , the quadratic _ plus _ ltt ephemeris currently provides a good representation of all the @xmath21@xmath22 residuals . this cyclical variation can be interpreted as an ltt effect due to an additional object with a mass of at least @xmath0=0.081 @xmath1 . the minimum mass of the third companion is close to the theoretical limit of about 0.072 @xmath1 for a brown dwarf star . because of noise in even the recent timings and also because only about 1.5 cycles of the 7.57-year period have been covered , these results are certainly preliminary . this interpretation is not unique . it is alternatively possible that a cyclical period modulation may be caused by a magnetic activity cycle in a binary system with at least one star having a spectral type later than about f5 . applegate ( 1992 ) and lanza et al . ( 1998 ) have both discussed this possibility . the latter reference constrains and elaborates the concepts of the former one but both developments remains too conservative for a binary such as ar boo . they are limited to considering detached objects such as rs cvn and ar lac and , for the present system , quadrupole gravitational moments do not express the complex mass distributions of the ar boo stars . if this concern is ignored , curve - fitting of the period residuals remains the same as for the mechanical model just described above , but no orbital meaning is attached to the shape and phasing of the curve . in order to calculate model parameters for magnetic activity , we applied the period ( @xmath24 ) and amplitude ( @xmath25 ) for the cyclical variation to applegate s formulae . the results are presented in table 4 , where @xmath26 , @xmath27 , @xmath28 , @xmath29 , @xmath30 , @xmath31 , @xmath32 , and @xmath4 denote the amplitude of orbital period modulation , the variation of the gravitational quadrupole moment , the rate of angular momentum transfer , the moment of inertia of the convective shell , the variable fraction of differential rotation , the work needed to transfer the angular momentum , the rms luminosity variation and mean subsurface magnetic field intensity , respectively . in addition to these paramaters , the quantity @xmath33 denotes a bolometric magnitude difference relative to the mean light level of ar boo converted to magnitude scale with equation ( 4 ) in the paper of kim et al . the light variation ( @xmath34 ) ascribed to the hotter , primary star is a factor of 5 greater than the value ( @xmath35 ) proposed by applegate , while that ( @xmath36 ) for the secondary is close to his theoretical value . ( actually , lanza et al . suggest that applegate s limiting values may be relaxed considerably . ) these results indicate that a magnetic dynamo seated in the secondary star can also adequately explain the observed period modulations of ar boo . distinguishing between the mechanical and magnetic interpretations of the 7.57-year oscillation requires additional information , which will be suggested later after the light curve fittings . in equation ( 1 ) , the coefficient of the quadratic term is positive and represents a continuous period increase with a rate of @xmath37/@xmath38 = 2.68@xmath810@xmath39 d yr@xmath40 . it is conventional to ascribe such a variation to conservative mass transfer between the stellar components in the system . the transfer rate from the less massive star to the more massive component of ar boo is calculated to be about 1.48@xmath810@xmath39 m@xmath41 yr@xmath40 from the relation between @xmath37/@xmath38 and the masses of the component stars calculated later . this value is larger than those recently derived for other w - subtype contact systems [ e.g. , 1.3@xmath810@xmath39 for ad cnc ( qian et al . 2007 ) , 2.9@xmath810@xmath42 for ty boo ( yang et al . 2007 ) , and 2.3@xmath810@xmath42 for cc com and 2.0@xmath810@xmath42 for bv dra ( yang et al . 2009 ) , all in m@xmath41 yr@xmath40 units ] and is a generally accepted one for classical semi - detached algols ( cf . richard & albright 1999 ) . one might imagine that the calculated @xmath37/@xmath38 could be apportioned between ( a ) mass transfer from the hot to the cool eclipsing member and ( b ) angular momentum transfer to the possible third companion but there is no independent support for this idea . as do previous light curves of ar boo , the new ones show continuously curving eclipses and also conspicuous seasonal light variations . within about 0.004 mag , the light curves of 2007 present equal light levels at the quadratures ( max i and max ii ) while those of 2008 show the inverse oconnell effect with max i fainter than max ii by about 0.028 , 0.023 , and 0.018 mag for the @xmath4 , @xmath2 , and @xmath11 bandpasses , respectively . moreover , the light maxima are slightly displaced to orbital phases 0.264 and 0.745 in 2007 and 0.257 and 0.745 in 2008 . these features are usually indicative of spot activity on the component stars and the striking seasonal light changes in ar boo indicate that that activity has varied significantly with time . photometric solutions of ar boo have been reported only by samec et al . ( 2006 ) and no spectroscopic observations have been published for the system . we analyzed our data in a manner similar to that for the cool binaries tu boo ( lee et al . 2007 ) and v432 per ( lee et al . 2008 ) . our light curves were normalized to unit light at phase 0.75 and were solved by using contact mode 3 of the wd code . the surface temperature of the more massive star was fixed at 5100 k from flower s ( 1996 ) table , according to ( @xmath43)=@xmath30.886@xmath120.010 given by samec et al . and @xmath44(@xmath43)=@xmath30.015 calculated following schlegel et al . the logarithmic bolometric ( @xmath45 , @xmath46 ) and monochromatic ( @xmath47 , @xmath48 ) limb - darkening coefficients were interpolated from the values of van hamme ( 1993 ) and were used in concert with the model atmosphere option . to avoid possible confusion , we refer to the primary and secondary stars as those being eclipsed at min i and min ii , respectively . we first conducted a detailed @xmath49-search procedure over the range 0.3 @xmath50 3.8 . in this process , we did not consider either a third light source ( @xmath51 ) or spots . the search was made simultaneously for all light curves of each season and the results are plotted in figure 3 . from the figure , we can see that the @xmath49-search result of 2007 is different from that of 2008 and that local minima occur around @xmath49=1.3 and @xmath49=2.7 in 2008 . this may be caused by the year - to - year variability of the system and the noisiness of the 2008 light curve . because of this confusion , the @xmath49-search of the 2008 observations is considered indeterminate . we did find that the light - curve synthesis in 2007 is slightly better if @xmath521.0 and that an acceptable solution is close to @xmath49=2.6 . this indicates that ar boo is an w - subtype contact binary . to obtain an unspotted solution , we analyzed all six light curves simultaneously by treating the initial value of @xmath49 as an adjustable parameter . the results are listed in table 5 and the residuals from the fitting are plotted in the left panels of figure 4 . as indicated by these panels , the computed light curves do not fit our data satisfactorily . in many contact binaries , second - order light variations have been attributed to local atmospheric inhomogeneities that may result from either magnetic dynamo - related activity or mass transfer between the component stars . because the second - order light effects are themselves seasonally variable for ar boo , they can not be caused by a gas stream from the less massive primary impacting the inner hemisphere of the more massive secondary component unless the flow rate were itself variable . also , our period study suggests checking for a third light source in the system . accordingly , we re - analyzed our light curves by considering both a spot and such a source . the result is given as the one - spot model of table 6 and the light residuals from the model are plotted in the middle panels of figure 4 . from these displays , we see that a single cool spot on the larger secondary component with a third light source does improve the light - curve fitting greatly but that large discrepancies still remain in the first half - period of the 2007 data . this detail can reasonably be modeled by an additional hot spot on the surface of the secondary . it remains an open question whether such a spot results from mass transfer or is an intrinsic feature of the cool star . the final result is listed as the two - spot model of table 6 and the residuals from this binary model are plotted in the right panels of figure 4 . our solutions for ar boo show that @xmath51 contributes 2.0 - 2.8 % light in @xmath4 , 1.2 - 1.5 % in @xmath2 , and 0.9 - 1.0 % in @xmath11 . these values seem to point to a source relatively hotter than the binary components but the problem with this interpretation is that the @xmath51 values are not individually statistically significant . a final light curve fitting with @xmath51=0 led to parameter evaluations which , within the errors , did not change from those given in the table . the second - order light variations are best fitted by the simultaneous existence of a cool spot and a hot spot on the cool secondary star . it should be noticed , however , that the two - spot model of 2008 gives a value of @xmath53 identical to that for the one - spot model and does not improve the light - curve fitting appreciably . although the effect is not clearly seen in fig . 1 , the 2008 light curves are appreciably noisier than those of the preceding year and the difference is evident in all the panels of fig . 4 . this difference makes a large contribution to the unchanged fitting criterion for 2008 and also to the large errors associated with the 2008 hot spot . although we could have ignored that spot in 2008 , we decided to leave its formal determination just as the wd code found it . to take the understanding of ar boo further and to examine whether our solutions can reasonably describe the samec et al . light curves , we analyzed them using our photometric parameters as initial values . the results are given in columns ( 2)-(3 ) of table 7 and plotted in figure 5 as continuous curves . within the errors yielded by the wd code , the stellar parameters from the samec et al . data are in satisfactory accord with those from our light curves . we see again that the formal @xmath51 values are not significant and that the differences among the ( now 3 ) seasonal light curves can be explained almost entirely by spot variability , especially by the longitude drifts of the spots with time . the light curves of wolf et al . ( 1998 ) do not have the observational weight of ours and we held only limited expectations for their study . since the samec et al . and our light curves had converged to consistent geometric and photometric parameters , we assumed these values to be appropriate for the wolf et al . ones and analyzed these latter only for spot descriptions . the results are listed in columns ( 4)-(5 ) of table 7 and lead to the continuous curves in figure 6 . these fit the observations satisfactorily . nonetheless , as in modelling the 2008 light curves , the wolf et al . data do not need a hot spot , which may be a main cause of the large errors for the spot parameters . the 4-season mean values of @xmath51@xmath54 2 % , if real , could come from a third object either gravitationally bound to or only optically related with the eclipsing pair ar boo . we do not see such an object optically resolved on our time - series ccd images . further , ar boo is at a very high galactic latitude so background or foreground contamination is not likely . a previous section of this report quantified a possible projected orbit for a bound third body . if that supposed object be identified with the third light " source , it would have to be more than 3 magnitudes fainter than the hot eclipsing component and somewhat bluer than that star , i.e. a sub - dwarf . the evidence of the errors in tables 6 and 7 forces the simpler understanding : the third light " values are only artifacts of the light curve fitting process . if such an object exists , it has not yet been photometrically identified from all these light curves . the more massive secondary component with a spectral type of k1 v was assumed to conform to harmanec s ( 1988 ) relation for masses and radii as functions of spectral type . the remaining absolute parameters were then derived from the two - spot model of 2007 in table 6 . all the stellar characteristics are listed in table 8 wherein the bolometric corrections were obtained from the relation between @xmath55 and bc given by kang et al . ( 2007 ) . the apparent visual magnitude @xmath2=+12.76 at max ii ( samec et al . 2006 ) and our computed light ratio lead to @xmath56=+13.87 , and @xmath57=+13.27 for the primary and secondary stars , respectively . the distance to the system was calculated to be about 350 pc from the interstellar absorption of @xmath58=0.05 and the values of @xmath57 and @xmath59 . the masses and radii in table 8 are not close to those of main sequence stars but they do conform to the general pattern of contact binaries shown long ago by hilditch et al . ( 1988 ) when it is remembered that those authors define primary and secondary stars by mass and not by temperature . if the assumption is made that the hot star is a normal main sequence one , the sum of the masses is unrealistically large . figure 3 shows that there can be considerable uncertainty in determining the mass ratio as was noted in that section of the text . with a presently unknowable impact , this uncertainty propagates into the geometrical parameters of the synthesis and ultimately into the stellar radii and masses . quite likely , radial velocities would diminish this uncertainty since the spectrum should be double - lined if the light ratio is correct . this paper gives a summary of the discontinuous history of light curve morphology of ar boo from 1997 into 2008 . one description of the second - order morphology emphasizes that cool spots have manifested themselves over a siderographic band from @xmath60 to @xmath61 in longitude and 80@xmath62 to 92@xmath62 in co - latitude on the cool star . simultaneously and in the same coordinate sense , hot spots have been seen over the band @xmath63 to @xmath64 and 99@xmath62 to 115@xmath62 on the same star . whether or not these hot spots are chromospheric plages in the solar sense is not known for ar boo but we will use that noun here even if they turn out to be photospheric features . with respect to the stellar equator , the cool spots and plages have appeared in opposite hemispheres but close to and nearly parallel to that great circle . when account is taken of errors , the co - latitude excursions have been so small that the features may be said to have shown no latitude drift at all . the longitudes of the disturbed regions , on the other hand , have shown significant changes with time . solar - type magnetograms can not be made for an unresolved star and a very large telescope would be needed to measure the normalized circular polarization component for ar boo . one must seek surrogates for these conventional indicators of a magnetic environment and the light curves are the only source of other markers . it is a reasonable working inference that the spots and plages are surface manifestations of an embedded magnetic field and activity consistent with the second understanding of the 7.57-year period variation . possible quantifications of this activity might be ( a ) the fluxes emitted by either or both of the hot and cool spots , ( b ) the sum of these fluxes and ( c ) the ratio of the fluxes . from the entries in tables 6 , 7 and 8 the appropriate parameters have been extracted so as to calculate the spot and plage fluxes and their sums and ratios on the assumption of black body emissivity . the samec et al . and our light curves fall upon the decreasing - period branch of the 7.57-year waveform and the models of these light curves show ordered relations for 3 of the criteria with time and , more suggestively , with amplitude in that waveform as follows . with advancing time , the progressively more negative excursions in the waveform are associated with ( a ) increases in the plage fluxes and decreases in the cool spot fluxes , ( b ) no monotonic change in the flux sum and ( c ) decreasing values of ( plage flux / cool spot flux ) . ( in the formulation by lanza et al . these points , falling as they do on the negative slope fraction of the 7.57-year cycle , signal an increase of the cool star s quadrupole moment with concomitant changes in the magnetic dynamo . ) the weakness of the associations is evident : the spot characterizations from wolf et al . do not conform to the relation defined by only the 3 points sampling the first half of the following cycle but , it is possible to extenuate this seeming failure in the following way . one does not know what would be the appearance of the entire cycle of spot and plage activity as a function of the 7.57-year oscillation . would it be a simple closed oval , a degenerate single curve , or a twisted closed curve ? perhaps then , the failure is only a seeming one and is really evidence that magnetic activity changes from one 7.57-year cycle to the next in the same manner that solar magnetic activity is not a constant of nature . if this would be so , there would also be no sense to the propagation of that cycle forward or backward in time as we have done in the middle panel of figure 2 . it may be imagined that the cyclical component of the period variability , discussed first in section 3 , could be produced by apparent phase shifts of the real conjunctions due to asymmetrical eclipse minima originating from starspot activity and/or the method of measuring the timings of minima ( cf , lee et al . 2004 , 2008 ) . to examine this possibility , we calculated anew the minimum epoch for each eclipse curve of wolf et al . ( 1998 ) , samec et al . ( 2006 ) , and loao with the wd code by means of adjusting only the ephemeris epoch ( @xmath65 ) and not allowing a phase shift as in table 6 . the results are listed in the second column of table 9 together with the previously - tabulated timings for comparison and are illustrated with the plus symbols in figure 2 . we can see that the differences among them are smaller than the observed amplitude ( about 0.003 d ) of the 7.57-year variation which therefore survives after spot correction . there are , moreover , systematic runs with time between the observed timings and the wd ones . in our data of table 9 , the column 4 differences in 2007 are positive for both minima , while those in 2008 are positive for min i except for one eclipse ( hjd 2454555.8471 ) and negative for min ii . in 2007 , a cool spot and a hot spot are displayed to the observer before and after the first quadrature , respectively , which cause the positive values for both eclipses . in 2008 , however , a cool spot is seen at the first quadrature . this spot produces positive values for min i but negative values for min ii . both the diminution in scatter among the timings and our spot interpretation of the timing displacements can be considered consistent with the expectation of maceroni & vant veer ( 1994 ) that minimum timings are systematically shifted by light curve asymmetries due to starspot activity . our results lend weight to the proposition that the light curve synthesis method leads to better timing determinations than does the kwee & van woerden method that has been almost universally applied to individual eclispses for more than 50 years . this work on ar boo presents the most convincing confirmation in the current literature of the predictions of those authors . our interpretation also implicitly expresses the observational precision and the time density of measures that have to be attained so as to prevent spot effects from contaminating keplerian ephemerides . only if this be done can one seek out higher - order period variations . we suggest that all future period studies of this star use the timings of column 2 of table 9 rather than those of table 2 . there is another recognition to be made . it will be impossible in general to correct casually - observed minimum timings for spot bias so some small systematic error will afflict the results of dedicated minimum monitoring programs . in this article , we have analyzed new ccd observations of the eclipsing binary system ar boo obtained during three successive seasons beginning in 2006 february . the light curves show partial eclipses and season - to - season light variability . our light - curve solution indicates that ar boo is a w - subtype contact binary composed of a hotter g9 primary star and a cooler k1 companion , the derived spectral types depend , of course , on the adopted temperature for the secondary component and the observed ( @xmath43 ) color index . the disturbed light curves of the system were best modeled by using a two - spot model with cool and hot spots on the secondary star . we speculate that both spots may be produced by magnetic dynamo - related activity . according to the thermal relaxation oscillation ( tro ) theory ( lucy 1976 ; lucy & wilson 1979 ) , contact binaries must oscillate cyclically between contact and non - contact conditions . because our detailed study of period and light variations show that the orbital period is increasing and that mass is transferring from the less massive primary to the more massive secondary star , ar boo may presently be in a transition state evolving from a contact to a non - contact configuration as suggested by this theory . our new orbital period study with all available timings reveals that the @xmath21@xmath22 residuals have varied in a cyclical oscillation superposed on a secular period increase . the semi - amplitude and period of the cyclical variation are low and short , respectively , and their values command modest confidence because the observed ccd timings cover only about 1.5 cycle of this period . the oscillation may be produced , in principle , either by the ltt effect due to a stellar- or substellar - mass companion of @xmath0=0.081 @xmath1 or by a active magnetic cycle in the more massive secondary component but not by the asymmetries of eclipse light curves due to starspot activity . the circumstance that there has existed even a limited association between phase in the 7.57-year cycle and the spot activity seen in the light curves diminishes confidence in the postulated 3rd body while suggesting that a more searching attempt be made to understand magnetic activity mediating angular momentum exchange between the spin of the stellar mass distribution and the orbit . there s a further message to be drawn here . three of the authors had never even heard of this binary before the loao observations were in train . it is a very nondescript object , yet the assiduous accumulation of the best possible observations has given weight to two ideas : starspots measurably inflect eclipse timings and magnetic cycles in stellar components of close binaries express themselves dynamically . every binary has its individual information encoded for us but , a priori , we have no idea of the richness of that information and we must uncover it and piece it together with information from all the others that we observe . in a perfect world there would be seasonal accumulation of light curves ( and radial velocities ) of ar boo over the next 8 years and then surely understanding of this system and its activity would be advanced . it is even possible that the declination of the binary permits accumulation of two independent light curves per observing season . a justification for such a strenuous effort may be found in looking back at the 2007 and 2008 light curves . whereas those of the earlier year were compiled over the 6 nights from march 11 through march 17 , the data for 2008 were observed over 17 nights from december 23 , 2007 through april 12 , 2008 . we wish to thank professor chun - hwey kim for his help in using the @xmath21@xmath22 database of eclipsing binaries . we also thank the staff of mt . lemmon optical astronomy observatory for assistance with our observations . this research has made use of the simbad database maintained at cds , strasbourg , france . applegate , j. h. 1992 , apj , 385 , 621 baki , v. , dogru , s. s. , baki , h. , dogru , d. , erdem , a. , cick , c. , & demircan , o. 2005 , inf . variable stars , no . 5662 binnendijk , l. 1970 , vistas in astronomy , 12 , 217 blttler , e. 1998 , bbsag bull . 117 blttler , e. 2002 , bbsag bull . 127 brt , l. , zejda , m. , & svoboda , p. 2007 , open european journal on variable stars , 74 , 1 diethelm , r. 1996 , bbsag bull . 112 diethelm , r. 1997 , bbsag bull . 115 diethelm , r. 1998 , bbsag bull . 117 diethelm , r. 2001 , inf . bull . variable stars , no . 5027 diethelm , r. 2003 , inf . bull . variable stars , no . 5438 diethelm , r. 2004 , inf . variable stars , no . 5543 diethelm , r. 2005 , inf . variable stars , no . 5653 diethelm , r. 2006 , inf . bull . variable stars , no . 5713 flower , p. j. 1996 , apj , 469 , 355 han , w. , et al . 2005 , pasj , 57 , 821 harmanec , p. 1988 czechoslovakia , 39 , 329 hilditch , r. w. , king , d. j. , & mcfarlane , t. m. 1988 , mnras , 231 , 341 houck , j. c. , & pollock , j. t. 1986 , pasp , 98 , 461 hbscher , j. , paschke , a. , & walter , f. 2005 , inf . variable stars , no . 5657 hbscher , j. , paschke , a. , & walter , f. 2006 , inf . bull . variable stars , no . 5731 hbscher , j. , steinbach , h .- m . , & walter , f. 2009 , inf . variable stars , no . 5874 irwin , j. b. 1952 , apj , 116 , 211 irwin , j. b. 1959 , aj , 64 , 149 kang , y .- w . , hong , k .- s . , & lee , j. 2007 , in asp conf . 362 , the seventh pacific rim conference on stellar astrophysics , ed . kang et al . ( san francisco : asp ) , 19 kim , c .- h . , jeong , j. h. , demircan , o. , myesserolu , z. , & budding , e. 1997 , aj , 114 , 2753 kim , c .- h . , lee , j. w. , kim , h .- , kyung , j .- m . , & koch , r. h. 2003 , aj , 126 , 1555 krajci , t. 2005 , inf . bull . variable stars , no . 5592 kreiner , j. m. , kim , c .- h . , & nha , i .- s . 2001 , an atlas of @xmath21@xmath22 diagrams of eclipsing binary stars ( krakow : wydawn . pedagogicznej ) kurochkin , n. e. 1960 , perem . zvezdy , 13 , 84 kwee , k. k. , & van woerden , h. 1956 , bull . netherlands , 12 , 327 lanza , a. f. , rodono , m. , & rosner , r. 1998 , mnras , 296 , 893 lee , j. w. , kim , c .- h . , han , w. , kim , h .- i . , & koch , r. h. 2004 , mnras , 352 , 1041 lee , j. w. , kim , h .- i . , & kim , s .- l . 2007 , pasp , 119 , 1099 lee , j. w. , youn , j .- h . , kim , c .- h . , lee , c .- u . , & kim , h .- i . 2008 , aj , 135 , 1523 lucy , l. b. 1976 , apj , 205 , 208 lucy , l. b. , & wilson , r. e. 1979 , apj , 231 , 502 maceroni , c. , & vant veer , f. 1994 , a&a , 289 , 871 nelson , r. h. 2007 , inf . bull . variable stars , no . 5760 nelson , r. h. 2009 , inf . bull . variable stars , no . 5875 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes ( cambridge : cambridge univ . press ) , chapter 15 pribulla , t. , & rucinski , s. m. 2006 , aj , 131 , 2986 qian , s .- b . , yuan , j .- z . , soonthornthum , b. , zhu , l .- y . , he , j .- j . , & yang , y .- g . 2007 , apj , 671 , 811 richards , m. t. , & albright , g. e. 1999 , apjs , 123 , 537 af , j. , & zejda , m. 2000 , inf . bull . variable stars , no . 4887 af , j. , & zejda , m. 2002 , inf . bull . variable stars , no . 5263 samec , r. g. , lofling , t. s. , & van hamme , w. 2006 , inf . bull . variable stars , no . 5696 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , apj , 500 , 525 van hamme , w. 1993 , aj , 106 , 2096 wilson , r. e. , & devinney , e. j. 1971 , apj , 166 , 605 wolf , m. , borovika , j. , arounov , l. , af , j. , & afov , e. 1998 , inf . bull . variable stars , no . 4601 yang , y .- g . , dai , j .- m . , yin , x .- , & xiang , f .- y . 2007 , aj , 134 , 179 yang , y .- g . , lu , g .- zhu , c .- h . , & nakajima , k. 2009 , aj , 137 , 236 zejda , m. 2002 , inf . variable stars , no . 5287 zejda , m. 2004 , inf . variable stars , no . 5583 , @xmath2 , and @xmath11 bandpasses . the differences between the two seasons are shown in the middle panel and the magnitude difference between the check and comparison stars in the lower panel . the open circle and plus symbols in the upper and lower panels are the individual measures of the 2007 and 2008 seasons , respectively . the dotted lines in the middle panel refer to values of 0.0 mag . ] @xmath22 diagram of ar boo . in the upper panel the continuous curve represents the final improved ephemeris . cc , pg , vi , and p stand for ccd , photographic , visual , and photographic plate minima , respectively . the middle and lower panels display the ccd residuals from the linear and quadratic terms and from the complete ephemeris , respectively . the 7.57-year cycle is shown propagated back in time only to indicate that it could not have been recognized in the noisy earlier minimum timings . in the two lower panels , plus symbols represent the minimum times obtained by re - analyzing individual eclipse curves with the wd code . ] ( the sum of the residuals squared ) of ar boo as a function of mass ratio @xmath49 . the filled and open circles represent the @xmath49-search results for the 2007 and 2008 data sets , respectively . the arrows indicate minimum values for each data set . ] cccccc 2,454,171.79150 & @xmath660.919 & 2,454,171.79340 & @xmath661.063 & 2,454,171.79482 & @xmath661.173 + 2,454,171.79648 & @xmath660.866 & 2,454,171.79842 & @xmath661.026 & 2,454,171.79984 & @xmath661.118 + 2,454,171.80151 & @xmath660.817 & 2,454,171.80350 & @xmath660.972 & 2,454,171.80501 & @xmath661.059 + 2,454,171.80671 & @xmath660.746 & 2,454,171.80872 & @xmath660.912 & 2,454,171.81020 & @xmath661.000 + 2,454,171.81192 & @xmath660.684 & 2,454,171.81392 & @xmath660.841 & 2,454,171.81541 & @xmath660.938 + 2,454,171.81712 & @xmath660.624 & 2,454,171.81913 & @xmath660.782 & 2,454,171.82061 & @xmath660.899 + 2,454,171.82232 & @xmath660.578 & 2,454,171.82432 & @xmath660.763 & 2,454,171.82582 & @xmath660.879 + 2,454,171.82754 & @xmath660.567 & 2,454,171.82956 & @xmath660.758 & 2,454,171.83105 & @xmath660.893 + 2,454,171.83276 & @xmath660.594 & 2,454,171.83476 & @xmath660.792 & 2,454,171.83626 & @xmath660.926 + 2,454,171.83797 & @xmath660.646 & 2,454,171.83998 & @xmath660.846 & 2,454,171.84147 & @xmath660.988 + lrrrcl 0,180.4108 & @xmath120.0028 & -6.0 & 0.00131 & i & safr & zejda ( 2000 ) + 0,182.4791 & @xmath120.0002 & 0.0 & 0.00036 & i & wolf et al . ( 1998 ) + 0,192.4789 & @xmath120.0030 & 29.0 & -0.00119 & i & diethelm ( 1996 ) + 0,200.4128 & @xmath120.0002 & 52.0 & 0.00060 & i & wolf et al . ( 1998 ) + 0,200.5850 & @xmath120.0005 & 52.5 & 0.00037 & ii & wolf et al . ( 1998 ) + 0,547.3555 & @xmath120.0002 & 1058.0 & -0.00015 & i & wolf et al . ( 1998 ) + 0,551.4936 & @xmath120.0018 & 1070.0 & -0.00054 & i & diethelm ( 1997 ) + 0,607.3635 & @xmath120.0004 & 1232.0 & -0.00025 & i & wolf et al . ( 1998 ) + 0,607.5346 & @xmath120.0003 & 1232.5 & -0.00159 & ii & wolf et al . ( 1998 ) + 0,611.5025 & @xmath120.0005 & 1244.0 & 0.00026 & i & wolf et al . ( 1998 ) + 0,638.4028 & @xmath120.0002 & 1322.0 & 0.00038 & i & wolf et al . ( 1998 ) + 0,888.4365 & @xmath120.0002 & 2047.0 & 0.00037 & i & wolf et al . ( 1998 ) + 0,899.4721 & @xmath120.0003 & 2079.0 & 0.00000 & i & wolf et al . ( 1998 ) + 0,923.4410 & @xmath120.0017 & 2148.5 & 0.00015 & ii & diethelm ( 1998 ) + 0,925.5099 & @xmath120.0014 & 2154.5 & -0.00019 & ii & diethelm ( 1998 ) + 0,926.3730 & @xmath120.0006 & 2157.0 & 0.00072 & i & blttler ( 1998 ) + 0,927.4073 & @xmath120.0003 & 2160.0 & 0.00040 & i & wolf et al . ( 1998 ) + 1,270.5551 & @xmath120.0021 & 3155.0 & -0.00137 & i & safr & zejda ( 2002 ) + 1,277.4539 & @xmath120.0025 & 3175.0 & -0.00005 & i & safr & zejda ( 2002 ) + 1,284.3511 & @xmath120.0015 & 3195.0 & -0.00033 & i & safr & zejda ( 2002 ) + 1,320.7360 & @xmath120.0008 & 3300.5 & 0.00038 & ii & diethelm ( 2001 ) + 1,580.5980 & @xmath120.0033 & 4054.0 & -0.00016 & i & zejda ( 2002 ) + 1,626.4677 & @xmath120.0015 & 4187.0 & 0.00121 & i & zejda ( 2002 ) + 1,641.2961 & @xmath120.0016 & 4230.0 & -0.00002 & i & brt et al . ( 2007 ) + 1,684.4062 & @xmath120.0025 & 4355.0 & 0.00063 & i & brt et al . ( 2007 ) + 1,956.5138 & @xmath120.0015 & 5144.0 & 0.00039 & i & brt et al . ( 2007 ) + 1,965.4811 & @xmath120.0022 & 5170.0 & 0.00088 & i & zejda ( 2004 ) + 1,968.5846 & @xmath120.0025 & 5179.0 & 0.00049 & i & brt et al . ( 2007 ) + 2,014.4530 & @xmath120.0022 & 5312.0 & 0.00025 & i & brt et al . ( 2007 ) + 2,043.4225 & @xmath120.0022 & 5396.0 & 0.00009 & i & brt et al . ( 2007 ) + 2,053.4230 & @xmath120.0015 & 5425.0 & -0.00084 & i & brt et al . ( 2007 ) + 2,287.5953 & @xmath120.0004 & 6104.0 & 0.00015 & i & blttler ( 2002 ) + 2,344.4994 & @xmath120.0004 & 6269.0 & -0.00039 & i & blttler ( 2002 ) + 2,365.5369 & @xmath120.0032 & 6330.0 & -0.00037 & i & zejda ( 2004 ) + 2,723.3454 & @xmath120.0009 & 7367.5 & -0.00120 & ii & diethelm ( 2003 ) + 2,723.5173 & @xmath120.0007 & 7368.0 & -0.00174 & i & diethelm ( 2003 ) + 2,730.4160 & @xmath120.0049 & 7388.0 & -0.00057 & i & zejda ( 2004 ) + 2,808.3591 & @xmath120.0003 & 7614.0 & 0.00048 & i & baki et al . ( 2005 ) + 3,090.4682 & @xmath120.0001 & 8432.0 & 0.00077 & i & krajci ( 2005 ) + 3,094.4343 & @xmath120.0013 & 8443.5 & 0.00079 & ii & diethelm ( 2004 ) + 3,117.5406 & @xmath120.0012 & 8510.5 & 0.00038 & ii & hbscher et al . ( 2005 ) + 3,131.8527 & @xmath120.0003 & 8552.0 & 0.00012 & i & samec et al . ( 2006 ) + 3,135.8183 & @xmath120.0013 & 8563.5 & -0.00036 & ii & samec et al . ( 2006 ) + 3,143.2339 & @xmath120.0003 & 8585.0 & 0.00040 & i & krajci ( 2005 ) + 3,165.8238 & @xmath120.0003 & 8650.5 & 0.00091 & ii & samec et al . ( 2006 ) + 3,203.4157 & @xmath120.0004 & 8759.5 & 0.00130 & ii & hbscher et al . ( 2005 ) + 3,351.7109 & @xmath120.0002 & 9189.5 & -0.00026 & ii & hbscher et al . ( 2006 ) + 3,409.6501 & @xmath120.0044 & 9357.5 & -0.00026 & ii & hbscher et al . ( 2005 ) + 3,463.4508 & @xmath120.0002 & 9513.5 & -0.00023 & ii & hbscher et al . ( 2006 ) + 3,515.5263 & @xmath120.0010 & 9664.5 & -0.00103 & ii & diethelm ( 2005 ) + 3,767.9772 & @xmath120.0005 & 10396.5 & 0.00059 & ii & this article + 3,769.0115 & @xmath120.0003 & 10399.5 & 0.00026 & ii & this article + 3,770.0456 & @xmath120.0004 & 10402.5 & -0.00027 & ii & this article + 3,797.9806 & @xmath120.0002 & 10483.5 & -0.00022 & ii & this article + 3,813.4999 & @xmath120.0018 & 10528.5 & -0.00034 & ii & hbscher et al . ( 2006 ) + 3,821.7770 & @xmath120.0003 & 10552.5 & -0.00027 & ii & nelson ( 2007 ) + 3,859.5407 & @xmath120.0003 & 10662.0 & -0.00049 & i & diethelm ( 2006 ) + 4,171.8267 & @xmath120.0001 & 11567.5 & 0.00038 & ii & this article + 4,172.8614 & @xmath120.0001 & 11570.5 & 0.00045 & ii & this article + 4,173.8959 & @xmath120.0001 & 11573.5 & 0.00032 & ii & this article + 4,174.7577 & @xmath120.0001 & 11576.0 & -0.00007 & i & this article + 4,176.8269 & @xmath120.0001 & 11582.0 & -0.00012 & i & this article + 4,177.8615 & @xmath120.0001 & 11585.0 & -0.00015 & i & this article + 4,521.0145 & @xmath120.0003 & 12580.0 & 0.00035 & i & this article + 4,522.9109 & @xmath120.0002 & 12585.5 & -0.00008 & ii & this article + 4,539.8093 & @xmath120.0002 & 12634.5 & -0.00074 & ii & this article + 4,540.8442 & @xmath120.0002 & 12637.5 & -0.00048 & ii & nelson ( 2009 ) + 4,551.8804 & @xmath120.0003 & 12669.5 & -0.00041 & ii & this article + 4,553.9497 & @xmath120.0003 & 12675.5 & -0.00038 & ii & this article + 4,554.8127 & @xmath120.0002 & 12678.0 & 0.00042 & i & this article + 4,554.9844 & @xmath120.0002 & 12678.5 & -0.00032 & ii & this article + 4,555.5023 & @xmath120.0014 & 12680.0 & 0.00026 & i & hbscher et al . ( 2009 ) + 4,555.8471 & @xmath120.0002 & 12681.0 & 0.00019 & i & this article + 4,556.8819 & @xmath120.0004 & 12684.0 & 0.00035 & i & this article + 4,568.7795 & @xmath120.0002 & 12718.5 & -0.00038 & ii & this article + 4,569.8142 & @xmath120.0002 & 12721.5 & -0.00031 & ii & this article + 4,569.9873 & @xmath120.0003 & 12722.0 & 0.00035 & i & this article + 4,598.4399 & @xmath120.0006 & 12804.5 & 0.00042 & ii & hbscher et al . ( 2009 ) + ccc @xmath65 & 2,450,182.47724(28 ) & hjd + @xmath67 & 0.344874257(32 ) & d + @xmath68 & 1.264(27)@xmath69 & d + @xmath70 & 0.259(67 ) & au + @xmath18 & 280(18 ) & deg + @xmath17 & 0.59(28 ) & + @xmath71 & 0.1301(17 ) & deg d@xmath40 + @xmath20 & 2,446,142(97 ) & hjd + @xmath72 & 7.573(99 ) & yr + @xmath25 & 0.00149(38 ) & d + @xmath73 & 0.000304(79 ) & @xmath1 + @xmath74 ( @xmath75=90 deg)@xmath76 & 0.081 & @xmath1 + @xmath74 ( @xmath75=60 deg)@xmath76 & 0.095 & @xmath1 + @xmath74 ( @xmath75=30 deg)@xmath76 & 0.170 & @xmath1 + @xmath37/@xmath38 & 2.676(58)@xmath77 & d yr@xmath40 + @xmath78/@xmath38 & 1.482@xmath77 & @xmath1 yr@xmath40 + cccc @xmath26 & 0.1069 & 0.1069 & s + @xmath79 & @xmath80 & @xmath80 & + @xmath27 & @xmath81 & @xmath82 & g @xmath83 + @xmath28 & @xmath84 & @xmath85 & g @xmath86 s@xmath40 + @xmath87 & @xmath88 & @xmath89 & g @xmath86 + @xmath30 & @xmath90 & @xmath91 & s@xmath40 + @xmath92 & @xmath93 & @xmath94 & + @xmath31 & @xmath95 & @xmath96 & erg + @xmath32 & @xmath97 & @xmath98 & erg s@xmath40 + & 0.162 & 0.088 & @xmath99 + & 0.506 & 0.144 & @xmath100 + @xmath33 & @xmath120.174 & @xmath120.085 & mag + @xmath4 & 20640 & 14576 & g ccc @xmath65 ( hjd ) & + @xmath67 ( d ) & + @xmath49 & + @xmath101 ( deg ) & + @xmath20 ( k ) & 5398(14 ) & 5100 + @xmath102 & 6.036(6 ) & 6.102 + @xmath68 & 0.5 & 0.5 + @xmath103 & 0.32 & 0.32 + @xmath45 , @xmath46 & 0.645 , 0.186 & 0.643 , 0.169 + @xmath104 , @xmath105 & 0.945(51 ) , 0.070 & 0.866(39 ) , 0.006 + @xmath106 , @xmath107 & 0.770(53 ) , 0.183 & 0.854(44 ) , 0.131 + @xmath108 , @xmath109 & 0.605(55 ) , 0.219 & 0.798(48 ) , 0.186 + @xmath110 , @xmath111 & 0.924(49 ) , 0.070 & 0.808(40 ) , 0.006 + @xmath112 , @xmath113 & 0.737(51 ) , 0.183 & 0.771(44 ) , 0.131 + @xmath114 , @xmath115 & 0.607(52 ) , 0.219 & 0.699(48 ) , 0.186 + @xmath116 & 0.376(2 ) & 0.624 + @xmath117 & 0.371(2 ) & 0.629 + @xmath118 & 0.366(2 ) & 0.634 + @xmath119 & 0.372(2 ) & 0.628 + @xmath120 & 0.366(2 ) & 0.634 + @xmath121 & 0.356(2 ) & 0.644 + @xmath122 ( pole ) & 0.2841(4 ) & 0.4412(4 ) + @xmath122 ( side ) & 0.2968(5 ) & 0.4725(6 ) + @xmath122 ( back ) & 0.3331(9 ) & 0.5007(7 ) + @xmath122 ( volume)@xmath123 & 0.3064 & 0.4729 + @xmath53 & + cccccccc @xmath124@xmath123 & -0.00027(10 ) & -0.00052(9 ) & & & + @xmath49 & 2.601(7 ) & 2.591(4 ) & & & + @xmath101 ( deg ) & 77.4(4 ) & 77.6(4 ) & & & + @xmath125 ( k ) & 5388(16 ) & 5378(14 ) & & & + @xmath126=@xmath127 & 6.007(11 ) & 5.982(9 ) & & & + @xmath128 ( % ) & 12.3 & 14.3 & & & + @xmath129 & 0.942(69 ) & 0.966(57 ) & & & + @xmath130 & 0.756(68 ) & 0.766(56 ) & & & + @xmath131 & 0.575(68 ) & 0.627(58 ) & & & + @xmath132 & 0.857(92 ) & 0.849(67 ) & & & + @xmath133 & 0.793(85 ) & 0.789(68 ) & & & + @xmath134 & 0.700(84 ) & 0.704(71 ) & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath139 & 0.361(4 ) & 0.360(3 ) & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath140 & 0.359(4 ) & 0.359(3 ) & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath141 & 0.357(4 ) & 0.352(4 ) & & & + _ @xmath142_@xmath143 & 0.036(18 ) & 0.025(14 ) & & & + _ @xmath144_@xmath143 & 0.018(17 ) & 0.012(14 ) & & & + _ @xmath145_@xmath143 & 0.006(17 ) & 0.007(14 ) & & & + @xmath146 ( pole ) & 0.2853(10 ) & 0.2865(8 ) & & & + @xmath146 ( side ) & 0.2981(12 ) & 0.2995(9 ) & & & + @xmath146 ( back ) & 0.3350(21 ) & 0.3371(16 ) & & & + @xmath146 ( volume ) & 0.3079 & 0.3095 & & & + @xmath147 ( pole ) & 0.4414(8 ) & 0.4420(7 ) & & & + @xmath147 ( side ) & 0.4728(11 ) & 0.4736(9 ) & & & + @xmath147 ( back ) & 0.5014(15 ) & 0.5026(12 ) & & & + @xmath147 ( volume ) & 0.4734 & 0.4743 & & & + + colatitude@xmath148 ( deg ) & 93.4(2.6 ) & 90.4(0.6 ) & & 91.5(2.8 ) & 103.0(2.4 ) & 83.2(1.0 ) & 103.5(1.5 ) + longitude@xmath148 ( deg ) & 172.5(1.3 ) & 97.4(1.3 ) & & 160.4(0.4 ) & 39.7(2.8 ) & 89.4(0.4 ) & 64.6(1.1 ) + radius@xmath148 ( deg ) & 14.99(46 ) & 16.07(21 ) & & 14.42(35 ) & 17.25(92 ) & 17.71(18 ) & 14.63(73 ) + @xmath20@xmath149/@xmath20@xmath150 & 0.928(3 ) & 0.917(3 ) & & 0.925(2 ) & 1.033(2 ) & 0.918(2 ) & 1.024(3 ) + @xmath53 & 0.0068 & 0.0084 & & & cccccccccc @xmath65 ( hjd ) & & & + @xmath67 ( day ) & & & + @xmath49 & & & + @xmath101 ( deg ) & & & + @xmath20 ( k ) & & & & & + @xmath102 & & & & & + @xmath128 ( % ) & & & + @xmath151 & & & & & + @xmath152 & & & & & + @xmath153 & & & & & + @xmath154 & & & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath139 & & & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath140 & & & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath141 & & & & & + @xmath135/(@xmath136+@xmath137+@xmath138)@xmath155 & & & & & + _ @xmath142_@xmath156 & & & + _ @xmath144_@xmath156 & & & + _ @xmath145_@xmath156 & & & + _ @xmath157_@xmath156 & & & + @xmath122 ( pole ) & & & & & + @xmath122 ( side ) & & & & & + @xmath122 ( back ) & & & & & + @xmath122 ( volume ) & & & & & + colatitude ( deg)@xmath143 & & 90.7(13.0 ) & 115.3(2.9 ) & & & 79.9(0.8 ) & 98.9(9.4 ) + longitude ( deg)@xmath143 & & 150.3(2.9 ) & 35.7(2.5 ) & & & 106.7(0.7 ) & 298(11.0 ) + radius ( deg)@xmath143 & & 11.4(1.4 ) & 19.0(0.8 ) & & & 20.8(0.4 ) & 13.1(2.9 ) + @xmath20@xmath158/@xmath20@xmath159@xmath143 & & 0.931(0.012 ) & 1.086(0.006 ) & & & 0.870(0.006 ) & 1.024(0.010 ) ccc @xmath160 ( @xmath1 ) & 0.35 & 0.90 + @xmath11 ( @xmath161 ) & 0.65 & 1.00 + @xmath162 @xmath103 ( cgs ) & 4.36 & 4.39 + @xmath163 ( @xmath99 ) & 0.32 & 0.61 + @xmath164 ( mag ) & 5.94 & 5.23 + bc ( mag ) & @xmath660.17 & @xmath660.26 + @xmath165 ( mag ) & 6.11 & 5.49 + distance ( pc ) & + ccccccl 0,607.3635 & 0,607.36271 & @xmath120.00021 & @xmath30.00079 & @xmath5 & i & wolf et al . + 0,607.5346 & 0,607.53490 & @xmath120.00022 & @xmath660.00030 & @xmath5 & ii & wolf et al . + 0,611.5025 & 0,611.50142 & @xmath120.00028 & @xmath30.00108 & @xmath11 & i & wolf et al . + 0,638.4028 & 0,638.40188 & @xmath120.00013 & @xmath30.00092 & @xmath11 & i & wolf et al . + 3,131.8527 & 3,131.85333 & @xmath120.00016 & @xmath660.00063 & @xmath6 & i & samec et al . + 3,135.8183 & 3,135.81836 & @xmath120.00016 & @xmath660.00060 & @xmath6 & ii & samec et al . + 3,165.8238 & 3,165.82275 & @xmath120.00020 & @xmath30.00105 & @xmath6 & ii & samec et al . + 4,171.8267 & 4,171.82622 & @xmath120.00006 & @xmath30.00048 & @xmath7 & ii & this article + 4,172.8614 & 4,172.86085 & @xmath120.00009 & @xmath30.00055 & @xmath7 & ii & this article + 4,173.8959 & 4,173.89546 & @xmath120.00005 & @xmath30.00044 & @xmath7 & ii & this article + 4,174.7577 & 4,174.75757 & @xmath120.00007 & @xmath30.00013 & @xmath7 & i & this article + 4,176.8269 & 4,176.82668 & @xmath120.00008 & @xmath30.00022 & @xmath7 & i & this article + 4,177.8615 & 4,177.86136 & @xmath120.00007 & @xmath30.00014 & @xmath7 & i & this article + 4,521.0145 & 4,521.01425 & @xmath120.00011 & @xmath30.00025 & @xmath7 & i & this article + 4,522.9109 & 4,522.91122 & @xmath120.00012 & @xmath660.00032 & @xmath7 & ii & this article + 4,539.8093 & 4,539.80960 & @xmath120.00012 & @xmath660.00030 & @xmath7 & ii & this article + 4,551.8804 & 4,551.88093 & @xmath120.00009 & @xmath660.00053 & @xmath7 & ii & this article + 4,553.9497 & 4,553.95013 & @xmath120.00015 & @xmath660.00043 & @xmath7 & ii & this article + 4,554.8127 & 4,554.81233 & @xmath120.00009 & @xmath30.00037 & @xmath7 & i & this article + 4,554.9844 & 4,554.98474 & @xmath120.00012 & @xmath660.00034 & @xmath7 & ii & this article + 4,555.8471 & 4,555.84711 & @xmath120.00011 & @xmath660.00001 & @xmath7 & i & this article + 4,556.8819 & 4,556.88168 & @xmath120.00015 & @xmath30.00022 & @xmath7 & i & this article + 4,568.7795 & 4,568.77993 & @xmath120.00009 & @xmath660.00043 & @xmath7 & ii & this article + 4,569.8142 & 4,569.81443 & @xmath120.00016 & @xmath660.00023 & @xmath7 & ii & this article + 4,569.9873 & 4,569.98727 & @xmath120.00011 & @xmath30.00003 & @xmath7 & i & this article +

we continue our study on corrections from canonical quantum gravity to the power spectra of gauge - invariant inflationary scalar and tensor perturbations . a direct canonical quantization of a perturbed inflationary universe model is implemented , which leads to a wheeler - dewitt equation . for this equation , a semiclassical approximation is applied in order to obtain a schrdinger equation with quantum - gravitational correction terms , from which we calculate the corrections to the power spectra . we go beyond the de sitter case discussed earlier and analyze our model in the first slow - roll approximation , considering terms linear in the slow - roll parameters . we find that the dominant correction term from the de sitter case , which leads to an enhancement of power on the largest scales , gets modified by terms proportional to the slow - roll parameters . a correction to the tensor - to - scalar ratio is also found at second order in the slow - roll parameters . making use of the available experimental data , the magnitude of these quantum - gravitational corrections is estimated . finally , the effects for the temperature anisotropies in the cosmic microwave background are qualitatively obtained .

1611.02932

the search for a theory of quantum gravity will only be successful if one eventually finds a way to test the candidate theories by experiment or observation @xcite . because of the extremely high energies at which quantum - gravitational effects are expected to be strong , many researchers have looked for features of quantum - gravitational origin in the anisotropies of the cosmic microwave background ( cmb ) ; see , for example , @xcite . these anisotropies are thought to originate during the inflationary phase of the primordial universe from quantum fluctuations of the metric and the scalar inflaton field . hence , they are in a sense already a consequence of the quantum nature of space - time and thus an effect of quantum gravity . what we are going to study in the following is to derive the corrections that arise from quantizing the universe as a whole in the canonical formalism that leads to the wheeler - dewitt equation . one way to obtain corrections to the power spectra of the inflationary scalar and tensor perturbations , which lead to the cmb anisotropies , is to perform a semiclassical approximation of the wheeler - dewitt equation @xcite . this procedure leads to a schrdinger equation with quantum - gravitational correction terms , which can be used to calculate the corrected power spectra . this was analyzed for a simple model containing only non - gauge - invariant perturbations of a scalar field in @xcite . an alternative semiclassical approximation was presented in @xcite and was used to calculate cmb corrections in @xcite . the results of both approaches were essentially the same , with differences only in the numerical factors . we have previously studied the corrections to the power spectra of gauge - invariant scalar and tensor perturbations and have made explicit calculations for the de sitter case @xcite . here , we shall extend our analysis and use a _ generic slow - roll approximation _ that is compatible with all observational data obtained so far and which encompasses a wide range of inflaton potentials . our paper is organized as follows . in sec . ii , we summarize the quantization procedure and the semiclassical approximation . this leads to the equations we use in sec . iii to obtain general expressions for the power spectra without and with quantum gravity corrections . in sec . iv , we introduce the slow - roll approximation , and in sec . v we calculate the corresponding uncorrected power spectra . vi represents the main part of the paper ; here , we calculate the slow - roll power spectra with the quantum - gravitational corrections . in sec . vii , we derive from this the corrections to the spectral index , its running , and to the tensor - to - scalar ratio . we also comment on the observability of the calculated effects and compute the correction for the cmb temperature anisotropies . finally , in sec . viii , we summarize and perform an outlook . we also compare our results with the results obtained in @xcite . here and in the next section , we present a brief self - contained summary of our earlier results from @xcite . we model an inflationary universe as a flat friedmann - lematre - robertson - walker ( flrw ) space - time plus fluctuations . the matter content is supposed to be a massive scalar inflaton field @xmath0 with potential @xmath1 , and a vanishing cosmological constant is assumed . for convenience , we will use conformal time @xmath2 , which is defined in terms of the cosmic time @xmath3 and the scale factor @xmath4 by @xmath5 . furthermore , introducing four space - time functions @xmath6 , @xmath7 , @xmath8 , @xmath9 to describe the scalar perturbations , as well as the symmetric spatial tensor @xmath10 to encode the tensorial degrees of freedom , the metric of the flrw space - time plus fluctuations reads ( see , for example , @xcite ) @xmath11\d x^i\d x^j\bigr\rbrace\ , . \nonumber\end{aligned}\ ] ] the additional scalar perturbations @xmath12 of the field @xmath0 can be combined with the scalar metric perturbations to construct a master gauge - invariant quantity called the _ mukhanov - sasaki variable _ @xcite : [ mukhdef ] v(,):=a , where we have indicated the derivative with respect to @xmath2 with a prime and used the definition @xmath13 . we denote the fourier transform of as @xmath14 and also introduce the fourier - transformed perturbation variable of the gauge - invariant tensor perturbations @xmath10 with polarization @xmath15 as v_^ ( ) : = . we note that an important feature in the definition of these variables is the rescaling with respect to @xmath16 . the action @xmath17 for our perturbed inflationary universe then takes the form @xmath18 \nonumber \\ + \,\frac{1}{\mathfrak{l}^3}\sum_{\vec{k } } & \,\bigl [ v_{\vec{k}}^\prime{v_{\vec{k}}^*}^\prime + { } ^\text{s}\omega^2_{\vec{k}}\,v_{\vec{k}}v_{\vec{k}}^*\bigr ] \biggr . \nonumber\\ + \,\frac{1}{\mathfrak{l}^3}\sum_{\lambda=+,\times}\sum_{\vec{k } } & \left [ v_{\vec{k}}^{(\lambda)\prime}v_{\vec{k}}^{(\lambda)*\prime } + { } ^\text{t}\omega^2_{\vec{k}}\,v_{\vec{k}}^{(\lambda)}v_{\vec{k}}^{(\lambda)*}\right]\biggr\rbrace\,.\end{aligned}\ ] ] here , we have introduced the `` frequencies '' @xmath19 and @xmath20 by [ defomega ] ^^2_():=k^2-,^^2_():=k^2- , where @xmath21 is defined as @xmath22 . in order to avoid the infinity arising from the volume integral of the action , it is necessary to introduce a maximum length scale @xmath23 , which can be thought of as the maximum size of the universe under consideration ( see e.g. the appendix of @xcite ) . we will later have to specify a value for @xmath23 when comparing our results to observations ; however , up to that point it is possible to remove @xmath23 from the notation by applying the following redefinitions @xcite : [ repl1 ] a _ = a _ , _ = k _ , v _ = . note that , after these rescalings , @xmath16 obtains the dimension of a length , whereas @xmath2 , @xmath24 , and @xmath14 are dimensionless . after a canonical quantization @xcite of , and choosing a product ansatz for the full wave function , we end up with a wheeler - dewitt equation of the following form , for _ each _ mode @xmath25 and for _ both _ the scalar and tensor perturbations , @xmath26 \nonumber \\ & -\,\frac{\partial^2}{\partial v_{\vec k}^2}+\omega^2_{\vec k}(\eta)\,v_{\vec{k}}^2\biggr\}\psi_{\vec{k}}(\alpha,\phi , v_{\vec{k } } ) = 0\ , . \end{aligned}\ ] ] we have combined the notation for the scalar and tensor modes and thus removed the superscripts s , t and @xmath27 . we also have set ( besides @xmath28 ) @xmath29 and introduced the dimensionless quantity @xmath30 , where @xmath31 is a reference scale factor . for simplicity , we will not write out this reference scale in the following ( or , equivalently , @xmath32 will be imposed ) ; but it is implicitly understood that a factor @xmath31 is associated with every factor of @xmath33 . we have also defined a rescaled planck mass @xmath34 in order to incorporate several numerical prefactors , [ mpdef ] m_^2 : = . we shall use @xmath35 as the parameter with respect to which the semiclassical approximation of is carried out @xcite . for this purpose , @xmath0 has to be rescaled to the dimensionless variable : = m_^-1 . more precisely , the reason behind this rescaling is that we will subsequently perform an expansion in inverse powers of @xmath36 and , in order to obtain the correct classical background equations at first order in that expansion , it is necessary that the first two terms in square brackets of have the same power of @xmath36 . in any case , once the expansion is made , we will revert and provide all expressions in terms of @xmath0 . the approximation is then performed by expanding the wave function @xmath37 using the functions @xmath38 , @xmath39 , as follows : [ wkbansatz ] _ = . after inserting this wkb - type ansatz into , the terms containing a certain power of @xmath40 are collected and their sum is set equal to zero . at order @xmath35 , we obtain the hamilton - jacobi equation of the minisuperspace background , [ hjeq ] -()^2 + m _ p^2()^2+()=0 . the planck mass occurs explicitly here because the expansion in is performed in the form of the equation using @xmath41 instead of @xmath0 . at the next order , @xmath42 , it is possible to write the corresponding equation for @xmath43 as a schrdinger equation for a related wave function @xmath44 in the following way , [ schreq ] _ ^(0)_=^(0 ) _ , where the conformal time is defined in terms of the minisuperspace variables by [ ctimedef ] : = ^-2 , and the perturbative hamiltonian operator is given as follows , @xmath45 the information at the next order , @xmath46 , can be encoded in a wave function @xmath47 , related to the function @xmath48 , which obeys the following corrected schrdinger equation : @xmath49\ , . \nonumber\end{aligned}\ ] ] here , we have defined an auxiliary potential as [ potential ] v(a , ) : = ( ) , which has the dimension of a length squared . in summary , the most important results of this section are the two wave equations and . the former one describes quantum fluctuations evolving on a classical ( flrw ) background spacetime , while the latter one encodes also corrections arising from the quantum behavior of that background . the aim of this paper is thus to solve these two equations with relevant initial conditions in order to obtain the power spectra with quantum - gravitational corrections for inflationary perturbations . in the next section , the ansatz to be used for such a purpose is described , as well as the explicit form of the power spectra . we use a gaussian ansatz for both the uncorrected schrdinger equation and the corrected one , with the normalization factor @xmath50 and the inverse gaussian widths @xmath51 @xcite , [ gaussianansatz ] ^(0,1)_(,v _ ) = n_^(0,1)()^-_^(0,1)()v_^2 . here , and in the following , the superscript @xmath52 stands for the uncorrected and @xmath53 for the corrected case . as will be shown below , in order to obtain the power spectra for the scalar and tensor perturbations , we have to find the solutions for @xmath54 . for the uncorrected schrdinger equation , we have to solve _ ^(0 ) ( ) = ( _ ^(0)())^2-^2 _ k ( ) , [ eqomega0 ] while for the corrected schrdinger equation , we have to find a solution to _ ^(1 ) ( ) = ( _ ^(1)())^2-^2 _ k ( ) . [ omegaeq ] in this last expression the corrected frequencies @xmath55 have been defined as follows : @xmath56.\end{aligned}\ ] ] as is explicitly written in this definition , in this paper we will only consider the real part of the correction terms for the frequencies , since the imaginary part tends to exhibit unphysical behavior ; in particular , unitarity violation . for a more detailed discussion about this point we refer the reader to @xcite . since the equation for @xmath57 is nonlinear , it is reasonable to linearize it around the background solution @xmath58 by defining @xmath59 . after neglecting the quadratic term , one obtains the following linear equation : @xmath60 since it is usually possible to find an analytical solution for @xmath58 , one is then left with only a linear equation for @xmath61 , which is easier to solve than the full non - linear equation . apart from the equations themselves , another important key aspect of the analysis is to choose appropriate and meaningful initial conditions . for the uncorrected case , as it is usually done in quantum field theory , the bunch - davies vacuum will be chosen , which in this setting means @xmath62 . the case of an initial non - bunch - davies state was recently discussed in @xcite . moreover , as we have discussed in @xcite , the most natural initial conditions for the corrected case should be those that best resemble the properties of the bunch - davies vacuum . namely , when the mode is well inside its hubble radius , it should be oscillating with a fixed constant frequency and amplitude . this is achieved by choosing @xmath63 since the imaginary part of the corrected frequencies has been neglected , these two relations can simply be written as @xmath64 . for the linearized function the above conditions are rewritten as @xmath65 . therefore , the latter is the condition that will be used as initial data at early times ( @xmath66 ) . the power spectrum for the scalar perturbations including the quantum - gravitational corrections is given by [ ps0omega ] p_^(1)(k ) = _ ^(0)(k)\{1 + _ } , where @xmath67 is the slow - roll parameter defined below in , and @xmath68 is the usual scalar power spectrum defined by inserting @xmath58 instead of @xmath57 into this expression . the quantum - gravitational effects are thus encoded in the term [ defdeltas ] _ : = - . this ratio has to be computed in the limit of super - hubble scales ( or late times ) , given by @xmath69 , when the perturbations get `` frozen '' . the power spectrum for the tensor perturbations is , consequently , obtained by [ pt0omega ] p_^(1)(k ) = _ ^(0)(k)\{1+_}. here , the same comment as in the scalar case about taking the super - hubble scale limit for @xmath70 applies , and the corrections are given by the term _ : = - . finally , the corrected tensor - to - scalar ratio @xmath71 is defined as [ tts0 ] r^(1 ) : = r^(0 ) ( 1+_- _ ) , where @xmath72 is the ratio found at the order of approximation that corresponds to quantum field theory on a fixed , curved background . in the de sitter case , since @xmath73 , both corrections are the same , @xmath74 , and thus in our previous work @xcite there appeared no correction to this quantity at the considered order of approximation . this will be different here . to recap , note that the task of obtaining quantum - gravitational corrections to the scalar and tensor power spectra reduces to solving equation with initial data . once this is done , one only needs to compute the corresponding power spectra using the relations and . in this section the well - known slow - roll approximation is described in order to clarify and set up the notation . in addition , several physical quantities , which appear in the key equations of motion and , will be explicitly given up to linear order in the slow - roll parameters , which we will define below . using the hubble parameter @xmath75 , we can define the _ first slow - roll parameters _ @xmath67 and @xmath76 ( see e.g. eqs ( 8.37 ) and ( 8.38 ) in @xcite ) , @xmath77 in terms of the slow - roll parameter @xmath67 , the equation @xmath78 and the hamiltonian constraint equation can be written as follows : @xmath79 or , using the conformal time @xmath2 , @xmath80 no approximation has been assumed yet , so these relations are exact . in order to recover from the hamilton jacobi equation of our semiclassical approximation , one has to use the action s_0 = - , which is valid in the slow - roll approximation to be applied from now on . this approximation consists in assuming ^2 , 3h . it can be shown that this is equivalent to assume that the slow - roll parameters @xmath67 and @xmath76 are small : @xmath81 and @xmath82 . if one assumes those to be vanishing , one would recover the de sitter case analyzed in our previous paper @xcite . since the pure de sitter case is often too restrictive to make contact with observations , we shall now consider the first - order slow - roll approximation , which assumes @xmath67 and @xmath76 to be small but nonvanishing , and drop quadratic terms in those . in order to solve the equations presented in the previous section in this approximation , we need to express the frequencies and , as well as the potential in terms of the conformal time @xmath2 and the slow - roll parameters @xmath67 and @xmath76 . it is well - known ( see e.g.@xcite ) that , in this approximation , used in can also be written as @xmath83 . ] the frequencies of the modes can be written as @xmath84 where @xmath85 stands for terms quadratic in @xmath67 and @xmath76 and where we have defined , for later convenience , the combination : = 2- . note that setting @xmath86 , or equivalently @xmath87 , converts the equation for scalar perturbations into the one for tensor perturbations . therefore we will perform all calculations only for the scalar perturbations and obtain the results for the tensor perturbations using this relation at the end . making use of the hamiltonian constraint as written in , it is possible to write the rescaled potential as v = a^4 h^2(1- ) . in order to write this expression explicitly in terms of the conformal time and the slow - roll parameters , we use the definition of the hubble factor in terms of the conformal time to write = . integrating this relation by parts twice , we get [ timeah ] = -+o(2 ) . now we can express the hubble parameter as @xmath88 and integrate this equation , which leads to a=+o(2 ) , with an integration constant @xmath89 . the hubble parameter then takes the form [ hubbleeta ] h = h_0 ( ) ^+ o(2)=h_0 + o(2 ) , in which the constant @xmath89 has been replaced by c = ( 1 + ) ; @xmath90 being the value of the hubble parameter at @xmath91 . physically , @xmath90 defines the de sitter space - time we are considering as our reference . below we will have to choose a value for @xmath90 , since the results will depend on it . the only physically meaningful point that exists in the evolution of different modes , is when they cross the hubble horizon . therefore , we will choose @xmath90 as the value of the hubble factor at the hubble - scale exit , that is , @xmath92 , with @xmath93 . following relation , this implies choosing @xmath94 on the right - hand side of . ( note that in this equation , at the level of approximation chosen , @xmath95 is only meaningful at order @xmath96 . ) this choice makes both @xmath90 and @xmath97 become @xmath24-dependent but , since the evolution of every @xmath24-mode is independent , there is no problem in assuming that they experience a different reference de sitter space - time . as a side remark , note that the present approximation is valid during times @xmath2 which obey |_0|^-1/|| |_0|^1/. due to the smallness of @xmath67 , this time range is very large . nonetheless , when considering the asymptotic value for different physical objetcs as @xmath98 , one should take into account that in this limit the approximation breaks down . finally , making use of all above results , the potential @xmath99 can be written in the following way : @xmath100 where @xmath97 has already been replaced by its value at horizon crossing . this expression can then be used in to obtain the explicit form of the corrected frequencies at first order in the slow - roll parameters . in this way , we can already write our main equations , and , at this level of approximation . in this section , we will obtain the solution to eq . , in order to construct the power spectra for scalar and tensor perturbations at the level of approximation that corresponds to the usual formalism of quantum field theory on classical background space - times . in @xcite , one can find a detailed computation in a formalism very similar to the present one . considering for now only scalar perturbations , the solution to eq . with @xmath101 given by can be written as follows : [ omegaeta ] ^_^(0 ) ( ) = - , where the mode functions @xmath102 can be expressed in terms of the bessel functions @xmath103 : @xmath104.\end{aligned}\ ] ] in order to obtain the bunch - davies vacuum for @xmath66 , and choosing the usual normalization of the wronskian ^y^(0)_k^y^(0)*_k - ^y^(0)_k^y^(0)_k= , it is necessary to set @xmath105}\ , . \nonumber\end{aligned}\ ] ] in the super - hubble limit @xmath106 , the real part of @xmath107 is given by [ reomegaeta ] ^^(0)_k()= ( -k)^2 ( 1 + ) . taking the inverse of this function and linearizing it in terms of the slow - roll parameters , we get the standard result for the power spectrum of the scalar perturbations ( see e.g. @xcite , p. 498 ) : @xmath108 , \label{ps0srk}\end{aligned}\ ] ] where @xmath109 is the euler mascheroni constant and the result should be evaluated at the horizon exit of the mode . up to a global multiplicative factor , the power spectrum for the tensor modes can be immediately obtained from the last expression by setting @xmath110 for the terms inside the square brackets , and reads @xmath111 . \label{pt0srk}\end{aligned}\ ] ] in the de sitter case , expression simplifies considerably to [ om0ds ] ^^(0)_():=+ . ( this corresponds to eq . ( 129 ) in @xcite . ) given the frequent appearance of the factor @xmath112 , we now define the quantity [ defxi ] : = -k , which allows us to isolate the @xmath24-dependence of this expression : [ om0ds ] ^_^(0)():=k . note that the @xmath113 , whose real part is given in , only depend on the combination @xmath114 of the slow - roll parameters . for later convenience , we linearize it around its de sitter counterpart , [ decomega0 ] ^^(0)_=^^(0)_+^(0 ) _ , . the term linear in the slow - roll parameters has the following form : [ om0gamma ] ^(0)_,():=k ; being the exponential integral function . in spite of the appearance of this special function , this form of @xmath115 is much more manageable than the definition above in terms of derivatives of bessel functions . from this point on , we will skip the index @xmath25 all along in order to simplify the notation . inserting the form of the potential into the definition ( [ defcorrectedomega ] ) and linearizing in the slow - roll parameters , we obtain the following form for the corrected frequencies for the scalar sector : [ omegadecomposition ] _ ^2=_^2+()(^2_+^2 _ + ^2 _ + ( 2 ) ) ; the classical frequencies @xmath116 have been defined above in , while the other terms stand for different quantum - gravity corrections and are defined as follows : @xmath117,\\\label{defomegagamma } \widetilde\omega_{\gamma}^2&:=&\frac{\xi ^4}{2 \left(\xi ^2 + 1\right)^4}\big\ { 2 \text{ci}(2 \xi ) \left[2 \xi \left(2 \xi ^6 - 9 \xi ^4 + 14 \xi ^2 - 11\right ) \sin ( 2 \xi ) -\left(6 \xi ^8 + 10 \xi ^6 + 53 \xi ^4 - 12 \xi ^2 + 11\right ) \cos ( 2 \xi ) \right]\nonumber\\ & & \qquad\qquad\qquad-7 \xi ^6 + 21 \xi ^4 - 89 \xi ^2 + 2 \left(2 \xi ^6 - 9 \xi ^4 + 14 \xi ^2 - 11\right ) \xi \cos ( 2 \xi ) ( \pi -2 \text{si}(2 \xi ) ) \nonumber\\&&\qquad\qquad\qquad+\left(6 \xi ^8 + 10 \xi ^6 + 53 \xi ^4 - 12 \xi ^2 + 11\right ) \sin ( 2 \xi ) ( \pi -2 \text{si}(2 \xi ) ) + 27 \big\}.\end{aligned}\ ] ] the corrected frequencies for the tensor sector can be obtained directly from ( [ omegadecomposition ] ) by setting @xmath87 . note that , when written in terms of this dimensionless time variable @xmath118 , the only explicit dependence of the corrected frequencies on @xmath24 is encoded in the global factor written in front of the quantum - gravity corrections in the decomposition . furthermore , @xmath119 and @xmath120 are , respectively , the sine and the cosine integral functions : @xmath121 finally , the logarithmic term that appears in the definition of @xmath122 [ see ] comes from the @xmath67 in the exponent in the form of the potential . note that the argument of this logarithm should be @xmath123 but , since @xmath124 is chosen as the hubble - scale exit time , it turns out that @xmath125 . let us comment now on the behavior of these corrected frequencies in the limit of late and early times . at late times @xmath126 , or equivalently @xmath127 , it is possible to check by direct computation that _ ^2=k^2-+o(^4 ) . as can be seen , all the quantum - gravity corrections disappear in this limit . on the other hand , for the limit at early times @xmath128 ( @xmath129 ) , one obtains that [ asymptoticsomega ] _ ^2=k^2 + + o(1/^2 ) . as commented in sec . iii , this last result will provide us with the natural initial data for @xmath130 given in as well as for its linearized version @xmath131 , which is defined as @xmath132 at @xmath129 . in order to solve the linear equation for @xmath131 , we proceed in a similar way as in the decomposition for the corrected frequency . we split it into three different functions , each of them corresponding to one kind of correction : [ decompositionomegatilde1 ] ^(1):=()^2(^(1 ) _ + ^(1 ) _ + ^(1 ) _ ) , where the overall prefactor has been written for convenience . their corresponding initial data , at @xmath129 , can be straightforwardly obtained by linearizing the asymptotic behavior of the corrected frequency [ the square root of the right - hand side of ] on @xmath133 : @xmath134{}&\frac{1}{4},\\\label{initialdata2 } \widetilde\omega^{(1)}_{\epsilon}&\xrightarrow[\xi\rightarrow\infty]{}&-\frac{1}{12}\left[23 - 6\ln\xi\right ] , \\\label{initialdata3 } \widetilde\omega^{(1)}_{\gamma}&\xrightarrow[\xi\rightarrow\infty]{}&\frac{3 } { 8}.\end{aligned}\ ] ] in the case of @xmath135 , due to the presence of the logarithmic term , the limit turns out to be divergent . this is obviously not a physical consequence of the quantum - gravity corrections but an artifact of the slow - roll approximation which , as mentioned in sec . iv , is only valid for a large , but finite , interval of time . thus , in practice , the condition will have to be imposed for a large , but finite , value of @xmath136 . in order to find the equations of motion for the different @xmath131 , it is enough to insert the form into equation , linearize everything in the slow - roll parameters , and collect terms with the same slow - roll coefficient ( either 1 , @xmath67 or @xmath114 ) . following this procedure , eq . is rewritten as three different equations : @xmath137 where @xmath138 and @xmath139 have been given above in eqs . and . equation for @xmath140 is coupled to @xmath141 , whereas equation for @xmath135 is independent of it . this is due to the presence of the slow - roll parameter @xmath114 in @xmath142 . note that , due to the definitions and decompositions we have performed , in particular and , we have been able to write our fundamental equation as three differential equations with initial data , which do not depend on any parameter : they only depend on the time variable @xmath118 . neither @xmath143 nor @xmath40 appear in these equations and also none of the slow - roll parameters . furthermore , we have written the equations in terms of the time variable @xmath118 , such that there is no explicit dependence on @xmath24 either . [ note that the explicit inverse of the @xmath24 factor that appears multiplying @xmath138 and @xmath139 in these equations , cancels exactly with their linear dependence on @xmath24 , see eqs . ] . therefore , the explicit dependence of the result on the different parameters is analytically known . in particular , one can already deduce that the correction for the spectrum of the scalar perturbations @xmath144 defined in will have the following form : [ defdeltas2 ] _ = , with certain numerical factors @xmath145 , @xmath146 and @xmath147 . these factors are given by using the decompositions and in and linearizing in the slow - roll parameters , @xmath148 note again that all these are @xmath24-independent numbers . the explicit @xmath24 in front of the limit cancels out with the global @xmath24 factors in the expressions of @xmath149 and @xmath150 [ see eqs . ] . the correction @xmath151 , corresponding to the tensorial sector , can be obtained from the scalar one by imposing @xmath87 : [ defdeltat ] _ finally , the meaningful correction for the tensor - to - scalar ratio @xmath152 will be given by the difference between both : [ deltatminusdeltas ] _ -_=(- ) _ , where we have reintroduced the second slow - roll parameter @xmath76 . in this subsection we have already achieved the principal goal of this paper ; namely , obtaining the specific forms of @xmath144 and @xmath151 . the only issue left is to obtain the values of the numerical coefficients . this will be performed in the rest of the present section by solving the three equations above . for this purpose , except for the de sitter part , which is analytically solvable , numerical simulations will have to be performed . by construction , the equation for the de sitter part @xmath141 is the same as we found in our previous paper @xcite ( see eq . ( 144 ) there ) , and it can be analytically solved . we will not repeat here the whole computation , but note that the behavior of the solution at @xmath129 is given by ^(1)_=+c^2 i + o((^-2 ) ) . in order to impose the initial condition as given by and , in fact , to pick up the only non - oscillating solution , the integration constant @xmath89 must be chosen to be vanishing . after analyzing the behavior of the solution at the super - horizon limit , one can show that the quantum - gravity correction term corresponding to the de sitter part takes the following specific value : [ valuebetads ] _ = 0.988 , cf . ( 147 ) in @xcite . ( continuous black line ) and its asymptotic value ( red dashed line).,scaledwidth=50.0% ] ( the right - hand side of eq . without taking the limit ) as a function of @xmath118.,scaledwidth=50.0% ] the difficulty to solve this part lies in the presence of the logarithmic term both in the source and the asymptotic limit . in fact , if one removes this term , it is possible to solve equation and obtain the value of @xmath146 analytically . this analysis is presented in the appendix , in order to show that the presence of the logarithmic term , even if it is divergent in both early - time ( @xmath129 ) and late - time ( @xmath153 ) limits , does not change the result drastically . the complete equation must be solved by using numerical methods . in order to impose the initial condition , one needs to choose an initial large value of @xmath154 and insert that value into the function ^(1)_(_)= - . note that it is very convenient to pick up different values for @xmath155 to check that the final result does not depend on it . in particular , we have chosen @xmath156 , @xmath157 , and @xmath158 . the absolute difference between the various @xmath159 , as computed for those different values , is less than @xmath160 during the whole evolution ( for any of the considered values of @xmath118 ) , which translates to a negligible difference of around @xmath161 in the value of @xmath146 . in fact , in fig . [ omega1epsilon ] it is possible to see that @xmath162 tends very quickly to its asymptotic value , which explains the very weak dependence of the numerical solution on the chosen value of @xmath155 . finally , in order to compute numerically @xmath146 , we have evaluated the expression that appears in its definition for a very small value of @xmath163 , which leads to [ valuebetaepsilon ] _ note that all the different @xmath164 tend to vanish as @xmath153 and thus one can not numerically compute the limit by evaluating it at @xmath165 . nonetheless , in fig . [ betaepsilon ] the evolution of the right - hand side of ( without the limit taken ) has been plotted , in order to show that the limit behaves smoothly and is well defined . equation for @xmath140 is the most intricate one . it is , in particular , coupled to the equation for @xmath141 . as in the previous subsection , for the computation of @xmath146 , there seems to be no way to write the solution analytically in terms of special functions and thus , it is necessary to resort to a numerical resolution . nevertheless , the presence of the sine and cosine integral functions in the source term makes it computationally highly demanding to deal with this equation numerically for large values of their argument . thus , at early times ( @xmath166 ) , it will be very convenient to consider instead their series expansion . therefore , we have applied the following procedure . first , these two functions are replaced by their series expansion in inverse powers of @xmath118 up to 20th order in the source term . then , equation is solved with this approximated source term and by imposing the initial data @xmath167 for a very large initial time @xmath168 . next , the obtained numerical solution is used as initial data at a still large , but smaller value , of @xmath169 , for eq . with the full form of its source term . finally , this latter equation is solved and , with this numerical solution at hand , one can compute the expression at the left - hand side of eq . , whose limit as @xmath153 will define our quantity of interest @xmath147 . as in the previous subsection , this limit is numerically computed by evaluating the corresponding expression for a small value of @xmath118 . in order to check the robustness of this method , different values for @xmath155 and @xmath170 have been used . in particular , we have chosen @xmath155 as @xmath158 , @xmath171 , and @xmath172 , while for @xmath170 the values 100 , 400 , and 800 have been picked up . we find that the largest absolute difference between @xmath173 , as computed with the solution corresponding to these different values , during the whole evolution turns out to be smaller than @xmath174 . this is translated to an absolute difference of a similar order for @xmath147 . in fact , a value of @xmath175 is found and , thus , the error due to assuming an approximate equation for early times is very small . the evolution of the real part of @xmath140 is shown in fig . [ omega1gamma ] , in combination with its asymptotic value @xmath176 . note that @xmath177 is a non - oscillating solution and approaches its asymptote from below very quickly . in fig . [ betagamma ] , the evolution of the relation that defines @xmath147 [ the right - hand side of relation without taking the limit ] is shown in terms of @xmath118 . as can be seen clearly in the figure , its late - time limit behaves smoothly and leads to the value [ valuebetagamma ] _ 2.56 . ( black continuous line ) and its asymptotic value 3/8 ( red dashed line).,scaledwidth=50.0% ] ( the right - hand side of eq . without taking the limit ) as a function of @xmath118.,scaledwidth=50.0% ] in this section we will summarize our results and obtain the explicit form of the different parameters of the power spectrum , in particular the spectral indices and their running . we will also comment on the magnitude of the obtained corrections and the possibility of observing them . finally , we will give the form of the correction for the @xmath178 coefficients used in the cmb data analysis . the main results of this paper are the corrected forms of the power spectra for gauge - invariant scalar and tensor modes : @xmath179 where @xmath68 and @xmath180 are the usual power spectra , which can be obtained in the approximation of quantum fields propagating on fixed cosmological backgrounds and which are explicitly given in and , respectively . it corresponds to the standard result derived usually by other means ( see e.g. @xcite ) . the quantum - gravity corrections are thus encoded in the @xmath144 and @xmath151 terms , with the @xmath145 , @xmath181 and @xmath182 computed , respectively , in , , and . inserting explicitly these numerical values , the corrections to the power spectra read as follows : @xmath183 here , we have reverted the rescaling applied in and a reference wave number @xmath184 has been defined as the inverse of the length scale introduced to regularize the spatial integral . since the dependence of the quantum - gravity corrections on the @xmath24 wave numbers have been analytically obtained , it is possible to derive the spectral indices and their runnings by direct computation . we will use the usual parametrization of the power spectra given by the following power - law relation , @xmath185 where higher - order terms in @xmath186 have been neglected in the exponent and the pivot scale @xmath187 has been introduced . in this way , at a linear level in the slow - roll parameters , we obtain for the scalar sector @xmath188 \nonumber \\ & \approx 2\delta -4 \epsilon-\frac{h_k^2}{m_\text{p}^2}\left(\frac{\bar{k}}{k_*}\right)^3\left ( 2.96 + 11.40\,\epsilon-7.68\,\delta\right ) , \nonumber\end{aligned}\ ] ] with the first two terms taken together being the usual first - order approximation of the spectral index . in order to obtain this result , the relation = - , has been used . similarly , for the tensor sector , one obtains @xmath189 \nonumber \\ & \approx & -\,2\epsilon-\frac{h_k^2}{m_\text{p}^2}\left(\frac{\bar{k}}{k_*}\right)^3\left(2.96 + 3.72\,\epsilon\right ) . \nonumber\end{aligned}\ ] ] finally , one can also obtain the running of the spectral indices , @xmath190 by taking into account the dependence of the slow - roll parameters on the wave number . in particular , up to second - order terms we have @xmath191 where the second - order slow - roll parameter @xmath192 is defined as : = . in this way , one gets by direct computation the results @xmath193 \nonumber\\ & \approx&4\epsilon(\delta-\epsilon)-2\theta \nonumber\\ & & + \,\frac { h_k^2}{m_\text{p}^2 } \left(\frac{\bar{k}}{k_*}\right)^3 \left(8.89 + 40.12\,\epsilon - 23.04\,\delta\right ) , \nonumber\\ \alpha_\text{t}&=&4\epsilon(\delta-\epsilon)\\ & & + \,\frac{3 h_k^2}{m_\text{p}^2}\left(\frac{\bar{k}}{k_*}\right)^3 \bigl[3\beta_{\text{ds}}+\epsilon ( 4\beta_{\text{ds}}+3\bigl(\beta_\epsilon+\beta_\gamma)\bigr ) \bigr ] \nonumber \\ & \approx&4\epsilon(\delta-\epsilon)+\,\frac { h_k^2}{m_\text{p}^2 } \left(\frac{\bar{k}}{k_*}\right)^3 \left(8.89 + 17.08\,\epsilon\right ) , \nonumber\end{aligned}\ ] ] where only the first - order slow - roll terms have been kept in the correction term . note that the prefactors in the correction term for the runnings get larger than for the spectral indices due to the fact that the usual parametrization is not a good fit to the obtained quantum - gravity correction . let us finally also give the quantum - gravitational correction to the @xmath152-parameter . since this correction is given by , we obtain [ scalartensorratio ] r^(1)16(1 + 2.56 ( ) ^3(- ) ) . in most models @xmath194 , thus the correction gives a small negative contribution to the scalar - to - tensor ratio . note that , in the pure de sitter case there is no correction for this quantity , so it arises entirely from the slow - roll part . in order to give an estimation of the magnitude of the corrections for different quantities , one necessarily needs to assume a specific value for the scale @xmath195 . it seems reasonable to understand this scale as an infrared cut - off and relate it to the largest scale that could influence the cmb ( @xmath196 ) . nonetheless , in the literature there are some analyses that try to fix its value by other means . in particular , in ref . @xcite , which is based on a similar semiclassical approach to the geometrodynamical quantization of the present problem , arbitrary parameters are considered in front of the correction factor @xmath197 and they are fitted with the observational data from the planck mission @xcite . in particular , the best fit obtained relates @xmath198 to the size of galaxies or galaxy clusters ( @xmath199 ) . the meaning of such a scale is , however , not clear . all the above been said , in order to give an estimate of the magnitude of the correction we have obtained , here we will assume @xmath198 to be equal to the pivot scale chosen by the planck mission , that is @xmath200 . at the end of this subsection , we will comment about the maximum possible value of @xmath198 not to contradict the experimental data . as we have shown in our previous work @xcite , taking into account the energy scale of inflation in combination with the upper bound given by the planck mission @xcite for the scalar - to - tensor ratio @xmath201 , it is possible to derive the following condition , [ maxvaluehmp ] 1.3 10 ^ -5 , @xmath202 being the average hubble parameter during inflation . in the following , in order to give estimates for the correction to the quantities derived in the previous section , we will assume that @xmath203 . in addition , the experimental constraint for the spectral index , @xmath204 ( see @xcite ) , implies that @xmath205 and @xmath206 . with these values at hand , we can give an estimate for the upper limits of the quantum - gravity correction for scalar and tensor perturbations as follows : @xmath207 since the value of the slow - roll parameters is so small and the dominant de sitter contribution @xmath208 has a value close to 1 , the approximated upper bounds for the corrections coincide with the approximated maximal value of the ratio @xmath209 . using the estimated values for the slow - roll parameters , we can , however , deduce that the corrections for both kinds of perturbations differ by about @xmath210 : 1.02 . inserting the estimated numbers for @xmath67 and @xmath76 , we can immediately see that the correction to the spectral index is significantly smaller than the statistical uncertainty in the planck data : _ k=|k - 3.1 -5 10 ^ -10 . in this case , the quantum - gravitational correction to the spectral index is also tiny , _ k=|k - 3.0 -5 10 ^ -10 . estimating the magnitude of the quantum - gravity correction for the running gives @xmath211_{k=\bar{k } } & \approx 9.2\,\frac{h_{\text{inf}}^2}{m_\text{p}^2 } \approx 2 \times 10^{-9}\ , , \\ \left[\alpha_\text{t}^{(1 ) } - \alpha_\text{t}^{(0)}\right]_{k=\bar{k } } & \approx 9.0\,\frac{h_{\text{inf}}^2}{m_\text{p}^2 } \approx 2 \times 10^{-9}\,.\end{aligned}\ ] ] finally , the upper bound for the correction of the scalar - to - tensor ratio can be estimated as _ k=|k -0.023 -4 10 ^ -12 . as can be seen , all corrections are very small and they are inside the current experimental error bars . let us finally comment on the maximally allowed value for @xmath198 by the experimental data . the fact that the experimental errors are larger than the corrections we have obtained leads to the following relation , _ max=()^1/3 k _ * , @xmath212 being the relative experimental error in the power spectrum . in order to give a rough estimate , we assume that this error is of order one , which , as can be seen in @xcite , is a very high bound . using the maximum value for the ratio @xmath213 found in , we get [ kmax ] |k_max100 mpc^-1 . this implies a minimum value for the length scale @xmath214 . note , even though , that a lower value of @xmath213 would increase @xmath215 . in this subsection we will obtain the correction for the cmb temperature anisotropies , which are usually expressed by the quantities @xmath178 as defined below . in order to obtain these coefficients , it is necessary to evolve the scalar power spectrum through subsequent phases of the universe from the end of inflation until today . in addition , one finally needs to project it on the celestial sphere . this whole procedure can be reduced to compute the following integral , which is usually done numerically , c^(i)_=_0^ ^(i)_(k)_^2(k ) , with @xmath216 denoting the uncorrected and corrected coefficients , respectively , and @xmath217 being the transfer function . for large scales ( small @xmath218 ) , however , it is possible to solve this integral analytically . in this regime , the fluctuations were well outside the horizon at the end of recombination and thus they were not affected by subhorizon physics . therefore , it is only necessary to take into account the primordial spectrum and perform the projection on the celestial sphere . in particular , the transfer function can in this case be given in terms of the spherical bessel functions @xmath219 as follows @xcite , _ ( k)=j_(k[_- _ ] ) , where @xmath220 is the conformal time at horizon crossing and @xmath221 the conformal time at recombination . let us define the quantum - gravitational correction to the temperature anisotropies in the following way , c_:=c_^(1)- c_^(0 ) . applying the results for the corrected scalar power spectrum , we get that , for large scales , this correction has the following form , c__0^()^4()^3 j^2_(k[_- _ ] ) , the approximate symbol in this equation stands for two reasons . on the one hand , we are assuming an approximated transfer function . and , on the other hand , the overall factor that appears in the correction term , which depends on the slow - roll parameters but is of order one , has been dropped . at this point , as explained in @xcite , we use the fact that the bessel function is strongly peaked around @xmath222 and effectively acts as a dirac delta mapping between @xmath24 and @xmath218 . therefore , one can integrate the explicit @xmath24 dependences in the last integral , which leads to c_()^4 , while the implicit @xmath24-dependences on @xmath143 and @xmath67 should be evaluated at @xmath223 . applying exactly the same approximations , it is straightforward to obtain also the well - known result for the uncorrected temperature anisotropies at large scales : c_^(0 ) ( ) ^2 , which does not have any explicit dependence on @xmath224 and , for a scale - invariant spectrum ( constant @xmath143 and @xmath67 ) is just proportional to the inverse of @xmath225 . thus , due to a correction proportional to @xmath226 in the power spectrum , the temperature anisotropies get a correction that goes as the inverse of a fifth - order polynomial in @xmath218 for large scales . however , since the uncorrected @xmath227 go with @xmath228 , the relative correction @xmath229 is of order @xmath230 . as commented in the previous section , specially due to the presence of @xmath198 is this correction , it is difficult to be certain about its absolute value . we have thus plotted the behavior of the relative correction with respect to @xmath218 without the physical prefactors in fig . [ pl - plot ] . to the uncorrected @xmath231 without the non - numerical prefactors , such that we have + @xmath232^{-1}$].,scaledwidth=50.0% ] we can also check the relative values and see that the correction drops very quickly with @xmath218 . for instance , comparing it for the first multipoles : 0.09,0.02 , 0.007 . it is important to note that the qualitative behavior derived in this section for correction of the temperature anisotropies essentially comes from the explicit @xmath226 dependence of the correction of the power spectrum , which has been obtained in several approaches ( see , for instance , @xcite ) . we can also give an estimate of how large the hubble parameter , i.e. the energy scale , during inflation would have to be such that one could see a correction of our type . given that cosmic variance behaves like = , one can conclude that for @xmath233 , where cosmic variance is about @xmath234 and our quantum - gravitational correction is given by 0.12()^2||k(_-_)|^3 , the remaining factors in the above expression would have to be larger than 5 in order to clearly see an effect in the cmb data . using @xmath235 , and given that @xmath236 can be estimated to be about @xmath237 ( see e.g. tab . i in @xcite ) , the factor @xmath238 turns out to be of order @xmath239 . therefore , in order to see an effect , @xmath240 would be required , which is by far outside the range allowed by the measured tensor - to - scalar ratio . moreover , if we use the maximum allowed value by the latter limit for the hubble factor , it would be necessary for @xmath198 to be around @xmath241 to get an observable effect . this value is significantly smaller than the maximum value @xmath215 derived in . in this paper quantum - gravitational corrections for the power spectra of the gauge - invariant scalar and tensor perturbations have been obtained in the slow - roll regime . in particular , the corrected form for the scalar and tensor power spectrum is given , respectively , by and , where @xmath144 is given in and @xmath151 in in terms of the numerical coefficients @xmath145 , @xmath181 and @xmath182 . these coefficients have been computed by solving the linearized evolution equation for the gaussian width @xmath130 with natural initial data ( in the sense that the initial state , which is constructed as a small deformation of the usual bunch davies vacuum , best describes in this context the expected properties of a freely - evolving mode ) . their values are given in , , and , respectively . the above results generalize the results for the de sitter case obtained earlier in @xcite to a more realistic scenario of slow - roll inflation . in particular , and as one would naively expect , the main part of the correction is due to the de sitter contribution ( which introduces an enhancement of the spectrum ) , whereas the slow - roll part slightly modifies it . let us at this point briefly comment on the results of @xcite , obtained from an alternative expansion of the wheeler - dewitt equation . as in our treatment , the authors find quantum - gravitational correction terms proportional to @xmath242 . nonetheless , their result for the slow - roll approximation is not just a small perturbation of the de sitter case , but can give a comparable contribution for large scales , which can even lead to a power - loss instead of an enhancement of the power spectra . moreover , let us stress that the kind of correction that has been obtained in this analysis being proportional to the factor @xmath243 has appeared in several different approaches in the context of quantum geometrodynamics @xcite . the form of this correction is not completely unexpected . in fact , it is possible to argue , already on dimensional grounds , that @xmath244 is the only non - dimensional parameter that one could use to include perturbatively ( as a power series expansion ) quantum - gravity corrections . furthermore since , due to the background homogeneity , one needs to introduce explicitly a volume ( @xmath245 ) in order to regularize the spatial integral in the action , another dimensionless quantity @xmath246 enters the game . nevertheless , in principle , the power of this latter quantity might have been different and thus it is very interesting to see how the same correction is explicitly realized in different specific models . in the last section we have also analyzed the magnitude of the obtained corrections and the possibility of observing them experimentally . the most difficult issue in order to give a precise estimation is that , due to the regularization of the spatial integral in the action , a length scale needs to be considered . the power spectrum then depends on that length scale and there seems to be nothing physical to fix it . as we have commented , in the main part of the paper , the most reasonable choice is to take it as an infrared cut - off , relating it to the largest observable scale in the cmb . in our case , just to give an approximated estimate , we have chosen it as the length scale of a typical mode that affects the cmb . in particular we have chosen the pivot scale selected by the planck mission . in this way , we have obtained that the corrections for all different parameters of the power spectra ( spectral indices and runnings ) are well inside the current experimental error bars . finally , we have also obtained the qualitative form of the correction induced in the cmb temperature anisotropies by this quantum - gravity effect . the analysis we have performed is valid for large scales ( small @xmath218 ) , for which quantum - gravity effects are expected to be more relevant . in particular , it shows that a correction of the form @xmath226 which , as commented above , seems very generic in this context , leads to a relative correction of the order @xmath230 for the anisotropies , which thus quickly declines with increasing @xmath218 . with this paper we conclude our investigations on quantum - gravitational corrections arising from a canonical quantization of a perturbed universe model using the wheeler - dewitt equation . the effects on large scales we have obtained are , for a reasonable choice of @xmath198 , not observable in the cmb data and since we have used a generic slow - roll model that encompasses a wide range of inflationary models , using more refined models that obey the slow - roll approximation would not enhance the corrections . nonetheless , it is still an open question whether such corrections can be observed in situations where cosmic variance is not present ; for example , in galaxy - galaxy correlation functions . we thank iaki garay and jon urrestilla for interesting discussions . d.b . is supported by project fis2014 - 57956-p of the spanish ministry of economy and competitiveness . the research of m.k . was financed by the polish national science center grant dec-2012/06/a / st2/00395 . is shown , for different range of values of @xmath118 , as given by its full equation ( black continuous line ) and as obtained by dropping the logarithmic term ( black dashed - dotted line ) . their corresponding asymptotes are shown as a red - dashed line and as a red - dotted line , respectively.,title="fig:",scaledwidth=50.0% ] is shown , for different range of values of @xmath118 , as given by its full equation ( black continuous line ) and as obtained by dropping the logarithmic term ( black dashed - dotted line ) . their corresponding asymptotes are shown as a red - dashed line and as a red - dotted line , respectively.,title="fig:",scaledwidth=50.0% ] in this appendix , eq . , with the logarithmic term dropped , will be solved in order to show that , even if this term is divergent in both late and early time limits , the value that is obtained for @xmath146 does not change dramatically . note that , if one removes the logarithmic term from the source term , one should also remove it from the initial condition . interestingly , in this case eq . can be analytically solved , and the solution takes the following form : @xmath247,\end{aligned}\ ] ] where @xmath89 is an integration constant . if we analyze the behavior of this solution at @xmath129 , we find that ^(1)_--c^2 . therefore , in order to have a non - oscillating solution , we choose @xmath248 . finally , we compute the limit defined in eq . and find the following value of @xmath146 : _ = 1+(^2 - 3 ^4 ( -2)-9(2))-2.62 . as commented above , this proves that the logarithmic term is indeed important to compute the precise value for @xmath146 , but it is not critical in the sense that qualitatively the same result is obtained if one drops it . in fig . [ comparisonomega1epsilon ] , the evolution of the real part of @xmath249 is shown for both the solution with and without the logarithmic term , in combination with their corresponding asymptotes . it can be seen that the tendency at late times is quite similar for both , which explains the weak dependence of @xmath181 on the commented term . c. kiefer , _ quantum gravity _ , international series of monographs on physics * 155 * , third edition ( oxford university press , oxford , 2012 ) . c. kiefer and m. krmer , int . d * 21 * , 1241001 ( 2012 ) . c. kiefer and m. krmer , phys . lett . * 108 * , 021301 ( 2012 ) . d. bini , g. esposito , c. kiefer , m. krmer , and f. pessina , phys . d * 87 * , 104008 ( 2013 ) . g. calcagni , ann . ( berlin ) * 525 * , 323 ( 2013 ) ; erratum _ ibid . d. bini and g. esposito , phys . d * 89 * , 084032 ( 2014 ) . g. l. alberghi , r. casadio , and a. tronconi , phys . d * 74 * , 103501 ( 2006 ) . a. y. kamenshchik , a. tronconi , and g. venturi , phys . b * 726 * , 518 ( 2013 ) . a. y. kamenshchik , a. tronconi , and g. venturi , phys . b * 734 * , 72 ( 2014 ) . a. y. kamenshchik , a. tronconi , and g. venturi , j. cosmol . astropart . phys . 04 ( 2015 ) 046 . a. y. kamenshchik , a. tronconi , and g. venturi , arxiv:1609.02830 [ gr - qc ] . i. agullo , a. ashtekar , and w. nelson , phys . lett . * 109 * , 251301 ( 2012 ) ; phys . d * 87 * , 043507 ( 2013 ) ; class . . grav . * 30 * , 085014 ( 2013 ) . d. martn de blas and j. olmedo , jcap * 06 * , 029 ( 2016 ) . b. bolliet , j. grain , c. stahl , l. linsefors , and a. barrau , phys . d * 91 * , 084035 ( 2015 ) . s. schander , a. barrau , b. bolliet , l. linsefors , j. mielczarek , and j. grain , phys . d * 93 * 023531 ( 2016 ) . a. ashtekar and b. gupt , arxiv:1608.04228 [ gr - qc ] . c. kiefer and t. p. singh , phys . d * 44 * , 1067 ( 1991 ) ; c. kiefer , lect . notes phys . * 434 * , 170 ( 1994 ) . c. bertoni , f. finelli , and g. venturi , class . quantum grav . * 13 * , 2375 ( 1996 ) . d. brizuela , c. kiefer and m. krmer , phys . d * 93 * , 104035 ( 2016 ) . v. f. mukhanov , h. a. feldman , and r. h. brandenberger , phys . rep . * 215 * , 203 ( 1992 ) . p. peter and j .- p . uzan , _ primordial cosmology _ ( oxford university press , oxford , 2009 ) . l. parker and d. toms , _ quantum field theory in curved spacetime _ , cambridge university press , cambridge , ( 2009 ) . b. j. broy , arxiv:1609.03570 [ hep - th ] . j. martin , v. vennin , and p. peter , phys . d * 86 * , 103524 ( 2012 ) . p. a. r. ade _ et al . _ ( planck collaboration ) , astron . astrophys . * 594 * , a13 ( 2016 ) . p. hunt and s. sarkar , jcap * 12 * , 052 ( 2015 ) . s. dodelson , _ modern cosmology _ , academic press , elsevier ( 2003 ) . d. baumann , _ physics of the large and the small _ , tasi 09 , proceedings of the theoretical advanced study institute in elementary particle physics , world scientific ( 2009 ) .

we have carried out large scale co observations with a mm / sub - mm telescope nanten toward a far infrared loop - like structure whose angular extent is about 20 degrees around ( , ) ( 109 , ) in pegasus . its diameter corresponds to 25 pc at a distance of 100 pc , adopted from that of a star hd886 ( b2iv ) near the center of the loop . we covered the loop - like structure in theco ( = 10 ) emission at 4 grid spacing and in theco ( = 10 ) emission at 2 grid spacing for theco emitting regions . theco distribution is found to consist of 78 small clumpy clouds whose masses range from 0.04 to 11 , and 83% of theco clouds have very small masses less than 1.0 . co observations revealed that 18 of the 78co clouds show significantco emission . co emission was detected in the region where the column density of h derived fromco is greater than 5 , corresponding tov of 1 mag , which takes into account that of hi . we find no indication of star formation in these clouds in iras point source catalog and 2mass point source catalog . the very low mass clouds , 1 , identified are unusual in the sense that they have very weakco peak temperature of 0.5 k2.7 k and that they aggregate in a region of a few pc with no main massive clouds ; contrarily to this , similar low mass clouds less than 1 in other regions previously observed including those at high galactic latitude are all associated with more massive main clouds of 100 . a comparison with a theoretical work on molecular cloud formation ( koyama & inutsuka 2002 ) suggests that the very low - mass clouds may have been formed in the shocked layer through the thermal instability . the star hd886 ( b2iv ) may be the source of the mechanical luminosity via stellar winds to create shocks , forming the loop - like structure where the very low - mass clouds are embedded .

astro-ph0601315

high galactic latitude molecular clouds ( hereafter hlcs ) are typically located at @xmath17b@xmath17 @xmath18 20@xmath430@xmath4 . since the gaussiun scale height of co is estimated to be @xmath3 100 pc in the inner galactic disk ( e.g. , magnani et al . 2000 ) , hlcs are likely located very close to the sun , within a few hundred pc or less . their proximity to the sun and the low possibility of overlapping with other objects along the line of sight enable us to study them with a high spatial resolution and to compare co data unambiguously with the data at other wavelengths . hlcs have lower molecular densities compared with dark clouds where the optical obscuration is significant . therefore , hlcs are often called as translucent clouds ( e.g. , van dishoeck & black 1988 ) and most of the known hlcs are not the sites of active star formation , although a few of them are known to be associated with t tauri stars ( e.g. , magnani et al . 1995 ; pound 1996 ; hearty et al . 1999 ) . given the very small distances of hlcs , it is a challenging task for observers to make a complete survey for hlcs over a significant portion of the whole sky . @xmath6co ( @xmath7 = 10 ) emission has been used to search for hlcs because the line emission in the mm band is strongest among the thermally or sub - thermally excited spectral lines of interstellar molecular species . it is however difficult to cover an area as large as tens of square degrees subtended by some of the hlcs because of the general weakness of the @xmath6co emission , typically @xmath3 a few k ( e.g. , magnani et al . 1996 ) , with existing mm - wave telescopes in a reasonable time scale . hlcs have been therefore searched for by employing various large - scale datasets at other wavelengths including the optical obscuration ( magnani et al . 1985 ; keto & myers 1986 ) , the infrared radiation ( reach et al . 1994 ) , and the far - infrared excess over hi ( = fir excess)(blitz et al . 1990 ; onishi et al . 2001 ) . on the other hand , unbiased surveys in co at high galactic latitudes have been performed at very coarse grid separations of 1@xmath4 resulting in a small sampling factor of a few % ( hartmann et al . 1998 ; magnani et al . most recently , onishi et al . ( 2001 ) discovered 32 hlcs or hlc complexes . this search was made based on the fir excess , demonstrating the correlation among fir excess clouds with co clouds is a useful indicator of co hlcs . previous co observations of individual hlcs at higher angular resolutions show that hlcs exhibit often loop - like or shell - like distributions having filamentary features with widths of several arc min or less ( hartmann et al . 1998 ; magnani et al . 2000 ; bhatt 2000 ) , and in addition that hlcs often compose a group , whose angular extent is @xmath3 10 degrees or larger . in order to better understand the structure of hlcs and to pursue the evolution of hlc complexes , co observations covering tens of square degrees at a high angular resolution are therefore crucial . the past observations of such complexes of hlcs are limited to a few regions including polaris flare ( heithausen & thaddeus 1990 ) , ursa major ( pound & goodman 1997 ) and the hlc complex toward mbm 53 , 54 , and 55 ( yamamoto et al . pound & goodman ( 1997 ) showed an arc - like structure of the molecular cloud system and suggested that the origin of such structures could be some explosive events . most recently , yamamoto et al . ( 2003 ) carried out extensive observations of the molecular cloud complex including mbm 53 , 54 , and 55 and suggest that the hlcs may be significantly affected by past explosive events based on the arc - like morphologies of molecular hydrogen ( see also gir et al . 1994 ) . the region of mbm 53 , 54 , and 55 is of particular interest among the three , because it is associated with a large hi cloud of @xmath3 590 @xmath10 at a latitude of @xmath535 degrees and because there is a newly discovered hlc of 330@xmath10 , hlcg92@xmath535 , which is significantly hi rich with a mass ratio @xmath15(h@xmath11)/@xmath15(hi ) of @xmath3 1 , among the known hlcs ( yamamoto et al . this cloud was in fact missed in the previous surveys based on optical extinction ( magnani et al . subsequent to these observations we became aware of that the region is also very rich in interstellar matter as shown by the 100@xmath19 m dust features ( kiss et al . there is a loop - like structure shown at 100 @xmath19 m around ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath545@xmath4 ) . toward the center of the loop , an early type star hd886(b2iv ) is located and may play a role in creating the loop . its proper motion is large at a velocity of a few km s@xmath20 , suggesting that the stellar winds of the star might have continued to interact with the surrounding neutral matter over a few tens of pc in @xmath3 a few myr . magnani et al . ( 1985 ) and onishi et al . ( 2001 ) yet observed only a small part of this region . in order to reveal the large scale co distribution of the region , we have carried out observations toward ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath545@xmath4 ) by @xmath6co ( @xmath7 = 10 ) and @xmath9co ( @xmath7 = 10 ) with nanten 4-meter millimeter / sub - mm telescope of nagoya university at las campanas , chile . we shall adopt the distance of 100 pc from the sun to the loop - like structure which is equal to the distance of the b2 star in the center of the loop , and is also a typical value for the hlcs . @xmath6co ( @xmath7 = 10 ) and @xmath9co ( @xmath7 = 10 ) observations were made with the 4-meter telescope , nanten , of nagoya university at las campanas observatory of carnegie institutions of washington , chile . the front - end was an sis receiver cooled down to 4 k with a closed - cycle helium gas refrigerator ( ogawa et al . 1990 ) . the backend was an acousto - optical spectrometer with 2048 channels , and the total bandwidth was 40 mhz . the frequency resolution was 35 khz , corresponding to a velocity resolution of @xmath3 0.1 km s@xmath20 . a typical system noise temperature was @xmath3 200 k ( ssb ) at 115.271 ghz and @xmath3 150 k ( ssb ) at 110.201 ghz . the half - power beam width was about 2@xmath216 , corresponding to 0.076 pc at a distance of 100 pc . the pointing accuracy was better than 20@xmath22 , as established by radio observations of jupiter , venus , and the edge of the sun in addition to optical observations of stars with a ccd camera attached to the telescope . the observed region in @xmath6co was @xmath3 240 square degrees toward the whole area of the loop - like structure centered at around ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath545@xmath4 ) shown in a 100 @xmath19 m map by schlegel et al . first , the @xmath6co observations were made at a grid spacing of 8@xmath8@xmath0cos(@xmath2 ) and 8@xmath8 in galactic longitude and latitude , respectively . then , the regions where the @xmath6co emission is significantly detected were observed at a grid spacing of 4@xmath8@xmath0cos(@xmath2 ) and 4@xmath8 in galactic longitude and latitude , respectively . the @xmath9co observations were made in and around the whole area where the peak temperature of @xmath6co emission is higher than 2.0 k at a grid spacing of 2@xmath8@xmath0cos(@xmath2 ) and 2@xmath8 in galactic longitude and latitude , respectively . the periods of @xmath6co observations were several sessions between 2002 may and november and those of @xmath9co were those between 2003 april and august . all the observations were made by frequency switching whose interval is 20 mhz , corresponding to @xmath3 50 km s@xmath20 . the integration times per point of @xmath6co and @xmath9co observations were typically @xmath3 30 s and @xmath3 75 s , respectively , resulting in typical rms noise temperatures per channel of @xmath3 0.35 k and @xmath3 0.15 k in the radiation temperature , @xmath23 , respectively . in reducing the spectral data , we subtracted forth - order polynomials for the emission - free parts in order to ensure a flat spectral baseline . total numbers of observed points of @xmath6co and @xmath9co are 16890 and 3100 , respectively . we employed a room - temperature blackbody radiator and the sky emission for the intensity calibration . an absolute intensity calibration and the overall check of the whole system were made by observing orion kl [ @xmath24(1950 ) = 5@xmath2532@xmath2647.@xmath270 , @xmath28(1950 ) = @xmath55@xmath424@xmath821@xmath22 ] every 2 hours . we assumed the @xmath23 of orion kl to be 65 k for @xmath6co and 10 k for @xmath9co . figure 1 shows the distribution of the velocity - integrated intensity map of @xmath6co emission . we defined a @xmath6co cloud as a collection of more than two contiguous observed positions whose integrated intensity exceeds 0.77 k km s@xmath20 ( 5@xmath30 ) . based on the definition , we identified 78 molecular clouds in this region . molecular clouds are concentrated from ( @xmath1 , @xmath2 ) @xmath3 ( 107@xmath4 , @xmath537@xmath4 ) to ( 116@xmath4 , @xmath545@xmath4 ) and around ( @xmath1 , @xmath2 ) @xmath3 ( 114@xmath4 , @xmath552@xmath4 ) . most of the molecular clouds are very small , having size of @xmath31 1@xmath4 . figure 2 shows the distribution of the co superposed on the sfd 100 @xmath19 m ( schlegel et al . 1998 ) , which was derived from a composite of the cobe / dirbe and iras / issa maps , with the foreground zodiacal light and confirmed point sources removed . co clouds are distributed along the infrared loop whose diameter is @xmath3 25 pc . we detected little co emission within the loop - like structure , while toward some of the local peaks of sfd 100 @xmath19 m there is no co emission . figure 3 shows the peak radial velocity distribution derived from the present @xmath6co data set . the velocity in figure 3 is derived by a single gaussiun fitting from all co spectra . the velocity range of the molecular clouds is from @xmath518.3 km s@xmath20 to 0.3 km s@xmath20 and there is no systematic large scale velocity gradients . some of the molecular clouds have already been known by previous observations . molecular clouds toward ( @xmath1 , @xmath2 ) @xmath3 ( 110@xmath3218 , @xmath541@xmath3223 ) and ( 117@xmath3236 , @xmath552@xmath3228 ) are identified by magnani et al . ( 1985 ) and named as mbm 1 and mbm 2 , respectively . dir117@xmath544 and dir105@xmath538 identified by reach et al . ( 1998 ) are also identified in co toward ( @xmath1 , @xmath2 ) @xmath3 ( 116@xmath325 , @xmath544@xmath320 ) and ( 105@xmath320 , @xmath538@xmath320 ) by onishi et al . magnani et al . ( 1986 ) detected co emission at ( @xmath1 , @xmath2 ) @xmath3 ( 112@xmath4 , @xmath540@xmath4 ) . magnani et al . ( 2000 ) also covered this region even though they made observations on a locally cartesian grid with 1@xmath4(true angle ) spacing in longitude and latitude for a beam size of 8@xmath218 , they detected co emission at eight positions of ( @xmath1 , @xmath2 ) @xmath3 ( 103@xmath322 , @xmath538@xmath320 ) , ( 103@xmath322 , @xmath539@xmath320 ) , ( 104@xmath324 , @xmath539@xmath320 ) , ( 106@xmath328 , @xmath537@xmath320 ) , ( 108@xmath320 , @xmath552@xmath320 ) , ( 109@xmath325 , @xmath551@xmath320 ) , ( 110@xmath324 , @xmath541@xmath320 ) , and ( 111@xmath320 , @xmath550@xmath320 ) in the present region while they missed the present small molecular clouds whose sizes are less than several arc min in figure 1 due to the coarse grid spacing . seventy - eight @xmath6co molecular clouds are identified in the present region . for each molecular cloud , @xmath33 derived from single gaussian fitting was from 0.5 to 3.7 km s@xmath20 , and the radial velocity , @xmath34 , ranges from @xmath515.7 to @xmath50.1 km s@xmath20 . the maximum brightness temperature , @xmath23(@xmath6co ) ranges from 0.5 to 5.7 k. the radius of a cloud , @xmath35 , which is defined as the radius of an equivalent circle having the same area , i.e. , and @xmath35(pc)@xmath36 where a is the total cloud surface area within the 5@xmath30contour level , ranges from 0.07 to 0.79 pc . the peak column density of molecular hydrogen , @xmath37(h@xmath11 ) , in each cloud derived by assuming a conversion factor of 1.0@xmath010@xmath12 @xmath13/(k km s@xmath20 ) ( magnani et al . 2000 ) ranges from 8.0@xmath010@xmath38 to 1.7@xmath010@xmath39 @xmath13 with the present detection limit , 7.7@xmath010@xmath38 @xmath13 , corresponding to mass detection limit of 0.014 @xmath10 . we estimate the molecular mass , @xmath15(@xmath6co ) , by using the following formula @xmath40,\ ] ] where @xmath19 is the mean molecular weight , assumed to be 2.8 by taking into account a relative helium abundance of 25% in mass , @xmath41 is the mass of the atomic hydrogen , @xmath42 is the distance from the sun to the molecular clouds , and @xmath43 is the solid angle subtended by a unit grid spacing of ( 4@xmath8)@xmath0(4@xmath8@xmath0cos(@xmath2 ) ) . @xmath15(@xmath6co ) ranges from @xmath3 0.04 to @xmath3 11 @xmath10 and the total mass of molecular clouds is @xmath3 64 @xmath10 . these physical properties are listed in table 1 and the histograms of @xmath23(@xmath6co ) , @xmath33 , log(@xmath35 ) , and log(@xmath37(h@xmath11 ) ) of these clouds are shown in figure 4 . histograms in figure 4 are divided into three different categories , _ usual cloud _ ( hereafter uc ) whose mass is greater than 1 @xmath10 , _ small cloud _ ( hereafter sc ) whose mass is between 0.1 and 1 @xmath10 , and _ very small cloud _ ( hereafter vsc ) whose mass is less than 0.1 @xmath10 . it is remarkable that there are a number of molecular clouds having mass less than 1 @xmath10 and that the fractions of sc and vsc are 43/78 @xmath3 55% and 22/78 @xmath3 28% in the present region , respectively . in addition , the sizes of sc and vsc are equal to or less than 0.1 pc . we also note that the peak temperatures of sc and vsc are typically in a range from 0.5 k to 2.7 k , well below that of uc in the same region . figure 5 shows the distribution of the velocity - integrated intensity map of the @xmath9co emission superposed on the @xmath6co distribution . the total area of the @xmath9co observations is @xmath3 29 square degrees toward 38 of the 78 @xmath6co clouds . we observed all of 13 ucs , 24 of 43 scs , and 3 of 22 vscs . we detected @xmath9co emission at 11 of the 13 ucs , 8 of the 24 scs , and none of the 3 vscs , indicating a trend that the @xmath9co intensity increases with @xmath6co cloud mass . a @xmath9co cloud is defined in the same way as for a @xmath6co cloud except for the lowest integrated intensity level , 0.3 k km s@xmath20 ( 3@xmath30 ) . based on the definition , we identified 33 @xmath9co clouds . for the 33 @xmath9co molecular clouds , @xmath44@xmath45 derived from single gaussian fitting is @xmath3 1.5 km s@xmath20 and @xmath34 of them ranges from @xmath513.1 to @xmath51.9 km s@xmath20 . other physical properties , the maximum brightness temperature , @xmath23(@xmath9co ) , and @xmath35 range from 0.3 to 2.3 k and from 0.04 to 0.21 pc , respectively . the physical parameters including the molecular column density and mass ( hereafter @xmath46 ) are derived on the assumption of local thermodynamic equilibrium ( lte ) . to derive the column density of molecular hydrogen , the optical depth of @xmath9co is estimated by using the following equations , @xmath47,\ ] ] where @xmath48 is the excitation temperature of the @xmath7 = 10 transition of co in k and was derived from @xmath49\right\}}.\ ] ] @xmath48 was estimated to be 9.4 k from our @xmath6co data . the @xmath9co column density , @xmath37(@xmath9co ) , is estimated by @xmath50 } ( { \rm cm^{-2}}).\ ] ] the ratio of @xmath37(h@xmath11)/@xmath37(@xmath9co ) was assumed to be 7@xmath010@xmath51 ( dickman 1978 ) . the @xmath46 of a cloud from @xmath37(h@xmath11 ) is derived by the same way as @xmath6co ( see equation ( 1 ) ) . the column density and @xmath46 range from 2.3@xmath010@xmath12 to 1.7@xmath010@xmath39 @xmath13 and 0.03 to 1.41 @xmath10 , respectively , where the detection limit in the column density is 2.0@xmath010@xmath12 @xmath13 , coressponding to mass limit of 0.009 @xmath10 , smaller than that of @xmath6co because the observations of @xmath9co were made by higher grid sampling and lower rms noise fluctuations than those of @xmath6co , respectively . figure 6 shows the histograms of each physical property . the virial mass , @xmath52 , of a cloud was derived by using the following equation , assuming isothermal , spherical , and uniform density distribution with no external magnetic pressure : @xmath53 where @xmath35 and @xmath54 are the radius ( pc ) and line width ( km s@xmath20 ) of the composite profile obtained by averaging all the spectra within a cloud , respectively ( for details of the line width of composite profiles , see yonekura et al . 1997 ; kawamura et al . 1998 ) . from this equation , @xmath52 is estimated to be in a range from 4.7 to 197 @xmath10 . these physical properties are also listed in table 2 . figure 7a and 7b show the mass spectrum of the present @xmath6co and @xmath9co clouds . the spectra have been fitted by the maximum - likelihood method ( crawford et al . 1970 ) , and it is found that they are well fitted by a single power law as follows ; @xmath55 @xmath56 @xmath57 for the @xmath6co clouds and @xmath55 @xmath56 @xmath58 for the @xmath9co clouds . these values of the spectral indices seem to be similar to those for the higher mass range ( e.g. , yonekura et al . 1997 ) . figure 8 shows a plot of size , @xmath35 , versus line width , @xmath33 , of the @xmath9co clouds in this region and for a comparison with other hlcs , mbm 53 , 54 , and 55 complex ( yamamoto et al . 2003 ) . we can make fitting as follows by using a least - squares fitting , log(@xmath33 ) = ( 0.22@xmath590.43 ) @xmath0 log(@xmath35 ) + ( 0.37@xmath590.52 ) ( c.c.=0.23 ) for the present region and log(@xmath33 ) = ( 0.43@xmath590.32 ) @xmath0 log(@xmath35 ) + ( 0.53@xmath590.28 ) ( c.c.=0.37 ) for mbm 53 , 54 , and 55 complex . the low correlation coefficient ( c.c . ) indicates that there is no significant correlation between @xmath35 and @xmath33 because of a small range of @xmath35 . here we do not show the same relationship for @xmath6co , because the non - circular shape of the @xmath6co clouds may not be appropriate to derive reliable @xmath35 . figure 9 shows a plot of @xmath46 versus @xmath52 . the present @xmath9co clouds are located far above the equilibrium line where @xmath46 is equal to @xmath52 , indicating that the @xmath9co clouds are not in the virial equilibrium . this indicates that none of the molecular clouds are gravitationally bound . these parameters can be fitted by using a least - squares fitting as follows , log(@xmath52 ) = ( 0.91@xmath590.30 ) @xmath0 log(@xmath46 ) + ( 2.23@xmath590.29 ) ( c.c.=0.66 ) for present molecular clouds and log(@xmath52 ) = ( 0.77@xmath590.13 ) @xmath0 log(@xmath46 ) + ( 0.16@xmath590.08 ) ( c.c.=0.74 ) for mbm53 , 54 , and 55 complex . as mentioned in yamamoto et al . ( 2003 ) , the present molecular clouds also tend to be more virialized as the mass increases . for the gemini and auriga , and cepheus - cassiopeia regions , the indices of @xmath46 for @xmath52 of @xmath9co clouds are estimated to be 0.72@xmath590.03 and 0.62@xmath590.03 for the cloud mass range of @xmath46 @xmath60 10@xmath61 @xmath10 and 10@xmath62 @xmath10 @xmath60 @xmath46 @xmath60 10@xmath51 @xmath10 , respectively ( kawamura et al . 1998 ; yonekura et al . 1997 ) . although the mass ranges of mbm 53 , 54 , and 55 complex and of this region are 10@xmath20 @xmath10 @xmath60 @xmath46 @xmath60 10@xmath62 @xmath10 and 10@xmath63 @xmath10 @xmath60 @xmath46 @xmath60 1 @xmath10 , respectively , difference in the power - law indices among these regions is small and a tendency that the scs have large ratios of @xmath52/@xmath46 is commonly seen . in order to look for sings of star formation associated with the present molecular clouds , we searched the iras point source catalog for candidates of protostellar objects satisfying the following criteria : ( 1 ) point sources having a data quality flag better than 2 in 4 bands , ( 2 ) flux ratios at 12 , 25 , 60 @xmath19 m satisfying both log(@xmath64/@xmath65 ) @xmath60 @xmath50.3 and log(@xmath65/@xmath66 ) @xmath60 0 , and not identified as galaxies or planetary nebulae and stars . we find that there are no cold iras point sources satisfying these criteria in the present molecular clouds . we also find that there are no iras point sources having a spectrum like a t - tauri type star or no ysos identified from point source catalog of two - micron all - sky survey in this region . here we select the 2mass sources whose signal to noise ratio of valid measurements in all bands are greater than 10 and extract the sources which have the spectra like t tauri stars in ( @xmath7@xmath5@xmath67)(@xmath67@xmath5@xmath68 ) color color diagram ( e.g. , meyer et al . these results suggest that the present molecular clouds are not the site of recent star - formation , or that the region is not remnants of past star formation . such a low level of star formation is similar to the other hlcs including mbm 53 , 54 , and 55 complexes . figure 10 shows the integrated intensity map of hi taken from a leiden - dwingeloo hi survey ( hartmann & burton 1997 ) superposed on the integrated intensity of co. the integrated velocity range is from @xmath516 to 0 km s@xmath20 , corresponding to the velocity range of the @xmath6co emission . because an angular resolution of @xmath3 30@xmath8 is coarser than that of the present co observations by a factor of @xmath3 10 , we discuss here only the overall comparison between co and hi distributions . the hi distribution is loop - like and the molecular clouds are distributed nearly along the hi loop . figure 11 shows the position - velocity diagram of hi integrated from @xmath540@xmath325 to @xmath539@xmath325 and @xmath552@xmath325 to @xmath551@xmath325 in galactic latitude , respectively . the hole like structures can be seen in hi , suggesting that these hi clouds are expanding . these expanding structures in figures 11(a ) and 11(b ) correspond to the galactic northern and southern hi shells shown as the thick dashed ( semi ) ellipses in figure 10 , respectively , while the expanding motion is not seen in co ( see figure 3 ) . these two expanding shells are also identified by an infrared radiation ( kiss et al . 2004 ) . from figure 10 , an hi cloud around ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath552@xmath4 ) seems to be located on the left side of the southern expanding shell and the molecular clouds are distributed nearby . it is difficult to distinguish with which hi shell the molecular clouds located around ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath552@xmath4 ) are associated because the hi velocities of the two shells are similar with each other . the shape of @xmath6co and sfd100 @xmath19 m radiation around ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath552@xmath4 ) in figure 2 is also similar to the left side of the expanding shell . if this is true , there may be two expanding structures in the present region . hd886 ( 109@xmath3243 , @xmath546@xmath3268 ) is located near the center of the northern expanding shell , indicating that hd886 may be affecting the northern expanding shell . the parallax of hd886 has been measured to be 9.79@xmath590.81 mas ( perryman et al . 1997 ) , corresponding to a distance from the sun of 102@xmath69 pc . the proper motion has also been measured to be @xmath70=0@xmath710047 yr@xmath20 , @xmath72=@xmath50@xmath710082 yr@xmath20 by perryman et al . ( 1997 ) . from this proper motion , the velocity of hd886 in the l - b map is estimated to be @xmath3 4.3@xmath010@xmath73 pc yr@xmath20 at @xmath3 100 pc ( see figure 10 ) . we use typical values of the stellar wind for b2(iv ) star on @xmath74/@xmath75=10@xmath76 @xmath10 yr@xmath20 and @xmath77 = 1000 km s@xmath20 ( e.g. , snow 1982 ) . from these parameters , the energy injected to the northern shell is estimated to be @xmath3 10@xmath78 ergs in a few @xmath010@xmath79 yr . the expanding energy of the northern shell is estimated to be @xmath3 10@xmath80 ergs from the atomic and molecular hydrogen , using that the masses of amomic and molecular hydrogen associated with the northern shell are @xmath3 400 and 42 @xmath10 , respectively , the expanding velocity is @xmath3 7 km s@xmath20 which is estimated from figure 11 and the equation of @xmath81=1/2@xmath15@xmath82 . since the expanding energy of the atomic and molecular hydrogen is comparable to the energy from hd886 , additional source of energy other than hd886 such as photo evaporation is needed to explain the expanding energy because the energy conversion efficiency of the stellar wind is @xmath31 10% . the energy of the southern shell is estimated to be @xmath3 10@xmath78 ergs , by using that the masses of atomic and molecular hydrogen are @xmath3 1000 and 16 @xmath10 , respectively , and that the expanding velocity is @xmath3 9 km s@xmath20 . we could not find possible candidates of the energy source for the expanding feature in the literature ( simbad ) and there are no counterparts in optical or x - ray wavelength . although we could not identify the possible candidates , we can not exclude a possibility that these objects may have escaped from the region in a few @xmath0 10@xmath83 yr after forming these structures . the present observations have revealed numerous molecular clouds having very small mass of less than 1 @xmath10 . it is of considerable interest to pursue the physical states of these scs from view points of cloud physics and chemistry as well as of the origin of molecular clouds . we shall hereafter focus on the low mass @xmath6co clouds whose mass is less than 1 @xmath10 . the total number of such clouds is 65 among 78 . the @xmath9co emission has been searched for toward 24 of the 43 @xmath6co low - mass clouds whose mass is in a range of 0.11.0 @xmath10 and has been detected from 8 of them . figure 12 shows correlations of the molecular column density , estimated from @xmath6co and @xmath9co , of the clouds where both @xmath6co and @xmath9co emission are detected . it is seen that almost all of the @xmath6co clouds having molecular column density greater than 5@xmath010@xmath12 @xmath13 , corresponding to the visual extinction of 0.55 mag if we use the relationship of @xmath84=9.4@xmath010@xmath12@xmath0@xmath14v @xmath13 ( bohlin et al . 1978 ; hayakawa et al . 1999 ) , show significant @xmath9co emission . we note that there are 22 @xmath6co clouds whose mass is less than 0.1 @xmath10 ; for those it is doubtful that the @xmath9co emission is so significant as those whose mass is a range of 0.11.0 @xmath10 although only 3 of them were searched for the @xmath9co emission in the present study . we note that we ignore the possible contribution of atomic hydrogen in the above relationship . if we take into account the distribution of atomic hydrogen as @xmath37(hi ) , the visual extinction would increase by 0.2 @xmath3 0.3 mag . this may be explained as that the @xmath9co emitting regions become significant when @xmath14v becomes larger than @xmath3 1 mag , marginally enough to shield the ultraviolet radiation to protect @xmath9co molecules ( e.g. , warin et al . 1996 ) , although @xmath9co molecules may be also affected from the ultraviolet radiation because the @xmath84 derived from @xmath9co is lower than that derived from @xmath6co in most of molecular clouds . the peak intensity ratio of @xmath6co and @xmath9co is around 5 , much smaller than the terrestrial abundance ratio of 89 , indicating that the @xmath6co emission is optically thick in the clouds where the both emissions are detected . the maximum @xmath6co peak temperature of the present brightest @xmath6co emission is 6 k , and this suggests that the excitation temperature is consistent with the kinetic temperature of @xmath3 10 k , typical to the local dark clouds . the low mass clouds show lower @xmath6co peak temperatures down to 1 k , significantly less than the brightest peak intensities . it is not clear if this is due to the lower excitation temperatures or due to smaller filling factors significantly less than 1 . in order to clarify this point we need observations of the present low mass clouds at much higher angular resolutions . figure 13 shows the histograms of mass and sizes of @xmath6co clouds . in the present region , we have detected a large number of molecular clouds of mass less than 0.1 @xmath10 and sizes less than 0.1 pc not detected so far in the other regions . we may ask why molecular clouds of mass less than 0.1 @xmath10 and size less than 0.1 pc such as the present clouds have not been detected so far . the main reason for this is perhaps the paucity of high - resolution observational studies of nearby molecular clouds . most of the observations of the local high latitude clouds were made at lower resolutions of 10 arc - min or at a coarse grid spacing of 1 degree , both of which are unable to detect and resolve the present low mass clouds . this suggests that the low mass clouds similar to the present ones may not be uncommon in the interstellar space and warrant more extensive searches for them in the other parts of the sky . it is interesting to compare the physical parameters of the present vscs with theoretical studies . the typical size of these hlcs , @xmath3 0.1 pc , is significantly less than the jeans length of 1.3 pc 7 pc for molecular gas with @xmath85=10 k and @xmath86(h@xmath11)=10100 @xmath87 . if we take temperature higher than 10 k , the length becomes even larger . figure 14 shows the radial distribution of mass surface density which is derived by dividing the mass in circular annulus of a radius by the area of the circular annulus . mass surface density in the present region is fairly flat in radius because there is no massive molecular cloud in the center of the vscs . on the other hand , mass surface density in the other regions has a gradient for radius , indicating that there are small molecular clouds around a massive molecular cloud whose mass is several dozen @xmath10 or greater ( e.g. , sakamoto 2002 , sakamoto & sunada 2003 ) . in these regions , the gravity of the massive molecular cloud may contribure to the formation of these small clouds by increasing the pressure in the surroundings . these suggest that mechanisms other than gravitational instability might contribute to the formation of present vscs . a theory of molcular cloud formation is discussed by koyama & inutsuka ( 2002 ) . according to them , molecular clouds smaller than the jeans - length can be formed in the shocked layer through the thermal instability . we shall present some considerations by comparing the observational results with their theoretical results . present vscs are likely to be affected by hd886 in the last few @xmath010@xmath79 yr ( for details , see section 5.2 ) . although the mechanical luminosity from hd886 injected to the loop - like structure during a few @xmath010@xmath79 yr is low to explain the expanding of the interstellar matter , it is a possibility that the stellar wind of hd886 is the source of shock . koyama & inutsuka ( 2002 ) assumed shock velocity of 26 km s@xmath20 and density of 0.6 @xmath87 as an initial pre - shock condition and find that the region of density greater than 100 @xmath87 grows in size to @xmath3 0.2 pc @xmath0 0.1 pc in 1.06@xmath010@xmath79 yr and the internal structure consists of some filaments . the velocity dispersion of co derived by koyama & inutsuka ( 2002 ) is a few km s@xmath20 . the size of the smallest molecular clouds and the velocity dispersion of co are comparable to those derived from koyama & inutsuka ( 2002 ) . we can not resolve the internal structure of the vscs because the present vscs are detected with only a few points for each . the typical column density derived by koyama & inutsuka ( 2002 ) is 2@xmath010@xmath12 @xmath13 , while the column density of the vscs is estimated to be @xmath3 1.6@xmath010@xmath12 @xmath13 ( see figure 4 ) . the surface filling factor of the region of density greater than 100 @xmath87 in koyama & inutsuka ( 2002 ) is roughly estimated to be 3040% . in this surface filling factor the column density estimated from the observations is consistent with that derived from koyama & inutsuka ( 2002 ) and the low temperature of the vscs is consistent with the result that their peak temperature is lower than that of the typical local dark clouds . these results indicate a potential that the present vscs are formed in the shock compressed layer through thermal instability . in order to compare the observational results with the theoretical simulation on internal temperature and density structure in more details , the observations of higher resolutions are needed . we have made a large - scale survey of high galactic latitude molecular clouds in the @xmath7 = 10 lines of @xmath6co and @xmath9co toward a large scale structure located around ( @xmath1 , @xmath2 ) @xmath3 ( 109@xmath4 , @xmath545@xmath4 ) with nanten . this survey spatially resolved the distribution of molecular gas associated with the large scale structure . the main conclusions of the present study are summarized as follows : 1 . the @xmath6co observation covered the entire large loop - like structure . the loop - like structure consits of very small clumpy clouds . the @xmath6co clouds are concentrated on the north to north - west of the loop - like structure and toward the south of that . we identified 78 @xmath6co clouds in the observed region . the total mass is estimated to be @xmath3 64 @xmath10 if we assume the conversion factor from co intensity to @xmath37(h@xmath88 ) as 1.0@xmath010@xmath12 @xmath13/(k km s@xmath20 ) . we performed @xmath9co observations in and around the whole area where the peak temperature of @xmath6co is more than 2.0 k. we identified 33 @xmath9co clouds and derived physical properties under the assumption of lte . the mass spectra are well fitted by a power law , @xmath55 @xmath56 @xmath57 for the @xmath6co clouds and @xmath55 @xmath56 @xmath58 for the @xmath9co clouds . these spectral indices are similar to those derived in the other regions . the size and the line width relation of @xmath9co clouds is fitted by a least - squares method , log(@xmath33 ) = ( 0.22@xmath590.43 ) @xmath0 log(@xmath35 ) + ( 0.37@xmath590.52 ) ( c.c.=0.23 ) , but the correlation is not good . present @xmath9co clouds are far from the virial equilibrium , indicating that @xmath9co clouds are not gravitationally bound . @xmath52 and @xmath46 relation can be fitted by a least - squares method as log(@xmath52 ) = ( 0.91@xmath590.30 ) @xmath0 log(@xmath46 ) + ( 2.23@xmath590.29 ) ( c.c.=0.66 ) . this index is slightly different from the indices in the other regions although the tendency that molecular clouds are more vilialized as the mass increases is consistent with the other regions . there is no sign of star formation from the comparison of iras point sources and point source catalog of two - micron all - sky survey in the present region . this suggests that molecular clouds in this region are not the site of present star formation or the remnants of past star formation . . there may be two expanding shells in the present region as inferred from hi although we can not identify them from co. the total mechanical luminosity of hd886 during the last few @xmath0 10@xmath79 yr is comparable to the expanding energy of the northern expanding hi shell . this indicates that some additional source of energy other than hd886 is needed to explain the expanding energy . @xmath9co emission is significantly detected in the @xmath6co clouds having molecular column density greater than 5@xmath010@xmath12 @xmath13 . this may be explained as that the @xmath9co emitting regions become significant when @xmath14v becomes larger than @xmath3 1 mag , marginally enough to shield the ultraviolet radiation to protect @xmath9co molecules . there is a possibility that very small clouds have been formed in the shoked layer through the thermal instability . the stellar wind of hd886 may be the source to creat shocks , forming the loop - like structure where the very small clouds are embedded . we greatly appreciate the hospitality of all staff members of the las campanas observatory of the carnegie institution of washington . the nanten telescope is operated based on a mutual agreement between nagoya university and the carnegie institution of washington . we also acknowledge that the operation of nanten can be realized by contributions from many japanese public donators and companies . three of the authors ( n.m . , t.o . , and y.f . ) acknowledge financial support from the scientist exchange program under bilateral agreement between jsps ( japan society for the promotion of science ) and conicyt ( the chilean national commission for scientific and technological research ) . this research has made use of the simbad astronomical database operated by cds , strasbourg , france . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . this research has made use of the iras point sources from the nasa / ipac infrared science archive , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . bhatt , h. c. 2000 , , 362 , 715 blitz , l. , bazell , d. , & desert , f. x. 1990 , , 352 , 13 bohlin , r. c. , savage , b. d. , & drake , j. f. 1978 , , 224 , 132 crawford , d. f. , jauncey , d. l. , & murdoch , h. s. 1970 , , 162 , 405 dickman , r. l. 1978 , , 37 , 407 gir , b. y. , blitz , l. , & magnani , l. 1994 , , 434 , 162 graedel , t. e. , langer , w. d. , & frerking , m. a. 1982 , , 48 , 321 hartmann , d. , & burton , w. b. 1997 , atlas of galactic neutral hydrogen . cambridge univ . press , cambridge hartmann , d. , magnani , l. , & thaddeus , p. , 1998 , , 492 , 205 hayakawa , t. , mizuno , a. , onishi , t. , yonekura , y. , hara , a. , yamaguchi , r. , & fukui , y. 1999 , , 51 , 919 hearty , t. , magnani , l. , & caillault , j. p. 1999 , , 341 , 163 heithausen , a. , & thaddeus , p. 1990 , , 353 , l49 kawamura , a. , ohnishi , t. , mizuno , a. , ogawa , h. , & fukui , y. , 1998 , , 117 , 387 keto , e. r. , & myers , p. c. 1986 , , 304 , 466 kiss , cs . , mo@xmath89r , a. , t@xmath89th , l. v. 2004 , 418 , 131 koyama , h. , & inutsuka , s. 2002 , , 564 , 97 magnani , l. , blitz , l. , & mundy , l. 1985 , , 295 , 402 ( mbm ) magnani , l. , blitz , l. , & lada , e. a. 1986 , , 301 , 395 magnani , l. , caillault , j. p. , buchalter , a. , & beichman , c. a. 1995 , , 96 , 159 magnani , l. , hartmann , d. , & speck , b , g. , 1996 , , 106 , 447 magnani , l. , hartmann , d. , holcomb , s. l. , smith , l. e. , & thaddeus , p. 2000 , , 535 , 167 meyer , m. r. , calvet , n. , & hillenbrand , l. a. 1997 , , 114 , 288 ogawa , h. , mizuno , a. , hoko , h. , & fukui , y. 1990 , int . j. infrared millimeter waves , 11 , 717 onishi , t. , yoshikawa , n. , yamamoto , h. , kawamura , a. , mizuno , a. , & fukui , y. 2001 , , 53 , 1017 perryman , m. a. c. , et al . 1997 , , 323 , l49(hipparcos catalogue ) pound , m. w. 1996 , , 457 , 35 pound , m. w. , & goodman , a. a. 1997 , , 482 , 334 reach , w. t. , koo , b. c. , & heiles , c. 1994 , , 429 , 672 reach , w. t.,wall , w. f. , & odegard , n. 1998 , , 507 , 507 sakamoto , s. 2002 , apj , 565 , 1050 sakamoto , s. , & sunada , k. 2003 , apj , 594 , 340 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 snow , t. p. jr 1982 , , 253 , l39 van dishoeck , e. f. , & black , j. h. , 1988 , , 334 , 771 warin , s. , benayounm , j. j. , & viala , y. p. 1996 , , 308 , 535 yamamoto , h. , onishi , t. , mizuno , a. , & fukui , y. 2003 , , 592 , 217 yonekura , y. , dobashi , k. , mizuno , a. , ogawa , h. , & fukui , y. , 1997 , , 110 , 21 ccccccccc @1 & @95.52 & @xmath550.00 & 3.2 & 1.3 & @xmath510.7 & 0.31 & @4.7 & @1.07 + @2 & @96.11 & @xmath553.13 & 1.1 & 1.3 & @@xmath59.4 & 0.09 & @0.8 & @0.07 + @3 & 101.43 & @xmath541.53 & 1.5 & 1.5 & @@xmath59.5 & 0.17 & @1.9 & @0.27 + @4 & 101.91 & @xmath541.20 & 0.7 & 0.5 & @@xmath59.5 & 0.08 & @0.9 & @0.04 + @5 & 102.09 & @xmath541.20 & 1.9 & 1.4 & @@xmath58.9 & 0.10 & @3.4 & @0.12 + @6 & 102.19 & @xmath541.07 & 0.9 & 1.1 & @@xmath58.9 & 0.08 & @1.0 & @0.04 + @7 & 102.80 & @xmath540.27 & 2.5 & 1.6 & @@xmath58.4 & 0.16 & @4.5 & @0.30 + @8 & 103.32 & @xmath540.33 & 1.1 & 1.3 & @@xmath58.3 & 0.11 & @1.6 & @0.11 + @9 & 103.47 & @xmath540.60 & 1.0 & 1.2 & @@xmath57.3 & 0.08 & @1.0 & @0.04 + 10 & 103.74 & @xmath539.33 & 5.2 & 2.9 & @xmath510.1 & 0.59 & 12.3 & @7.02 + 11 & 104.20 & @xmath538.93 & 0.8 & 1.0 & @xmath515.7 & 0.10 & @2.0 & @0.09 + 12 & 104.48 & @xmath538.53 & 1.4 & 3.5 & @@xmath58.6 & 0.16 & @4.5 & @0.44 + 13 & 104.81 & @xmath538.80 & 1.4 & 1.2 & @@xmath55.0 & 0.10 & @1.5 & @0.08 + 14 & 105.07 & @xmath538.80 & 1.1 & 1.4 & @@xmath55.7 & 0.08 & @1.8 & @0.06 + 15 & 105.08 & @xmath552.27 & 1.0 & 1.2 & @@xmath50.1 & 0.07 & @1.2 & @0.04 + 16 & 105.10 & @xmath538.07 & 2.5 & 2.4 & @xmath512.7 & 0.26 & @5.7 & @1.19 + 17 & 105.22 & @xmath553.20 & 1.2 & 1.3 & @@xmath53.7 & 0.09 & @1.1 & @0.05 + 18 & 105.40 & @xmath548.13 & 1.6 & 1.8 & @@xmath59.0 & 0.12 & @3.0 & @0.20 + 19 & 105.77 & @xmath538.40 & 5.6 & 1.9 & @@xmath55.0 & 0.47 & 10.5 & @5.60 + 20 & 105.80 & @xmath539.60 & 1.7 & 1.8 & @@xmath59.8 & 0.24 & @3.1 & @0.70 + 21 & 105.82 & @xmath554.00 & 1.1 & 1.4 & @@xmath54.8 & 0.07 & @1.3 & @0.04 + 22 & 106.76 & @xmath536.53 & 2.7 & 3.0 & @@xmath59.8 & 0.53 & @8.3 & @4.29 + 23 & 106.97 & @xmath537.87 & 1.7 & 1.7 & @@xmath51.2 & 0.12 & @3.4 & @0.17 + 24 & 107.32 & @xmath537.60 & 1.2 & 1.5 & @@xmath52.3 & 0.12 & @1.6 & @0.11 + 25 & 107.38 & @xmath552.00 & 1.0 & 2.6 & @@xmath57.5 & 0.12 & @2.1 & @0.13 + 26 & 107.87 & @xmath553.93 & 0.8 & 1.6 & @@xmath54.9 & 0.11 & @1.8 & @0.11 + 27 & 108.21 & @xmath553.73 & 2.2 & 1.4 & @@xmath55.1 & 0.16 & @2.9 & @0.25 + 28 & 108.33 & @xmath553.13 & 2.2 & 1.2 & @@xmath50.2 & 0.07 & @3.5 & @0.08 + 29 & 108.78 & @xmath552.60 & 5.7 & 2.4 & @@xmath56.6 & 0.79 & 16.6 & 11.13 + 30 & 109.00 & @xmath552.13 & 3.3 & 1.9 & @@xmath51.5 & 0.31 & @5.1 & @1.43 + 31 & 109.00 & @xmath550.07 & 2.5 & 1.6 & @@xmath54.7 & 0.13 & @5.1 & @0.26 + 32 & 109.11 & @xmath550.60 & 0.9 & 2.2 & @@xmath55.9 & 0.09 & @1.6 & @0.06 + 33 & 109.17 & @xmath537.60 & 4.0 & 1.5 & @@xmath54.4 & 0.36 & @5.6 & @1.83 + 34 & 109.55 & @xmath552.87 & 1.3 & 2.0 & @@xmath57.1 & 0.14 & @2.5 & @0.16 + 35 & 109.68 & @xmath538.27 & 4.6 & 1.4 & @@xmath58.1 & 0.17 & @5.4 & @0.52 + 36 & 109.76 & @xmath538.00 & 2.7 & 2.0 & @@xmath54.9 & 0.21 & @4.7 & @0.63 + 37 & 109.78 & @xmath553.33 & 2.8 & 1.5 & @@xmath56.4 & 0.18 & @2.6 & @0.30 + 38 & 109.84 & @xmath550.73 & 4.0 & 2.6 & @@xmath57.8 & 0.18 & @5.0 & @0.40 + 39 & 109.85 & @xmath538.60 & 0.8 & 1.8 & @@xmath58.1 & 0.12 & @2.1 & @0.13 + 40 & 110.05 & @xmath550.40 & 2.1 & 3.1 & @@xmath57.4 & 0.10 & @3.6 & @0.14 + 41 & 110.06 & @xmath541.27 & 4.4 & 2.9 & @@xmath56.1 & 0.47 & 10.2 & @4.21 + 42 & 110.32 & @xmath548.93 & 2.9 & 0.9 & @@xmath54.2 & 0.12 & @2.6 & @0.17 + 43 & 110.94 & @xmath541.00 & 1.3 & 3.4 & @@xmath56.3 & 0.17 & @5.7 & @0.67 + 44 & 110.98 & @xmath539.13 & 1.2 & 2.2 & @@xmath59.6 & 0.12 & @2.9 & @0.20 + 45 & 111.08 & @xmath550.13 & 1.6 & 1.0 & @@xmath57.2 & 0.13 & @1.7 & @0.11 + 46 & 112.04 & @xmath539.87 & 3.4 & 2.3 & @@xmath55.4 & 0.26 & @6.4 & @1.11 + 47 & 112.40 & @xmath540.07 & 1.3 & 2.6 & @@xmath57.2 & 0.10 & @3.6 & @0.19 + 48 & 112.55 & @xmath539.60 & 1.2 & 1.9 & @@xmath53.6 & 0.16 & @3.0 & @0.29 + 49 & 112.64 & @xmath550.07 & 0.7 & 1.5 & @@xmath56.5 & 0.14 & @1.7 & @0.14 + 50 & 112.72 & @xmath541.13 & 1.2 & 1.9 & @@xmath56.2 & 0.11 & @2.4 & @0.13 + 51 & 112.72 & @xmath541.13 & 1.2 & 1.9 & @@xmath56.2 & 0.11 & @2.4 & @0.13 + 52 & 112.72 & @xmath539.67 & 1.2 & 2.6 & @@xmath55.3 & 0.10 & @1.9 & @0.09 + 53 & 112.84 & @xmath540.27 & 1.3 & 1.7 & @@xmath55.1 & 0.08 & @2.9 & @0.11 + 54 & 113.15 & @xmath539.47 & 1.5 & 1.2 & @@xmath54.7 & 0.16 & @2.5 & @0.23 + 55 & 113.23 & @xmath552.07 & 4.4 & 1.3 & @@xmath57.3 & 0.30 & @5.7 & @1.32 + 56 & 113.42 & @xmath542.33 & 3.4 & 2.4 & @xmath510.5 & 0.23 & @5.9 & @0.89 + 57 & 113.47 & @xmath539.20 & 0.5 & 1.9 & @@xmath55.7 & 0.10 & @1.4 & @0.08 + 58 & 113.56 & @xmath549.93 & 1.3 & 0.9 & @@xmath57.5 & 0.12 & @1.1 & @0.09 + 59 & 113.94 & @xmath551.67 & 1.4 & 1.6 & @@xmath59.7 & 0.07 & @3.2 & @0.07 + 60 & 114.38 & @xmath551.73 & 3.9 & 1.6 & @@xmath59.7 & 0.17 & @6.7 & @0.55 + 61 & 114.49 & @xmath550.80 & 1.5 & 1.8 & @@xmath57.8 & 0.09 & @3.2 & @0.12 + 62 & 114.52 & @xmath541.53 & 2.2 & 1.1 & @@xmath57.8 & 0.15 & @2.1 & @0.20 + 63 & 114.83 & @xmath551.07 & 0.8 & 1.8 & @@xmath58.1 & 0.10 & @1.4 & @0.07 + 64 & 114.90 & @xmath541.73 & 1.5 & 1.2 & @@xmath58.3 & 0.08 & @1.1 & @0.04 + 65 & 115.10 & @xmath543.80 & 1.6 & 1.0 & @@xmath52.1 & 0.11 & @1.5 & @0.09 + 66 & 115.16 & @xmath545.33 & 1.2 & 1.6 & @@xmath58.4 & 0.17 & @1.6 & @0.25 + 67 & 115.24 & @xmath543.40 & 1.8 & 2.0 & @@xmath53.5 & 0.23 & @2.7 & @0.57 + 68 & 115.63 & @xmath543.60 & 0.8 & 1.0 & @@xmath53.3 & 0.08 & @1.6 & @0.06 + 69 & 115.74 & @xmath546.20 & 1.5 & 1.0 & @@xmath56.8 & 0.13 & @2.3 & @0.18 + 70 & 116.20 & @xmath543.73 & 3.2 & 1.5 & @@xmath52.6 & 0.16 & @3.6 & @0.31 + 71 & 116.33 & @xmath544.80 & 4.2 & 2.3 & @@xmath53.9 & 0.59 & 10.0 & @7.13 + 72 & 116.45 & @xmath550.53 & 2.3 & 1.1 & @@xmath57.3 & 0.17 & @3.5 & @0.35 + 73 & 116.64 & @xmath552.33 & 0.9 & 1.5 & @@xmath57.6 & 0.07 & @2.1 & @0.05 + 74 & 116.86 & @xmath543.87 & 0.8 & 3.7 & @@xmath54.7 & 0.18 & @2.9 & @0.41 + 75 & 117.01 & @xmath550.73 & 2.7 & 0.8 & @@xmath57.5 & 0.10 & @2.0 & @0.10 + 76 & 117.11 & @xmath544.33 & 2.3 & 1.1 & @@xmath52.6 & 0.14 & @5.1 & @0.25 + 77 & 118.12 & @xmath552.13 & 1.1 & 1.1 & @@xmath56.7 & 0.12 & @1.8 & @0.12 + 78 & 118.23 & @xmath552.67 & 4.2 & 2.0 & @@xmath57.8 & 0.43 & @8.8 & @3.12 + ccccccccccc 16a & 105.11 & @xmath538.03 & 0.7 & 1.7 & @xmath511.4 & 0.06 & 0.11 & @4.0 & 0.08 & @34.6 + 16b & 105.12 & @xmath537.80 & 0.7 & 1.0 & @xmath513.1 & 0.07 & 0.12 & @3.6 & 0.09 & @13.5 + 19a & 105.54 & @xmath538.63 & 0.8 & 1.1 & @@xmath55.5 & 0.09 & 0.14 & @2.5 & 0.17 & @23.2 + 19b & 105.77 & @xmath538.37 & 2.3 & 1.7 & @@xmath54.0 & 0.21 & 0.46 & 16.6 & 1.41 & 129.8 + 20@ & 103.70 & @xmath539.33 & 1.2 & 1.5 & @@xmath59.7 & 0.13 & 0.22 & 10.0 & 0.56 & @57.9 + 22@ & 106.93 & @xmath536.40 & 0.7 & 2.1 & @@xmath59.4 & 0.05 & 0.13 & @3.7 & 0.06 & @45.2 + 26@ & 107.87 & @xmath553.93 & 0.3 & 1.8 & @@xmath57.7 & 0.07 & 0.13 & @2.3 & 0.09 & @46.4 + 27@ & 108.10 & @xmath553.77 & 0.8 & 3.1 & @@xmath56.2 & 0.09 & 0.13 & @3.7 & 0.21 & 176.1 + 29a & 107.82 & @xmath551.70 & 1.2 & 1.6 & @@xmath55.0 & 0.08 & 0.21 & @7.0 & 0.21 & @40.7 + 29b & 108.78 & @xmath552.63 & 1.4 & 2.0 & @@xmath54.6 & 0.14 & 0.26 & 11.1 & 0.87 & 120.6 + 29c & 108.84 & @xmath552.03 & 1.2 & 0.7 & @@xmath56.2 & 0.11 & 0.21 & @5.9 & 0.29 & @10.6 + 29d & 108.89 & @xmath552.17 & 1.3 & 1.1 & @@xmath56.9 & 0.08 & 0.23 & @3.8 & 0.13 & @18.4 + 29e & 109.00 & @xmath552.40 & 1.1 & 2.1 & @@xmath57.8 & 0.12 & 0.20 & @6.0 & 0.41 & 110.6 + 29f & 109.00 & @xmath552.13 & 0.9 & 1.6 & @@xmath51.9 & 0.04 & 0.15 & @5.8 & 0.05 & @22.5 + 29 g & 109.16 & @xmath551.87 & 1.0 & 1.7 & @@xmath55.8 & 0.10 & 0.17 & @2.3 & 0.25 & @57.6 + 29h & 109.53 & @xmath551.17 & 0.8 & 0.9 & @@xmath56.5 & 0.05 & 0.15 & @2.9 & 0.04 & @@8.7 + 29i & 109.64 & @xmath551.57 & 0.3 & 2.4 & @@xmath57.0 & 0.05 & 0.12 & @3.2 & 0.04 & @59.7 + 29j & 109.80 & @xmath551.37 & 0.4 & 1.5 & @@xmath56.7 & 0.05 & 0.14 & @2.5 & 0.04 & @22.3 + 33a & 109.17 & @xmath537.60 & 0.8 & 1.0 & @@xmath54.4 & 0.05 & 0.13 & @2.4 & 0.05 & @10.7 + 33b & 109.21 & @xmath537.80 & 0.9 & 0.9 & @@xmath54.1 & 0.06 & 0.16 & @4.7 & 0.08 & @10.6 + 33c & 109.34 & @xmath537.80 & 0.7 & 0.7 & @@xmath54.5 & 0.05 & 0.13 & @3.2 & 0.05 & @@4.7 + 35@ & 109.72 & @xmath538.27 & 0.9 & 0.7 & @@xmath58.1 & 0.05 & 0.16 & @3.2 & 0.05 & @@5.1 + 38@ & 109.90 & @xmath550.73 & 1.1 & 1.8 & @@xmath57.7 & 0.11 & 0.19 & 10.5 & 0.48 & @75.3 + 41a & 110.11 & @xmath541.27 & 1.0 & 2.9 & @@xmath55.6 & 0.11 & 0.23 & 10.1 & 0.41 & 197.4 + 41b & 110.19 & @xmath541.07 & 0.6 & 2.0 & @@xmath57.6 & 0.04 & 0.10 & @2.3 & 0.03 & @33.1 + 43@ & 111.12 & @xmath541.00 & 0.7 & 1.1 & @@xmath55.6 & 0.04 & 0.12 & @4.3 & 0.03 & @@9.4 + 46@ & 112.08 & @xmath539.87 & 0.8 & 0.9 & @@xmath55.1 & 0.06 & 0.14 & @3.7 & 0.07 & @@9.3 + 55@ & 113.17 & @xmath552.03 & 1.2 & 0.7 & @@xmath57.2 & 0.11 & 0.22 & @4.3 & 0.24 & @10.6 + 56@ & 113.17 & @xmath542.60 & 0.8 & 0.9 & @xmath510.9 & 0.04 & 0.14 & @3.7 & 0.04 & @@6.5 + 60@ & 114.32 & @xmath551.70 & 0.8 & 1.2 & @@xmath59.7 & 0.08 & 0.15 & @4.8 & 0.19 & @24.9 + 71a & 115.49 & @xmath544.40 & 0.9 & 1.0 & @@xmath53.5 & 0.10 & 0.16 & @3.7 & 0.20 & @22.2 + 71b & 116.21 & @xmath544.97 & 1.2 & 1.6 & @@xmath53.9 & 0.18 & 0.22 & @9.1 & 1.04 & @93.9 + 78@ & 118.19 & @xmath552.70 & 1.5 & 1.2 & @@xmath58.1 & 0.12 & 0.28 & @9.6 & 0.53 & @36.7 + note col . ( 1 ) : cloud number of @xmath9co taken from that of @xmath6co cloud with which the @xmath9co cloud is associated . if plural @xmath9co clouds are associated with one @xmath6co cloud , a sequential alphabet is added , col . ( 2)(3 ) : cloud peak ( @xmath1 , @xmath2 ) position in degree , col . ( 4 ) : peak temperature in k , col . ( 5 ) : line width of the composite spectrum in km s@xmath20 , col . ( 6 ) : peak velocity of the composite spectrum in km s@xmath20 , col . ( 7 ) : radius of the molecular cloud in pc , col . ( 8) : optical depth of @xmath9co , col . ( 9 ) : column density of peak position in 10@xmath12 @xmath13 , col . ( 10 ) : mass of the molecular cloud assuming the lte in @xmath10 , col . ( 11 ) : virial mass of the molecular cloud in @xmath10 . ( 4 ) to ( 6 ) are derived by using a single gaussiun fitting .

infrared magnitude - redshift relations ( k - z relations ) for the 3cr and 6c samples of radio galaxies are presented for a wide range of plausible cosmological models , including those with non - zero cosmological constant . variations in the galaxy formation redshift , metallicity and star formation history are also considered . the results of the modelling are displayed in terms of magnitude differences between the models and no - evolution tracks , illustrating the amount of k - band evolution necessary to account for the observational data . given a number of plausible assumptions , the results of these analyses suggest that : ( i ) cosmologies which predict ( where denotes the current age of the universe ) can be excluded ; ( ii ) the star formation redshift should lie in the redshift interval , values towards the lower end of the range being preferred in cosmologies with larger values of ; ( iii ) the einstein - de sitter model ( ) provides a reasonable fit to the data ; ( iv ) models with finite values of can provide good agreement with the observations only if appropriate adjustments of other parameters such as the galaxy metallicities and star - formation histories are made . without such modifications , even after accounting for stellar evolution , the high redshift radio galaxies are more luminous ( that is , more massive ) than those nearby in models with finite , including the favoured model with . for cosmological models with larger values of , the conclusions are the same regardless of whether any adjustments are made or not . the implications of these results for cosmology and models of galaxy formation are discussed . # 1to 0pt#1 [ firstpage ] galaxies active galaxies radio galaxies galaxy evolution cosmological parameters .

astro-ph0110054

the models used to account for the infrared magnitude - redshift relation of luminous radio galaxies are sensitive to the assumptions made concerning the evolution of their stellar content , as well as the choice of cosmological model . galaxies at a particular redshift are observed at different cosmic epochs in different cosmological models and so their predicted spectra at a given redshift can vary quite dramatically . accurate modelling of the evolution of galaxy spectra with cosmic time is required if the infrared k magnitude - redshift relation is to be used to constrain the range of acceptable world models . in this paper the spectral evolution of galaxies is modelled for different plausible cosmological models using the most recent spectral synthesis codes of bruzual and charlot ( 2001 ) , gissel2000 , and these results compared with the observational data . even for the simplest cases , numerous cosmological and evolutionary effects need to be taken into account and it is often not intuitively obvious how changes in the assumptions made change the predicted relations one of the objectives of this paper is to provide these insights . although our prime concern is the k - z relation for radio galaxies , the considerations of this paper are relevant to all such cosmological studies . the radio galaxies associated with the most powerful extragalactic radio sources seem to form a remarkably uniform class of galaxy out to large redshifts . following the early work of lilly and longair ( 1984 ) , best _ et al . _ ( 1998 ) carried out a much more detailed analysis of the photometric properties of a complete sample of 3cr radio galaxies , the brightest radio galaxies in the northern sky selected at 178 mhz ( laing _ et al . they modelled the k - z relation for world models with @xmath7 and @xmath8 , in both cases assuming @xmath9 . these models included passive evolution of the stellar populations of the galaxies with formation redshifts of 3 , 5 and 20 . the passive evolution models in the @xmath8 universe provided a reasonable fit to the observations , whereas those with @xmath10 gave a poor fit . the models with low redshifts of formation , @xmath11 , resulted in more luminous galaxies than those observed at @xmath12 , suggesting that the stellar populations of the galaxies must have formed at @xmath13 . eales and rawlings ( 1996 ) and eales _ et al . _ ( 1997 ) determined the k - z relation for a complete sample of 6c radio galaxies which were selected at 151 mhz and are about six times fainter in flux density than the 3cr sample ( eales 1985 ) . the 6c host galaxies have similar k - luminosities to those of the 3cr galaxies at @xmath14 , but are about 0.6 magnitudes fainter at larger redshifts . nevertheless , the high redshift 6c galaxies are significantly brighter than expected in world models in which no evolution of the stellar populations of the galaxies is assumed . the objectives of this paper are to review the current status of the k - z relation for 3cr and 6c luminous radio galaxies ( section 2 ) and then to carry out a more extensive analysis of the resulting k - z relation for a wider range of plausible world models than those considered by best _ ( 1998 ) and eales _ et al . _ ( 1997 ) , including models with a non - zero cosmological constant . the predicted evolution of the stellar populations of the galaxies is derived from the most recent spectral synthesis codes of bruzual & charlot ( 1993 , 2001 ) , gissel2000 ; a wide range of stellar synthesis models is considered , as well as the effects of changes in metallicity . the details of the different cosmological models , the spectral synthesis codes and procedures used to predict the k - z relations are described in section 3 . the results of these calculations are described in section 4 , indicating which parameters have the greatest impact upon the form of the predicted k - z relation . also included in this section are the results of varying the star formation history and metallicity in plausible ways . most massive galaxies at small redshifts have high metallicities and colour - magnitude relations show that more massive galaxies are redder in colour , suggesting a mass - metallicity correlation ( see , for example , arimoto & kodama , 1996 ) . on the other hand , at high redshifts , we might expect a lower metal content , as suggested by the analyses of pei _ et al . _ ( 1999 ) and pettini _ et al . _ ( 1997 ) . in section 5 the results of these analyses are discussed and our conclusions are summarised in section 6 . throughout the paper , @xmath15 km s@xmath16 mpc@xmath16 is adopted , unless otherwise stated . the observational data used in this analysis consist of infrared k magnitudes for 72 3cr galaxies in the redshift interval @xmath17 ( best _ et al . _ 1998 ) , and 57 6c galaxies in the range @xmath18 ( eales _ et al . _ 1996 , 1997 ) . the 3cr galaxies comprise a complete sample of radio sources with flux densities @xmath19 jy at 178 mhz , with a mean flux density of about 15 jy ; the 6c sample is complete in the flux density interval @xmath20 jy at 151 mhz , with an equivalent mean flux density at 178 mhz of about 2.6 jy . the 3cr k magnitudes were obtained for @xmath21 from observations with the ircam3 array on the uk infrared telescope ( best _ et al . _ 1997 ) , and for @xmath21 from the photometry of lilly & longair ( 1984 ) . the 6c data points were mostly derived from ircam observations , but also from colour corrected k@xmath22 observations with the redeye camera of the canada - france - hawaii telescope ( eales _ et al . _ 1997 ) . various factors had to be included to create a self - consistent set of k magnitudes . firstly , aperture corrections were made , reducing the magnitudes to a standard circular aperture of metric diameter 63.9kpc . these aperture corrections are dependent on the cosmological model , due to differences in the angular diameter versus redshift relation ; however , the changes from the standard @xmath8 , @xmath9 model are generally small , @xmath23 mag , and so the data were converted to 63.9kpc in the @xmath24 , @xmath9 model , and the model to model variations were included in the predicted model tracks . the 63.9kpc aperture was chosen since this had been adopted for the 6c observations wherever possible . in cases where there was a nearby companion galaxy , or other problems prevented the use of the full aperture , magnitudes for the 6c galaxies were determined for smaller apertures and aperture corrections applied by eales _ ( 1997 ) ; for galaxies at redshifts @xmath21 they assumed that the radio galaxies follow an elliptical galaxy growth curve ( sandage 1972 ) , and at redshifts @xmath25 they assumed that the growth curve for the luminosity @xmath26 of the galaxy within radius @xmath27 could be represented by @xmath28 . the k magnitudes of the 3cr galaxies were originally determined within a 9 arcsec diameter aperture . at large redshifts , this is almost equivalent to 63.9 kpc , and so the aperture corrections were small . larger aperture corrections were necessary at low redshifts since the 9 arcsec aperture included only the central regions of the galaxy ( lilly & longair 1984 ) . a de vaucouleurs profile for giant ellipticals was used to correct these magnitudes to the standard diameter , since the analysis of best _ et al . _ ( 1998 ) showed that this is a good representation of the variation of the k surface brightness with radius . it is also consistent with the radial optical light profile found for low redshift radio galaxies by lilly , mclean and longair ( 1984 ) . the resulting k - z relation for 3cr and 6c galaxies , represented by open boxes and crosses respectively , is shown in fig . [ fig : 1 ] . new versions of the predicted k - z relations are shown for an @xmath24 , @xmath9 world model . these involve the use of the most recent set of spectral synthesis codes ( gissel2000 ) , for which a scalo initial mass function was adopted , with upper and lower mass cut - offs of 100m@xmath29 and 0.1m@xmath29 respectively ; solar metallicity was assumed . in addition , improved k - band transmission data for the k - band filter of ukirt were used . details of these procedures are discussed in section 3.2 . the differences as compared with the analysis carried out by best _ ( 1998 ) are very small , and so their conclusions for the @xmath7 and @xmath8 models are unchanged . in particular , the @xmath24 , @xmath30 model with passive evolution of the stellar content of the galaxies provides a satisfactory fit to the observations and will be used as a fiducial model in the new computations . one of the intriguing features of the comparison of the 3cr and 6c data is that , although they are in excellent agreement at redshifts @xmath31 , there is a progressive shift in their mean magnitudes at greater redshifts , the mean k - magnitude of the 6c galaxies at @xmath32 being about 0.6 magnitudes fainter on average than the 3cr galaxies ( eales _ et al . this can be seen in fig . [ fig : 1 ] , where the boxes , representing 3cr galaxies , lie towards the low limit of the distribution of the crosses , 6c galaxies , at @xmath32 . given the quite large dispersion about the mean k - z relation , bayesian statistics have been used to investigate the significance of this magnitude difference and how it evolves with redshift without making restrictive assumptions about the underlying distribution ( see figure [ fig : 2 ] ) . the results show clearly the mean difference between the two data sets of about 0.6 magnitudes at high redshift / early times and its decrease to zero at later times . the mean value of @xmath33 increases slightly at the lowest redshifts to about 0.15 magnitudes , but this increase is not significant . at high redshifts the difference in k - magnitudes between 3cr and 6c galaxies is significant at the 95% confidence limit . this result for the 6c galaxies is mirrored by that of the still fainter 7c radio galaxies ( willott , priv comm ) . however , mccarthy ( 1999 ) and van breugel ( 1998 ) have studied the k - z relation for radio galaxies in the mrc 1jy sample , which contains sources with flux densities @xmath34jy at 408mhz ( @xmath35jy ) from the molonglo reference catalogue ( mccarthy et al 1996 ) , and they find that the mean k - band magnitudes of the mrc radio galaxies to be only 0 to 0.2 magnitudes fainter than the 3cr , a much smaller difference than for the 6c sources . the @xmath36 spectroscopic incompleteness of this sample may introduce a bias in that it is more difficult to obtain redshifts for intrinsically fainter galaxies , but mccarthy argues that this bias is unlikely to be sufficiently great to alter the mean magnitude of the mrc galaxies by 0.6 magitudes . the important aspect of this discussion is whether or not there is a correlation between radio luminosity and k - absolute magnitude , as would be the case if it is accepted that there is a real difference between the k - z diagrams for the 3cr and 6c samples . for the purpose of the present analysis , we have assumed that the difference between the 6c and 3cr samples , being smaller than the scatter within either sample , can be ignored . therefore , we have taken mean values and dispersions for the combined sample . these are indicated by the solid triangles and associated error bars in fig . [ fig : 1 ] . a literal interpretation of the k - z diagram shown in fig . [ fig : 1 ] suggests that to a good approximation the luminous radio galaxies belong to a single population of passively evolving galaxies of the same stellar mass . the galaxies also have the same characteristic radii of about 10kpc over the entire range @xmath37 ( best et al 1998 , mclure & dunlop 2000 ) , which supports this conclusion . as discussed by best _ _ , this is a somewhat surprising result , since the preferred hierarchical clustering picture of galaxy formation predicts that the masses and radii of galaxies would grow significantly over the redshift range considered . evidence suggests that galaxy environments also differ with redshift ; high redshift 3cr galaxies seem to belong to proto - cluster environments , whereas low redshift galaxies are more commonly found in small groups or otherwise isolated environments . intuitively , it seems difficult to reconcile these points with closed box passive evolution of the stellar populations of the 3cr galaxies . recognised this problem and argued that the luminous radio galaxies must be observed at a similar stage in their evolution , specifically that at which they had attained a certain stellar mass . provided that stars brought together by on going mergers were formed at the same early epoch , the old stellar populations of galaxies formed by hierarchical clustering would appear to be passively evolving . the small scatter in the colour magnitude relations of elliptical galaxies observed out to redshifts @xmath38 ( stanford _ et al . _ 1998 , bower _ 1998 ) justifies this assumption . a further objective of our analysis is to determine whether it remains true that the luminous galaxies present on the k - z relation have the same stellar mass , and how different assumptions about the underlying cosmology and the evolution of the stellar populations change this conclusion . because of the availability of the gissel codes of bruzual and charlot , the opportunity has been taken to consider a wider range of evolutionary histories of the star formation rate of radio galaxies and to couple this with an expanded range of cosmological models , in particular , those with non - zero values of the cosmological constant ( @xmath39 ) . friedman world models with zero cosmological constant and @xmath40 have been considered , as have a range of models with non - zero @xmath0 . the relation between cosmic time and redshift is : @xmath41^{1/2}}.\ ] ] for the world models with flat geometries , @xmath42 , corresponding to those favoured by recent analyses of the power spectrum of fluctuations in the cosmic microwave background radiation as observed in the boomerang and maxima experiments ( e.g. jaffe _ et al . _ 2000 ) , there is a parametric solution : @xmath43 numerical integrations have been used for other models , and for determining the variation of co - moving radial distance coordinate with redshift . the models with non - zero @xmath0 considered in this analysis are listed in table 1 . two of these have flat geometries and the other two models have open geometries with low values of both @xmath44 and @xmath0 . this range of models with finite values of @xmath0 spans those which would be consistent with the results found in the supernova cosmology project ( perlmutter _ et al . _ 1999 ) . the gissel codes of bruzual and charlot provide a means of modelling galaxy spectra through isochrone synthesis . the models used in the present analysis all include the passive evolution of the stellar populations of the galaxies . the standard model is taken to have solar metallicity , @xmath45 , and the stellar population of the galaxy was formed in an initial uniform starburst with a duration of 1.0 gyr . the gissel code enabled the spectral evolution of this model to be determined for ages up to 20 gyr and this is shown in fig . [ fig : 3](a ) at a number of ages which are equally spaced logarithmically in cosmic time . generally , only the late stages of evolution of the starburst ( @xmath46 years ) are of interest in this analysis . for one of the models , @xmath47 and @xmath48 , the present age of the universe , @xmath2 , is greater than 20 gyr and evolution in this model was terminated at a low redshift limit of @xmath49 . cccc @xmath44 & @xmath0 & + & & ( @xmath4 ) & @xmath2 ( gyr ) + 1.0 & 0.0 & 0.67 & 13.0 + 0.3 & 0.0 & 0.81 & 15.8 + 0.1 & 0.0 & 0.90 & 17.6 + 0.0 & 0.0 & 1.00 & 19.6 + + 0.3 & 0.7 & 0.96 & 18.9 + 0.3 & 0.3 & 0.86 & 16.9 + 0.1 & 0.9 & 1.28 & 25.0 + 0.1 & 0.3 & 0.98 & 19.1 + it is instructive to note that , in fig . 3(a ) , at optical and infrared wavelengths the spectra are roughly equally spaced logarithmically in luminosity at ages greater than @xmath50 years , indicating that , at a given wavelength , the spectral energy decreases roughly as a power - law with cosmic time . as shown by gunn ( 1978 ) , longair ( 1998 ) and others , this power - law behaviour can be understood in terms of the rate at which stars evolve from the main sequence onto the red giant branch , if the initial mass function is assumed to be of power - law form . such simple arguments predict that @xmath51 for the @xmath52 model . this is not so different from the evolution seen in fig . 3(a ) , but there are important variations in the power - law index as a function of wavelength . at 0.55 @xmath53 m , the evolution can be roughly described by @xmath54 , whereas at 2.2 @xmath53 m the exponent decreases to about @xmath55 . therefore , the simple approximation is not adequate for the present analysis . for essentially all our computations , only evolution at epochs greater than @xmath50 years is required . it can be seen however that at earlier times the k - band luminosity increases with age . this is associated with the evolution of stars onto the asymptotic giant branch . the stellar evolution models of maeder and megnet ( 1989 ) and girardi _ et al . _ ( 2000 ) show that , for stars with @xmath56 , as the stellar mass decreases , the luminosity of the upper agb stars increases and their temperature decreases . thus , the simple monotonically decreasing spectral luminosity is only applicable for times greater than @xmath50 years when the turn - off mass drops below 2m@xmath29 and the agb contribution becomes less important than evolution onto the red giant branch . in previous analyses , it was assumed that the metallicity of the galaxies corresponded to solar values . the evidence of the madau plot in its various incarnations ( see , for example , madau _ et al . _ 1998 ) and the analyses of the metallicities of the absorption line systems in distant quasars have suggested that the chemical abundances of the elements have evolved very significantly over the redshift interval @xmath57 ( see , for example , pettini _ 1998 ) . to understand the effects of metallicity upon the predicted k - z relation , models have been created for four different metallicities : @xmath58 , @xmath59 , @xmath60 , and @xmath61 . different star formation histories have also been considered . these include a continuous exponentially decreasing star formation rate , with an e - folding time of 1 gyr , and a simple delta - function starburst . combined models could then be constructed involving multiple star formation events occurring at different cosmic times . for example , if the onset of radio source and quasar activity were contemporaneous with ( or stimulated ) an active starburst , the additional starburst component might affect the k - magnitude of the galaxy . to illustrate how such events could impact the k - z relation , model starbursts can be added to the underlying passive evolution of the majority old stellar population . two model star bursts were considered . in the first , the starburst took place instantaneously , and was observed at times @xmath62 , @xmath63 and @xmath64 years after that event ; this spectral evolution is shown in fig . [ fig : 3](b ) . the intriguing aspect of these spectra is that the starburst is most luminous in the infrared waveband after about @xmath63 years . after about this time the most massive stars complete their evolution on the main sequence and join the asymptotic giant branch . it is interesting that the time scale of @xmath63 years is of the same order as the typical age of the radio source events associated with the 3cr and 6c sources and so such a star burst could contribute to the k - luminosity of the galaxy . in the second example , the case of a constant star burst lasting @xmath64 years is considered , but observed at @xmath62 , @xmath63 and @xmath64 years after the start of that event ; these spectra are shown in fig . [ fig : 3](c ) . in this case , the star burst is expected to have a much smaller effect upon the k - luminosity of the galaxy . it can be seen that , after @xmath62 years , the starburst has the same spectrum as the first model , but with only one hundredth of the luminosity since only that fraction of the stars have formed by that time . on the other hand , after @xmath64 years , the starburst is more luminous than in the first model since there is a significant contribution of recently formed stars to the k - luminosity . in either case , it can be seen that starbursts observed between @xmath63 and @xmath64 years after the onset of the starburst can have a significant impact upon the k - luminosity of the galaxy . models have been considered in which the mass fractions of newly formed stars range between 0.5% and 5.0% of the total stellar mass of the galaxy . + the resulting spectral energy distributions ( seds ) were then redshifted and convolved with the transmission function of the standard k - filter . these were then combined with the different world models to create k - z diagrams . a catalogue of the range of models considered is given in tables 3 and 4 . it is simplest to illustrate the effects of changing the parameters of the cosmological models by subtracting the predicted k - z track for the chosen cosmology from the corresponding predictions of the @xmath65 , @xmath66 model : @xmath67 is defined as @xmath68 so that negative values of @xmath67 mean that a galaxy of a given absolute luminosity would be observed to have a fainter apparent magnitude . the results for the cosmologies listed in table 1 are shown in fig . [ fig : 4 ] and tabulated in table 2 . it is instructive to distinguish those contributions to @xmath69 due to the choice of cosmological model and those due to the evolving spectrum of the galaxy . the latter can be thought of as the k - correction including the effects of the evolution of the stellar population . following the notation of longair ( 1998 ) , the flux density of the galaxy is given by @xmath70}{4\pi d_i^2(z)(1 + z)}\end{aligned}\ ] ] where @xmath71 labels the choice of cosmological model and @xmath72 is a distance measure which is related to the standard luminosity distance @xmath73 by @xmath74 . the luminosities of the modelled galaxies in the different cosmologies are normalised by varying their stellar mass , such that at zero redshift they have k - band luminosities @xmath75 matching the observed k - z relation . therefore if @xmath76 $ ] is the luminosity of the galaxy per solar mass , @xmath77 = l^{\prime}[\nu_0(1 + z),t_i(z ) ] \times \frac{l_{\rm{k(z=0)}}}{l^{\prime}[\nu_0,t_i(0)]}\ ] ] the ratio of the flux densities in the _ i_th model to that in the fiducial model , for the same observed flux density at small redshifts is given by @xmath78}{l^{\prime}[\nu_0 , t_i(0)]}\times \frac{l^{\prime}[\nu_0 , t_0(0)]}{l^{\prime}[\nu_0(1 + z ) , t_0(z ) ] } \times \frac{d_0 ^ 2(z)}{d_i^2(z)}\ ] ] thus , the magnitude difference associated with the choice of cosmological model is @xmath79,\ ] ] while that associated with observing different parts of the evolving spectral energy distribution at different cosmic times is @xmath80}{l^{\prime}[\nu_0 , t_i(0)]}\times \frac{l^{\prime}[\nu_0 , t_0(0)]}{l^{\prime}[\nu_0(1 + z ) , t_0(z)]}\right]\ ] ] the total change in magnitude is given by @xmath81 . table 2 shows these differences for the models listed in table 1 . cc@=lc@=r@.lc@=r@.lc@=r@.lc@=r@.lc@=r@.lc@=r@.lc@=r@.l & & & & & & & & + & & & & & & & & + @xmath82 & & & & & & & & + & @xmath83 & 0.00 & @xmath83 & 0 & 00 & @xmath83 & 0 & 00 & @xmath83 & 0 & 00 & @xmath83 & 0 & 00 & @xmath83 & 0 & 00 & @xmath83 & 0 & 00 & @xmath83 & 0 & 00 + & @xmath84 & 0.00 & @xmath84 & 0 & 00 & @xmath84 & 0 & 00 & @xmath84 & 0 & 00 & @xmath84 & 0 & 00 & @xmath84 & 0 & 00 & @xmath84 & 0 & 00 & @xmath84 & 0 & 00 + & @xmath85 & 0.00&@xmath85 & 0 & 00&@xmath85 & 0 & 00&@xmath85 & 0 & 00&@xmath85 & 0 & 00&@xmath85 & 0 & 00&@xmath85 & 0 & 00&@xmath85 & 0 & 00 + & @xmath86 & 0.00 & @xmath86 & 0 & 00 & @xmath86 & 0 & 00 & @xmath86 & 0 & 00 & @xmath86 & 0 & 00 & @xmath86 & 0 & 00 & @xmath86 & 0 & 00 & @xmath86 & 0 & 00 + + @xmath87 & & & & & & & & + & @xmath83 & 0.59 & @xmath83 & 0 & 69 & @xmath83 & 1 & 11 & @xmath83 & 0 & 75 & @xmath83 & 0 & 77 & @xmath83 & 0 & 72 & @xmath83 & 0 & 90 & @xmath83 & 0 & 77 + & @xmath84 & 0.00 & @xmath84 & @xmath88 & 34 & @xmath84 & @xmath88 & 47 & @xmath84 & @xmath88 & 54 & @xmath84 & @xmath88 & 60 & @xmath84 & @xmath88 & 44 & @xmath84 & @xmath88 & 92 & @xmath84 & @xmath88 & 59 + & @xmath85 & 0.00 & @xmath85 & @xmath88 & 13 & @xmath85 & 0 & 00 & @xmath85 & @xmath88 & 11 & @xmath85 & @xmath88 & 02 & @xmath85 & @xmath88 & 04 & @xmath89 & @xmath88 & 21 & @xmath85 & @xmath88 & 07 + & @xmath86 & 0.00 & @xmath86 & @xmath88 & 47 & @xmath86 & @xmath88 & 47 & @xmath86 & @xmath88 & 65 & @xmath86 & @xmath88 & 62 & @xmath86 & @xmath88 & 48 & @xmath86 & @xmath90 & 13 & @xmath86 & @xmath88 & 66 + + @xmath91 & & & & & & & & + & @xmath83 & 0.85 & @xmath83 & 1 & 10 & @xmath83 & 1 & 24 & @xmath83 & 1 & 33 & @xmath83 & 1 & 12 & @xmath83 & 1 & 15 & @xmath83 & 1 & 54 & @xmath83 & 1 & 32 + & @xmath84 & 0.00 & @xmath84 & @xmath88 & 59 & @xmath84 & @xmath88 & 84 & @xmath84 & @xmath88 & 99 & @xmath84 & @xmath88 & 78 & @xmath84 & @xmath88 & 67 & @xmath84 & @xmath90 & 30 & @xmath84 & @xmath88 & 97 + & @xmath85 & 0.00 & @xmath85 & @xmath88 & 29 & @xmath85 & @xmath88 & 13 & @xmath85 & @xmath88 & 39 & @xmath85 & @xmath88 & 10 & @xmath85 & @xmath88 & 12 & @xmath85 & @xmath88 & 40 & @xmath85 & @xmath88 & 21 + & @xmath86 & 0.00 & @xmath86 & @xmath88 & 88 & @xmath86 & @xmath88 & 97 & @xmath86 & @xmath90 & 38 & @xmath86 & @xmath88 & 88 & @xmath86 & @xmath88 & 79 & @xmath86 & @xmath90 & 70 & @xmath86 & @xmath90 & 18 + + @xmath92 & & & & & & & & + & @xmath83 & 1.00 & @xmath83 & 1 & 43 & @xmath83 & 1 & 68 & @xmath83 & 1 & 88 & @xmath83 & 1 & 49 & @xmath83 & 1 & 46 & @xmath83 & 1 & 98 & @xmath83 & 1 & 77 + & @xmath84 & 0.00 & @xmath84 & @xmath88 & 77 & @xmath84 & @xmath90 & 13 & @xmath84 & @xmath90 & 37 & @xmath84 & @xmath88 & 86 & @xmath84 & @xmath88 & 82 & @xmath84 & @xmath90 & 48 & @xmath84 & @xmath90 & 24 + & @xmath85 & 0.00 & @xmath85 & @xmath88 & 72 & @xmath85 & @xmath88 & 42 & @xmath85 & @xmath88 & 85 & @xmath85 & @xmath88 & 42 & @xmath85 & @xmath88 & 37 & @xmath85 & @xmath88 & 81 & @xmath85 & @xmath88 & 39 + & @xmath86 & 0.00 & @xmath86 & @xmath90 & 49 & @xmath86 & @xmath90 & 55 & @xmath86 & @xmath93 & 22 & @xmath86 & @xmath90 & 28 & @xmath86 & @xmath90 & 19 & @xmath86 & @xmath93 & 29 & @xmath86 & @xmath90 & 63 + fig . 4 shows that galaxies are expected to be fainter at large redshifts in all the models considered as compared with the model with @xmath8 , @xmath9 ; this means that for the same observed k - magnitude , the galaxies would have to be intrinsically more luminous . there are several effects which contribute to this decrease . the most important of these is the fact that , for a given redshift , galaxies are observed at greater radial co - moving coordinate distances as compared with the model with @xmath8 , @xmath94 . in addition , the cosmic times at which the spectral energy distributions are observed are significantly different . what is important , however , are the relative epochs at which the galaxies are observed . it turns out that , in the passively evolving models , the relative changes in observed luminosity ( @xmath95 ) are significantly less than the effects of the cosmological model ( @xmath96 ) . this can be understood from table 2 which shows the ages and distance coordinates at which galaxies at different redshifts are observed in the different cosmologies listed in table 1 assuming @xmath15 km s@xmath16 mpc@xmath16 ; included in this table are the contributions of @xmath97 , the total change in k - magnitude @xmath98 and the difference between them @xmath95 , which represents the contribution of the changing spectral energy distribution of the galaxies with cosmic epoch . table 2 demonstrates quantitatively the important point that , at all redshifts , the choice of cosmological model is always much more important than the evolution corrections within the context of passively evolving models for the stellar populations of the galaxies , so far as the _ differences _ from the fiducial model are concerned . it is interesting that it is possible to reproduce quite accurately the results of the exact computations by interpolating the luminosity changes at the appropriate redshifted wavelength in the template spectra shown in fig . it is found that this approach generally produces results accurate to about 0.1 mag , which is normally accurate enough to draw useful conclusions . the value of this approach is that it enables the investigator to use fig . 3(a ) to work out quite precisely the effects of the k - corrections for any other world model with any given value of hubble s constant and for alternative assumptions relative to the template spectral evolution shown in fig . poggianti ( 1997 ) provides tables of k - corections and evolutionary corrections for observations in a number of specific filters , assuming an @xmath99 cosmology . we have compared these results with those of our two most similar cosmologies , i.e. @xmath100 , @xmath9 and @xmath101 , @xmath9 . despite a different set of spectral synthesis models , the evolutionary corrections for an elliptical galaxy formed with an e - folding time of 1 gyr given by poggianti are in good general agreement with our own , suggesting that the choice of stellar synthesis code does not have a large impact on the evolutionary corrections determined . the k corrections given by poggianti for their single cosmology in the k waveband also agree well with our own results . considering first the range of models shown in fig . 4 , it is evident that only the passively evolving @xmath102 model provides a reasonable fit to the data . the error bars shown in the figure represent the standard errors in the mean of the k - magnitude of the galaxies , within the given redshift intervals , and are less than the standard deviations shown in fig . 1 by a factor of @xmath103 , where @xmath104 . for illustrative purposes , a quantitative comparison between four of the models plotted in fig . [ fig : 4 ] is given in table 3 . these are the models with @xmath105 ; @xmath106 ; @xmath107 ; and @xmath108 . in the first part of the table , the magnitude difference , @xmath67 , between the data points and the no - evolution tracks at redshifts @xmath109 and 2.0 are listed , and in the second the difference in @xmath67 between the data points and the passive evolution models of fig . [ fig : 4 ] are shown . these data are also included as the first two entries in table 4 . the k - z relations have been normalised to the same k - magnitude at low redshift , for both the passive evolution and the no - evolution tracks since the mean magnitudes of the galaxies at @xmath110 are consistent with a very well - defined linear k - z relation , as can be seen from fig . 1 . positive values of @xmath67 indicate that the predicted magnitudes correspond to brighter galaxies than are needed to account for the data . for example , the @xmath111 model results in slightly brighter tracks than the mean of the data points at @xmath112 . the importance of these results is that , in general , except for the fiducial model , @xmath102 , the galaxies are predicted to be fainter at large redshifts than they are observed to be , even when passive evolution is taken into account . this is a somewhat unexpected result . first of all , according to the hierarchical clustering model , galaxies at large redshifts should be less luminous than those nearby since they contain fewer stars . the present results go in entirely the wrong direction . for the most extreme @xmath4 models the galaxies would need to contain about ten times more stars to account for these differences . second , it might be argued that this result would be found if there were a correlation between radio luminosity and the k - luminosity of the galaxy . in this case , the galaxies at large redshift would be expected to be larger than those at small redshifts . the analysis of mclure and dunlop ( 2000 ) showed , however , that the half - width diameters of the luminosity profiles of the radio galaxies are invariant as a function of redshift , suggesting that the galaxies have attained a similar evolutionary state so far as their total stellar populations are concerned . because of these somewhat surprising conclusions , the impact of changing the assumptions made in the fiducial model in plausible ways has been analysed in more detail . ccr@ @xmath113 lr@ @xmath113 lr@ @xmath113 l cosmological & sample & + model & & & & + @xmath114 , @xmath115 & 3cr data & 0.597 & 0.140 & 1.022 & 0.126 & 1.780 & 0.114 + & 6c data & 0.706 & 0.097 & 0.433 & 0.145 & 1.136 & 0.233 + & combined data & 0.646 & 0.089 & 0.742 & 0.115 & 1.222 & 0.210 + @xmath116 & combined data & 0.778 & 0.091 & 1.110 & 0.118 & 1.945 & 0.220 + @xmath117 & combined data & 0.712 & 0.091 & 1.076 & 0.119 & 1.925 & 0.220 + @xmath118 & combined data & 0.789 & 0.091 & 1.101 & 0.117 & 1.758 & 0.214 + & model & + @xmath119 & standard initial starburst & -0.177 & 0.089 & 0.162 & 0.115 & 0.336 & 0.210 + @xmath116 & standard initial starburst & -0.303 & 0.091 & -0.262 & 0.118 & -0.696 & 0.220 + @xmath117 & standard initial starburst & -0.352 & 0.091 & -0.310 & 0.119 & -0.799 & 0.220 + @xmath118 & standard initial starburst & -0.420 & 0.091 & -0.295 & 0.117 & -0.454 & 0.214 + lcccc section & & & @xmath67 from the observational data : & + & & z=0.5&z=1.0&z=2.0 + & standard passive evolution model : & -0.18 & 0.16 & 0.34 + & @xmath65 , @xmath120 , @xmath121 , @xmath122 & & & + & 1gyr starburst , @xmath123 . & & & + & change from standard passive evolution model & + & & z=0.5 & z=1.0 & z=2.0 + ( 4.1 ) & @xmath47 , @xmath120 & -0.13 & -0.42 & -1.03 + & @xmath47 , @xmath124 & -0.18 & -0.47 & -1.14 + & @xmath125 , @xmath126 & -0.24 & -0.46 & -0.79 + + ( 4.2 ) & @xmath127=2 & 0.10 & 0.27 & - + & @xmath127=3 & -0.00 & 0.10 & 0.57 + & @xmath127=5 & -0.02 & 0.04 & 0.14 + & @xmath127=10 & 0.00 & 0.00 & 0.00 + & @xmath127=20 & -0.04 & -0.05 & -0.14 + & @xmath127=100 & 0.00 & -0.02 & -0.12 + + ( 4.3 ) & @xmath128 & 0.00 & 0.00 & 0.00 + & @xmath129 & -0.02 & 0.04 & -0.01 + & @xmath130&-0.01 & -0.05 & 0.12 + + ( 4.4 ) & @xmath131 & 0.05 & 0.22 & 0.45 + & @xmath132 & 0.09 & 0.10 & 0.32 + & @xmath133 & 0.00 & 0.00 & 0.00 + & @xmath134 & 0.13 & -0.04 & -0.16 + + ( 4.5)&instantaneous sfr & -0.04 & -0.09 & -0.21 + & 1gyr constant rate & 0.00 & 0.00 & 0.00 + & exponential sfr & 0.05 & 0.07 & 0.05 + + ( 4.5)&extra starburst , 2.5% mass , 10@xmath135years old & 0.04 & 0.07 & 0.17 + & extra starburst , 2.5% mass , 10@xmath136years old & 0.32 & 0.28 & 0.22 + & extra starburst , 2.5% mass , 10@xmath137years old & 0.04 & 0.04 & 0.05 + + ( 4.5)&extra starburst , 10@xmath136years old , 0.5% mass & 0.07 & 0.06 & 0.05 + & extra starburst , 10@xmath136years old , 1.0% mass & 0.14 & 0.12 & 0.09 + & extra starburst , 10@xmath136years old , 2.5% mass & 0.32 & 0.28 & 0.22 + & extra starburst , 10@xmath136years old , 5.0% mass & 0.56 & 0.50 & 0.40 + the epoch at which the stellar populations were formed was allowed to vary between redshifts of 2 and 100 . the sense of the changes to the k - z relations are similar for all cosmological models . fig . [ fig : 5 ] shows the k - band magnitude differences of the models relative to the no - evolution track in the same cosmology for models with @xmath102 and @xmath138 . the magnitude differences relative to the fiducial model are listed in table 4 for @xmath102 . results have not been tabulated for models with @xmath138 , as the magnitude differences within these models were not dissimilar to those in the einstein de sitter cosmology . in fig . [ fig : 5 ] , the values of @xmath67 are plotted against cosmic time as a fraction of @xmath2 , where @xmath2 is the present age of the universe listed in table 1 for @xmath139 . in fig . [ fig : 5 ] , it is assumed that the galaxies have solar metallicity . all models have been normalised to the same magnitude at the lowest redshifts , at which the mean k - z relation is well defined . the models diverge at earlier times . as expected , if the epoch of star formation is less than the reference value @xmath140 the predicted tracks are more luminous at redshifts @xmath141 . the largest differences occur in the cases in which the star formation is assumed to begin at @xmath142 and 3 , these being less than that at which radio galaxies are known to exist . nonetheless , it is noteworthy that , unless the epoch of star formation were as unreasonably low as @xmath143 , changing the epoch of star formation does not result in a particularly large change in @xmath67 . the reason for this can be understood from fig . 3(a ) . by adopting 2.2 @xmath53 m as the wavelength at which the magnitude - redshift relation is determined , even at redshift @xmath92 , the changes in the luminosity of the galaxies at a rest wavelength of 500 nm are quite small and change monotonically with time indeed , the difference between observing the starburst at ages 0.32 to @xmath144 years amounts to less than a factor of three at 5000@xmath145 in the rest frame . this behaviour contrasts dramatically with what would be expected if the redshift - magnitude relation were determined in , say , the r waveband . then , fig . 3(a ) shows that the variations in luminosity at rest frame wavelengths of 1500@xmath145 would be very sensitive to even small amounts of ongoing star formation . these calculations illustrate the considerable advantages of carrying out studies of the redshift magnitude relation in the k - waveband the observations provide information more directly about the bulk star formation activity of the galaxies . as expected , as @xmath127 increases the predicted k - z relations become fainter at large redshifts and there is little difference between the tracks in both models if @xmath127 is greater than 10 . in the case of the model @xmath146 , the fit to the data can be improved considerably if the star formation redshift @xmath147 is adopted . values of @xmath67 for the examples given in fig . 5(a ) at redshifts of 0.5 , 1.0 and 2.0 are included in table 4 . intuitively , it would seem that the choice of hubble s constant would have a significant impact upon the form of the k - z relation since the cosmic time - scale is significantly different if , say , @xmath148 were are large as @xmath149 . then , all cosmic time - scales would be reduced by a factor of two , allowing much less time for the evolution of the stellar populations of the galaxies and also observing them at very much earlier times at large redshifts . these changes , as well as those for @xmath150 , are given in table 4 . in the cases of models with larger values for @xmath148 , an initial starburst of very short duration was adopted so that the galaxy is observed during the passive evolution phase ; the tabulated values in table 4 have included this change in star formation history . it is noteworthy that changing the value of hubble s constant has a remarkably small effect upon the predicted evolutionary tracks . the reasons for this can be understood from inspection of fig . as was emphasised in section 3.2 , so long as the analysis is restricted to redshifts less than 3 , the spectral evolution of the stellar energy distribution is monotonic with time and can be reasonably described by power - laws at different wavelengths from 0.5 to 2.2 @xmath53 m . thus , if the value of hubble s constant is changed , the stellar mass of the galaxies required to normalise the k - z relation at low redshift is changed , but the differences in the values of @xmath151 depend primarily upon the changing _ relative _ cosmic times observed at different redshifts . in general , these are quite small and so changing the value of @xmath148 does not greatly change the evolutionary corrections . if the redshift magnitude relation were determined at rest - frame uv wavelengths , this would certainly not be the case . the metallicity of the galaxies can significantly influence their evolutionary tracks . rather than attempting to adopt some arbitrary chemical history , the evolutionary codes have been run for metallicities of @xmath152 , @xmath153 , @xmath60 and @xmath154 . the results are displayed in fig . [ fig : 6 ] for the same two cosmologies discussed in section 4.2 and for @xmath155 . for the einstein de sitter cosmology , these results are also tabulated in table 4 . all four k - z tracks are normalised to the same magnitude scale as the fiducial model , rather than to the same magnitude at the present epoch . this procedure has been adopted so that at all times the variations due to changes in metallicity are clearly displayed . the sense of the changes is that , the lower the metallicity , the more luminous the galaxy at early epochs . it is apparent from fig . [ fig : 6](b ) , that assuming a lower metallicity at large redshifts improves the fit of the @xmath108 model as compared with the fiducial model . the lowest metallicity track is initially the brightest ( largest @xmath67 ) , and the @xmath154 track the faintest , but the tracks cross over at low redshift . the greatest variation between the tracks occurs at early epochs and , as we have seen from section 3.2 , it is at these young ages that the galaxies undergo the greatest amount of evolution . the early evolution of the spectral energy distribution is highly sensitive to the metallicity of the galaxy because , for a star of a given mass , changing the metallicity leads to a change in the age at which it moves from the main sequence onto the agb ; for lower values of @xmath156 the stars evolve more rapidly since there is less line blanketing and lower opacity . this has two consequences : firstly , the more rapid evolution means that lower metallicity stars were more luminous at early times ; secondly , the stars were also shorter lived and at later times there are fewer stars present in the older galaxies , which therefore appear fainter than a higher metallicity galaxy of a similar age . it is interesting that the divergence of the different metallicity tracks occurs when the universe was roughly half its present age , corresponding to the redshift at which the k - z relations for the 3cr and 6c samples begin to diverge . it is just conceivable that such an effect could account for the difference between these samples , but it would require those 6c galaxies which are less luminous than the 3cr galaxies to have higher metallicities . there is no obvious reason why greater radio luminosity should be correlated with lower metallicity of the stellar populations of the radio galaxies . the diagrams in fig . [ fig : 6 ] can be used to construct approximate evolving galaxy models in which the metallicities of the galaxies increase with cosmic epoch . such evolution tracks can be found by changing from one locus to another as the metallicity increases . the range of models shown in fig . [ fig : 6 ] illustrates how such changes would alter the predicted k - z relations . the results so far have involved an initial starburst during which the star formation rate has been assumed to be constant . it is important to consider also models in which there is ongoing star formation , and in particular those in which the occurrence of a radio source event might be associated with a major star formation event in the host galaxy . this could occur if , for example , the central black hole were fuelled following some form of galaxy merger . to understand the significance of these results , we need to look more closely at the origin of the k - band flux density . the k - band luminosity of evolving galaxies arises from two stellar populations . stars with mass m@xmath1572m@xmath29 evolve beyond core and shell helium burning to become asymptotic giant stars ; those with m@xmath1582m@xmath29 become red giant stars forming the old red giant branch seen , for example , in globular clusters . for the present study , we are mainly interested in the evolution of the overall stellar population of the galaxies over cosmic time - scales greater than @xmath50years . this age corresponds to the main sequence lifetime of a star of mass 2@xmath159 . therefore , at ages greater than @xmath50years , the evolution is dominated by stars evolving onto the standard red giant branch and the arguments given in section 3.2 can account for the decreasing luminosity of the galaxy . on the other hand , if there are epochs of rapid star formation , the bulk of the near - infrared luminosity is associated with stars on the asymptotic giant branch . for the most luminous stars , evolution onto the asymptotic giant branch takes place after a few times @xmath62years . figure [ fig : 3](b ) shows dramatically the large increase in k - luminosity of an instantaneous starburst when it is about @xmath63years old associated with the formation of agb stars . on the other hand , if the starburst is of longer duration , say @xmath64years as illustrated in figure [ fig : 3](c ) , the effect is somewhat diluted by the continued formation of all types of massive stars . considering first variations in the long - term star formation history of the galaxy populations , fig . [ fig : 7 ] displays three different types of star formation history : ( i ) the standard 1-gyr starburst , ( ii ) an instantaneous starburst of negligibly short duration , and ( iii ) a model in which the star formation rate decays exponentially with an e - folding time of @xmath50 years . there is very little difference between the three models , all three of which have been normalised to the same stellar mass at late cosmic times . the brightest model ( largest @xmath67 ) except at the very earliest times is the exponential star formation model , and the faintest the instantaneous burst model . these results are not surprising since , it is only the behaviour after @xmath50years which is relevant to the k - z relation for @xmath160 . once star formation ceases , the continued star formation of the exponential model , and sustained production of agb stars ( albeit in small numbers ) , is sufficient to make this model the brightest track at low redshift . the most evolved model , the instantaneous starburst , has the lowest luminosity at low redshift . these differences between the models are very small , as can be seen from the values given in table 4 . [ fig : 8 ] shows the standard model coupled with the seds of instantaneous starbursts at three different ages ( @xmath62 , @xmath63 and @xmath64 years old ) , the additional starburst accounting for 2.5% of the total stellar mass of the galaxy . the model with the additional burst occurring the longest time before observations , @xmath64 years , gives the least change from the predictions of the underlying model . similarly , the @xmath62 year old recent burst also results in only a slight difference at low redshift . in the first case , the additional young ( @xmath64 years old ) high mass stars have evolved beyond the agb , and in the second case ( @xmath62 years old ) they have not yet had enough time for their evolution to reach that stage . [ fig : 3 ] shows that the rest frame k - band flux of an evolved instantaneous burst at these ages is much closer to that for the underlying star formation model than the @xmath63 year old track . at high redshift , however , we need to consider a somewhat bluer rest - frame wavelength region . whilst there is a small difference for the @xmath64 year old model , the @xmath62 year old recent burst is slightly brighter still , which is seen in fig . [ fig : 8 ] . the model with the @xmath63 year old recent burst component shows the greatest change from the standard model , as expected from fig . [ fig : 3 ] . the @xmath67 values for these models given in table 4 are solely due to the additional recent burst component . the additional burst consistently brightens the k - z tracks at late times . the reason for the effect being most pronounced at small redshifts is that the luminosity of the starburst is assumed to be independent of redshift . since the k - luminosity increases in roughly power - law fashion with increasing redshift , the luminosity of the starburst is soon swamped by the majority population of red giant branch stars . [ fig : 9 ] displays the results for the same set of models as in fig . [ fig : 8 ] , but the mass fraction of the additional starburst is varied rather than its age . the tracks in this figure involve an additional @xmath63 year starburst and four mass fractions . if the mass of the starburst as a function of redshift remains constant , the overall shape of the k - z relation is flattened at late times and does not result in an improved fit to the observational data . some scatter may be introduced into the k - z relation , because the radio galaxies are observed at a range of ages from @xmath62 to @xmath64 years . in order for starbursts to change significantly the shapes of the k - z relations , the additional starburst mass would need to change with redshift . this is quite possible , as there is more gas available in any merger - induced starburst at high redshift . the same would be true if the starburst were radio jet induced ( e.g. rees 1989 ) , which provides a possible explanation of the alignment effect ( mccarthy _ et al . _ 1987 , chambers _ 1987 ) . figs . [ fig : 8 ] and 9 can be used to estimate how significantly the k - band luminosity could have changed if the magnitude of the starburst were to vary with cosmic epoch . just as in the case of evolving metallicity , so in the case of evolving starbursts , the evolutionary history can be estimated by moving from one track to another in figs . [ fig : 8 ] and 9 . for example , in fig . [ fig : 9 ] , if the early starbursts involved 5% of the galaxy mass , the evolutionary track would begin on the dotted line ; if this subsequently decreased to 2.5% and 1% , the track would change to the dot - dashed and dashed lines respectively . a 5% starburst by mass is the absolute maximum allowable given the constraints of the observed ultraviolet luminosities for the 3cr radio galaxies at @xmath38 ( e.g. best _ et al . _ 1998 ) . in this extreme case the magnitude increase in k is at most about 0.5 magnitudes . the above analyses of the k - z relations for a wide range of plausible cosmological models have established a number of useful general rules concerning the potential use of the luminosities of massive galaxies as standard candles . the value of carrying out these studies in the k - waveband at 2.2@xmath53 m for redshifts @xmath161 is clearly illustrated by the synthesised spectral energy distributions shown in fig . [ fig : 3](a ) in which it can be seen that the evolution of the dominant stellar populations can be predicted with some confidence . as has been discussed by previous authors , the dominant evolutionary trends are associated with the evolution of stars in the mass range 1 2 @xmath162 from the main sequence onto the red giant branch . an equally important result is that , over this range of redshifts , the choice of cosmological model is much more important than the corrections associated the evolution of the stellar populations of the galaxies , as is vividly illustrated by the values in table 2 . thus , provided standard galaxies can be identified over the redshift range @xmath163 , there are good prospects for obtaining useful information about the choice of cosmological model . the problem with this programme is the identification of such a class of standard galaxy , particularly in the light of the favoured picture of hierarchical clustering to account for the formation of structures such as galaxies and clusters of galaxies over the same redshift interval . the analysis of best _ et al . _ ( 1998 ) showed that the 3cr radio galaxies seemed to form a population of galaxies of constant mass within the framework of the @xmath164 world model , although they described a number of reasons why this was a somewhat surprising result . the present calculations have extended that analysis to a much wider range of cosmological models and considered in more detail the effects of the evolution of the stellar populations of the galaxies . the results shown in fig . [ fig : 4 ] and documented in table 2 show that all the new models are a significantly poorer fit to the large redshift data . in particular , the principal cause of the poorer fits is the choice of cosmological model . expressed in physical terms , in the critical @xmath102 model , galaxies at a given redshift are observed at significantly smaller distance measures than in all the other models considered and so have the brightest k - magnitudes . in all the other models considered , a literal interpretation of the results would be that the radio galaxies at redshifts @xmath32 have greater stellar masses than the radio galaxies nearby . the analysis has shown that this result is independent of the choice of hubble s constant . it is of particular interest to find ways in which the favoured flat cosmological model with @xmath165 can be reconciled with the observational data . as the results in table 4 indicate , to produce an improved fit to the observational data one or more of the following changes to the standard evolution picture must be introduced : 1 . a lower redshift of star formation , @xmath166 , provides a much improved fit to the observed k - z relation for this cosmological model ( fig . [ fig : 5]b ) . a problem with this improved fit is that the galaxies would then be expected to be very much more luminous in optical wavebands with significant evidence for younger stellar populations , whilst the red colours of distant radio galaxies suggest a much larger luminosity weighted stellar age ( e.g. cf dunlop _ et al . _ , 1996 ) . further , the observation of well formed galaxies which are powerful radio sources out to redshift @xmath167 ( van breugel _ et al . _ , 1999 ) indicates that at least some of these galaxies formed at higher redshifts . the metallicities of the galaxies may have been significantly less at redshifts @xmath168 than they are at the present epoch . as shown in fig . 6(b ) , decreasing the metallicity of the galaxy to only about 20% of the present cosmic value improves significantly the fit of the model to the observations . starbursts associated with the radio source events can lead to a significant enhancement of the k - band luminosities of the galaxies , due to the rapid evolution of luminous massive stars onto the asymptotic giant branch . this could potentially improve the fit of the models in lambda cosmologies if associated starbursts brighten the distant galaxies , but not those at redshifts less than 1 . this is quite plausible since such starbursts could be enhanced at high redshifts due to the extra gas available at these epochs for forming stars , either in a galaxy merger or triggered by the radio jet ; indeed additional blue luminoisity is observed in powerful radio galaxies at high redshifts ( the alignment effect ) . there is clearly much scope for arbitrarily adjusting the masses , timescales and cosmic evolution of the starbursts to improve the fit to the k band data . however , the excess blue luminosity associated with the alignment effect requires only a few @xmath169 of newly formed stars ( @xmath170% of the galaxy mass ) , even were it all to be associated with star formation activity ( e.g. dunlop _ et al . _ 1989 ; best _ et al . _ , 1998 ) : this sets a strong constraint upon the luminosities and durations of the bursts . the alternative approach is to accept that , even after accounting for stellar evolution , the radio galaxies at @xmath171 may indeed be more luminous , that is , have a greater stellar mass , than those at small redshifts . this conclusion would mean that the low and high redshift radio galaxies would not form a single evolving population , and would be surprising in terms of hierarchical clustering scenarios according to which such massive galaxies at high redshift are extremely rare . given the growing evidence that the distant radio galaxies belong to cluster environments , it may be plausible that they had undergone more mergers than the nearby radio galaxies , which are found in more rarified environments : this possibility could be tested by carrying out cross - correlation analyses of radio galaxies of different k - luminosities with field galaxies , although this becomes a demanding project for the radio galaxies at @xmath171 . the similar characteristic sizes of the 3cr radio galaxies at redshifts 1 and 0 , however , sets constraints upon the intrinsic differences in luminosities which effectively rule out the most extreme cosmological models , those with @xmath172 . this work was supported in part by the formation and evolution of galaxies network set up by the european commission under contract erb fmrx ct96086 of its tmr programme . kji acknowledges the support of a pparc research studentship . pnb would like to thank the royal society for generous financial support through its university research fellowship scheme . aragon - salamanca a. , baugh c.m . , kauffmann g. , 1998 , mnras , 297 , 427 arimoto n. , kodama t. , 1996 , ` galaxy scaling relations ' , eds : l.n . da costa & a. renzini , p132 best p.n . , longair m.s . , rttgering h.j.a . , 1997 , mnras , 292 , 758 best p.n . , longair m.s . , rttgering h.j.a . , 1998 , mnras , 295 , 549 best p.n . , longair m.s . , rttgering h.j.a . , 1999 , ` the most distant radio galaxies ' , eds : h. rttgering et al . , p59 best p.n . , rttgering h.j.a . , bremer m.n . , cimatti a. , mack k.h . , miley g.k . , pentericci l. , tilanus r.p.j . , van der werf p.p . , 1998 , mnras , 301 , l15 bower r.g . , kodama t. , terlevich a. , 1998 , mnras , 299 , 1193 bruzual g. , charlot s. , 1993 , apj , 405 , 538 bruzual g. , charlot s. , 2001 , ( in prep . ) chambers k.c . , miley g.k . , van breugel w.j.m . , 1987 , nat , 329 , 604 charlot s. , private communication collins c.a . , mann r.g . , 1998 , mnras , 297 , 128 dunlop j.s . , peacock j.a . , savage a. , lilly s.j . , heasley j.n . , simon a.j.b . , 1989 , mnras , 238 , 1171 dunlop j. , peacock j. , spinrad h. , dey a. , jimenez r. , stern d. , windhorst r. , 1996 , nature , 381 , 581 eales s.a . , 1985 , mnras , 217 , 149 eales s.a . , rawlings s. , 1996 , apj , 460 , 68 eales s.a . , rawlings s. , law - green d. , cotter g.,lacy m. , 1997 , mnras , 291 , 593 gunn j.e . , longair m.s . , rees m.j . , 1978 , ` observational cosmology ' , 8th advanced course of the swiss society of astronomy and astrophysics ( ssaa ) girardi l. , bressan a. , bertelli g. , chiosi c. , 2000 , a&as , 141 , 371 jaffe et al . , 2001 , physrevl , 86 , 3475 laing r.a . , riley j.m . , longair m.s . , 1983 , mnras , 204 , 151 lilly s.j . , longair m.s . , 1984 , mnras , 211 , 833 lilly s.j . , mclean i.s , longair m.s . , 1984 , mnras , 209 , 401 longair m.s . , 1998 , ` galaxy formation ' , springer - verlag , berlin mackay d.j.c . , 1995 , icann 95 proceedings , p331 madau p. , pozzetti l. , dickinson m. , 1998 , apj , 498 , 106 maeder a. , meynet g. , 1989 , a&a , 210 , 155 mccarthy p.j . , 1999 , ` the most distant radio galaxies ' , eds : h. rttgering et al . , p5 mccarthy p.j . , van breugel w.j.m . , spinrad h. , djorgovski s. , 1987 , apj . , 321 , l29 mccarthy p.j . , spinrad h. , van breugel w.j.m . , 1995 , apjs , 99 , 27 mccarthy p.j . , kapahi v.k . , van breugel w. , persson s.e . , athreya r.m . , subrahmanya c.r . , 1996 , apjs , 107 , 19 mclure r.j . , dunlop j.s . , 2000 , mnras , 317 , 249 pei y.c . , fall s.m . , hauser m.g . , 1999 , apj , 522 , 604 perlmutter s. et al . , 1999 , apj , 517 , 565 pettini m. , smith l.j . , king d.l . , hunstead r.w . , 1997 , apj , 486 , 665 poggianti b.m . , 1997 , a&as , 122 , 399 rees m.j . , 1989 , mnras , 239 , 1p sandage a. , 1972 , apj , 173 , 485 stanford s.a . , eisenhardt p.r . , dickinson m. , 1998 , apj , 492 , 461 thomas a. , spiegelhalter d.j . , gilks w.r . , 1992 , ` bayesian statistics 4 ' , eds : bernardo j.m et al . , p837 van breugel w.j.m . , stanford s.a . , spinrad h. , stern d. , graham j.r . , 1998 , apj , 502 , 614 van breugel w.j.m . , de breuck c. , stanford s.a . , stern d. , rttgering , h.j.a . , miley g. , 1999 , apj , 518 , l61 willott , private communication +

we investigate the interaction between a giant planet and a viscous circumstellar disk by means of high - resolution , two - dimensional hydrodynamical simulations . we consider planet masses that range from to jupiter masses ( ) and initial orbital eccentricities that range from to . we find that a planet can cause eccentricity growth in a disk region adjacent to the planet s orbit , even if the planet s orbit is circular . disk - planet interactions lead to growth in a planet s orbital eccentricity . the orbital eccentricities of a and a planet increase from to within about orbits . over a similar time period , the orbital eccentricity of a planet grows from to . for a case of a planet with an initial eccentricity of , the orbital eccentricity grows to over orbits . radial migration is directed inwards , but slows considerably as a planet s orbit becomes eccentric . if a planet s orbital eccentricity becomes sufficiently large , , migration can reverse and so be directed outwards . the accretion rate towards a planet depends on both the disk and the planet orbital eccentricity and is pulsed over the orbital period . planet mass growth rates increase with planet orbital eccentricity . for the mass growth rate of a planet increases by above the value for . for , most of the accretion within the planet s roche lobe occurs when the planet is near the apocenter . similar accretion modulation occurs for flow at the inner disk boundary which represents accretion toward the star .

astro-ph0608355

[ sec : introduction ] a striking property of the extrasolar planets known to date is the range of their orbital eccentricities , which is far wider than that of planets in the solar system . eccentricities are typically @xmath19@xmath20 , but very low and high values are also found ( e.g. , * ? ? ? a variety of explanations have been proposed to explain the eccentricities . the high eccentricity cases may owe their explanation to the presence of a distant binary companion @xcite . in fact , the large eccentricity ( @xmath21 ) of hd 80606b could be the result of a three - body interaction process , known as `` kozai cycle '' @xcite . moreover , numerical experiments show that the observed distribution of orbital eccentricities for @xmath22 can be reproduced by assuming the action of a kozai - type perturbation @xcite . for more typical eccentricities , other processes are likely to be at work . planet - planet interactions involving scattering on dynamical timescales is a possibility @xcite . however , numerical experiments indicate that interactions between equal mass planets would produce more isolated planets on low - eccentricity orbits than those observed @xcite . secular interactions between planets is another possible means of eccentricity excitation @xcite . this mechanism assumes that the planets have appropriate initial separations for evolution to occur on secular timescales ( @xmath23 years ) . disk - planet interactions can also give rise to planetary eccentricities @xcite . the evolution of orbital eccentricity depends on a delicate balance between lindblad and corotation resonances . if the planet is massive enough to clear a gap , lindblad resonances promote eccentricity growth , while corotation resonances damp eccentricity . if the corotation resonances operate at maximal efficiency , they dominate over lindblad resonances by a slight margin , and the result is eccentricity damping @xcite . two mechanisms have been proposed to weaken the effects of corotation resonances and thereby provide eccentricity growth . the first mechanism relies on a large enough gap about a planet s orbit to exclude the corotation resonances from the disk , while leaving a remaining lindblad resonance , the 1:3 outer resonance . this mechanism requires a massive enough companion and/or low enough viscosity to open a wide enough gap . such a mechanism has been demonstrated for binary stars @xcite . the second mechanism relies on the delicate nature of corotational resonances in their ability to weaken ( saturate ) when the disk viscosity is sufficiently small and the planet eccentricity sufficiently large @xcite . a small finite initial eccentricity , typically @xmath24 , is required for the eccentricity to grow by this mechanism . moreover , a companion object on a circular orbit can drive eccentricity in a circumstellar disk . this process is believed to occur in disks involving @xmath25@xmath26 jupiter - mass ( @xmath2 ) companions @xcite and binary stars @xcite . the disk transfers eccentricity to the planet s orbit by exchanging energy and angular momentum . the disk - planet system undergoes a coupled eccentricity evolution . one goal of this paper is to determine whether the eccentricity growth by disk - planet interactions occurs . simulations by @xcite suggested that orbital eccentricity is excited when the mass of the embedded body is larger than about @xmath25@xmath26 jupiter - masses , while lower mass companions experienced eccentricity damping . by applying higher resolution and simulating over longer timescales , we aim to see whether eccentricity growth can occur for lower mass , planetary mass , companions . in addition , we are interested in the effects of eccentricity on planet migration . accretion of gas onto a planet is likely affected by the planet s orbit eccentricity . circumbinary disks surrounding eccentric orbit binaries undergo pulsed accretion on orbital timescales @xcite . a similar process could occur with eccentric orbit planets . the mean accretion rate could also be modified by the orbital eccentricity . another goal of this paper is to investigate this accretion process . g. dangelo , s. lubow , & m. bate we apply high - resolution hydrodynamical simulations to investigate disk - planet interactions over several thousand orbital periods . in section [ sec : model_description ] we describe the physical model . the numerical aspects of the calculations are detailed in section [ sec : numerical_issues ] . results on the growth of the disk eccentricity for a fixed planet orbit are presented in section [ sec : disk_eccentricity ] . similar results on the growth of disk eccentricity were recently reported by @xcite . we describe results on the planet s orbital evolution in section [ sec : orbital_evolution ] . results on the pulsed accretion are described in section [ sec : planet_accretion ] . finally , our conclusions are given in section [ sec : conclusions ] we assume that the disk matter and planet are in coplanar orbits . in order to describe the dynamical interactions between a circumstellar disk and a giant planet , we approximate the disk as being two - dimensional , by ignoring dynamical effects in the direction perpendicular to the orbit plane ( vertical direction ) . this approximation is justified by the fact that the disk thickness is smaller than the vertical extent of the planetary roche lobes for the cases we consider . comparisons between two- and three - dimensional models of jupiter - mass planets embedded in a disk indicate that the two - dimensional geometry provides a sufficiently reliable description of the system @xcite . we employ a cylindrical coordinate frame @xmath27 in which the disk lies in the plane @xmath28 and the origin , @xmath29 , is located on the primary . the reference frame rotates about the disk axis ( i.e. , the @xmath30-axis ) with an angular velocity @xmath31 and an angular acceleration @xmath32 . the set of continuity and momentum equations , describing the evolution of the disk , is written in a conservative form ( see , e.g , * ? ? ? * ) and is solved for the surface density , @xmath33 , and the velocity field of the fluid , @xmath34 . the turbulent viscosity forces in the disk are assumed to arise from a standard viscous stress tensor for a newtonian fluid with a constant kinematic viscosity , @xmath35 , and a zero bulk viscosity . a locally isothermal equation of state is adopted by requiring that the vertically integrated pressure is equal to @xmath36 and that the sound speed , @xmath37 , is equal to the disk aspect ratio , @xmath38 , times the keplerian velocity . in this study , we use a constant aspect ratio throughout the disk . the gravitational potential of the disk , @xmath39 , is given by @xmath40 where @xmath41 is the mass of the central star whereas @xmath42 and @xmath43 are the mass and the vector position of the planet , respectively . the length @xmath44 represents a softening length that is needed to avoid singularities in the gravitational potential of the planet . the third term on the right - hand side of equation ( [ eq : phi ] ) accounts for the acceleration of the origin of the ( non - inertial ) coordinate frame caused by the planet . we ignore disk self - gravity . for the disks we consider , the toomre-@xmath45 parameter never drops below @xmath46 during the simulations . the orbit of the planet evolves under the gravitational forces exerted by the central star and the disk material . non - inertial forces arising from the rotation of the reference frame also need to be taken into account . therefore , the equation of motion of the planet reads @xmath47 where the angular velocity and acceleration vectors of the rotating frame are defined as @xmath48 and @xmath49 , respectively . the second , third , and fourth terms on the right - hand side of the equation are the centrifugal , coriolis , and angular accelerations , respectively . the angular velocity of the reference frame relative to a fixed frame , @xmath50 , is chosen so as to compensate for the azimuthal motion of the planet . details of how this is achieved can be found in @xcite . since the azimuthal position of the planet is constant , a circular orbit reduces to a fixed point in the rotating frame . for a planet orbit with @xmath51 , the planet radially oscillates between the pericenter distance , @xmath52 , and the apocenter distance , @xmath53 . in this frame , the trajectory is a straight line of length @xmath54 with center at @xmath55 . this scheme has the advantage that the planet does not drift across the grid in the azimuthal direction . it always maintains a symmetric azimuthal position with respect to the zone centers ( where eqs . [ [ eq : phi ] ] , [ [ eq : a2 ] ] , and [ [ eq : a1 ] ] are evaluated ) while it moves radially . this method helps reduce artificially unbalanced forces that act on the protoplanet ( see discussion in * ? ? ? * ; * ? ? ? the last two terms in equation ( [ eq : pme ] ) , which represent the forces per unit mass exerted by the disk material on the planet and the star , are @xmath56 and @xmath57 these two vectors are included in equation ( [ eq : pme ] ) only when the disk back - reaction is accounted for and the protoplanet is allowed to adjust its trajectory in response to the disk torques . when these terms are applied , the integration in equations ( [ eq : a2 ] ) and ( [ eq : a1 ] ) is performed over the disk mass comprised in the simulated domain , @xmath58 . in these calculations , the stellar mass , @xmath41 , is the unit of mass and the initial semi - major axis of the planet s orbit , @xmath59 , represents the unit of length . the unit of time , @xmath60 , is defined such that @xmath61 . when we refer to the `` orbital period '' or `` orbit '' as length of time , we actually mean the initial orbital period @xmath62 . to provide estimates of various quantities in physical units , we adopt @xmath63 and @xmath64 , thus one orbit is @xmath65 years . in order to treat the strong perturbations induced by giant protoplanets and limit the influence of the imposed radial boundaries , we considered an extended disk whose radial borders are at @xmath66 and @xmath67 . the disk extends over the entire @xmath68 radians in azimuth around the star . in physical units , the disk models cover a region from @xmath69 to @xmath70 . the mass of the disk , in the absence of the planet , is @xmath71 within the radial limits of the simulated region . we used a constant disk aspect ratio , @xmath72 . the unperturbed initial surface density is axisymmetric and scales as @xmath73 , which produces an unperturbed density , at @xmath74 , equal to @xmath75 . however , given the large planetary masses considered in this investigation , we also included an initial gap along the orbit of the planet that accounts for an approximate balance between viscous and tidal torques ( e.g. , * ? ? ? * ) . the initial gap profile is based on equation ( 5 ) of @xcite . the initial gap width is modified by a factor of @xmath76 , in order to account for the planet s initial orbital eccentricity , @xmath77 . the initial velocity field in the disk is a keplerian one that is centered on the star and corrected for the rotation of the frame of reference . in order to account for the effects due to turbulence in the disk , a constant kinematic viscosity , @xmath35 , was used . in terms of the shakura & sunyaev parameter , we have @xmath78 , where @xmath79 ( in physical units , @xmath80 ) . although spatial variations and time fluctuations consistent with the mhd turbulence are not included , this relation yields a magnitude of @xmath81 that is in the range found in mhd simulations @xcite . the influence of viscosity was explored by performing a few calculations with other @xmath82 values ( @xmath83 and @xmath84 ) . ccccccc @xmath0 & @xmath85 & @xmath85 & @xmath85 & @xmath85 & @xmath85 & @xmath85 + @xmath86 & @xmath85 & & & & & + @xmath1 & @xmath85 & & @xmath85 & @xmath85 & @xmath85 & @xmath85 three planetary masses were considered : @xmath87 and @xmath6 ( i.e. , the mass ratio @xmath88 ranges from @xmath89 to @xmath90 ) . the planets were set on initially circular or eccentric orbits about one solar mass stars . we examined configurations with initial eccentricities , @xmath77 , up to @xmath4 . a complete list is given in table [ tbl : eccentricities ] . at time @xmath91 the planet starts from the pericenter position , while its azimuth , @xmath92 , remains constant ( in the rotating frame ) and equal to @xmath93 throughout the calculation . in order to allow the disk to adjust to the presence of the planet , we impose two stages to the evolution . during the first phase , the planet s orbit is static and terms ( [ eq : a2 ] ) and ( [ eq : a1 ] ) are not included in equation ( [ eq : pme ] ) . during the second phase , the protoplanet is `` released '' from the fixed orbit and is allowed to react to the disk torques via the full form of equation ( [ eq : pme ] ) . in the models presented here the first phase lasts until the release time , @xmath94 , which ranges from @xmath95 to @xmath96 orbits . the second phase lasts from several hundred to several thousand orbits . the value adopted for smoothing radius @xmath44 ( in eqs . [ [ eq : phi ] ] and [ [ eq : a2 ] ] ) resulted from numerical experiments in each orbital eccentricity configuration . the chosen value of @xmath44 was the smallest that prevented the integration time - step of the hydrodynamics equations from getting shorter than @xmath97 orbits . in models involving @xmath9 and @xmath5 planets , we set @xmath98 , where @xmath99 @xcite is planet s hill radius ( or sometimes called roche radius ) . in models involving @xmath6 planets , we applied softening lengths between @xmath100 and @xmath101 . the latter value was used at the highest initial orbital eccentricities , @xmath102 and @xmath4 . we found that torques within the roche lobe do not dominate the planet orbital evolution . moreover , the smoothing radius does not significantly affect planet accretion . thus , @xmath44 does not likely play an important role in these calculations . for simulations that account for the disk torques on the planet , an additional approximation is made , which is described at the beginning of section [ sec : orbital_evolution ] . the equations of motion of the disk are solved numerically by means of a finite - difference scheme that uses a directional operator splitting procedure . the method is second - order accurate in space and semi - second - order in time @xcite . hydrodynamic variables are advected by means of a transport scheme that uses a piecewise linear reconstruction of the variables with a monotonised slope limiter @xcite . high numerical resolution in an extended region around the planet is achieved by using a nested - grid technique ( see * ? ? ? * ; * ? ? ? * for details ) with fully nested subgrid patches , whereby each subgrid level increases the resolution by a factor @xmath86 in each direction . tests on the behavior of the nested - grid technique applied in a reference frame rotating at a variable rate @xmath50 are reported in the appendix of @xcite . the equation of motion of the planet ( eq . [ [ eq : pme ] ] ) is solved by using a high - accuracy algorithm described in @xcite . ccccc @xmath0 & @xmath103 & @xmath103 & @xmath104 & @xmath104 + @xmath86 & @xmath105 & @xmath105 & @xmath106 & + @xmath1 & & @xmath107 & & + the excitation or damping of a protoplanet s orbital eccentricity depends on a delicate balance between lindblad and corotation resonances ( * ? ? ? * ; * ? ? ? * and references therein ) . to study this balance , it is necessary to resolve the width of all the resonances involved in the process . the locations of first order eccentric lindblad resonances reside in a region that ranges radially from approximately @xmath108 to @xmath109 . furthermore , calculations on the saturation of isolated noncoorbital corotation resonances , performed by @xcite , suggest that a minimum resolution requirement to avoid spurious damping of eccentricity is that @xmath110 where @xmath111 is the azimuthal wavenumber and @xmath112 are coefficients of order unity given in @xcite . although nested grids provide a linear resolution @xmath113 in a region @xmath114 , the effects of these corotational resonances are not localized in azimuth , so the nested grids do not substantially improve their overall resolution . calculations that follow the orbital evolution of the planet were executed with grid systems gs1 and gs2 ( see table [ tbl : grids ] for a description of all grid systems employed in this study ) . according to equation ( [ eq : minres ] ) , the global radial resolution we apply @xmath115 could produce spurious damping in the outer disk for orbital eccentricities @xmath116 , with @xmath117 . in the @xmath9 case , simulations with @xmath118 could undergo spurious eccentricity damping for the most distant outer resonance @xmath119 only . for @xmath120 , there is no spurious damping . in the @xmath6 case , simulations with @xmath121 could undergo spurious damping for @xmath119 only . for @xmath122 , there is no spurious damping . accretion onto the protoplanet was simulated by removing material within a distance of @xmath123 . mass is removed by means of a two - step procedure according to which the removal timescale , @xmath124 , is @xmath125 orbital periods for @xmath126 and @xmath12 orbital periods in the outer part of the accretion region , @xmath127 . notice that @xmath124 is about equal to the keplerian period around the protoplanet ( i.e. , in the circumplanetary disk ) at @xmath128 . tests were carried out to evaluate the sensitivity of the procedure to both parameters @xmath129 and @xmath124 . these tests indicate that , if @xmath129 is reduced by a factor of @xmath130 , the average mass accreted during an orbit varies by less than @xmath25% . increasing the removal timescale by a factor of @xmath131 only affects the accretion rate by @xmath132% ( see also * ? ? ? we also checked whether the accretion parameters influence the orbital evolution of the planet . using the same tests , we found no significant differences over a few hundred periods of evolution . we do not add the removed mass to the planet mass over the course of the simulations . doing so would increase the planet mass by @xmath133 for the simulations reported here . boundary conditions at the radial inner boundary , @xmath134 , allow outflow of material , i.e. , the accretion flow toward the central star . the inner boundary conditions exclude inflow away from the star into the computational domain . two types of boundary conditions were used at the outer boundary ( @xmath135 ) : reflective and non - reflective . with the first kind , neither inflow nor outflow of material is permitted at the outer border , which behaves as a rigid wall . although @xmath136 is much larger than the apocenter radii of the planets in the simulations , wave reflection was observed at the outer border in this case , especially when the planet orbit was eccentric . in order to lessen the amount of wave reflection at the outer disk edge , we also applied non - reflective boundary conditions , following the approach of @xcite . in these circumstances , boundary conditions are not directly imposed on the primitive variables ( i.e , @xmath33 and @xmath34 ) , but rather on the characteristic variables ( i.e. , the riemann invariants of one - dimensional flows ) . the basic idea is to let outflowing ( inflowing ) characteristics propagate through the grid boundary . the correct propagation of the flow characteristics across the border depends on how accurately the adopted solution exterior to the grid approximates the exact solution . at both disk radial boundaries , the flow is assumed to be keplerian around the central star . this choice could lead to some small - amplitude wave excitation at @xmath135 , since the fluid tends to orbit about the center - of - mass of the system , rather than around the central star ( see also * ? ? ? the effects of such waves were checked to be unimportant as they tend to dissipate within a short distance from the outer disk boundary the global surface density in the disk is plotted in figure [ fig : globalsigma ] for different planet s masses and orbital eccentricities . left panels refer to configurations with @xmath137 , and right panels refer to configurations with @xmath138 . the orbital eccentricity increases from top ( @xmath17 ) to bottom ( @xmath139 ) . in all of the panels , the planet is at pericenter ( @xmath140 ) . for sufficiently long evolutionary times , the inner disk ( disk interior to the planet ) is largely depleted because of the tidal gap produced by the planet and because the grid does not have sufficient dynamic range to cover regions close to the star where an inner disk would reside . the outer disk is tidally truncated at a radial distance that depends on both @xmath42 and @xmath141 . figure [ fig : globalsigma ] shows that the size of the truncation radius increases with planet mass and eccentricity , analogous to the case for circumbinary disks @xcite . in addition , a wave or wake propagates in the outer disk . the disk truncation also be seen in figure [ fig : avsigma ] , which shows the azimuthally averaged surface density for various cases . the outer gap edge becomes less steep as the orbital eccentricity increases and does not change significantly over an orbital period , as illustrated in figure [ fig : gapedge_inertial ] . after a few hundred orbits , the disk region near the planet becomes eccentric , even if the planet is on a circular orbit . this effect can be seen in figure [ fig : rcmvstime ] , which illustrates the motion ( relative to the star ) of the center - of - mass of the disk interior to @xmath142 for different planetary masses and orbital eccentricities . we performed a mode decomposition of the surface density distribution of the disk adopting an approach along the lines of @xcite . we defined a mode component at each radial location in the disk as @xmath143 where @xmath144 , the angle @xmath145 is the azimuth relative to an _ inertial _ reference frame , and @xmath146 . the time interval @xmath147 is a ten - orbit period interval , beginning at a pericenter passage of the planet . the time integration is repeated every interval @xmath147 . the functions @xmath148 and @xmath149 are either sine or cosine . this decomposition corresponds to a fourier transform in both azimuth and time . the amplitude ( or strength ) of the mode is @xmath150^{2}+ \left[\mathcal{m}_{(k , l)}^{(\cos,\sin)}\right]^{2}+\right . \nonumber\\ & & \left . \left[\mathcal{m}_{(k , l)}^{(\sin,\cos)}\right]^{2}+ \left[\mathcal{m}_{(k , l)}^{(\sin,\sin)}\right]^{2 } \right\}^{1/2}.\end{aligned}\ ] ] in order to obtain the strength of a mode integrated over a radial interval @xmath151 $ ] , the mode components @xmath152 are first averaged between @xmath153 and @xmath154 and then substituted into equation ( [ eq : modeskl ] ) . a tidally disturbed non - eccentric disk has modes present that all have @xmath155 . for a disk ring to be eccentric , the mode strength associated with the pair @xmath156 , known as eccentric mode , must be non - zero , i.e. , @xmath157 . to follow the eccentricity evolution , we analyze @xmath158 as a function of the time . the amplitude of the eccentric mode , as defined in equation ( [ eq : modeskl ] ) , is shown in figure [ fig : s10_vs_t ] during the evolutionary phase in which the planet is kept on a fixed orbit . this figure refers to the configurations with @xmath137 , @xmath5 , and @xmath6 planets having @xmath17 . the disk region that undergoes a substantial eccentricity growth is that between @xmath159 and @xmath160 ( fig . [ fig : s10_vs_t ] , left panel ) . the relatively small initial value of @xmath158 is likely due to a transient effect , as the initially circular disk adjusts to the presence of the planet . farther away from the planet s orbit , the mode strength drastically decreases , and the disk is nearly circular ( fig . [ fig : s10_vs_t ] , middle and right panels ) . the eccentricity growth proceeds very rapidly during the first @xmath161 orbits and oscillates afterwards with some reinforcement in the @xmath6 case . the eccentricity driven in the region @xmath162 by the @xmath9 planet ( @xmath17 ) is rather small compared to that driven by the other two planetary masses . however , with an initial orbital eccentricity @xmath163 , a @xmath9 planet is able to sustain eccentricity growth in the disk , as indicated by the solid line in the left panel of figure [ fig : s10_vs_mp ] . in the @xmath6 case , an initial orbital eccentricity of up to @xmath139 induces a larger amplitude eccentricity perturbation only at the beginning of the evolution . but the long - term evolution of @xmath158 is not greatly affected ( fig . [ fig : s10_vs_mp ] ) . the simulations for mode analyses were generally performed with the grid gs4 . we also ran simulations and computed modes by using the grid system gs1 and found results consistent to those obtained with the higher resolution grid . the sensitivity of disk eccentricity growth to the disk kinematic viscosity was examined for a system containing a @xmath6 planet on an initially circular orbit . two additional values of @xmath82 ( i.e. , @xmath81 at @xmath74 ) were considered : @xmath164 and @xmath165 , which are respectively a factor of @xmath1 smaller and larger than the standard value @xmath79 . in the lowest viscosity model , the overall evolution of the amplitude of the eccentric mode closely resembles that of the model with standard viscosity ( see fig . [ fig : s10_vs_t ] , solid line ) , but was roughly @xmath166 larger . in the highest viscosity model , the eccentric mode strength reached a maximum of @xmath20 at @xmath167 orbits and then decays . around @xmath168 orbits , @xmath169 and continues to decline . the effects of viscosity were also investigated with a calculation involving a @xmath9 planet on a circular orbit and @xmath164 . in this case , the mode amplitude @xmath158 is sustained at about @xmath170 for @xmath171 orbits , unlike the case of declining eccentricity at higher viscosity that is displayed as a short - dashed line in the left panel of figure [ fig : s10_vs_t ] . these results suggest that disks with @xmath81 less than a few times @xmath172 experience sustained eccentricity , while disks with @xmath81 in excess of @xmath173 do not . the weakening of disk eccentricity with viscosity was also found in simulations by @xcite . in the case of superhump binaries , disk precession is dominated by the gravitational effects of the companion which causes prograde precession @xcite . on the other hand , pressure provides a retrograde contribution which is somewhat weaker @xcite . in the case of a circular orbit planet , the gravitational contribution to precession is expected to be weaker than the disk s pressure contribution . the magnitude of the pressure induced precession rate is @xmath174 . for the disks simulated in this paper , the precession timescale is then @xmath175 orbits . we estimated the gravitational precession due to a @xmath6 planet for the case plotted in figure [ fig : s10_vs_t ] at @xmath95 orbits and find that this rate is about @xmath25 times smaller than the pressure precession rate . we then expect the precession to be pressure - dominated and retrograde , with a timescale of about @xmath176 orbits . this estimate in accord with the simulation results for a @xmath6 planet in a circular orbit in figure [ fig : rcmvstime ] . the group velocity for an eccentric mode is estimated by using the dispersion relation for an @xmath177 disturbance in a keplerian disk perturbed by pressure . the group velocity is @xmath178 , where we assume the radial wavenumber @xmath179 and that the pattern speed is small compared to @xmath31 . the group velocity leads to eccentricity propagation timescales of order @xmath180 orbits . this timescale is consistent with the localization of the eccentricity over course of the simulation of @xmath175 orbits to within a region of order @xmath181 , as seen in figure [ fig : s10_vs_mp ] . over longer timescales the eccentricity would spread further . we analyze growth of eccentricity in an outer disk that is perturbed by a planet on a circular orbit . disk eccentricity growth via the 3:1 resonance for inner disks of superhump binaries occurs on a timescale @xmath182 binary orbit periods , where @xmath183 is the radial width of the eccentric region @xcite . we now consider the case for outer disks perturbed by planets . for simplicity , we assume that the eccentric corotational resonances in the disk are saturated ( i.e. , of zero strength ) . this situation is likely to hold if the disk eccentricity is of order @xmath11 , by analogy with the case of an eccentric orbit planet interacting with a circular disk @xcite . following the mode coupling analysis for eccentric lindblad resonances involving a circular orbit planet @xcite , we find that the disk eccentricity growth rate associated with a particular eccentric outer lindblad resonance is given by @xmath184 where @xmath185 @xmath186 @xmath187 @xmath188\,,\ ] ] where @xmath183 denotes the radial extent of the eccentric region and prime denotes differentiation in @xmath189 . all quantities are evaluated at the location of the eccentric lindblad resonance associated with azimuthal wavenumber @xmath190 , having @xmath191 and @xmath192^{2/3}\,a$ ] . quantities @xmath193 and @xmath194 are the velocity components in a circular disk associated with potential @xmath195 , and @xmath196 is the kronecker delta function in the indirect potential term . the combined effects of all the resonances is determined by the local disk density at each resonance . the growth rate of the mass - weighted eccentricity is then @xmath197 where @xmath198 is the mass of the eccentric region and the sum is taken over the active eccentric lindblad resonances , and @xmath199 due to torque cut - off effects @xcite . the contribution of each resonance in the above sum for @xmath200 for a @xmath6 planet is shown in figure [ fig : lam ] . we consider times beyond several hundred orbits , when the width of the eccentric region @xmath183 @xmath201 . figure [ fig : lam ] shows that there is a weak contribution from the outermost resonance , the 1:3 resonance , corresponding to @xmath177 . even though the density at this resonance is the largest , a nearly complete cancellation occurs in @xmath202 , due to effects of the indirect term in potential @xmath203 . if only the @xmath177 resonance were involved , then the eccentricity growth timescale would be very long , @xmath204 orbits . the growth rate contributions from regions closer to the planet are weakened by the lower density , but strengthened by the larger values of @xmath205 . at a time of 500 orbits , the density near the planet is small enough that the outermost resonances ( @xmath206 ) provide most of the growth rate . at a time of 500 orbits , the eccentricity growth timescale @xmath207 is about @xmath208 orbits , which is in very rough agreement with the average growth rate implied by figure [ fig : s10_vs_t ] for the innermost region , although there are considerable fluctuations in the simulations . with simulations , it is better to adopt a similar mass - weighted eccentricity . this requires a slightly different definition of @xmath158 than given by equation ( [ eq : modemkl ] ) . when we apply that definition , we obtain eccentric disk evolution in the @xmath6 case that is similar to the @xmath209 results in figure [ fig : s10_vs_t ] , except that the magnitude of @xmath158 in the innermost region is about a factor of 8 smaller . the average growth rate of @xmath158 in the innermost region is about the same over @xmath95 orbits . ] at earlier times , the growth rate is higher because @xmath183 is smaller . we conclude that the disk eccentricity growth is possible over @xmath175 orbits in the case of planets because of contributions from several resonances that lie in the disk edge / gap region . this situation differs from the superhump binary case where only a single resonance is involved , since the disk extends relatively closer to the perturber in the planet case , due to its weaker tidal barrier . increased viscosity affects the disk eccentricity in multiple ways . it leads to further disk penetration of the planet s tidal barrier , which could lead to stronger eccentricity growth . on the other hand , viscosity also acts to unsaturate ( or strengthen ) the corotation resonances which act to damp eccentricity . furthermore , the viscosity acts to damp the non - circular motions associated with the eccentricity . at time @xmath94 ( between @xmath95@xmath96 orbits ) the planet was allowed to adjust its orbit in response to the gravitational forces exerted by the surrounding disk material . we generally neglected torques on the planet due to gas within a distance of @xmath210 from the planet . however , analyses of the torque distribution at the release time indicate that torques within the roche lobe do not dominate the orbital evolution of the planet . so we believe this procedure does not likely lead to major errors in the planet s orbital evolution . all calculations presented in this section employed grid gs2 , except for the @xmath6 cases on initially circular orbits ( @xmath211 ) , which employed grid gs1 . in order to analyze the orbital evolution of the planet after release , we calculated the osculating elements of the orbit each few hydrodynamics time - steps ( approximately every @xmath11 orbits ) . to remove short - period oscillations , we computed the mean orbital elements @xcite by using an averaging period of one orbit . throughout the paper , the planetary orbital eccentricities and semi - major axes in the simulations refer to the mean orbital elements . simulations by @xcite showed that the interaction between an initially circular disk and a circular orbit planet with mass @xmath212 can lead to the growth of disk eccentricity and planetary orbital eccentricity . they also found that this interaction can be more efficient at augmenting orbital eccentricity than direct wave excitation at the outer 1:3 lindblad resonance in a non - eccentric disk ( e.g. , * ? ? ? * ) . we aim at determining whether a similar phenomenon can occur also in the jupiter - mass range . figure [ fig : et_mp23 ] shows that the interaction between a planet and a disk leads to orbital eccentricity growth for the @xmath5 ( dashed line ) and @xmath6 ( solid line ) cases . during the initial growth of @xmath141 for the @xmath6 planet , the rate is @xmath213 . this value is @xmath214 times that exhibited by the @xmath5 planet . the eccentricity growth stalls when @xmath215 for both planet masses . the planets may be experiencing some variation in their eccentricity forcing due to the phasing of their eccentricities relative to the disks. after @xmath216@xmath217 orbits from the release time , the orbital eccentricity starts to increase again with a growth timescale that is comparable to the initial growth timescale , @xmath218 orbits , for both planetary masses . the average growth timescale is shorter than the standard type ii migration ( or viscous diffusion ) timescale ( see [ sec : radial_migration ] ) . over the last @xmath95 orbits of the simulation , the eccentricity of the @xmath6 planet increases very slowly , at a rate @xmath219 . the model with @xmath137 and initial zero - eccentricity ( fig . [ fig : et_mp1vse ] , dashed line ) shows a much slower orbital eccentricity growth , reaching @xmath220 after @xmath8 orbits from the release time . as described in section [ sec : eccentricity_growth ] , the eccentric perturbation induced by the planet on the disk is also rather weak compared to that excited by the @xmath5 planet . at the average growth rate @xmath221 , it would take on the order of the viscous diffusion timescale to reach @xmath222 . in order to evaluate to the extent of orbital eccentricity growth , we used configurations with fixed non - zero planet eccentricities prior to release , @xmath77 . figure [ fig : et_mp1vse ] also shows the orbital eccentricity evolution of a @xmath137 planet with @xmath223 . after release , @xmath141 oscillates about the initial value . the oscillation grows in amplitude and , during one of these cycles , @xmath141 increases from @xmath3 to @xmath12 within about @xmath224 orbits . in this case , @xmath225 is of order the viscous diffusion timescale . we simulated several models with @xmath226 ( not plotted ) and found that there was generally a reduction of the orbital eccentricity , with some exceptions though . for example , in a model with a @xmath9 planet and @xmath227 , @xmath141 underwent small amplitude oscillations about the initial value , with periods of a few hundred orbits . this occurrence may be related to the relatively large eccentricity driven in the outer disk . some models with @xmath228 showed a rate of change of @xmath141 that diminishes in time . in these cases the evolution was generally monitored for less than @xmath95 orbits . longer time coverage simulations are required to assess the long - term behavior of these configurations . we describe here some results on the migration of eccentric orbit planets . we plan to explore this issue further in a future paper . radial migration of planets in the mass range considered in this study is expected to be in the standard type ii regime , which is characterized by an orbital decay timescale @xmath229 @xcite . the type ii migration rate depends only on the viscous timescale of the disk near the location of the planet and is independent of the disk density , provided the disk is locally more massive than the planet . type ii migration is based on the assumption that the gap , which separates the inner and outer disks , is devoid of material . in this case , the planet torques approximately balance the viscous torques at the gap edges . figure [ fig : at_mp123 ] plots the evolution of the semi - major axes , after the release time , of models with three planetary masses : @xmath137 ( short - dashed line ) , @xmath230 ( long - dashed line ) , and @xmath138 ( solid line ) . the figure shows that the initial migration rate depends on the planet s mass , which is inconsistent with the type ii prediction . we may expect some dependence of the migration rate on planet mass , because ratio of planet to disk mass is non - zero @xcite . in order to explore this result further , we have used a one - dimensional disk evolution code , along the lines of @xcite . we used the torque density per unit mass given in equation ( 4 ) of @xcite . we checked that the results are insensitive to the details of the torque density , provided that it is large enough to produce a gap . increasing the torque density everywhere by a factor of @xmath86 produced a small change in the migration rate ( less than @xmath231 ) . we adopted the same disk and planet parameters as in the two - dimensional simulations with zero planet eccentricity . in short , we find that the largest contributing factor to this non - type ii behavior is the lack of a substantial inner disk in the two - dimensional calculations . there is also some effect due to the non - zero planet - to - disk mass ratio . a comparison of orbital migration between one - dimensional and two - dimensional models is shown in figure [ fig : at_1dvs2d ] . in this comparison we used the azimuthal averaged surface density distributions in figure [ fig : avsigma ] as initial conditions for the one - dimensional models . as seen in figure [ fig : at_1dvs2d ] , the one - dimensional ( zero eccentricity ) migration rates , for a very low density inner disk , agree well with the two - dimensional rates at early times after release , while the planet eccentricity is small . for undepleted initial inner disks , we find that one - dimensional models have about the same migration rates for these two planet masses . it is possible that the two - dimensional simulations have an inner boundary @xmath232 that is too large to resolve the inner disk . more complete zone coverage of the inner region in the two - dimensional calculations might reveal an inner disk that acts to make the migration rate less dependent on mass , as indicated by the one - dimensional simulations . in spite of these possible limitations of our two - dimensional simulations , we describe below some interesting aspects of the migration of eccentric orbit planets in two - dimensional disks . as a planet s orbital eccentricity grows toward values of about @xmath233 , the rate of migration slows significantly ( see fig . [ fig : at_mp123 ] ) . over the last @xmath95 orbital periods of the calculated evolution , the @xmath6 planet exhibits a migration speed @xmath234 per orbit , with a tendency towards further reduction . this migration rate is about a factor of @xmath235 smaller than the rate at release time . the @xmath5 shows an even more drastic reduction of the migration rate that actually reverses and becomes positive around @xmath236 orbits . the outward migration speed is @xmath237 per orbit at the end of the simulation . the migration speed of the @xmath9 planet ( @xmath211 ) is much more constant over the course of the simulation as its orbital eccentricity remains small ( @xmath238 ) . the migration of a @xmath9 with @xmath223 proceeds as indicated by the short - dashed line in figure [ fig : at_mp123 ] . over the last @xmath239 orbits of evolution , however , the migration rate undergoes a decrease by a factor @xmath240 . during that time , the eccentricity grows to @xmath12 ( see solid line in fig . [ fig : et_mp1vse ] ) . the effect of non - zero orbital eccentricity on planet migration can also be seen in figure [ fig : at_mpvse ] , where the evolution of the semi - major axis is plotted for simulations with @xmath137 and @xmath138 and different initial orbital eccentricities . there is a clear trend towards slower migration rates for larger orbital eccentricities . in particular , when @xmath241 , the direction of migration is reversed . in all the calculations that show outward migration , the angular momentum of the eccentric orbit planet increases in time . we have conducted some preliminary investigations on the cause of this outward torque . one possibility is that it is due to the outer disk , as suggested in @xcite . when the planet eccentricity is large enough , the angular motion of the planet at apocenter can be slower than that of the inner parts of the outer disk , resulting in form of dynamical friction that increases the angular momentum of the planet . for the situation we wish to consider , it is not clear how the outer disk inner edge is maintained , when this model is applied . this disk material loses angular momentum from viscous torques and gains angular momentum from the planet , in the usual torque balance for a gap . the latter implies that the planet should lose , rather than gain , angular momentum . a preliminary analysis suggests that the outward torque may arise in the coorbital region . this region is supplied by material that flows from the outer disk across the gap , as discussed in section [ sec : planet_accretion ] . in any case , further analysis is required to understand this situation . we conducted a convergence test on the model with a @xmath9 planet and @xmath242 , which exhibits outward migration . the test involved a comparison of the migration rates obtained from the grid system gs2 ( see table [ tbl : grids ] ) to those obtained from a grid system whose linear resolution was a factor @xmath243 larger everywhere ( in both directions on each grid level ) . however , since computing resources were only available to run the higher resolution simulation for about @xmath244 orbits , we used the gauss perturbation equations ( e.g. , * ? ? ? * ) to compute @xmath245 resulting from the disk s gravitational forces , while keeping the planet s orbit fixed . the result of the test is that the migration rates , averaged from @xmath161 to @xmath244 orbits , differed by only @xmath246% at the two resolution levels . as a check on our use of the gauss equation , we also compared the migration rate determined from the gauss equation , averaged over the last @xmath247 orbits before release , to the initial @xmath245 after release , evaluated by integrating the equations of motion of the planet ( see the left panel of figure [ fig : at_mpvse ] ) . the two rates , both determined on grid system gs2 , differed by less than @xmath248 . as we discussed in section [ sec : viscosity_influence ] , the disk eccentricity decreases with viscosity . for a coupled disk - planet system , we similarly expect that the planet eccentricity would decrease with @xmath81 , since the eccentric corotation resonances become stronger . the orbital eccentricity evolution of @xmath6 planets in disks with different @xmath81 values is shown in the top - left panel of figure [ fig : eat_mp3vsnu ] . the model with standard viscosity ( solid line ) is the same as that in figure [ fig : et_mp23 ] . over a period of @xmath249 orbits , the orbital eccentricity of the model with @xmath165 ( short - dashed line ) remains small and never exceeds @xmath250 . on the other hand , the models with lowest viscosities , @xmath164 ( long - dashed line ) and @xmath251 exhibit a generally growing eccentricity . the trend towards faster growth for smaller viscosities is confirmed by the model with @xmath9 and @xmath164 , as indicated by the long - dashed line in the bottom - left panel of figure [ fig : eat_mp3vsnu ] . the orbital evolution of the semi - major axis of @xmath6 planets for the three disk viscosities is displayed in the top - right panel of figure [ fig : eat_mp3vsnu ] . the radial inward migration is faster for larger @xmath81 , as expected in a type ii - like regime . the plot supports the contention that planet eccentricity slows migration . around @xmath13 orbits after the release time , the two calculations with smallest viscosities ( @xmath79 and @xmath83 ) produce migration rates respectively equal to @xmath252 ( @xmath253 ) and @xmath254 ( @xmath255 ) per orbit . the first rate is a factor @xmath256 smaller , while the second a factor @xmath257 smaller , than the initial migration speed ( i.e. , when @xmath258 ) . for these cases , the average eccentricity growth rate at a time of about @xmath259 orbits is @xmath260 . the migration of a @xmath9 planet ( fig . [ fig : eat_mp3vsnu ] , bottom - right panel ) also shows that , as the orbital eccentricity approaches @xmath261 , @xmath262 starts to reduce . when @xmath164 , the average migration rate , over the last @xmath95 orbits , is about a factor @xmath132 smaller than it is during the first @xmath95 orbits of evolution after release . for eccentric orbit binary star systems , the accretion from a circumbinary disk onto the stars pulsates over the orbital period of the binary @xcite . this effect is related to the pulsating character of the equipotential surfaces of the elliptical restricted three - body problem @xcite . we measured the mass accretion rate ( which we denote as @xmath263 ) into the inner portion of the planet s roche lobe ( within @xmath123 ) , following the prescription described in section [ sec : accretion_procedure ] , as a function of the planet s orbit phase . we refer to this rate as the planet accretion rate although the flow is not resolved on the scale of the planet s radius and thus the rate at which the planet would accrete mass may be modulated somewhat differently from @xmath263 . quantity @xmath263 was determined by folding the mass accretion rate over planet orbital phase and averaging over @xmath264 orbital periods ( from @xmath264 to @xmath95 ) . figure [ fig : avmpdot ] shows the resulting averaged accretion rate , @xmath265 , for @xmath9 and @xmath6 planets , versus the true anomaly ( i.e. , the azimuthal position relative to pericenter ) of the planet and as a function of the orbital eccentricity . the simulations show pulsed accretion in cases of eccentric orbit planets . the amplitude of the variability , and to a lesser extent the phase , of @xmath263 depends on the orbital eccentricity . the accretion onto a @xmath9 planet with @xmath163 has two asymmetric peaks , the taller of which is around the apocenter position ( true anomaly equal to @xmath93 ) . the secondary peak is about @xmath266% of the primary peak and is located close to the pericenter position . for @xmath267 , the modulation of @xmath263 increases with @xmath141 . for larger orbital eccentricities , modulation decreases . this effect may be a consequence of the gap becoming broader and shallower with increasing @xmath141 ( see fig . [ fig : avsigma ] ) . a similar phase variability is found for the mass accretion onto a @xmath6 planet ( figure [ fig : avmpdot ] , right panel ) . even when the planet s orbit is circular , @xmath265 smoothly varies between @xmath268 and @xmath269 per orbit . the phasing in this case is arbitrary , since the planet s orbit is circular . this behavior is related to the eccentricity of the disk . the mass accretion is markedly peaked around the apocenter position when @xmath270 . the mass accretion modulation is again greatest for @xmath271 . when the planet eccentricity is between @xmath139 and @xmath4 , the highest accretion rate occurs roughly @xmath272 orbits after apocenter . this delay may be related to the time required by material to be captured once it has been perturbed near the apocenter . due to such a delay , accretion on binaries occurs near pericenter @xcite . the density distribution in the vicinity of an eccentric orbit planet varies strongly with its orbital phase . this variation is illustrated in the panels of figure [ fig : density_zoom ] , which depict the situation at the pericenter ( left ) and apocenter ( right ) , for a @xmath9 ( top ) and a @xmath6 planet ( bottom ) on an eccentric orbit with @xmath271 . the spiral waves have a regular pattern at pericenter , when the planet is orbiting in the low - density gap ( or cavity ) . as the planet approaches the apocenter , the outer spiral wake penetrates higher density regions , which causes fluid elements along the wake to lose angular momentum and flow through the gap . there are streams of material that extend inwards ( at @xmath273 and @xmath274 ) which appear in the right panels of figure [ fig : density_zoom ] . the mass accretion rate onto a planet with a fixed circular orbit decreases with increasing planet mass when @xmath275 @xcite . the average accretion rate in the simulations , at @xmath276 orbits , of a @xmath5 planet is @xmath277 times that of a @xmath9 planet . the ratio decreases to @xmath278 for a @xmath6 planet on a circular orbit and at @xmath279 orbits . these ratios agree within @xmath280 with the values given by @xcite , who used an independent code . as the disk eccentricity grows , the mass accreted during an orbit increases . at later times @xmath281 orbits , when @xmath282 , the accretion rate onto a @xmath5 planet is @xmath283 times the rate onto a @xmath9 planet and for a @xmath6 planet is @xmath284 times the rate onto a @xmath9 planet . notice that , for a @xmath6 planet , this implies a @xmath285% increase over the accretion rate at early stages , when the disk is circular . these results suggest that the eccentricity driven in the disk by a massive planet can augment the mass accretion rate onto the planet and hence shorten its growth timescale . mass accretion over an orbit period can also be enhanced by the planet s orbital eccentricity . for a @xmath9 planet on a fixed orbit , the mass growth timescale ( defined here as the ratio of @xmath42 to the average accretion rate between @xmath264 and @xmath95 orbits ) decreased by about @xmath286% when @xmath141 is increased from @xmath3 to @xmath287 . the reduction of mass growth timescale from @xmath17 to @xmath288 is only @xmath289% , perhaps as a result of the wider gap and its smoother outer edge at larger orbital eccentricities ( see fig . [ fig : avsigma ] ) . for a @xmath6 planet on a fixed orbit with @xmath139 , the mass growth timescale is reduced relative to the @xmath17 case by @xmath290% . while for @xmath288 there was a @xmath257% reduction . for cases with @xmath291 , the growth rate was not substantially different than the @xmath17 case . the mass growth timescale of a @xmath9 and a @xmath6 planet on circular orbit was also estimated for different values of the disk viscosity . for a @xmath6 planet an increase in @xmath81 by a factor of @xmath1 over the standard value ( see [ sec : physical_par ] ) reduced the growth timescale by about @xmath292% . for both planetary masses , a decrease in @xmath81 by a factor of @xmath293 below the standard value , lengthened the growth timescale by a factor @xmath86 . planetary accretion rates were determined by means of the grid system gs2 . simulations executed with grid systems gs1 and gs3 resulted in very similar outcomes ( within @xmath294% ) for both the modulation of @xmath263 and its average value . the results are not sensitive to the smoothing length , @xmath44 , although this parameter can affect the small - scale structure of the flow around the planet . we performed a calculation for an @xmath44 value that was reduced by a @xmath295% , with @xmath138 and @xmath139 , and obtained essentially the same accretion rates ( within @xmath0% ) . the region interior to a planet s orbit likely contains an inner disk which can not be resolved in the current two - dimensional simulations ( see discussion in [ sec : radial_migration ] ) . we estimate the modulation of mass onto this disk as a function of orbital phase by considering the mass flow rate across the inner boundary . as material accretes through the inner disk , the modulation would be expected to weaken . it is unclear whether the modulation would be reflected as a variable accretion at the surface of the star . perhaps it could be manifested as variability in emission from the region where the inflow meets the outer edge of the inner disk . figure [ fig : avmsdot ] plots the accretion rate at the inner boundary @xmath296 versus the true anomaly of the planet for cases with @xmath137 and @xmath6 . the case involving the more massive planet produces an accretion rate through the inner boundary that is largest when the planet is close to the apocenter . the maximum of @xmath296 occurs before the apocenter passage for the case involving the @xmath9 planet . the average mass accreted by the star during one orbital period of the planet is @xmath297 and @xmath298 for the @xmath9 and the @xmath6 cases , respectively . the same models with no orbital eccentricity show relatively constant @xmath296 values of @xmath299 ( @xmath137 ) and @xmath300 ( @xmath138 ) per orbit . as a comparison , the mass accretion through a steady - state @xmath81-disk is @xmath301 @xcite which , using the initial unperturbed surface density and standard viscosity ( @xmath79 ) , yields @xmath302 per orbit or @xmath303 . the ratio of the accretion on the star to the accretion in the disk , outside the gap , can be expressed as @xmath304 , where the subscript `` @xmath305 '' denotes is the integral of the respective accretion rate over a planetary orbit . for both the @xmath9 and the @xmath6 planets on circular orbit , this ratio is @xmath306 . when @xmath271 , the ratio is @xmath12 and @xmath307 for the @xmath9 and the @xmath6 planet , respectively . the reduced mass transfer across the planet s orbit , when @xmath137 and @xmath271 , can be attributed to the increased accretion rate onto the planet , as reported above . in the @xmath6 case , @xmath308 does not vary significantly as @xmath141 varies from @xmath3 to @xmath287 . instead , the mass flux across the gap is likely enhanced by the radial excursion of the planet ( see [ sec : planet_accretion ] ) . we simulated the orbital evolution of circular and eccentric orbit giant planets embedded in circumstellar disks . the disks were analyzed using a two - dimensional hydrodynamics code that utilizes nested grids to achieve high resolution in a large region ( @xmath309 ) around the planet . the disks were modeled as an @xmath81-disk and a few values of @xmath81 were considered . we investigated planet masses of @xmath9 , @xmath5 , and @xmath6 and initial orbital eccentricities that ranged from @xmath3 to @xmath4 . disk gaps become broader and shallower as the planet eccentricity increases ( see fig . [ fig : globalsigma ] and [ fig : avsigma ] ) . the density near the orbit of the planet is very small compared with the density in the disk for all eccentricities considered . a planet on a fixed circular orbit can cause an initially circular disk to become eccentric ( see fig . [ fig : rcmvstime ] ) . the disk eccentricity is suppressed at lower planet masses ( @xmath310 ) and higher disk viscosities ( @xmath311 ) , as also found by @xcite and by @xcite at higher planet masses . we attribute the eccentricity growth to a tidal instability associated with a series of eccentric outer lindblad resonances in the inner parts of the outer disk ( fig . [ fig : lam ] ) . the same type of instability , involving an inner disk , is thought to be responsible for the superhump phenomena in binary star systems @xcite . the simulations indicate that planet eccentricity can grow , as a consequence of disk - planet interactions ( fig . [ fig : et_mp23 ] and [ fig : et_mp1vse ] ) . the growth is stronger in the @xmath5 and @xmath312 cases than for @xmath9 , and for lower disk viscosity ( @xmath313 ) . planet eccentricities of @xmath314 were found in the simulations over the course of a few thousand orbits for @xmath315 and @xmath312 planets . a similar eccentricity growth is obtained for a @xmath316 planet in a disk with viscosity @xmath317 . the planet and disk both acquire eccentricity as they interact , which may lead to complicated time - dependent behavior of their eccentricities . the planet eccentricity growth is likely aided by the disk eccentricity growth . the results suggest that the eccentric growth found for @xmath318 planets by @xcite also occurs for lower planet masses . the higher resolution achieved by our calculations may be playing a role in obtaining this growth . for circular orbit planets , migration occurs on roughly the local viscous timescale , as expected for type ii migration . however , it is slowed for eccentric orbit planets . this result appears for several configurations with either dynamically determined ( fig . [ fig : at_mp123 ] and [ fig : eat_mp3vsnu ] ) or imposed planet eccentricities ( fig . [ fig : at_mpvse ] ) . for a @xmath5 case , even migration reversal ( outward migration ) is found for a dynamically determined eccentricity ( fig . [ fig : at_mp123 ] ) . migration slowing or reversal would have important consequences for the planet formation process . the cause is not yet clear . it may involve torques from outer disk @xcite or instead from the coorbital region . some preliminary evidence suggests the latter . mass accretion both within a planet s roche lobe and through a gap can be strongly modulated with orbital phase for eccentric orbit planets or eccentric disks ( fig . [ fig : avmpdot ] , [ fig : density_zoom ] , and [ fig : avmsdot ] ) . the modulation was largest for planet eccentricity @xmath319 . this pulsating accretion is similar to what is found for eccentric orbit binary stars embedded in a circumbinary disk @xcite , although the phasing is different . both disk and planet eccentricity also lead to enhanced accretion onto the planet . this enhancement likely helps planets achieve higher masses . the simulations lend support to the idea that disk - planet interactions cause planet eccentricity growth , along the lines of @xcite . the simulations suggest that planet eccentricities are easier to achieve for higher mass planets ( @xmath320 ) . our results are subject to the usual limitations in approximate initial conditions , simulation time , radial range for coverage of the disk ( likely resulting in the lack of an inner disk ) , the @xmath81-disk model , and the use of various numerical devices . we also neglected disk self - gravity , which may affect migration especially for higher mass disks @xcite . however , it is not clear that typical extra - solar planet eccentricities of @xmath287@xmath20 can be achieved through disk - planet interactions . the eccentricity growth at later times shows indications of slowing and possibly stalling for @xmath321 ( see fig [ fig : eat_mp3vsnu ] . ) perhaps higher eccentricities can be achieved for disks with different properties ( e.g. , lower viscosity and smaller disk s aspect ratio ) . eccentricity may be limited by damping due to high order eccentric inner lindblad resonances that lie outside a planet s orbit . simulations of eccentric orbit binary star systems suggest that little eccentricity growth occurs for @xmath322 @xcite . although the simulated planets do not achieve orbital eccentricities in excess of @xmath170 over the duration of the simulated evolution ( for configurations that start from circular orbits ) , the simulation times correspond to less than @xmath323 years . migration slowing and reversal may permit the planets to achieve higher eccentricities on longer timescales while avoiding orbit decay into the disk center / host star . we have not yet investigated eccentricity evolution of sub - jupiter mass planets . they may also provide important constraints . other simulations suggest that disks with standard viscosity have only mild gaps for smaller planet masses of @xmath324 ( e.g. , * ? ? ? * ; * ? ? ? * ) . under those conditions , disk - planet interactions likely lead to eccentricity damping , due to the dominance of the coorbital lindblad resonance @xcite . the observational determination of eccentricities for small mass planets would help constrain these models . the planet around hd 49674 is close to this regime . it has a minimum mass of @xmath325 and a best - fit eccentricity of @xmath326 ( p. butler , private communication ) . since it is close to the central star ( the period is @xmath327 days ) , it is possible that the eccentricity evolution could be more complicated , especially if it became trapped in a central disk hole . examples of isolated planets like this , but at longer periods would provide useful constraints . we thank gordon ogilvie and jim pringle for useful discussions . the computations reported in this paper were performed using the uk astrophysical fluids facility ( ukaff ) . gd was supported by the leverhulme trust through a ukaff fellowship , by the nasa postdoctoral program , and in part by nasa s outer planets research program through grant 811073.02.01.01.20 . sl acknowledges support from nasa origins of solar systems grant nng04gg50 g . mrb is grateful for the support of a philip leverhulme prize . , g. 2005 , methods of celestial mechanics . i : physical , mathematical , and numerical principles ( methods of celestial mechanics . i / gerhard beutler . in cooperation with leos mervart and andreas verdun . astronomy and astrophysics library . berlin : springer , isbn 3 - 540 - 40749 - 9 , 2005 , xvi , 464 pp . 99 figures , 11 in color , 32 tables and a cd - rom . ) , s. h. , & artymowicz , p. 1992 , in binaries as tracers of stellar formation . proceedings of a workshop held in bettmeralp , switzerland , sept . 1991 , in honor of dr . roger griffin . editors , antoine duquennoy , michel mayor ; publisher , cambridge university press , cambridge , england , new york , ny , 1992 . lc # qb821 .b55 1991 . isbn # 0521433584 . , 1992 , ed . a. duquennoy & m. mayor , 145

we perform a fully relativistic calculation of the reaction in the impulse approximation . we employ the gross equation to describe the deuteron ground state , and we use the said parametrization of the full nn scattering amplitude to describe the final state interactions ( fsis ) . we include both on - shell and positive - energy off - shell contributions in our fsi calculation . we show results for momentum distributions and angular distributions of the differential cross section , as well as for various asymmetries . we identify kinematic regions where various parts of the final state interactions are relevant , and discuss the theoretical uncertainties connected with calculations at high missing momenta . = 10000 = 10000

0805.3115

exclusive electron scattering from the deuteron target is very interesting by itself , and also as a very relevant stepping stone towards understanding exclusive electron scattering from heavier nuclei . the @xmath0 reaction at gev energies allows us - and requires us - to carefully study the reaction mechanism . it is necessary to consider final state interactions ( fsis ) between the two nucleons in the final state , two - body currents , and isobar contributions . of these , the fsis can be expected to be the most relevant part of the reaction mechanisms at the gev energy and momentum transfers relevant to the study of the transition from hadronic to quark - gluon degrees of freedom . for some recent reviews on this exciting topic , see e.g. @xcite . even though the deuteron is the simplest nucleus , and it has been the subject of considerable attention for a long time , there are several open questions : is it possible to experimentally determine the high - momentum components of the deuteron wave function even though the nuclear wave function or the momentum distribution are not observables ? will conventional nuclear physics break down or become too cumbersome at some point , and will a description involving quarks become necessary ? are there any six - quark admixtures in the deuteron wave function , and do they have an unambiguous experimental signal ? what influence does short range physics have on conventional wave functions , and how can this influence be removed to find an effective potential @xmath1 , provided one is interested in low energy scenarios only @xcite ? relativistic wave functions are available @xcite for the deuteron . and while the calculational effort is still considerable , it can be managed without resorting to super computers . the interest and importance of the @xmath0 process is reflected in the fact that recently , a deuteron benchmarking project has been started to investigate the differences between various calculations that are based on non - relativistic wave functions @xcite . anything that we learn about the reaction mechanism of the @xmath0 reaction has implications for heavier targets or experiments where the deuteron is used as a lab . some examples for the latter are the measurement of the neutron magnetic form factor by measuring a ratio of @xmath2 and @xmath3 cross sections . this allows for a significant reduction in the model dependence of the extracted form factor , but some theoretical input is still required @xcite . another example for using the nucleus as a lab is color transparency . while meson production from nuclei @xcite is the main thrust of color transparency investigations , color transparency in @xmath4 reactions is very interesting and topical , too @xcite . in order to study color transparency , one first needs to establish a firm understanding of all the conventional nuclear effects . several experiments on deuteron targets have been performed in the last few years , both at jefferson lab and at mit bates , and these data have either been published recently or are currently under analysis @xcite . there are also new proposals for d(e , ep ) experiments at jefferson lab @xcite . apart from these exciting new data , there are very interesting open questions posed by the data that have been available for some years now @xcite . regardless of the momentum transfers involved , there has been a discrepancy between data and calculations at low missing momentum . we discuss calculations in kinematics relevant to the new experiments as well as the low missing momentum puzzle in our results section . the experimental activity in @xmath4 reactions on the deuteron and other light nuclei has been matched by theoretical efforts . these calculations typically are performed using glauber theory , the generalized eikonal approximation @xcite , or a diagrammatic approach @xcite , although there are rare exceptions @xcite . even a second - order correction to the eikonal approximation has been suggested recently @xcite . many of these calculations focus on the differential cross section only , and use just the central part of the nn scattering amplitude . currently , almost all calculations for @xmath0 reactions are unfactorized @xcite , but factorized approaches are used for heavier targets @xcite . a common feature first introduced in glauber calculations is the assumption that the momentum transfer in the rescattering of the two nucleons is purely transverse . this has consequences both for the profile function , and the argument for which the nn scattering amplitude is evaluated . in this paper , we present a new calculation with several important features : we use a fully relativistic formalism , and describe the ground state with a solution of the gross equation@xcite ; we include all parts of the nucleon - nucleon scattering amplitude , including all the spin dependent parts , and use a realistic , modern parametrization ; the only approximation that we make is to neglect the negative energy states , as discussed at the end of the first section . this new calculation can be used at all energy and momentum transfers , provided that an appropriate full set of @xmath5 scattering data is available . this paper is organized as follows : first , we review the theoretical framework for our calculations , then , we show numerical results for the cross section and asymmetries . we conclude with a summary and outlook . the feynman diagrams representing the impulse approximation are shown in fig . [ impulse ] . figure [ impulse]a represents the plane wave contribution and fig . [ impulse]b represents the contribution from final state interactions . the plane wave contribution to the current matrix element is given by @xmath6 where the target deuteron has four - momentum @xmath7 and spin @xmath8 , the final proton has four - momentum @xmath9 and spin @xmath10 and the final neutron has four - momentum @xmath11 and spin @xmath12 . the single - nucleon propagator is @xmath13 and the current operator is chosen to be of the free dirac - plus - pauli form @xmath14 the deuteron vertex function with nucleon 2 on shell can be written as @xmath15 where @xmath16 , @xmath17 , @xmath18 is the charge - conjugation matrix and @xmath19 is the deuteron polarization four - vector . the invariant functions @xmath20 are given by @xmath21\\ g_2(p_2 ^ 2,p_2\cdot p)&=&\frac{2e_k - m_d}{\sqrt{8\pi}}\left [ \frac{m}{e_k+m}u(k)+\frac{m(2e_k+m)}{\sqrt{2}k^2}w(k)+\sqrt{\frac{3}{2}}\frac{m}{k}v_t(k)\right]\\ g_3(p_2 ^ 2,p_2\cdot p)&=&\sqrt{\frac{3}{16\pi}}\frac{m e_k}{k}v_t(k)\\ g_4(p_2 ^ 2,p_2\cdot p)&=&-\frac{m^2}{\sqrt{8\pi}m_d}\left[(2 e_k - m_d)\left(\frac{1}{e_k+m}u(k ) -\frac{e_k+2m}{\sqrt{2}k^2}w(k)\right)+\frac{\sqrt{3}m_d}{k}v_s(k)\right]\,,\nonumber\\\end{aligned}\ ] ] where @xmath22 is the magnitude of the neutron three - momentum in the deuteron rest frame and @xmath23 the functions @xmath24 , @xmath25 , @xmath26 and @xmath27 are the s - wave , d - wave , singlet p - wave and triple p - wave radial wave functions of the deuteron in momentum space . the radial wave functions are normalized in the absence of energy - dependent kernels such that @xmath28=1\,.\ ] ] for convenience , the spectator deuteron wave function can be defined as @xmath29 we choose to normalize this wave function such that in the deuteron rest frame @xmath30 which is correct only in the absence of energy - dependent kernels . the plane wave contribution to the current matrix element can then be written as @xmath31 the contribution from final state interactions represented by fig . [ impulse]b requires an integration for the loop four - momentum @xmath32 which involves both the deuteron vertex function and the @xmath5 scattering amplitude . an equivalent approach is to formulate the problem in terms of the spectator or gross equations @xcite where the equations for the scattering amplitude and vertex function are rewritten such that one particle is always taken to be on mass shell . this approach is manifestly covariant and has been successfully applied to elastic electron scattering from the deuteron @xcite . using this approach , the contribution of the final state interaction to the current matrix element is given by @xmath33 where @xmath34 is the @xmath5 scattering amplitude , @xmath35 is the positive energy projection operator and the dirac indices for the various components are shown explicitly . since the single - nucleon propagator can be decomposed as @xmath36\nonumber\\ & = & -\frac{m}{e_p}\left[\frac{\lambda^+(\bm{p})}{p^0-e_p+i\epsilon } -\frac{\lambda^-(-\bm{p})}{p^0+e_p - i\epsilon}\right ] \label{g0ofp}\end{aligned}\ ] ] and @xmath37 ( [ j_fsi ] ) can be written as @xmath38 where @xmath39 @xmath40 and @xmath41 these contributions are represented by the diagrams in fig . [ jabc ] . , @xmath42 and @xmath43 . the insertion of a cross on an internal propagator indicates that the particle has been placed on the positive - energy mass shell . the insertion @xmath44 indicates that the positive - energy principal value part of the propagator is used and the insertion of @xmath45 indicates the negative - energy part of the propagator is used . , title="fig:",height=240 ] , @xmath42 and @xmath43 . the insertion of a cross on an internal propagator indicates that the particle has been placed on the positive - energy mass shell . the insertion @xmath44 indicates that the positive - energy principal value part of the propagator is used and the insertion of @xmath45 indicates the negative - energy part of the propagator is used . , title="fig:",height=240 ] , @xmath42 and @xmath43 . the insertion of a cross on an internal propagator indicates that the particle has been placed on the positive - energy mass shell . the insertion @xmath44 indicates that the positive - energy principal value part of the propagator is used and the insertion of @xmath45 indicates the negative - energy part of the propagator is used . , title="fig:",height=240 ] equation ( [ j_fsia ] ) , represented by fig . [ jabc]a , has all four legs of the @xmath5 scattering amplitude on mass shell . for this case , the scattering amplitude can be parameterized in terms of five fermi invariants as @xmath46 where @xmath47 and @xmath48 are the usual mandelstam variables . the calculation of the invariant functions from helicity amplitudes is described in appendix [ appnn ] . we use the helicity amplitudes available from said as input for our calculations @xcite . for this calculation we have constructed a table of the invariant functions in terms of @xmath47 and the center of momentum angle @xmath49 . the table is then interpolated to obtain the invariant functions at the values required by the integration . an alternative two - dimensional representation of the scattering amplitudes is in terms of the saclay amplitudes . in this case , the scattering amplitude as an operator in two - dimensional spinor space is given by @xmath50 \label{saclaynndef}\end{aligned}\ ] ] where @xmath51 for @xmath52 and @xmath53 as the initial and final momenta of the proton . the first term is the central contribution , the next three terms are double - spin - flip terms and the final term is a single - spin - flip term . we can determine the sensitivity of the @xmath4 observables to these terms by determining the saclay amplitudes @xmath54 , @xmath55 , @xmath56 , @xmath57 and @xmath58 from the helicity matrix elements as described in the appendix , setting some of the amplitudes to zero and then transforming the result to give new invariant amplitudes for the fermi form . a common approximation is to use only the central part of the amplitude generated from a prescription involving the total cross section . the contribution to the current matrix element given by ( [ j_fsib ] ) , represented by fig . [ jabc]b , involves a principal value integral over off - mass - shell momenta for one leg of the scattering amplitude . the proton propagator for this leg contains only the positive energy contribution . determination of the off shell behavior of the scattering amplitude requires a dynamical model of the amplitude . such a model is not currently available to us in the range of energies required for the experiments being performed at jefferson lab . in order to estimate the possible effects of this contribution to the current matrix elements , we use a simple prescription for the off - shell behavior of the amplitude . although additional invariants are possible when the nucleon is allowed to go off shell , we keep only the forms in ( [ fermi ] ) . the center - of - momentum angle is calculated using @xmath59 the invariants are then replaced by @xmath60 where @xmath61 and the @xmath62 are obtained from interpolation of the on - shell invariant functions with the center - of - momentum angle obtained from ( [ thetacm ] ) . the form factor ( [ nff ] ) was used as a cutoff in calculating the gross vertex function used in this paper with @xmath63 gev . however , there should be an intrinsic fall off of the scattering amplitude due to the dynamics of the scattering which would be expected to be faster than provided by this cutoff mass . figure [ offshell ] shows the off - shell form factor for cutoff masses of @xmath64 1.0 , 1.1 and 1.2 gev . in the absence of a dynamical model of the scattering amplitudes , the effect of possible off - shell contributions on various observables can be reasonably estimated by using cutoff masses in this range . 1.0 , 1.1 and 1.2 gev . , height=240 ] the contribution to the current matrix elements from ( [ j_fsic ] ) , represented by fig . [ jabc]c , contains the effect of negative energy propagation of the off - shell leg of the scattering amplitude . since the denominator of this part of the propagator will be large compared with that of the positive energy part at large momentum transfers , we neglect this contribution for the present . this is the only approximation involved in our calculation . note , however , that by dropping this contribution the current matrix elements are no longer covariant . we have chosen to calculate the matrix elements in the laboratory frame . the @xmath2 cross section for unpolarized deuterons and protons in the lab frame can be written as @xcite @xmath65 \ , , \label{xsdef}\end{aligned}\ ] ] where @xmath66 , @xmath67 and @xmath68 are the masses of the deuteron , proton and neutron , @xmath69 and @xmath70 are the momentum and solid angle of the ejected proton , @xmath71 is the energy of the detected electron and @xmath72 is its solid angle . the helicity of the electron is denoted by @xmath73 . the mott cross section is @xmath74 and the recoil factor is given by @xmath75 the kinematic coefficients @xmath76 are @xmath77 if the response tensor is defined as @xmath78 the response functions @xmath79 are defined by @xmath80 where @xmath81 and @xmath82 with @xmath83 for our calculations , we have chosen the following kinematic conditions : the z - axis is parallel to @xmath84 , and the missing momentum is defined as @xmath85 . the representation of the cross section in terms of response functions is due to the mixed polarization of the virtual photon which varies with the electron kinematics and polarization . the transverse - transverse response function is the result of interference between the @xmath86 helicity states while the longitudinal - transverse response functions @xmath87 and @xmath88 are the result of interference between the deuteron charge and two linear combinations of the @xmath86 helicity states . as an alternative to a complete separation of the cross section into response functions , the interference response functions can be accessed through linear combinations of differential cross sections to produce three interference asymmetries defined as @xmath89 @xmath90 and @xmath91 where , for conciseness , @xmath92 note that while @xmath93 can be obtained by measuring protons in the electron scattering plane symmetrically about the direction of the three - momentum transfer , the asymmetries @xmath94 and @xmath95 require measurements to be made out of the scattering plane . the asymmetry @xmath95 is defined as an electron single spin asymmetry and can , therefore , be easily obtained by flipping the beam helicity . while @xmath96 and @xmath97 are independent of photon - helicity - dependent phases , the interference response functions are not . as a result , the interference response functions can be very sensitive to phase differences generated by non - nucleonic currents and final state interactions . this is particularly true of @xmath88 which can be shown to be zero in the pwia . the interference response function , @xmath87 , is very sensitive to the relativity included in the current operator , due to the various interference contributions from the charge and transverse current operators @xcite . the observable @xmath95 has recently been measured in jefferson lab s hall b @xcite . due to the large solid angle coverage in hall b , and the averaging over @xmath98 , the transverse - transverse interference response that is multiplied with a factor of @xmath99 in the cross section , drops out of the measured asymmetry @xmath95 @xcite . therefore , from now on in this paper , we calculate @xmath100 the difference between the asymmetry calculated with and without @xmath101 is very small in practice , due to the small size of the transverse - transverse response . in this section , we discuss our numerical results for several observables . we will investigate the effect of the final state interactions ( fsis ) , and in particular , we will point out the contributions of spin - dependent fsis , both the single - spin - flip contributions and the double - spin - flip contributions . we will also discuss the relative importance of on - shell and off - shell contributions to fsi . the most natural observable to investigate is the differential cross section . in fig . [ figpaul ] our calculations are compared to the published @xmath2 data from jefferson lab s hall a. the data from ulmer et al . @xcite are shown together with our pwia , on - shell fsi , and full fsi curves as a function of the missing momentum @xmath102 . for all of the calculations in the paper we use the mmd @xcite proton electromagnetic form factors . the data are presented as a reduced cross section , which is defined as @xmath103 gev , @xmath104 , @xmath105 , and @xmath106 . the data are from . , title="fig : " ] gev , @xmath104 , @xmath105 , and @xmath106 . the data are from . , title="fig : " ] at lower missing momenta , the effects of fsis are small . for missing momenta larger than @xmath107 mev , the fsi effects become visible . first , the pwia curve is slightly above the fsi results , then , at @xmath108 mev , the fsi curves become larger than the pwia contribution . the agreement with the data is quite nice overall . the off - shell fsi is small , and leads to final results a little below the data at larger missing momenta . this is a sensible result , as we do expect meson exchange currents ( mecs ) to play a role at the relatively low @xmath109 at which these data were taken . indeed , in @xcite , the calculation by arenhoevel @xcite agreed with the data at large missing momenta after mec contributions were included ; the fsi - only calculation was a bit below the data . we also have added a panel with a linear plot of just the low missing momentum data . the full fsi curve and the on - shell fsi curve coincide in this region . it has been observed in several previous @xmath2 measurements that at very low missing momenta , the calculations are somewhat above the data . this is quite puzzling as at these low missing momenta , effects like fsis , mecs etc are supposed to be small and well under control . for a nice compilation on this topic , see @xcite . in @xcite , 1 shows the deviation of the reduced cross section data and the calculations . here , we observe the same type of deviation at very low @xmath110 . the largest discrepancy appears at @xmath111 , where our calculation overpredicts the data by @xmath112 . overall , comparing with the previous results , our low missing momentum results seem to be an improvement , even though the discrepancy has not been fully removed . one main difference between the calculation presented in this article and the calculations in@xcite is the fully relativistic approach we take here . , @xmath113 , @xmath114 , and @xmath106 is shown in pwia ( dotted curve ) , with on - shell fsi ( solid curve ) , and with on - shell and off - shell fsis ( dashed curve ) , as a function of the missing momentum . bottom panel : the ratio of the off - shell calculations with varying cut - off @xmath115 1 gev ( short - dashed ) , 1.1 gev ( dash - dotted ) , and 1.2 gev ( long - dashed ) to the on - shell fsi calculation , in the same kinematics as the top panel.,title="fig:",height=384 ] , @xmath113 , @xmath114 , and @xmath106 is shown in pwia ( dotted curve ) , with on - shell fsi ( solid curve ) , and with on - shell and off - shell fsis ( dashed curve ) , as a function of the missing momentum . bottom panel : the ratio of the off - shell calculations with varying cut - off @xmath115 1 gev ( short - dashed ) , 1.1 gev ( dash - dotted ) , and 1.2 gev ( long - dashed ) to the on - shell fsi calculation , in the same kinematics as the top panel.,title="fig:",height=384 ] in figure [ figcsmomdis ] , the upper panel shows the cross section as a function of the missing momentum , with a beam energy of @xmath116 gev , @xmath113 , @xmath114 , and @xmath106 . the choice of bjorken - x implies roughly quasi - free kinematics . the azimuthal angle of the detected proton has been chosen to maximize the cross section . the pwia and on - shell fsi curves are almost identical at very low missing momenta , up to @xmath117 gev . then , the fsi curve reduces the cross section compared to the pwia result , roughly from @xmath118 gev to @xmath119 gev . for larger missing momenta , there is a marked increase in the differential cross section when fsi is included . these results are quite typical and have been seen in other calculations @xcite . the differential cross section decreases by several orders of magnitude with increasing missing momentum , and a small reduction due to fsi at lower missing momenta and larger differential cross section can lead to a very large increase at larger missing momenta and smaller differential cross sections . the inclusion of the off - shell fsi contributions leads to a slight reduction of the cross section for medium missing momenta . the lower panel shows the ratios of the off - shell fsi calculations with @xmath64 1.0 , 1.1 and 1.2 gev to the on - shell fsi . the off - shell effects are small for small @xmath102 , but become increasingly large as @xmath102 increases . these effects are not particularly sensitive to the cutoffs chosen here , but must be quite sensitive to lower cutoff masses . , @xmath120 , @xmath114 , and @xmath106 is shown calculated with on - shell fsi , as a function of the missing momentum . the solid line shows the result calculated with the full @xmath121 scattering amplitude , the dashed line shows the result without the double - spin - flip terms of the @xmath121 scattering amplitudes , and the dotted line shows the result with the central @xmath121 amplitude only . ] in fig . [ figcsmdspin ] , we investigate the role played by the various parts of the proton - neutron scattering amplitude contributing to ( [ fermi ] ) . the dashed line shows the results without the three double - spin - flip terms in the amplitude . one can see that for missing momenta from @xmath122 gev to @xmath123 gev , the double - spin - flip contribution is quite relevant . its omission in this region leads to a significantly smaller cross section . the single - spin - flip - or spin - orbit - part of the proton - neutron scattering amplitude becomes relevant only at higher missing momenta than the double - spin - flip terms . from roughly @xmath124 gev on , omitting the spin - orbit contribution leads to a decrease in the differential cross section . this clearly shows that while it is possible to parameterize the central part of the @xmath121 scattering amplitude , and to reproduce the @xmath121 cross section data this way , this type of parametrization effectively includes some physics that stems from the spin - dependent parts of the @xmath121 amplitude . here , it is interesting to see the influence of the spin - dependent parts of the @xmath121 amplitude on the unpolarized cross section . while the logarithmic scale necessary for the momentum distribution conceals the effects , it is important to note that the relative importance of the spin - dependent fsi contributions changes with missing momentum . we will return to the effects of spin - dependent fsi with fig . [ figcsadspin ] . gev , @xmath113 , @xmath125 , and @xmath106 is shown in pwia ( dotted curve ) , with on - shell fsi ( solid curve ) , and with on - shell and off - shell fsis ( dashed curve ) , as a function of the missing momentum . bottom panel : the ratio of the off - shell calculations with varying cut - off @xmath115 1 gev ( short - dashed ) , 1.1 gev ( dash - dotted ) , and 1.2 gev ( long - dashed ) to the on - shell fsi calculation , in the same kinematics as the top panel.,title="fig:",height=384 ] gev , @xmath113 , @xmath125 , and @xmath106 is shown in pwia ( dotted curve ) , with on - shell fsi ( solid curve ) , and with on - shell and off - shell fsis ( dashed curve ) , as a function of the missing momentum . bottom panel : the ratio of the off - shell calculations with varying cut - off @xmath115 1 gev ( short - dashed ) , 1.1 gev ( dash - dotted ) , and 1.2 gev ( long - dashed ) to the on - shell fsi calculation , in the same kinematics as the top panel.,title="fig:",height=384 ] in fig . [ figcsmdx1p3 ] , we display our results for the same four - momentum transfer , but higher @xmath125 . at this value of @xmath126 , and larger values , strong short - range @xmath5 correlations have been reported by inclusive electron scattering experiments on deuterium @xcite . the deviation of the on - shell fsi from the pwia is small for medium missing momenta , and seems to disappear altogether for missing momenta between 0.5 gev and 0.6 gev . however , the off - shell contribution to the fsi gains in relevance for larger missing momenta , and leads to a significant increase over the pwia results at @xmath127 gev . the lower panel again shows the ratio of off - shell fsi to the on - shell result for three values of the off - shell cutoff . the importance of the off - shell fsi here is larger than for @xmath114 ( as discussed above when considering fig . [ figcsmomdis ] and the sensitivity to the value of the cutoff is much greater . this is the expected behavior , as the deviation from @xmath114 corresponds to a deviation from the quasi - elastic kinematics , and stresses the off - shell region more . a recent new proposal @xcite suggests a measurement of the cross section at somewhat larger @xmath128 , but the same value of @xmath125 and a beam energy of @xmath129 gev . the results of our calculation for these kinematics are similar to what is displayed above in fig . [ figcsmdx1p3 ] . the on - shell fsi in this case deviates a bit more from the pwia result than for the kinematics displayed here . the off - shell contribution is just as significant in the proposed kinematics . gev , @xmath113 , @xmath130 gev , and @xmath106 is shown in pwia ( dotted curve ) , with on - shell fsi ( solid curve ) fsi ( dashed curve ) , as a function of the angle of the missing momentum . the top axis shows the corresponding values of bjorken - x . ] in figure [ figcsangdis ] , we show the cross section as a function of the angle @xmath131 of the missing momentum , with fixed @xmath113 and @xmath132 gev . the beam energy and azimuthal angle of the proton are the same as for the momentum distribution graphs . the angular distribution of the cross section shows much less variation in the magnitude , and therefore can be shown on a linear plot , allowing for a better look at the relevance of various parts of the cross section . while the region beyond @xmath133 is kinematically accessible to experiment , a calculation in this region requires the knowledge of the @xmath134 scattering amplitude above lab kinetic energies of @xmath135 gev in the @xmath5 system , which are not available from said . we therefore stay below @xmath136 . as is obvious from the plot , the most interesting features of the calculation are located below this angle : while the pwia results are gently sloping upwards and then downwards for angles larger than @xmath137 , the fsi results initially follow this behavior , but then show a pronounced peak at around @xmath138 . this value corresponds to @xmath114 , i.e. quasi - free kinematics for the knocked - out nucleon . for the lower angles , the fsi simply leads to a reduction in the cross section , but the shape is unchanged . for larger angles , around @xmath114 , the diffractive nature of the fsi leads to a redistribution of strength from smaller missing momenta , causing a large peak . if we consider the ratio of fsi to pwia cross section , as is sometimes done when comparing various methods of calculation @xcite , this ratio peaks at @xmath139 , too . our calculation clearly shows the same shift from a peak at @xmath140 , as seen in glauber theory calculations @xcite , to a lower angle , as seen in the generalized eikonal approximation ( gea ) @xcite and the diagrammatic approach of laget @xcite . gev , @xmath120 , @xmath141 gev , and @xmath106 is shown with on - shell fsi ( solid curve ) , and with on - shell and off - shell fsis at various cut - offs , as a function of the angle of the missing momentum . ] for the angular distribution , the influence of the off - shell fsi cut - off is clearly visible and can be studied easily . one does not expect a large contribution from far off - shell nucleons . the cut - off we use here serves to impose that constraint . for the kinematics displayed in fig . [ figcsadoffshell ] , we investigate the effects of various cut - off values . the cut - off at @xmath142 gev leads to a very small increase at small angles and a very small decrease at large angles , but the overall result is hardly different from the on - shell fsi only result . the primary effect for larger cutoff masses is to fill in the minimum in the on - shell result from @xmath143 to @xmath144 . gev , @xmath113 , @xmath130 gev , and @xmath106 is shown calculated with on - shell fsi , as a function of the angle of the missing momentum . the solid line shows the result calculated with the full @xmath121 scattering amplitude , the dashed line shows the result without the double - spin - flip terms of the @xmath121 scattering amplitudes , and the dotted line shows the result with the central @xmath121 amplitude only . ] in fig . [ figcsadspin ] , we show the effects of the various spin - dependent parts of the @xmath145 scattering amplitude . it is very interesting to observe that in the peak region , the contribution of the spin - dependent fsis ( both single and double - spin - flip ) amounts to about one quarter of the cross section . this is certainly a rather significant contribution . the contribution from the single - spin - flip term and the three double - spin - flip terms is about equal in the peak region . the figure shows that the double - spin - flip contribution to the cross section at small angles is almost negligible . it becomes noticeable at @xmath146 , and then leads to a sizable increase of the differential cross section in the peak region . the omission of the single - spin - flip contribution leads to a noticeable reduction in the cross section for all angles . the effect is most pronounced in the peak region and in the very shallow dip just before the peak region . gev , @xmath113 , @xmath130 gev , and @xmath106 is shown with on - shell fsi as a function of the angle of the missing momentum . the solid curve shows the contribution of the full @xmath121 scattering amplitude , the other curves show the results for just one invariant term in the @xmath121 scattering amplitude . ] to this point , we have considered the different contributions of the spin - dependent parts of the amplitude , i.e. of the @xmath121 amplitude split up following the saclay convention ( [ saclaynndef ] ) . using the saclay formalism with its classification according to the spin - dependence is quite useful , as it allows one to understand the new contributions from different parts of the current operator when adding the single - spin - flip term and the double - spin - flip terms , see @xcite . even though we did not rewrite our current operator to distinguish e.g. between magnetization current and convection current , seeing the @xmath121 scattering amplitude in terms of its spin - dependence is a very natural view point , and allows for a certain intuitive understanding of the numerical results . it is also interesting to investigate the @xmath121 amplitude in terms of the five fermi invariants , which are so practical for actual calculations . > from ( [ hetoinv],[hetosa ] ) , it is clear that every invariant amplitude contains several different , spin - dependent pieces . in fig . [ figcsadinv1 ] , we show the results of the calculation with on - shell fsi if only one of the five invariants is included . for comparison , the solid lines depicts the result obtained with the full @xmath121 amplitude . the result obtained with just the pseudoscalar part is slightly above the full result for smaller angles , and continues smooth and almost straight towards larger angles . it does not exhibit a peak structure at large angles . the tensor and axial vector contributions are fairly close to the pseudoscalar contribution at large angles , and neither exhibits a peak at large angles . at small angles , however , these two contributions lead to a new peak , much larger than the original peak at large angles in the full result . the scalar and vector contributions show peaks at small angles , and a large peak at large angles . these results already show that there are large interference effects present between the various invariant amplitudes . there is no straightforward and intuitive explanation available for why these contributions interfere in such a way . gev , @xmath113 , @xmath147 gev , and @xmath106 is shown in pwia ( dotted curve ) , with on - shell fsi ( solid curve ) , and with on - shell and off - shell fsis ( dashed curve ) , as a function of the angle of the missing momentum . ] in fig . [ figcsadlowpm ] , we show the angular distribution of the differential cross section for a lower missing momentum , @xmath148gev . the lower missing momentum implies that the limiting value of @xmath135 gev lab energy for the @xmath121 system is reached at larger angles than for @xmath132 gev . while the pwia curve here is very similar in shape to the curve at the higher missing momentum value , the fsi curve looks rather different now : instead of a fairly sharp peak around @xmath138 , we now observe a broad , shallow dip at larger angles . at lower missing momenta , the fsis lead to a reduction in the cross section . part of this strength is redistributed to larger missing momenta , as discussed above . also including the off - shell fsis has no effect at very small angles , roughly below @xmath149 , and then tends to shift the overall result towards lower angles . gev , @xmath113 , @xmath147 gev , and @xmath106 is shown calculated with on - shell fsi , as a function of the angle of the missing momentum . the solid line shows the result calculated with the full @xmath121 scattering amplitude , the dashed line shows the result without the double - spin - flip terms of the @xmath121 scattering amplitudes , and the dotted line shows the result with the central @xmath121 amplitude only . ] in fig . [ figcsadlowpmspin ] , the influence of the different kinds of fsis is shown at low missing momentum . switching off the double - spin - flip contribution leads to a small reduction in the cross section for medium and large angles , roughly @xmath150 in the region of the shallow dip . switching off the single - spin - flip term , too , changes practically nothing . the fsis in this kinematic region are overall smaller than for higher missing momenta . the influence of spin - dependent fsis is smaller here , too . however , it is interesting to note that the double - spin - flip terms are actually more relevant here than the single - spin - flip terms . we have observed this already when discussing the momentum distributions shown in fig . [ figcsmdspin ] . figure [ figrtt ] shows the interference asymmetry @xmath94 for @xmath151 gev and @xmath152 gev . the upper panel is for @xmath153 and the lower for @xmath154 . is shown for @xmath114 in pwia ( dotted line ) , with on - shell fsi ( solid line ) , and with off - shell fsi for various values of the cutoff - @xmath155 gev ( short - dashed line ) , @xmath156 gev ( dash - dotted line ) , and @xmath157 gev ( long - dashed line ) . bottom panel : the asymmetry @xmath94 is shown for @xmath125 . the curves have the same meaning as in the top panel.,title="fig:",height=384 ] is shown for @xmath114 in pwia ( dotted line ) , with on - shell fsi ( solid line ) , and with off - shell fsi for various values of the cutoff - @xmath155 gev ( short - dashed line ) , @xmath156 gev ( dash - dotted line ) , and @xmath157 gev ( long - dashed line ) . bottom panel : the asymmetry @xmath94 is shown for @xmath125 . the curves have the same meaning as in the top panel.,title="fig:",height=384 ] in both cases , the fsi result in a substantial change from the pwia result , both with respect to size and shape of the asymmetry . for @xmath153 , the variation with the cutoff mass for the three values shown here is small , but it is somewhat larger for @xmath154 reducing the asymmetry almost to zero for @xmath157 gev around @xmath158 gev . is shown for @xmath114 in pwia ( dotted line ) , with on - shell fsi ( solid line ) , and with off - shell fsi for various values of the cutoff - @xmath155 gev ( short - dashed line ) , @xmath156 gev ( dash - dotted line ) , and @xmath157 gev ( long - dashed line ) . bottom panel : the asymmetry @xmath93 is shown for @xmath125 . the curves have the same meaning as in the top panel . , title="fig:",height=384 ] is shown for @xmath114 in pwia ( dotted line ) , with on - shell fsi ( solid line ) , and with off - shell fsi for various values of the cutoff - @xmath155 gev ( short - dashed line ) , @xmath156 gev ( dash - dotted line ) , and @xmath157 gev ( long - dashed line ) . bottom panel : the asymmetry @xmath93 is shown for @xmath125 . the curves have the same meaning as in the top panel . , title="fig:",height=384 ] figure [ figrlt ] shows results for the interference asymmetry @xmath93 for the same kinematics as in the previous figure . again the sensitivity to final state interactions is substantial . sensitivity to off - shell contributions is relatively modest at @xmath153 but is much larger at @xmath154 . the single - spin asymmetry @xmath95 is shown in fig . [ figrltp ] for the same kinematics . this asymmetry is identically zero in the pwia , so the fsi are responsible for any asymmetry . a more detailed discussion of fsi effects is given below for kinematics relevant to a recent experiment . here , in these kinematics , we focus on comparing the behavior of the three asymmetries , in the same kinematics . at @xmath153 there is very little sensitivity to off - shell contributions , but it is large at @xmath154 . as stated above , this behavior is expected as @xmath159 corresponds to quasi - free , i.e. on - shell , kinematics , while @xmath125 probes nucleons that are much more off - shell . from the plots for @xmath125 , one can see clearly that the off - shell contribution to the nn scattering amplitude introduces a certain amount of ambiguity , especially at medium to high missing momenta . while great progress has been made on the experimental side , with measurements at very high missing momenta , there clearly are some theoretical uncertainties in these kinematic regions . is shown for @xmath114 in pwia ( dotted line ) , with on - shell fsi ( solid line ) , and with off - shell fsi for various values of the cutoff - @xmath155 gev ( short - dashed line ) , @xmath156 gev ( dash - dotted line ) , and @xmath157 gev ( long - dashed line ) . bottom panel : the asymmetry @xmath95 is shown for @xmath125 . the curves have the same meaning as in the top panel . , title="fig:",height=384 ] is shown for @xmath114 in pwia ( dotted line ) , with on - shell fsi ( solid line ) , and with off - shell fsi for various values of the cutoff - @xmath155 gev ( short - dashed line ) , @xmath156 gev ( dash - dotted line ) , and @xmath157 gev ( long - dashed line ) . bottom panel : the asymmetry @xmath95 is shown for @xmath125 . the curves have the same meaning as in the top panel . , title="fig:",height=384 ] asymmetry for a beam energy of @xmath160 gev and four - momentum transfer @xmath161 is shown calculated with on - shell fsi , as a function of the missing momentum . the solid line shows the result calculated with the full @xmath121 scattering amplitude , the dashed line shows the result without the double - spin - flip terms of the @xmath121 scattering amplitudes , and the dotted line shows the result with the central @xmath121 amplitude only . ] we are in the fortunate situation that the asymmetry @xmath95 has been measured over a wide range of kinematics , from very low four - momentum transfers @xmath162 up to medium values of @xmath163 @xcite . the data are currently under analysis . in this range , the proton - neutron scattering amplitudes from said are available , so there are no limits to our ability to calculate for these kinematics . here , we discuss our results for two representative kinematics : @xmath164 , and @xmath165 . in both cases , we assume a beam energy of @xmath160 gev . fig . [ figatlpspin ] shows our results for @xmath166 . the pwia result is zero , and not shown on the plot . the full , on - shell fsi result starts out negatively , dips around @xmath167 gev , and then increases and changes sign around @xmath168 gev . then , the asymmetry peaks around @xmath169 gev , and then decreases and changes sign again . the second zero of the asymmetry occurs at @xmath170 gev . one can see that the double - spin - flip contributions to the fsi are not very relevant : they just lead to small modifications in the dip and peak regions . the spin - orbit ( i.e. the single - spin - flip ) contribution is very important , though . switching it off so that only the central fsi remains leads to a completely different picture : the asymmetry is tiny , and remains positive for large missing momenta . asymmetry for a beam energy of @xmath160 gev and four - momentum transfer @xmath171 is shown calculated with on - shell fsi , as a function of the missing momentum . the solid line shows the result calculated with the full @xmath121 scattering amplitude , the dashed line shows the result without the double - spin - flip terms of the @xmath121 scattering amplitudes , and the dotted line shows the result with the central @xmath121 amplitude only . ] in fig . [ figatlploqspin ] , we show the corresponding results for the spin - dependence of the fsis for lower four - momentum transfer , @xmath171 . at these kinematics , we do expect the influence of meson exchange currents and isobar states to be relevant . these effects are not included in the present calculation . however , the fsis are crucial for @xmath95 , and we can investigate them within our model . the full calculations are qualitatively very similar at both @xmath109 values , showing a negative dip around @xmath167 gev and then an increase into positive values , with a peak around @xmath172 gev . a very interesting difference , however , is the size of the contribution of the double - spin - flip terms to the fsi . while their influence is small , almost negligible , at the higher @xmath109 value , it is quite significant for the low @xmath109 value : the double - spin - flip terms serve to partially fill in the negative dip , and are also responsible for pushing the asymmetry back towards positive values . one obvious difference between the calculation presented here and the traditional glauber and generalized eikonal approximation ( gea ) is the evaluation of the argument of the nucleon - nucleon scattering amplitude for the fsis . as described in section [ sectheo ] , we evaluate the five terms of the nn scattering amplitude([eqdefnn ] ) at the values of the mandelstam variables @xmath47 and @xmath48 computed from the particular kinematics . in glauber and gea settings , one typically finds expressions where the nn scattering amplitude is evaluated assuming a purely transverse momentum transfer when evaluating @xmath48 , even though the longitudinal momentum transfer is taken into account in the gea . this is typically denoted with expressions like @xmath173 . with the kinematics variables as defined in fig . [ impulse ] , the mandelstam @xmath48 is given by @xmath174 , whereas assuming a purely transverse momentum transfer implies : @xmath175 . using this we can define the cm scattering angle as @xmath176 in fig . [ fignnratiocsmd ] , we show the ratio of the transverse - momentum approximation to the full , on - shell calculation . the kinematics are identical to the kinematics used for fig . [ figcsmomdis ] : a beam energy of @xmath116 gev , @xmath120 , @xmath114 , and @xmath106 . up to missing momenta of @xmath177 gev , the two approximation works well , leading to small deviations of less than @xmath150 . beyond @xmath177 gev , the deviation from the full result grows , and for missing momenta larger than @xmath178 gev , the quality of the approximation deteriorates quickly . we have performed the same calculation for the angular distribution shown in fig . [ figcsangdis ] . here , we found that the approximation does well - the deviations are less than @xmath150 for any angle . this corresponds to our findings for the momentum distribution . as an illustrative example for the other observables discussed in this article , we also show our results for the asymmetry @xmath95 in fig . [ fignnratioaltp ] . the kinematics correspond to fig . [ figatlpspin ] , we have used a beam energy of @xmath160 gev and four - momentum transfer @xmath161 . the spike seen in the ratio around @xmath132 gev stems from the sign change in the asymmetry , they are not relevant to our discussion . again , we see that the approximation is doing well up to missing momenta of roughly @xmath177 gev . then , the approximation considerably overestimate the full result , and for very large missing momenta , @xmath179 gev , it even fails to reproduce the correct sign of the asymmetry . it is interesting to see that the effects of the approximation are visible even in a quantity that is a ratio of quantities that are both affected by the approximation . in conclusion , approximating the argument of the nn scattering amplitude with the popular `` transverse momentum transfer only '' works well up to missing momenta of @xmath177 gev for various observables . for missing momenta higher than that , the approximation becomes questionable . argument , @xmath48 , and with the purely transverse momentum , @xmath180 . the kinematics are identical to fig . [ figcsmomdis ] : beam energy of @xmath181 gev , @xmath120 , @xmath114 , and @xmath106 . the calculation was performed using on - shell fsi . ] evaluated with the full @xmath121 argument , @xmath48 , and with the purely transverse momentum , @xmath180 . the kinematics are identical to fig . [ figatlpspin ] : the calculations were performed using on - shell fsi . ] in this paper , we have presented a fully relativistic @xmath2 calculation in impulse approximation . we have steered clear from a number of common , simplifying assumptions . the only approximation made in this paper is neglecting the negative - energy contributions to the propagator of the off - shell nucleon . these contributions can realistically expected to be very small compared to the positive - energy contributions . we have used a parametrization of experimental @xmath121 data from said to describe the full @xmath5 scattering amplitude for the final state interaction . this leads to certain limits in the kinematics we can access , as these parametrizations are available only for lab kinetic energies of @xmath135 gev or less . in our calculations , we have investigated the effects of the different contributions to the nn scattering amplitude : the central , spin - orbit , and double - spin - flip parts , using the saclay formalism to describe the different contributions . many other fine calculations using the generalized eikonal approximation @xcite or a diagrammatic approach @xcite use the central part only . while the central part of the amplitude is clearly dominant in almost all observables , the spin - orbit and double - spin - flip parts do contribute visibly to the cross section in the peak area of the angular distribution , increasing the peak height roughly by a quarter . for the out - of - plane asymmetry @xmath95 , which is non - zero only in the presence of fsis , the spin - orbit part is clearly the most relevant . depending on the kinematics , the double - spin - flip can also play a relevant role for this asymmetry . we also showed the different contributions of the nn amplitude in terms of the five invariant amplitudes . interestingly , they are all relevant , and a lot of interference effects contribute to the full result . it is not possible to identify a single , dominant contribution in this framework for the description of the nn amplitudes . in the spirit of avoiding all unnecessary approximations , we have used the full argument for the calculation of the nn scattering amplitudes . in glauber theory and its variants , one often encounters the assumption of a transverse momentum transfer only , and this changes the value of mandelstam @xmath48 . we have investigated the validity of this assumption , and found that it is a very good approximation for missing momenta up to @xmath177 gev . for higher missing momenta , the quality of this approximation deteriorates quickly , and it should probably not be used . we have also compared the influence of the off - shell fsi contributions to the on - shell fsi contributions . the former are expected to not be too large , and they require some interpolation of on - shell amplitudes and the introduction of a regulator function to suppress very far off - shell contributions . the off - shell fsi contributions tend to be negligible to small for lower missing momenta , @xmath182 gev for any observable . beyond that , their importance varies depending on the specific kinematics : the off - shell fsi is very small for the momentum distribution calculated for the quasi - free value @xmath114 , but it is large for @xmath125 . this pattern was observed both for the cross section and the asymmetries @xmath94 , @xmath93 , and @xmath95 . the size of the off - shell contribution does depend on the chosen cut - off , with a larger cut - off admitting a sometimes much - larger contribution . the main purpose of showing figures with the ratios of off - shell calculations with different cut - offs to the on - shell result is to identify `` safe '' kinematics and observables , where the off - shell fsi contributions are definitely small . in regions where they are relevant , a certain amount of theoretical uncertainty can not be avoided , until reliable and believable models of the off - shell nn interaction at the relevant energies are developed . this is particularly relevant for the interpretation of new data taken at high missing momentum at jefferson lab . the current calculations will be applied or already have been applied to the forthcoming experimental data from jefferson lab @xcite , and calculations for the blast data from mit bates @xcite are planned . our calculation also does a nice job of improving the agreement with the low missing momentum data of ulmer et al . @xcite , even though the `` low missing momentum puzzle '' @xcite is not completely resolved . logical next steps for enhancing our calculations are the inclusion of meson exchange currents , and isobar states . * acknowledgments * : we thank paul ulmer for quickly providing us with his data from @xcite in tabulated form . sj thanks charlotte elster for interesting discussions . this work was supported in part by funds provided by the u.s . department of energy ( doe ) under cooperative research agreement under no . de - ac05 - 84er40150 and by the national science foundation under grants no . phy-0354916 and phy-0653312 . the invariant functions @xmath183 can be obtained from scattering data . for example , the helicity amplitudes are defined as @xmath184 where @xmath185 is the helicity spinor for helicity @xmath186 . the helicity matrix elements for @xmath5 scattering in the center - of - momentum frame can be obtained from the program said for laboratory kinetic energies of up to 1.3 gev . if the amplitudes are extracted in units of @xmath187 , the helicity amplitudes consistent with the conventions used here are related to the said amplitudes by @xmath188 parity , time - reversal and particle interchange symmetries can be used to show that there are only five independent helicity amplitudes defined as @xmath189 using ( [ fermi ] ) in ( [ helicity ] ) and solving for the invariant functions gives @xmath190 where @xmath191}{\sqrt{s}\sin\theta}\\ a_{13}&=&\frac{m^2[2m^2(1+\cos\theta)-s(3+\cos\theta)]}{s(1+\cos\theta ) } \\ a_{22}&=&\frac{4m^{2}(1+\cos\theta)}{\sqrt{s}\sin\theta}\\ a_{23}&=&\frac{2m^{2}[m^{2}(1+\cos\theta)+s]}{s(1+\cos\theta)}\\ a_{32}&=&-\frac{m\sqrt{s}(1-\cos\theta)}{2\sin\theta}\\ a_{33}&=&-\frac{m^2[m^2(1+\cos\theta)+s(1-\cos\theta)]}{s(1+\cos\theta)}\\ a_{42}&=&-\frac{m[8m^{2}+s(3+\cos\theta)]}{\sqrt{s}\sin\theta}\\ a_{43}&=&\frac{m^{2}[2m^2(1+\cos\theta)-s(3+\cos\theta)]}{s(1+\cos\theta)}\\ a_{45}&=&-\frac{m^{2}[2m^2(1-\cos\theta)+s(7+\cos\theta)]}{s(1-\cos\theta)}\\ a_{52}&=&-\frac{4m^{2}(1-\cos\theta)}{\sqrt{s}\sin\theta}\\ a_{53}&=&\frac{2m^{2}[m^{2}(1+\cos\theta)-s]}{s(1+\cos\theta)}\,.\end{aligned}\ ] ] the saclay amplitudes are given in terms of the helicity amplitudes by @xmath192 equations ( [ hetoinv ] ) and ( [ hetosa ] ) can then be used to obtain the transformation from the saclay amplitudes to the fermi invariant functions . r. a. gilman and f. gross , j. phys . g * 28 * , r37 ( 2002 ) [ arxiv : nucl - th/0111015 ] . i. sick , prog . part . phys . * 47 * , 245 ( 2001 ) [ arxiv : nucl - ex/0208009 ] . s. k. bogner , r. j. furnstahl and r. j. perry , phys . c * 75 * , 061001 ( 2007 ) [ arxiv : nucl - th/0611045 ] . w. cosyn , m. c. martinez and j. ryckebusch , arxiv:0710.4837 [ nucl - th ] ; w. cosyn , m. c. martinez , j. ryckebusch and b. van overmeire , phys . c * 74 * , 062201 ( 2006 ) [ arxiv : nucl - th/0701029 ] ; b. van overmeire and j. ryckebusch , phys . b * 644 * , 304 ( 2007 ) [ arxiv : nucl - th/0608040 ] ; j. ryckebusch , w. cosyn , b. van overmeire and c. martinez , eur . phys . j. a * 31 * , 585 ( 2007 ) . b. clasie _ et al . _ , arxiv:0707.1481 [ nucl - ex ] . j. ryckebusch , w. cosyn , b. van overmeire and c. martinez , eur . j. a * 31 * , 585 ( 2007 ) ; j. ryckebusch , p. lava , m. c. martinez , j. m. udias and j. a. caballero , nucl . phys . a * 755 * , 511 ( 2005 ) ; p. lava , m. c. martinez , j. ryckebusch , j. a. caballero and j. m. udias , phys . b * 595 * , 177 ( 2004 ) [ arxiv : nucl - th/0401041 ] ; l. l. frankfurt , w. r. greenberg , g. a. miller , m. m. sargsian and m. i. strikman , z. phys . a * 352 * , 97 ( 1995 ) [ arxiv : nucl - th/9501009 ] . jefferson lab experiment e01 - 020 , spokespersons w. boeglin , m. jones , a. klein , p. ulmer , j. mitchell , e. voutier . k. s. egiyan _ et al . _ [ clas collaboration ] , phys . lett . * 96 * , 082501 ( 2006 ) [ arxiv : nucl - ex/0508026 ] ; k. s. egiyan _ et al . _ [ clas collaboration ] , phys . c * 68 * , 014313 ( 2003 ) [ arxiv : nucl - ex/0301008 ] . g. gilfoyle , spokesperson , jefferson lab hall b , e5 run period ; g.p . gilfoyle , ( the clas collaboration ) , out - of - plane measurements of the fifth structure function of the deuteron , bull . am . phys . , fall dnp meeting , df.00010(2006 ) . p. e. ulmer _ et al . _ , phys . lett . * 89 * , 062301 ( 2002 ) . werner boeglin , talk at the 2005 workshop on `` probing microscopic structure of the lightest nuclei in electron scattering at jlab energies and beyond '' , trento , italy july 25 - 30 , 2005 , http://www.fiu.edu/ sargsian / ect05 m. m. sargsian , int . j. mod . e * 10 * , 405 ( 2001 ) [ arxiv : nucl - th/0110053 ] ; m. m. sargsian , t. v. abrahamyan , m. i. strikman and l. l. frankfurt , phys . rev . c * 71 * , 044614 ( 2005 ) [ arxiv : nucl - th/0406020 ] ; l. l. frankfurt , m. m. sargsian and m. i. strikman , phys . c * 56 * , 1124 ( 1997 ) [ arxiv : nucl - th/9603018 ] . j. ryckebusch , d. debruyne , p. lava , s. janssen , b. van overmeire and t. van cauteren , nucl . a * 728 * , 226 ( 2003 ) [ arxiv : nucl - th/0305066 ] ; d. debruyne , j. ryckebusch , w. van nespen and s. janssen , phys . c * 62 * , 024611 ( 2000 ) [ arxiv : nucl - th/0005058 ] . c. c. degli atti and l. p. kaptari , phys . c * 71 * , 024005 ( 2005 ) ; c. ciofi delgi atti and l. p. kaptari , arxiv:0705.3951 [ nucl - th ] ; c. ciofi degli atti , l. p. kaptari and d. treleani , phys . rev . c * 63 * , 044601 ( 2001 ) [ arxiv : nucl - th/0005027 ] . r. schiavilla , o. benhar , a. kievsky , l. e. marcucci and m. viviani , phys . c*72 * , 064003 ( 2005 ) [ arxiv : nucl - th/0508048 ] . b. van overmeire and j. ryckebusch , phys . b * 650 * , 337 ( 2007 ) [ arxiv:0704.0705 [ nucl - th ] ] . s. jeschonnek , phys . c * 63 * , 034609 ( 2001 ) [ arxiv : nucl - th/0009086 ] . c. ciofi degli atti , l. p. kaptari and h. morita , nucl . a * 782 * , 191 ( 2007 ) [ arxiv : nucl - th/0609029 ] . f. gross , phys . rev . * 186 * , 1448 ( 1969 ) ; phys . d * 10 * , 223 ( 1974 ) , phys . c * 26 * , 2203 ( 1982 ) . j. w. van orden , n. devine and f. gross , phys . lett . * 75 * , 4369 ( 1995 ) . r. a. arndt , w. j. briscoe , i. i. strakovsky and r. l. workman , phys . c * 76 * , 025209 ( 2007 ) [ arxiv:0706.2195 [ nucl - th ] ] ; data available through said , http://gwdac.phys.gwu.edu/ h. arenhoevel , w. leidemann and e. l. tomusiak , phys . c * 46 * , 455 ( 1992 ) ; h. arenhovel , w. leidemann and e. l. tomusiak , phys . c * 52 * , 1232 ( 1995 ) ; w. leidemann , e. l. tomusiak and h. arenhovel , phys . c * 43 * , 1022 ( 1991 ) ; f. ritz , h. goller , t. wilbois and h. arenhovel , phys . c * 55 * , 2214 ( 1997 ) . a. bianconi , s. jeschonnek , n. n. nikolaev and b. g. zakharov , _ phys . lett . _ * b343 * , 13 ( 1995 ) ; a. bianconi , s. jeschonnek , n. n. nikolaev and b. g. zakharov , _ phys . c _ * 53 * , 576 ( 1996 ) . a. bianconi , s. jeschonnek , n. n. nikolaev and b. g. zakharov , nucl . * a608 * , 437 ( 1996 ) .

degenerate optomechanical parametric oscillators are optical resonators in which a mechanical degree of freedom is coupled to a cavity mode that is nonlinearly amplified via parametric down - conversion of an external pumping laser . below a critical pumping power the down - converted field is purely quantum - mechanical , making the theoretical description of such systems very challenging . here we introduce a theoretical approach that is capable of describing this regime , even at the critical point itself . we find that the down - converted field can induce significant mechanical cooling and identify the process responsible of this as a cooling - by - heating mechanism . moreover , we show that , contrary to naive expectations and semi - classical predictions , cooling is not optimal at the critical point , where the photon number is largest . our approach opens the possibility for analyzing further hybrid dissipative quantum systems in the vicinity of critical points .

1508.05245

degenerate optical parametric oscillators ( dopos ) consist of a driven optical cavity containing a crystal with second - order optical nonlinearity @xcite . down - conversion in the crystal can generate a field at half the frequency of the driving laser and classical electrodynamics predicts that such field will start oscillating inside the cavity only if the external laser power exceeds some threshold value , where the nonlinear gain can compensate for the cavity losses . a fully quantum - mechanical theory , on the other hand , reveals that even below threshold the down - converted field is not vacuum , but a squeezed field whose quantum correlations increase as the threshold is approached . recent developments in the fabrication of crystalline whispering gallery mode ( cwgm ) resonators @xcite have opened the way to study the intracavity interplay between down - conversion and optomechanics @xcite , a setup that we will refer to as _ degenerate optomechanical parametric oscillator _ ( dompo ) . so far , it has been shown that the presence of down - conversion in an optomechanical cavity can help enhancing mechanical cooling @xcite , normal mode splitting @xcite , sensitivity in position measurements @xcite , or even bringing optomechanics close to the strong coupling regime with additional bath engineering @xcite . in all these works , however , the nonlinear crystal is operated as a parametric amplifier , providing a nonlinear gain to some external field that is injected in the cavity at the down - converted frequency ( _ stimulated _ down - conversion ) . in contrast , the description of the interaction between the field generated via _ spontaneous _ down - conversion and the mechanical mode is much more challenging , since ( below threshold ) the former is purely quantum mechanical @xcite , so that the optomechanical coupling can not be linearized and does not admit a simple gaussian description . in this work we provide a theory for the dompo which can be trusted all the way to threshold , and is obtained by combining traditional adiabatic elimination techniques with our recently developed self - consistent mori projector ( c - mop ) theory @xcite . to this end we first introduce the master equation which models the dompo , and perform an adiabatic elimination of the optical modes by neglecting the mechanical backaction . the mechanical state is found to stay approximately thermal for parameters compatible with current cwgm resonators , with an effective temperature dependent on the steady - state value and two - time correlation function of the down - converted photon number , which we derive in two ways . first by treating the pump mode as a classical field ( semi - classical approach ) , allowing us to obtain simple analytical expressions and provide a physical explanation for the regions of significant cooling , showing that the system provides a realistic implementation of the cooling - by - heating mechanism @xcite below threshold . second , by using c - mop theory on the optical dynamics to find reliable results at threshold and justify the absence of the mechanical backaction onto the optics . remarkably , this accurate approach allows us to prove that the semi - classical predictions break down when working very close to threshold , where cooling is shown to disappear . these results might have strong implications not only for future analysis , but also for previous results which make use of semi - classical approaches while working very close to threshold @xcite . in the final section we apply c - mop theory on the full optomechanical problem , and identify the region of the parameter space where the mechanical backaction on the optics is negligible ( which contains the parameters of our interest ) . the system we consider can be schematically represented as in fig . [ figdompo ] . a crystal with second - order optical nonlinearity is shared by two cavities with relevant resonances at frequencies @xmath0 ( _ pump _ ) and @xmath1 ( _ signal _ ) . the pump cavity is driven by a resonant laser , so that photons in the signal cavity can be generated via _ spontaneous down - conversion _ in addition , one of the mirrors of the signal cavity can oscillate , and is therefore _ optomechanically _ coupled to the down - converted field via radiation pressure @xcite . let us define annihilation operators @xmath2 for the pump ( @xmath3 ) , signal ( @xmath4 ) , and mechanical ( @xmath5 ) modes . including losses of the optical modes at rate @xmath6 ( assumed the same for pump and signal without loss of generality ) , as well as the irreversible energy exchange of the mechanical mode with its thermal environment at rate @xmath7 ( with which , in the absence of light , it is in thermal equilibrium with @xmath8 phonons ) , the master equation governing the evolution of the dompo s state @xmath9 can be written as @xmath10&+&\gamma_0 \mathcal{d}_{a_s}[\rho]+\gamma_0 \mathcal{d}_{a_p}[\rho ] \\ & + & \gamma_m(\bar{n}_\mathrm{th}+1)\mathcal{d}_{a_m}[\rho]+ \gamma_m\bar{n}_\mathrm{th}\mathcal{d}_{a^\dagger_m}[\rho].\nonumber\end{aligned}\ ] ] we have defined lindblad superoperators @xmath11=2j(\cdot ) j^\dagger - j^\dagger j(\cdot)-(\cdot ) j^\dagger j$ ] , and the hamiltonian @xmath12 where we normalize the optomechanical coupling @xmath13 to the frequency of the mechanical oscillation @xmath14 . the optical hamiltonian can be written in a picture rotating at the laser frequency as @xmath15 where @xmath16 is the detuning of the signal mode ( which we will take positive in this work ) , @xmath17 is the down - conversion rate , and @xmath18 is proportional to the square root of the injected laser power . ( color online ) sketch of the degenerate optomechanical parametric oscillator . we take crystalline whispering gallery mode resonators @xcite as a reference for our choice of parameters , but electromechanical implementations are also possible using superconducting circuits in the degenerate parametric oscillation configuration @xcite coupled to drum - shaped oscillating capacitors @xcite ( see also @xcite ) . ] in the classical limit , the steady - state phase diagram of the dompo features a variety of phases @xcite . here we focus on the regime where the state of the signal field is fully quantum , i.e. where the trivial solution @xmath19 is the only stable one , henceforth referred to as the _ monostable phase _ , which requires two conditions . first , defining the _ injection parameter _ @xmath20 and the normalized detuning @xmath21 , the trivial solution becomes unstable in favour of a nontrivial one @xmath22 for @xmath23 @xcite . hence , we write @xmath24 and focus on the @xmath25 $ ] region . the second condition , @xmath26 , guarantees that the nontrivial solution does not enter the @xmath25 $ ] region @xcite . in this expression we have introduced the dimensionless down - conversion coupling @xmath27 as well as the sideband - resolution parameter @xmath28 . we emphasize that the vanishing signal field amplitude excludes the possibility of using a linearization approach similar to those applied in @xcite . in the following , we provide a theory that works in all the @xmath29 $ ] region , and use it to predict the action of the down - converted field on the mechanical state . despite the complexity of the problem , we remarkably find with the help of c - mop theory that for typical system parameters the optical modes do not receive considerable backaction from the mechanics . this property , which we justify in sec . [ backaction ] , allows us to simplify the problem significantly via an adiabatic elimination of the optical modes @xcite leading to an effective master equation for the reduced mechanical state @xmath30 . as we show in appendix [ app1 ] , the mechanical steady state can then be approximated by a thermal state ( displaced by @xmath31 ) characterized by its phonon number @xmath32 where @xmath33 is the cooling efficiency and @xmath34 the fundamental limit for the phonon number . all the information about the optical modes is contained in the heating and cooling rates @xmath35 through the optical correlation function @xmath36 where @xmath37 is the signal photon number and @xmath38 the bare cooperativity with @xmath39 the mechanical quality factor . here , @xmath40 is the steady state of the optical liouvillian @xmath41=-\mathrm{i}[h_\mathrm{opt},\cdot]+\sum_{j={s , p}}\gamma_0 \mathcal{d}_{a_j}[\cdot],\ ] ] that is , @xmath42=0 $ ] . in the following we study the behaviour of the steady - state phonon number as we approach the dompo s threshold . from eq . ( [ ss phonon number ] ) it is clear that optimal cooling is then found by simultaneously maximizing @xmath43 and minimizing @xmath44 . the nonlinear nature of the parametric down - conversion process in eq . ( [ loptnew ] ) and a potential back action of the mechanical mode preclude an exact treatment of the optical correlation function in eq . ( [ two time corr ] ) . to get simple analytic expressions that enable physical insight , we first apply standard linearization to the optical problem , which we denote by _ semi - classical _ approach and has been the method of choice in previous works @xcite . next , applying c - mop theory @xcite we show the failure of the semi - classical approach close to the critical point and find more accurate expressions at criticality , which will also allow us to justify the adiabatic elimination of the optical modes . ( color online ) steady - state phonon number as a function of @xmath45 and @xmath46 , as obtained from a semi - classical description of the dopo and a thermal phonon number @xmath47 ( achievable either with cryogenics @xcite or via a standard sideband pre - cooling stage ) . we choose typical parameters for cwgm resonators @xcite : @xmath48 , @xmath49 , and @xmath50 . note that typical optical decay rates are on the mhz range , although only normalized quantites are relevant for our results . the signal steady - state photon numbers @xmath51 corresponding to the ticked @xmath46-values are shown in the upper axis , showing that cooling is effective even with just @xmath52 photons . on the other hand , c - mop theory has allowed us to prove that the cooling region closer to threshold disappears once such a more accurate approach is used , see fig . [ nf cmop vs classical ] . ] below threshold , the linearization of the dopo is accomplished by treating the pump mode as a classical stationary source , that is , by performing the replacement @xmath53 @xcite . within this approximation , the optical problem is governed by a gaussian single - mode liouvillian @xmath54=-\mathrm{i}[\delta a_s^\dagger a_s+\sigma ( a_s^{\dagger \,2}-a_s^2)/2,\cdot]+\mathcal{d}_{a_s}[\cdot],\ ] ] from which any correlation function can be easily found , allowing us to obtain analytical expressions for the relevant quantities in eq . ( [ ss phonon number ] ) , as we show in appendix [ sec semi - class supmat ] . for the fundamental limit , we find @xmath55/8\omega\delta,\ ] ] while the cooling efficiency can be written as @xmath56 where we have defined the function @xmath57}{(4+\omega^2)[\omega^4 + 16 ( 1+\delta_\mathrm{eff}^2)^2 + 8 \omega^2 ( 1-\delta_\mathrm{eff}^2)]},\nonumber\ ] ] and a parameter @xmath58 that will be shown later to play the important role of an effective optical detuning . the photon number @xmath59 is fully due to quantum fluctuations and increases hyperbolically until threshold @xmath60 where it diverges in this semi - classical approach . in fig . [ nf classical ] we show the steady - state phonon number as a function of the two control parameters , detuning @xmath45 and distance to threshold @xmath46 , fixing the rest of parameters to typical values of cwgm resonators @xcite . there are two regions where significant cooling effects appear . one is in the vicinity of the threshold point and can be traced back to the vast increase of the photon number @xmath61 which makes @xmath62 for virtually any value of the remaining parameters . however , as we will show below with the c - mop approach , so close to threshold this semi - classical approach breaks down , hence rendering this prediction incorrect . the other region , which turns out to be of major significance when aiming for optimal cooling , corresponds to @xmath63 , see the black solid line in fig . [ nf classical ] . the c - mop approach will confirm this prediction in the next section . moreover , it can be understood in physical terms by moving to a new picture defined by the squeezing operator @xmath64 $ ] with @xmath65 ( note that such transformation requires @xmath66 , which corresponds in fig . [ nf classical ] to the region above the black dashed line ) . this transformation diagonalizes the hamiltonian in the optical liouvillian ( [ ls ] ) , so that , defining the parameters @xmath67 and @xmath68 , the transformed state @xmath69 evolves according to @xmath70+(\bar{n}_\mathrm{eff}+1 ) \mathcal{d}_{a_s}[\tilde{\rho}]+\bar{n}_\mathrm{eff}\mathcal{d}_{a_s^\dagger}[\tilde{\rho } ] \nonumber \\ & & -[\mathrm{i}\omega a_m^\dagger a_m,\tilde{\rho}]+(\omega / q)\{(\bar{n}_\mathrm{th}+1)\mathcal{d}_{a_m}[\tilde{\rho}]+\bar{n}_\mathrm{th}\mathcal{d}_{a_m^\dagger}[\tilde{\rho}]\ } \nonumber \\ & & + [ \omega\eta_\mathrm{om}m(a_s^2 a_m^\dagger - a_s^{\dagger 2 } a_m),\tilde{\rho}],\end{aligned}\ ] ] within the rotating - wave approximation , valid under the conditions @xmath71 and @xmath72 , see appendix [ cooling by heating app ] . therefore , in this picture the signal field is turned into a bosonic mode with oscillation frequency @xmath73 and thermal occupation @xmath74 . the optomechanical coupling is dressed by the squeezing parameter @xmath75 , similarly to the dressing by the intracavity field amplitude in standard sideband cooling @xcite . however , at difference with that case the interaction exchanges phonons with pairs of photons ( rather than single photons ) , thus explaining why @xmath76 is the resonance condition for cooling . within this condition , assuming @xmath77 and @xmath78 , we furthermore find @xmath79 . the cooling efficiency @xmath43 thus receives an additional contribution @xmath80 from the effective thermal photon number , which is a direct consequence of the nonlinear nature of the effective optomechanical coupling in ( [ newpictureme ] ) that can not be found in standard sideband cooling , as we discuss in appendix [ cooling by heating app ] . this represents a natural example of the so - called _ cooling - by - heating _ effect @xcite , where heating up the optical field can contribute to making optomechanical cooling more efficient . however , as it is well known from standard sideband cooling @xcite , thermal photons also contribute to the fundamental limit , which indeed can be approximated by @xmath81 in our scenario . when the term @xmath82 dominates over @xmath44 in eq . ( [ ss phonon number ] ) the thermal optical background @xmath74 can then be interpreted as `` good noise '' , while as soon as the fundamental limit is reached it becomes `` bad noise '' and heats up the mechanical motion ( see the thick black curve in fig . [ nf classical ] ) . we discuss this phenomenon in more detail in appendix [ cooling by heating app ] . it is interesting to examine the limits that this `` bad noise '' imposes on cooling . for the parameters of fig . [ nf cmop vs classical ] , compatible with current cwgm resonators , it is found that the cooling efficiency and the fundamental limit can be simultaneously optimized to about @xmath83 and @xmath84 as a function of the experimentally - tunable parameters @xmath45 and @xmath85 , precluding ground - state cooling . however , the foreseeable increase of the mechanical quality factor and the optomechanical coupling by one order of magnitude in next - generation cwgm resonators @xcite will improve these numbers to @xmath86 and @xmath87 . ( color online ) terms @xmath82 and @xmath44 contributing to the steady - state phonon number ( [ ss phonon number ] ) as a function of the distance - to - threshold parameter @xmath46 . we fix the detuning to @xmath88 , corresponding to the straight thin solid line in fig . [ nf classical ] , taking @xmath89 @xcite and the rest of parameters as in that figure . the inset displays a close up of the steady - state phonon number @xmath90 for @xmath91 $ ] , but without applying the rotating - wave and markov approximations , see [ mech moments ] . note that c - mop gives finite results equivalent to those found within the rotating - wave approximation , while the semi - classical predictions diverge at threshold , which can be taken as further evidence that the theory breaks down there . ] the simplest way of obtaining the optical correlator @xmath111 is by using standard linearization on @xmath171 . in this approach , we move to a displaced picture in which the large coherent background of the pump mode is removed , and then keep terms of the transformed optical liouvillian only up to second order in the the bosonic operators . the displacement operator @xmath172 $ ] allows us to move to the new picture , in which the transformed optical state @xmath173 evolves according to a transformed liouvillian @xmath174 . removing terms beyond quadratic order , this transformed liouvillian can be written as a sum of independent liouvillians for the pump and signal modes , @xmath175 , with @xmath176 and @xmath177+\mathcal{d}_{a_s}[\cdot],\ ] ] with the injection parameter @xmath178 and normalized detuning @xmath21 . consequently , the optical steady state in the original picture becomes the separable state @xmath179 where @xmath180 is a coherent state of amplitude @xmath181 and @xmath182 is the gaussian state satisfying @xmath183=0 $ ] . the latter is completely characterized by its first and second moments , which are trivially found to be @xmath19 , @xmath184 , and @xmath185 , where we use the usual notation @xmath186 for any operator @xmath107 acting on the signal subspace . the optical correlation function simplifies to @xmath187 where we have defined a traceless operator @xmath188 . using again the fact that the liouvillian @xmath189 is gaussian , it is simple to evaluate the correlation function @xmath111 . to this aim , let us define the column vector @xmath190 where the expectation value of an operator @xmath107 with the tilde is defined as @xmath191 . taking the derivative of this vector with respect to @xmath192 , we find the linear system @xmath193 , where the matrix @xmath75 reads @xmath194 it is straightforward to solve this linear system , for example by diagonalizing @xmath195 . we write @xmath196 , with a similarity matrix @xmath197 that can be found analytically ( but its expression is too lengthy to be reported here ) , and a diagonal matrix @xmath198 containing the eigenvalues of @xmath195 , @xmath199 , and @xmath200 . notice that for @xmath201 the square root becomes imaginary , making @xmath202 , and in fact @xmath203 at threshold , @xmath204 . consequently , we call the region with @xmath201 the _ critical slowing down regime_. the solution of the linear system is then found as @xmath205 where we have defined the initial condition vector @xmath206 and the matrices @xmath207 , where @xmath208 . note that the vector @xmath209 is formed by fourth order moments . in order to find them , we simply exploit the gaussian structure of @xmath189 , which allows us to express moments of any order as products of moments of first and second order . specifically , concerning third and fourth order moments we simply use [ gsamoments ] @xmath210 where @xmath211 . note that the optical correlation function we are looking for is given by the first component of the vector , @xmath212_1 $ ] , and the integrals appearing in @xmath213 and @xmath214 in eq . ( [ d1 and d2 ] ) can be easily evaluated due to the exponential @xmath192-dependence of @xmath215 in eq . ( [ v_sol ] ) . the semi - classical approach has allowed us to get analytical and physical insight into the problem . it is however well known that this approximation fails close to the critical point , although there is no systematic way of checking where exactly within the semi - classical formalism itself . hence , to determine where it exactly breaks down and to find more accurate results for those parameters , we make use of the recently developed c - mop technique @xcite , which allows finding reduced equations for the constituent parts of a composite system , even in situations where there is significant backaction among its parts and no time - scale separation between their dynamics . for parameters compatible with cwgm resonators , the theory is already regularized by using c - mop only in the optical problem ( dopo ) , which provides a more accurate description for the optical correlation function ( [ two time corr ] ) and photon number that enter the effective mechanical dynamics . the application of c - mop to the dopo has been detailed in @xcite , but we review its most relevant steps for completeness in appendix [ c - mop theory ] . specifically , we use a combination of c - mop and a gaussian - state approximation , which provides an efficient and accurate tool capable of regularizing the divergencies and unphysical predictions of the semi - classical approach . in particular , we show in appendix [ c - mop theory ] that at threshold the decay rate of the optical correlator scales as @xmath92 , while the photon number as @xmath93 , in contrast to semi - classical results in which the former goes to zero while the latter diverges . we show a very representative case for the phonon number @xmath90 as a function of the distance to threshold @xmath46 in fig . [ nf cmop vs classical ] . the method finds perfect agreement with the semi - classical predictions sufficiently below threshold , in particular verifying the cooling - by - heating effect presented above . most importantly , we find that the absolute minimum phonon number is indeed reached when the resonance condition @xmath94 is met . on the other hand , close to threshold we find a significant correction to the semi - classical predictions for the fundamental limit @xmath44 . in particular , while this is independent of the distance to threshold @xmath46 in the semi - classical picture , c - mop shows that it actually increases very rapidly as the critical point is approached , and hence no cooling is found no matter how much the efficiency @xmath43 is increased . this is consistent with the fact that when @xmath95 ( as happens at threshold ) , @xmath73 becomes imaginary and there is no resonance for the optomechanical interaction . the adiabatic elimination of the optical fields which we have used throughout the work relies on the time - scale separation between the optical and mechanical degrees of freedom . in particular , such an approach neglects mechanical backaction onto the optics , which is a good approximation as long as the rate of any mechanical perturbation is much smaller than the intrinsic relaxation rate of the optics @xmath96 . far from the critical point the optical relaxation rate is @xmath6 , which usually dominates over any other rate in the system . however , as the critical point is approached the dopo dynamics exhibits a critical slowing down , and its relaxation rate becomes smaller and smaller . hence , in our work that considers parameters close to threshold , it is very important to check that the desired time - scale separation is present . an intuitive argument supporting such a time - scale separation follows from relating the mechanical backaction rate with the optical frequency - shift induced by the optomechanical interaction , @xmath97 , where we have used ( [ displacement ] ) . hence , using the scaling of @xmath61 and @xmath96 obtained in the previous section at threshold , the condition @xmath98 becomes @xmath99 , which is very well satisfied for the parameters that we work with . moreover , note that this condition is automatically satisfied when working within the monostability condition @xmath100 as long as @xmath101 . we can set more rigorous bounds to the region where mechanical backaction is negligible by using c - mop theory @xcite , since , in contrast to adiabatic elimination methods , it does not rely on the concept of time - scale separation or absence of backaction effects . hence , we apply this theory to the dompo system by using the time - dependent self - consistent mori projectors @xmath102 and @xmath103 , that is , using a bipartition `` optics @xmath104 mechanics '' for the system . this approach will allow us to identify the terms contributing to the mechanical backaction and find upper bounds to their scaling . following the procedure introduced in previous works @xcite , the c - mop equations for the reduced optical and mechanical states in the asymptotic @xmath105 or steady - state limit are easily found to read [ cmopfordompo ] @xmath106\label{cmop for m } \\ & -\omega_m^2\eta_\mathrm{om}^2\left[x_m,\int_0^\infty d\tau \,e^{\mathcal{l}_m\tau}(\delta x_m\bar{\rho}_m s(\tau)-\mathrm{h.c.})\right ] , \nonumber \\ & \frac{d\bar{\rho}_\mathrm{opt}}{dt}=0=\mathcal{l}_\mathrm{opt}\bar{\rho}_\mathrm{opt}+\mathrm{i}\omega_m\eta_\mathrm{om } \langle x_m\rangle[a_s^\dagger a_s,\bar{\rho}_\mathrm{opt } ] \label{cmop for optics } \\ & -\omega_m^2\eta_\mathrm{om}^2\left[a_s^\dagger a_s,\int_0^\infty \hspace{-2mm}d\tau \,e^{\mathcal{l}_\mathrm{opt}\tau}(\delta n_s\bar{\rho}_\mathrm{opt}\ , s_m(\tau)-\mathrm{h.c.})\right],\nonumber\end{aligned}\ ] ] where , for any operator @xmath107 , we have introduced the usual definitions @xmath108 and @xmath109 , while @xmath110 . @xmath111 is the usual optical correlation function ( [ two time corr ] ) , and we have defined the mechanical correlation function @xmath112,\nonumber\end{aligned}\ ] ] where the final expression is easily found by following a similar procedure as the one shown in appendix [ sec semi - class supmat ] for the semi - classical optical correlation function , since @xmath113 is quadratic . the last two terms on the right - hand side of eq . ( [ cmop for optics ] ) account for the mechanical backaction on the optics , and we proceed now to bound their effect . the second to last term describes precisely the optical detuning @xmath114 induced by the optomechanical interaction that we already discussed at the beginning of the section , concluding that it is negligible within the parameter regime we work with . then we focus on the last term in eq . ( [ cmop for optics ] ) , the born term . for this purpose we derive the steady - state equation for the moment @xmath115 ( note that the equation of motion of @xmath116 receives no explicit mechanical backaction even within c - mop theory ) which reads @xmath117\right\ } \nonumber \\ & \hspace{-1mm}+\hspace{-0.5mm}2\mathrm{i}\omega_m^2\eta_\mathrm{om}^2 \int_0^\infty \hspace{-3mm}d\tau \,\mathrm{im}\{s_m(\tau)\ } \,\mathrm{tr}_\mathrm{opt}\left\{a_s^{\dagger 2}e^{\mathcal{l}_\mathrm{opt}\tau}\{\delta n_s,\bar{\rho}_\mathrm{opt}\}\right\}\hspace{-1 mm } , \nonumber\end{aligned}\ ] ] where @xmath118 denotes the anticommutator . note first that the correlation functions @xmath119\right\}$ ] and @xmath120 , which are of similar structure as the optical correlation function @xmath111 , decay to zero at a rate @xmath121 . next , we derive upper bounds for the last two terms in eq . ( [ as2 eq ] ) . for the second to last term we find @xmath122\right\}\right| \nonumber \\ & \leq \left| 2\omega_m^2\eta_\mathrm{om}^2 \frac{\mathrm{re}\{s_m(0)\ } \,\mathrm{tr}_\mathrm{opt}\left\{a_s^{\dagger 2}e^{\mathcal{l}_\mathrm{opt}0}[a_s^\dagger a_s,\bar{\rho}_\mathrm{opt}]\right\}}{\gamma_m+\gamma_\mathrm{opt}}\right| \nonumber \\ & \leq \underbrace{\frac{4\omega_m^2\eta_\mathrm{om}^2\bar{n}_m}{\gamma_m+\gamma_\mathrm{opt}}}_{\gamma_\mathrm{back}^\prime } |\langle a_s^{\dagger 2}\rangle|,\end{aligned}\ ] ] where in the last step we have used @xmath123 ( note that we expect the mechanical state to stay approximately thermal , and hence @xmath124 ) and @xmath125\right\}=\langle[a_s^{\dagger 2},a_s^\dagger a_s]\rangle=2\langle a_s^{\dagger 2}\rangle$ ] . similarly , for the last term in eq . ( [ as2 eq ] ) we find @xmath126 where for the last expression we have used @xmath127 , within the gaussian state approximation ( [ gsamoments ] ) , that is , @xmath128 , noting that @xmath19 below threshold . a sufficient condition for mechanical backaction to be negligible is then @xmath129 . we pass to check whether this is the case in our work . note first that @xmath130 even at threshold , since @xmath131 for the parameters we are interested in . using the scalings @xmath92 and @xmath93 at threshold ( where these bounds are the tightest ) , we can then write the conditions under which backaction is negligible as @xmath132 for the parameter set of fig . 2 these lead to the conditions @xmath133 and @xmath134 , respectively . for the large values of @xmath45 that we use during most of the work , these conditions are very well satisfied . for small @xmath45 they seem to be too tight , but we need to stress here that we have been extremely conservative when estimating the born terms ( [ estimate1 ] ) and ( [ estimate ] ) , meaning that in practice backaction should be negligible even in a much broader region of the parameter space . overall , c - mop theory has allowed us to quantify the mechanical backaction on the optics in a rigorous manner . we have obtained very conservative bounds that the system parameters must satisfy in order for such backaction to be negligible , showing that this is indeed the case for the parameters used in our work , which are compatible with an implementation based on cwgm resonators . it is however foreseeable that such devices , as well as their electromechanical counterparts , will be able to study regions where backaction is significant , in which case the c - mop approach presented in this section will be very useful . by exploiting adiabatic elimination techniques , semi - classical methods , and c - mop theory , we have provided a theoretical analysis of the dompo which works even at the critical point . we have focused on the region where the optical field is fully quantum , showing that such a quantum - correlated field with no coherent component can induce significant mechanical cooling through a cooling - by - heating mechanism . c - mop techniques have allowed us to check the validity of the optical adiabatic elimination as well as the semi - classical approximation , whose predictions have indeed been shown to break down at threshold , showing the potential of c - mop to treat dissipative quantum - optical problems in the vicinity of critical points . we thank yue chang , tao shi , alessandro farace , eugenio roldn , germn j. de valcrcel , florian marquardt , christoph marquardt , and j. ignacio cirac for useful discussions and comments . this work has been supported by the german research foundation ( dfg ) via the crc 631 and the grant ha 5593/3 - 1 . acknowledge funding from the alexander von humbolt foundation through their fellowship for postdoctoral researchers . acknowledges support by uk engineering and physical sciences research council ( epsrc ) under ep / n009428/1 . here we present a derivation of the effective mechanical master equation leading to the phonon number of eq . ( 4 ) in the main text . our starting point is the master equation governing the evolution of the state @xmath135 of the dompo , eq . ( 1 ) , which for convenience we rewrite here as @xmath136+\mathcal{l}_m[\rho]+\mathcal{l}_\mathrm{om}[\rho]\ ] ] with @xmath137 & = \left[-\mathrm{i}\delta_s a_s^\dagger a_s + \left(\epsilon_p a_p^\dagger + \frac{\chi}{2}a_p a_s^{\dagger \,2}-\text{h.c.}\right),\rho\right ] \nonumber \\ & + \gamma_0\mathcal{d}_{a_p}[\rho]+\gamma_0\mathcal{d}_{a_s}[\rho ] , \\ \mathcal{l}_m[\rho ] & = [ -\mathrm{i}\omega_m a_m^\dagger a_m,\rho ] \nonumber \\ & + \gamma_m ( \bar{n}_\mathrm{th}+1)\mathcal{d}_{a_m}[\rho]+\gamma_m \bar{n}_\mathrm{th}\mathcal{d}_{a_m^\dagger}[\rho ] , \\ \mathcal{l}_\mathrm{om}[\rho ] & = [ \mathrm{i}\omega_m\eta_\mathrm{om } a_s^\dagger a_s ( a_m+a_m^\dagger),\rho].\end{aligned}\ ] ] all the quantities are defined in the main text , and we remind the notation @xmath138=2j\rho j^\dagger - j^\dagger j\rho-\rho j^\dagger j$ ] for superoperators in lindblad form . in order to eliminate the optical modes and find an effective master equation for the mechanical state @xmath30 we proceed as follows . we first define the projector superoperator @xmath139=\bar{\rho}_\mathrm{opt}\otimes \mathrm{tr}_\mathrm{opt}\{\cdot\}$ ] whose action on the full state @xmath135 of the dompo is @xmath140=\bar{\rho}_\mathrm{opt}\otimes\rho_m(t)$ ] . here , @xmath40 is the steady state of the optical liouvillian , that is , @xmath141=0 $ ] . applying this superoperator and its complement @xmath142 onto the master equation , and formally integrating the latter , we obtain an exact equation of motion for @xmath30 , the so - called _ nakajima - zwanzig _ equation @xcite . such equation is not solvable , and therefore we apply a _ born approximation _ which takes into account terms up to second order in the optomechanical interaction . the resulting equation reads @xmath143 \\ & -\omega_m^2 \eta_\mathrm{om}^2 \left[x_m,\int_0^t d\tau \,e^{\mathcal{l}_m\tau } [ x_m \rho_m(t-\tau ) s(\tau)-\mathrm{h.c.}]\right ] , \nonumber\end{aligned}\ ] ] where we have defined the mechanical position quadrature @xmath144 , the photon number in the signal mode @xmath145 , and the optical correlation function @xmath146\}-\bar{n}_s^2.\ ] ] it is well known that the steady state @xmath40 of the dopo is unique ( which intuitively comes from the fact that both the pump and signal modes have local dynamics leading to unique steady states , and the parametric interaction preserves that uniqueness ) , and hence @xmath147 is a _ relaxing map _ @xcite , mapping all optical operators @xmath148 into the steady state , that is , @xmath149=\mathrm{tr}_\mathrm{opt}\{o\}\bar{\rho}_\mathrm{opt}$ ] . thus , the optical correlation function @xmath111 will always decay to zero within some finite _ memory time _ which we denote by @xmath150 . hence , in the asymptotic limit we can write @xmath151 in the integral kernel of eq . ( [ nze2 ] ) , obtaining an equation for @xmath152 which is quadratic in the operators @xmath153 and therefore allows for a gaussian - state solution @xcite . in other words , the equations for the first and second steady - state mechanical moments form a closed linear algebraic set [ mech moments ] @xmath154,\nonumber\end{aligned}\ ] ] where we used the abbreviations @xmath155 , @xmath109 , and [ d1 and d2 ] @xmath156 these equations can be solved for the steady - state moments as functions of the optical photon number @xmath61 and correlation function @xmath111 without the need of further approximations . however , in order to obtain more physical insight into the mechanical steady state @xmath152 we apply both markov approximation and a _ rotating - wave approximation _ to eq . ( [ nze2 ] ) . the markov approximation is based on the assumption that within the optical memory time @xmath150 all the mechanical dynamics can be neglected except for the evolution provided by the free hamiltonian @xmath157 . as a result we can write @xmath158\approx x_m(\tau ) \rho_m(t)$ ] with @xmath159 . on the other hand , the rotating - wave approximation consists in neglecting all the terms proportional to @xmath160 and @xmath161 in the effective mechanical master equation , under the assumption that their rotation at frequency @xmath162 is much larger than the rates they are weighted by . after applying these approximations in eq . ( [ nze2 ] ) we are left with an effective mechanical master equation given by @xmath163 \\ & + \gamma_m\gamma_- \mathcal{d}_{a_m}[\rho_m]+\gamma_m\gamma_+ \mathcal{d}_{a_m^\dagger}[\rho_m],\nonumber\end{aligned}\ ] ] where the heating and cooling rates @xmath164 coincide precisely with those defined in the main text . this master equation has a very simple gaussian steady state @xmath152 corresponding to a displaced thermal state with mean @xmath165 phonon number @xmath166 , where @xmath90 is given by eq . ( 4 ) in the main text , and @xmath167 . we note that , starting from a thermal state , the mechanical mode relaxes to this steady state with a rate @xmath168 , where @xmath33 is what we have called the cooling efficiency in the main text , since the equations of motion for the phonon number fluctuations and the mechanical field amplitude are given by @xmath169\,{\langle}a_m{\rangle}\hspace{-0.5mm}+\hspace{-0.5mm}\mathrm{i}\eta_\mathrm{om}\omega_m { \langle}a_s^\dagger a_s{\rangle}.\end{aligned}\ ] ] we have checked that this rate @xmath170 is smaller than the decay rate of the optical correlator @xmath111 for the parameters of interest , hence making markov a valid approximation . let us remark that throughout the work we have been using both eq . ( [ nze2 ] ) and eq . ( [ me born - markov plus rwa ] ) to obtain the steady - state moments of the mechanical oscillator . we have never observed any notable differences between them , except when working extremely close to threshold within the semi - classical approach , see the inset of fig . 3 . in these cases , however , the failure of eq . ( [ me born - markov plus rwa ] ) can be directly attributed to the failure of the semi - classical approach , and not to a failure of the rotating - wave approximation itself , which indeed is very well satisfied as shown by the c - mop approach . thus , we conclude that for the parameter regime studied in this work ( compatible with current cwgm resonators ) the state of the mechanical oscillator is indeed a displaced thermal state , with a phonon number that can only be evaluated once the optical photon number @xmath61 and correlation function @xmath111 are known . in the main text we have shown that significant cooling can be obtained when working in the resolved sideband regime @xmath216 and close to the resonance condition @xmath217 . we have interpreted such a phenomenon as a `` cooling by heating '' effect , for which we have moved to a new picture defined by the squeezing operator @xmath218 $ ] within the semi - classical approach explained above . here we want to explicitly perform an adiabatic elimination of the optical mode in this picture , what will allow us to get more insight into the cooling mechanism . let us first write the master equation in this `` squeezed picture '' . the transformation diagonalizes the hamiltonian in the optical liouvillian ( [ signal liouvillian app ] ) , turning it into @xmath219 = \mathrm{i } [ \delta_{\text{eff } } a_s^\dagger a_s,\cdot ] \\ & & \hspace{3mm}+(1+\bar{n}_\mathrm{eff})\mathcal{d}_{a_s}[\cdot]+\bar{n}_\mathrm{eff } \mathcal{d}_{a_s^\dagger}[\cdot]+\mathrm{i}m \mathcal{k}_{a_s}[\cdot]-\mathrm{i}m \mathcal{k}_{a_s^\dagger}[\cdot],\nonumber\end{aligned}\ ] ] where we have defined the superoperator @xmath220=2 j(\cdot)j - j^2(\cdot)-(\cdot)j^2 $ ] , as well as the parameters @xmath67 and @xmath68 . note that the @xmath221 terms rotate at frequency @xmath222 , and hence are highly suppressed when we work within the cooling condition @xmath223 and with @xmath224 ( rotating - wave approximation ) . therefore , as mentioned in the main text , in this picture the signal field is turned into a bosonic mode with oscillation frequency @xmath73 , and at thermal equilibrium with occupation @xmath74 . on the other hand , the photon - number operator is transformed into @xmath225 , and hence the optomechanical interaction can be approximated by @xmath226 within the rotating - wave approximation as long as @xmath227 . hence , within these conditions , the transformed state @xmath69 evolves according to a master equation that we write as @xmath228+\mathcal{l}_m[\tilde{\rho}]+\tilde{\mathcal{l}}_\mathrm{om}[\tilde{\rho}]\ ] ] with @xmath229 & = & \left[-\mathrm{i}\gamma_0\delta_\mathrm{eff } a_s^\dagger a_s,\tilde{\rho}\right ] \\ & & + \gamma_0(\bar{n}_\mathrm{eff}+1 ) \mathcal{d}_{a_s}[\tilde{\rho}]+\gamma_0\bar{n}_\mathrm{eff}\mathcal{d}_{a_s^\dagger}[\tilde{\rho}],\nonumber \\ \mathcal{l}_m[\tilde{\rho } ] & = & [ -\mathrm{i}\omega_m a_m^\dagger a_m,\tilde{\rho } ] \\ & & + \gamma_m ( \bar{n}_\mathrm{th}+1)\mathcal{d}_{a_m}[\tilde{\rho}]+\gamma_m \bar{n}_\mathrm{th}\mathcal{d}_{a_m^\dagger}[\tilde{\rho}],\nonumber \\ \tilde{\mathcal{l}}_\mathrm{om}[\tilde{\rho } ] & = & [ \omega_m\eta_\mathrm{om}m(a_s^2 a_m^\dagger - a_s^{\dagger 2 } a_m),\tilde{\rho}],\end{aligned}\ ] ] where @xmath74 and @xmath75 are defined in the main text . the structure of this master equation is similar to the original one , eq . ( [ app full master equation ] ) , with the only difference that the optical liouvillian is replaced by @xmath230 , corresponding to a single - mode at finite temperature , and the optomechanical interaction @xmath231 by @xmath232 . the adiabatic elimination of the optical mode can be carried out exactly in the same way as we did in appendix [ app1 ] , and under the cooling condition @xmath76 it would lead to the heating and cooling rates @xmath233 where @xmath234 is the bare optomechanical cooperativity , and @xmath235 is a thermal state with mean photon number @xmath74 . the cooling efficiency is then given by @xmath236\tilde{\rho}_s\}=2cm^2\hspace{-1mm}\left(\hspace{-0.5mm}\bar{n}_\mathrm{eff}\hspace{-0.5mm}+\hspace{-0.5mm}\frac{1}{2}\hspace{-0.5mm}\right)\hspace{-1mm}.\ ] ] the `` cooling by heating '' effect is clearly seen because the cooling efficiency increases with the effective thermal photon number @xmath74 . but it is important to note that this enhancement of the cooling efficiency is a direct consequence of the commutator appearing in the trace , contributing as @xmath237=4a_s^\dagger a_s+2 $ ] , which in turn comes from the fact that the effective optomechanical interaction @xmath238 corresponds to the exchange of phonons with pairs of photons . in the usual sideband laser cooling scenario , the effective optomechanical interaction is bilinear , e.g. @xmath239 , meaning that the commutator in the expression above is replaced by @xmath240=1 $ ] , and hence the thermal photonic background does not enter the cooling efficiency . let us finally note that the fundamental limit can be written as @xmath241\tilde{\rho}_s\}}=\frac{\bar{n}_\mathrm{eff}^2}{2\bar{n}_\mathrm{eff}+1}\;\xrightarrow[]{\bar{n}_\mathrm{eff}\gg1 } \ ; \frac{\bar{n}_\mathrm{eff}}{2},\ ] ] which increases linearly with the effective thermal photon number . hence , as explained in the main text , the `` cooling by heating '' mechanism is optimized by finding a proper trade off between the increase in the cooling efficiency ( `` good noise '' ) and the increase in the fundamental limit ( `` bad noise '' ) . it is to be noted that within the usual sideband laser cooling , any thermal background will still contribute to this fundamental limit , @xmath242\rho_s\}\xrightarrow[]{\bar{n}_\mathrm{eff}\gg1}\bar{n}_\mathrm{eff}$ ] , but , as explained above , it provides no enhancement of the cooling efficiency @xmath43 . in other words , in standard sideband cooling the thermal background acts only as `` bad '' noise . we emphasize that these expressions for @xmath43 and @xmath44 agree with the ones provided in the main text , which have been first calculated exactly within the semi - classical approach , and then approximated to leading order in @xmath243 for @xmath76 . despite the analytical and physical insight that it provides , the semi - classical approach suffers from several issues , most importantly its divergent character at threshold , which shows that it can not be trusted when working close to such point . unfortunately , there is no systematic way of checking within the formalism itself where exactly it fails . this question can only be answered by comparing it to a more accurate approach . to this aim , we have applied our recently developed self - consistent mori projector ( c - mop ) theory @xcite . in particular this approach has allowed us to characterize optical steady - state observables such as the photon number @xmath61 or the correlation function @xmath111 in all relevant parameter space , including threshold . we detailed the application of c - mop theory @xcite to the dopo problem in @xcite , including its combination with the gaussian - state approximation that we use in this work , which was shown to be quite accurate both for steady - state quantities and dynamics . let us now briefly introduce such an approach here for completeness , keeping in mind that details can be looked up in @xcite . our main goal consists in finding the optical correlation function ( [ two time corr ] ) , for which we need to solve the dynamics generated by the optical liouvillian @xmath171 , that is , the dopo dynamics @xmath244 this is a nonlinear two - mode problem with no analytic solution and whose direct numerical simulation becomes unfeasible for moderate photon numbers already . in contrast , c - mop theory works with the reduced states of the pump and signal modes , @xmath245 and @xmath246 , respectively , whose coupled dynamics are approximated by the set of nonlinearly coupled equations [ cmop eq app ] @xmath247\nonumber \\ & & + \left ( \frac{\chi}{2}\right)^2\left\{\left[a_s^2\,,\,h_s(t)\right]+\text{h.c.}\right\ } , \\ \dot{\rho}_p(t)&=&\mathcal{l}_p \rho_p(t)+\frac{\chi}{2}\left[a_p { \langle}a_s^{\dagger\,2}{\rangle}(t)-a_p^\dagger{\langle}a_s^2{\rangle}(t)\,,\rho_p(t)\right]\nonumber \\ & & + \left(\frac{\chi}{2}\right)^2\left\{\left[a_p\,,\sum_{n=1}^3 h_{p , n}(t)\right]+\text{h.c.}\right\ } , \\ \dot h_s(t)&= & ( -\gamma_p + \mathcal l_s ) h_s(t ) + \mathcal k _ s(t , t)\rho_s(t ) , \\ \dot h_{p , n}(t)&= & ( \lambda_n + \mathcal l_p ) h_{p , n}(t)+\mathcal k_{p , n}(t , t)\rho_p(t),\end{aligned}\ ] ] where @xmath248 and @xmath249 are auxilliary operators acting on the signal and pump subspaces ( introduced to turn the c - mop equations into ordinary differential equations , since originally they have an integro - differential structure @xcite ) , and we refer to @xcite for the definitions of the superoperators @xmath189 , @xmath250 , @xmath251 , and @xmath252 . denoting by @xmath253 and @xmath254 the dimensions of the truncated hilbert spaces of signal and pump in a numerical simulation , we see that the original problem ( [ masterdopo ] ) requires solving a @xmath255 system , whereas there are only @xmath256 c - mop equations ( or even less by exploiting the quadratic or gaussian structure of the pump equations ) . nevertheless , even though c - mop allows to get numerical insight in a larger region of parameter space , the simulation of problems with very large photon numbers ( such as the ones we work with close to threshold ) is still challenging . it is in those regions where a gaussian - state approximation becomes extremely useful . as the name suggests , such approximation consists in assuming that the reduced signal and pump states are gaussian , meaning that they are completely characterized by first and second order moments . under such circumstances , we can approximate third and fourth order moments of @xmath182 as in ( [ gsamoments ] ) , and the c - mop equations provide a closed set of nonlinear equations for the first and second order moments of the operators @xmath257 , @xmath258 , @xmath248 and @xmath249 . the steady - state moments can then be efficiently found simply by finding the stationary solutions of these equations . as an example , in fig . [ scaling]a we show the steady - state photon number @xmath61 at the critical point @xmath259 as a function of the normalized detuning @xmath45 . it shows a clear linear dependence for @xmath260 which , together with the well - known @xmath261 scaling with the down - conversion coupling @xcite , provides an overall @xmath93 scaling of the signal photon number at threshold . the knowledge of this scaling plays an important role for the determination of the conditions under which mechanical backaction on the optics can be neglected , as we have explained in sec . [ backaction ] . let us now explain how the optical correlation function can be evaluated within this framework . first , note that we can rewrite it as @xmath262 where @xmath263 can be interpreted as an operator with evolution equation @xmath264 and initial condition @xmath265 . since this evolution equation is formally equivalent to the optical master equation ( [ masterdopo ] ) , we can apply c - mop theory directly on @xmath266 , approximating it by a separable operator @xmath267 , with @xmath268 and @xmath269 evolving according to equations ( [ cmop eq app ] ) with @xmath270 replaced by @xmath271 . under a gaussian approximation for @xmath272 similar to ( [ gsamoments ] ) but with expectation values defined with respect to @xmath272 , the evolution equations for the first and second moments of @xmath272 , @xmath273 , @xmath248 , @xmath249 , and their hermitian conjugates ( note that @xmath274 is not hermitian ) form a closed nonlinear system which we can solve again efficiently . note that the initial conditions for these moments , e.g. , @xmath275 , are found from the gaussian c - mop steady - state solutions as explained above . in fig . [ scaling]b we show the evolution of the absolute value of the correlation function @xmath111 at the critical point and for different values of the normalized detuning @xmath45 . time is normalized to @xmath276^{-1}$ ] , and hence the fact that all the curves decay on the same time scale proves that the optical relaxation time scales as @xmath277 at threshold . this again plays a fundamental role when proving that mechanical backaction is negligible , as we discussed in sec . [ backaction ] . _ nonlinear optics _ , r. w. boyd ( academic press , new york , 2003 ) . _ nonclassical effects in quantum optics _ , edited by p. meystre and d. f. walls ( american institute of physics , new york , 1991 ) . _ statistical methods in quantum optics 2 _ , h. j. carmichael ( springer verlag , berlin , 2008 ) . c. navarrete - benlloch , arxiv:1504.05917 . v. s. ilchenko , a. a. savchenkov , a. b. matsko , and l. maleki , j. opt . b * 20 * , 333 ( 2003 ) . v. s. ilchenko , a. a. savchenkov , a. b. matsko , and l. maleki , phys . lett . * 92 * , 043903 ( 2004 ) . a. a. savchenkov , a. b. matsko , m. mohageg , d. v. strekalov , and l. maleki , opt . lett . * 32 * , 157 ( 2007 ) . j. u. frst , d. v. strekalov , d. elser , m. lassen , u. l. andersen , c. marquardt , and g. leuchs , phys . * 104 * , 153901 ( 2010 ) . j. u. frst , d. v. strekalov , d. elser , a. aiello , u. l. andersen , ch . marquardt , and g. leuchs , phys . lett . * 105 * , 263904 ( 2010 ) . j. hofer , a. schliesser , and t. j. kippenberg , phys . rev , a * 82 * , 031804(r ) ( 2010 ) . j. u. frst , d. v. strekalov , d. elser , a. aiello , u. l. andersen , ch . marquardt , and g. leuchs , phys . * 106 * , 113901 ( 2011 ) . t. beckmann , h. linnenbank , h. steigerwald , b. sturman , d. haertle , k. buse , and i. breunig , phys . * 106 * , 143903 ( 2011 ) . c. s. werner , t. beckmann , k. buse , and i. breunig , opt . lett . * 37 * , 4224 ( 2012 ) . m. frtsch , j. u. frst , c. wittmann , d. strekalov , a. aiello , m. v. chekhova , c. silberhorn , g. leuchs , and c. marquardt , nature commun . * 4 * , 1818 ( 2013 ) . c. marquardt , d. strekalov , j. frst , m. frtsch , and g. leuchs , opt . news * 24 * , 38 ( 2013 ) . m. frtsch , g. schunk , j. u. frst , d. strekalov , t. gerrits , m. j. stevens , f. sedlmeir , h. g. l. schwefel , s. w. nam , g. leuchs , and c. marquardt , phys . rev . a * 91 * , 023812 ( 2015 ) . m. frtsch , t. gerrits , m. j. stevens , d. strekalov , g. schunk , j. u. frst , u. vogl , f. sedlmeir , h. g. l. schwefel , g. leuchs , s. w. nam , and c. marquardt , j. opt . * 17 * , 065501 ( 2015 ) . m. aspelmeyer , t. j. kippenberg , and f. marquardt , rev . 86 * , 1391 ( 2014 ) . s. huang and g. s. agarwal , phys . a * 79 * , 013821 ( 2009 ) . s. huang and g. s. agarwal , phys . a * 80 * , 033807 ( 2009 ) . v. peano , h. g. l. schwefel , ch . marquardt , and f. marquardt , phys . lett . * 115 * , 243603 ( 2015 ) . l , y. wu , j. r. johansson , h. jing , j. zhang , and f. nori , phys . * 114 * , 093602 ( 2015 ) . s. pina - otey , f. jimenez , p. degenfeld - schonburg , and c. navarrete - benlloch , arxiv:1412.2521 . p. degenfeld - schonburg and m. j. hartmann , phys . b * 89 * , 245108 ( 2014 ) . p. degenfeld - schonburg , c. navarrete - benlloch , and m. j. hartmann , phys . a * 91 * , 053850 ( 2015 ) . a. mari and j. eisert , phys . lett . * 108 * , 120602 ( 2012 ) . z. leghtas , s. touzard , i. m. pop , a. kou , b. vlastakis , a. petrenko , k. m. sliwa , a. narla , s. shankar , m. j. hatridge , m. reagor , l. frunzio , r. j. schoelkopf , m. mirrahimi , m. h. devoret , science * 347 * , 853 ( 2015 ) . j. d. teufel , t. donner , d. li , j. w. harlow , m. s. allman , k. cicak , a. j. sirois , j. d. whittaker , k. w. lehnert , and r. w. simmonds , nature * 475 * , 359 ( 2011 ) . j. d. teufel , d. li , m. s. allman , k. cicak , a. j. sirois , j. d. whittaker , and r. w. simmonds , nature * 471 * , 204 ( 2011 ) . t. a. palomaki , j. w. harlow , j. d. teufel , r. w. simmonds , and k. w. lehnert , nature * 495 * , 210 ( 2013 ) . p. d. nation , j. suh , and m. p. blencowe , arxiv:1507.00115 . f. marquardt , j. p. chen , a. a. clerk , and s. m. girvin , phys . rev * 99 * , 093902 ( 2007 ) . c. genes , d. vitali , p. tombesi , s. gigan , and m. aspelmeyer , phys . a * 77 * , 033804 ( 2008 ) . s. rips , m. kiffner , i. wilson - rae , and m. j. hartmann , new j. phys . * 14 * , 023042 ( 2012 ) . c. navarrete - benlloch , e. roldn , y. chang , and t. shi , optics express * 22 * , 024010 ( 2014 ) . m. abdi and m. j. hartmann , new j. phys . * 17 * , 013056 ( 2015 ) . _ the theory of open quantum systems _ breuer and f. petruccione ( oxford university press , new york , 2002 ) . _ quantum noise _ , c. w. gardiner and p. zoller ( springer - verlag , berlin heidelberg , 2004 ) . _ statistical methods in quantum optics 1 _ , h. j. carmichael ( springer - verlag , berlin heidelberg , 2002 ) . i. wilson - rae , n. nooshi , w. zwerger , and t. j. kippenberg , phys . rev . lett . * 99 * , 093901 ( 2007 ) . r. rivire , s. delglise , s. weis , e. gavartin , o. arcizet , a. schliesser , and t. j. kippenberg , phys . a * 83 * , 063835 ( 2011 ) . j. m. dobrindt , i. wilson - rae , and t. j. kippenberg , phys . * 101 * , 263602 ( 2008 ) . _ open quantum systems . an introduction _ , a. rivas and s. f. huelga ( springer , 2011 ) . s. g. schirmer and xiaoting wang , phys . a * 81 * , 062306 ( 2010 ) . c. weedbrook , s. pirandola , r. garcia - patron , n. j. cerf , t.c . ralph , j.h . shapiro , and s. lloyd , rev . phys . * 84 * , 621 ( 2012 ) . _ an introduction to the formalism of quantum information with continuous variables _ , c. navarrete - benlloch , ( morgan & claypool publishers and iop publishing , 2015 ) . m. wolinsky and h. j. carmichael , phys . lett . * 60 * , 1836 ( 1988 ) . p. kinsler and p. d. drummond , phys . a * 52 * , 783 ( 1995 ) . o. veits and m. fleischhauer , phys . a * 55 * , 3059 ( 1997 ) .

for a given stochastic process the _ future drawdown process _ is defined by for fixed non - negative , two path - functionals that describe the fluctuations of are the running supremum and running infimum of , the path - functionals and are of interest in various areas of application , including financial mathematics and queueing theory . in the case that is a lvy process with strictly positive mean , we find the exact asymptotic decay as of the tail probabilities and of and both when the jumps satisfy the cramr assumption and in the heavy - tailed case . furthermore , when the jumps of the lvy process are of single sign , we identify the one - dimensional distributions in terms of the scale function of . by way of example , we derive explicit results for the black scholes samuelson model .

1409.3780

in recent times , various pricing models with jumps have been put forward which address the shortcomings of diffusion models in representing the risk related to large market movements ( see e.g. @xcite ) . such models allow for a more realistic representation of price dynamics and a greater flexibility in calibration to market prices and in reproducing a wide variety of implied volatility skews and smiles . an important indicator for the riskiness and effectiveness of an investment strategy is the drawdown , which is the distance of the current value away from the maximum value it has attained to date . the drawdown may also be seen as a measure of regret as it represents the hypothetical gain the investor would have made had he sold the asset at the moment the current maximum was attained . a number of commonly used trading rules are based on the drawdown ( see e.g. @xcite ) , while ( maximum ) drawdowns have also been applied as risk - measures ( see @xcite ) . drawdown processes ( also called reflected processes ) are also encountered in various other areas , such as applied probability , queueing theory and mathematical genetics . see @xcite and references therein for further applications and results concerning drawup and drawdown processes . in this paper we consider a number of path - functionals of the increments of a given stochastic process @xmath0 that are closely related to the drawdown and drawup processes . in particular , we consider a _ future drawdown _ process @xmath17 which is defined as a two - parameter process with value at @xmath18 equal to the smallest increment @xmath19 for @xmath20 in the interval @xmath21 $ ] , _ i.e. _ @xmath22 together with its running supremum and infimum described by @xmath8 similarly , we define a _ future drawup process _ of @xmath11 by @xmath23 with its extrema @xmath24 we denote the corresponding quantities for an infinite horizon by @xmath25 note that for any nonnegative @xmath3 we have @xmath26 and @xmath27 , where @xmath28 denotes the quantity calculated for the dual process @xmath29 , so that in the analysis we may restrict to either the future drawdown or drawup process . in this paper we focus on the case that @xmath11 is a lvy process . the future drawdown and drawup processes are of interest in various applications , including in financial risk analysis and queueing models . we note that , under an exponential lvy model @xmath30 for the stock price , the random variables @xmath9 and @xmath10 are path - dependent risk indicators : @xmath4 is the _ lowest _ future log - return @xmath31 achieved at some @xmath20 in the time - window @xmath21 $ ] , while @xmath32 and @xmath33 are the maximal and minimal such future returns for @xmath34 ranging over @xmath35 $ ] . another application comes from telecommunications and queueing models , where @xmath36 and @xmath37 describe the supremum and the infimum of the workload process over a finite time horizon @xmath34 in a fluid model with netput @xmath11 , respectively ( see for details @xcite ) . in the mentioned applications it is of interest to characterise the laws , and in particular the tail - probabilities , of the random variables @xmath9 , @xmath10 , @xmath38 and @xmath39 for finite and infinite horizons @xmath3 . restricting ourselves to the case @xmath40 we identify the _ exact _ asymptotic decay as @xmath12 of the tail probabilities @xmath41 and @xmath42 of @xmath9 and @xmath10 . we do so in the distinct cases of a light - tailed and a heavy - tailed lvy measure . in the former setting we also consider the asymptotics when @xmath43 and @xmath3 tend to infinity in a fixed proportion . more specifically , in the case of light - tailed jumps ( i.e. , assumptions [ a2 ] and [ a3 ] below are satisfied ) we prove that @xmath44 for some constant @xmath45 ( which is identified explicitly ) and @xmath46 solving @xmath47 where @xmath48 denotes the laplace exponent of @xmath11 . in the case that @xmath3 and @xmath43 tend to infinity in fixed proportion ( i.e. , if @xmath49 such that @xmath50 for some feasible proportion @xmath51 ) , then we show that for @xmath52 @xmath53 for some other constant @xmath54 . similar results are obtained for @xmath55 . we also describe the law of @xmath56 _ conditional _ on @xmath57 or @xmath58 being large , which gives rise to `` asymptotic drawup measures '' @xmath59 and @xmath60 . similar results are established for @xmath39 and @xmath61 leading to `` asymptotic drawdown measures '' @xmath62 and @xmath63 . in a complementary case that the lvy measure @xmath64 of @xmath11 is in the class @xmath65 of tail - equivalent measures then , under some additional regularity conditions , we show @xmath66 for some explicit constant @xmath67 , and derive a similar result for @xmath55 . when the jumps of @xmath11 are all of single sign , we explicitly identify the laplace transform in time of the one - dimensional distributions in terms of the scale function of @xmath11 . as example , we analyze in detail ( future ) drawdowns and drawups under the black - scholes model , identifying in particular the mean of the value @xmath68 under the measure @xmath62 and the laws of @xmath69 and @xmath70 . the remainder of the paper is organized as follows . in section [ main ] we present the main representation of drawup and drawdown processes . in section [ sec : ascramer ] the cramr asymptotics is identified which produce the drawup and drawdown measures described in section [ sec : asmeasures ] . the heavy - tailed case is analyzed in section [ sec : heavy ] . in section [ examplesexact ] we find further results in the spectrally one - sided case and we conclude the paper with more explicit expressions of future drawups and drawdowns in the black scholes samuelson model . let @xmath71 be a general lvy process ( i.e. a process with stationary and independent increments with cdlg paths such that @xmath72 ) defined on some filtered probability space ( @xmath73 with @xmath74 denoting the completed filtration generated by @xmath11 . the law of @xmath11 is determined by its characteristic exponent @xmath75 which is the map @xmath76 of @xmath11 that satisfies @xmath77= \exp\psi(\theta)$ ] . the drawdown and drawup processes of @xmath11 , @xmath78 and @xmath79 , are path - functionals of the increments of @xmath11 given by @xmath80 with @xmath81 and @xmath82 . we note that the drawdown @xmath83 and drawup @xmath84 at time @xmath34 are equal to the largest of all increments @xmath85 and the negative of the smallest increment of such increments . by replacing the interval @xmath86 $ ] by intervals of the form @xmath21 $ ] , we arrive at the following generalizations which we will call future drawdown and drawup processes : the _ future drawdown _ and _ future drawup _ process of @xmath11 , @xmath87 and @xmath88 , are given by @xmath89 further , we will use the following notation : @xmath90 and @xmath91 . before turning to the analysis of the future drawdown and drawup processes , we first note a number of facts concerning drawup and drawdown processes that follow from the fluctuation theory of lvy processes . the marginal distributions of the drawup @xmath84 and drawdown @xmath83 , @xmath92 , can be expressed in terms of the marginal distributions of @xmath11 by deploying the wiener hopf factorisation of @xmath11 , according to which the characteristic exponent @xmath75 of @xmath11 is related to the marginal distributions of the running supremum and running infimum of @xmath11 at an exponential random time @xmath93 ( independent of @xmath11 ) as follows : @xmath94 { \mathbb{e}}[e^{i\theta \unl x_{\mathbf e_q}}],\qquad q>0 , \theta\in\mathbb r.\ ] ] using the duality lemma ( see e.g. ( * ? ? ? vi.3 ) ) , the wiener hopf factorization may be expressed in terms of the drawdown and drawup processes by @xmath95 { \mathbb{e}}[e^{-i\theta d_{\mathbf e_q}}],\qquad q>0 , \theta\in\mathbb r.\ ] ] if @xmath96 $ ] is strictly negative , it follows from the duality lemma that @xmath84 converges in distribution to a random variable @xmath97 as @xmath34 tends to infinity , and that @xmath97 is in distribution equal to @xmath98 . the laplace transform of @xmath98 is given explicitly in terms of the laplace exponent @xmath99 of the ascending ladder process @xmath100 . the ladder time process @xmath101 is the right - continuous inverse of the local time @xmath102 of @xmath103 at zero , while the corresponding ladder - height process @xmath104 is given by @xmath105 for all @xmath106 for which @xmath107 is finite , and defined to be @xmath108 otherwise . the laplace transform of @xmath98 ( and thus @xmath97 ) is given as follows : @xmath109 = { \mathbb{e}}[\te{-u u^*_0 } ] = \frac{\kappa(0,0)}{\kappa(0,u)},\q \text{with}\\ & & \kappa(\beta , \theta ) = -\log { \mathbb{e}}[\exp\{-\beta l^{-1}_1-\theta h_1\}\i_{\{h(1)<\infty\ } } ] , \nonumber\end{aligned}\ ] ] where @xmath110 denotes the indicator of the set @xmath111 . the laplace exponent of the ascending ladder process @xmath112 corresponding to the dual process @xmath29 of @xmath11 we denote by @xmath113 $ ] . the fluctuations of the processes @xmath114 and @xmath115 for fixed nonnegative @xmath3 are described by their running supremum processes @xmath116 , @xmath117 and running infimum processes @xmath118 , @xmath119 which are defined for @xmath106 by @xmath120 we denote the infinite - horizon versions by @xmath25 a first step in the study of the random variables @xmath38 , @xmath39 , @xmath9 and @xmath10 is the following result . in the cases that ( i ) the largest value of @xmath11 up to time @xmath121 has already been attained before time @xmath122 so that the path - functional @xmath123 is zero ( top picture ) or ( ii ) @xmath11 attains a new maximum between @xmath122 and @xmath121 and the path - functional @xmath123 is strictly positive ( bottom picture ) . these pictures help clarify the expression for @xmath123 in proposition [ repr ] . for further details , see its proof below . [ fig : exc],title="fig : " ] in the cases that ( i ) the largest value of @xmath11 up to time @xmath121 has already been attained before time @xmath122 so that the path - functional @xmath123 is zero ( top picture ) or ( ii ) @xmath11 attains a new maximum between @xmath122 and @xmath121 and the path - functional @xmath123 is strictly positive ( bottom picture ) . these pictures help clarify the expression for @xmath123 in proposition [ repr ] . for further details , see its proof below . [ fig : exc],title="fig : " ] [ repr ] let @xmath124 . we have the following representations : @xmath125 where @xmath126 denotes equality in distribution , @xmath127 , @xmath128 and where @xmath129 and @xmath130 are independent of @xmath131 . as a consequence , when @xmath96\in[-\infty,0)$ ] @xmath132\in(0,\infty])$ ] , then @xmath133 and @xmath134 @xmath135 and @xmath136 are finite @xmath137-a.s . , respectively . [ dualextrema ] * as noted in the introduction , it suffices to analyze @xmath38 and @xmath39 , since we have @xmath26 and @xmath27 , where @xmath138 and @xmath139 denote the extrema @xmath140 and @xmath141 corresponding to @xmath142 . * we observe that , as @xmath11 has independent increments , @xmath84 and @xmath83 are independent of @xmath143 and @xmath4 for any @xmath144 $ ] . furthermore , we note that , for any @xmath106 , @xmath145 is finite a.s . if and only if @xmath96 $ ] is strictly positive , as this condition implies that @xmath146 converges to infinity a.s . as @xmath147 similarly , @xmath148 is finite a.s . precisely if @xmath96 $ ] is strictly negative . * extending @xmath11 from @xmath149 to a two - sided version on @xmath150 and using a time - reversal argument we find that @xmath151 indeed , using the change of variables @xmath152 and @xmath153 we see that @xmath154 the result for @xmath155 follows similarly . + the random variables @xmath156 and @xmath155 arise in a queueing application . indeed , the workload process @xmath157 of a queue with net input process @xmath11 ( i.e. input less output ) evolves according to the process @xmath11 reflected at its infimum , _ i.e. _ , @xmath158 . if we assume that the workload process is stationary ( _ i.e. _ , @xmath159 follows the stationary distribution , which is equal to @xmath160 ( see @xcite ) ) , then the workload @xmath157 is given by : @xmath161 and @xmath156 and @xmath155 describe the supremum and infimum of the workload process @xmath162 over a finite time horizon @xmath122 , respectively . for details on queues driven by a lvy process ; see the review paper @xcite . * we note @xmath163 ( see for example ( * ? ? ? 15 , p. 170 ) and ( * ? ? ? 13 , p. 169 ) ) . in view of remark [ dualextrema](i ) it suffices to establish the statements concerning @xmath164 and @xmath165 . writing @xmath166=[t , t]\cup [ t , t+s]$ ] for given @xmath167 leads to @xmath168}(x_s - x_t)+x_t - x_t,\sup_{t\leq s\leq t}(x_s - x_t)\ri\}.\end{aligned}\ ] ] since @xmath169}(x_s - x_t)$ ] is independent of @xmath170 and is equal in distribution to @xmath171 , we find that @xmath172 is equal in distribution to @xmath173 which proves the first identity in . for the second identity in note that , for fixed nonnegative @xmath174 , the function @xmath175 attains its infimum over @xmath35 $ ] at @xmath176 or @xmath177 where @xmath178 . in the case that @xmath179 ( i.e. , when @xmath180 ) we have @xmath181 ( see figure [ fig : exc ] , top picture ) and @xmath182 while in the case that @xmath183 ( see figure [ fig : exc ] , bottom picture ) we find @xmath184 combining these two observations we deduce that @xmath185 where @xmath186 denotes an independent copy of @xmath11 . since @xmath187 is equal in distribution to @xmath171 , the expression for @xmath188 follows . taking @xmath189 in and noting that @xmath190 is finite @xmath137-a.s . if @xmath96\in[-\infty,0)$ ] we conclude that also @xmath156 and @xmath155 are @xmath137-a.s . in this section we study the asymptotics of the tail probabilities @xmath191 and @xmath192 in the case that the lvy measure is light - tailed . more precisely , in this section we wile make the following assumptions . [ a2]the cramr condition holds , _ i.e. _ , @xmath193=1$,}\ ] ] the mean of @xmath194 is negative and finite , @xmath96\in(-\infty,0)$ ] , and @xmath195<\infty$ ] . [ a3 ] @xmath11 has non - monotone paths and either @xmath196 is regular for @xmath197 or the lvy measure of @xmath11 is non - lattice . [ a4 ] the @xmath198-exponential moment of @xmath11 is finite , _ @xmath199 < \infty.\ ] ] under condition the characteristic exponent @xmath200 can be extended to the strip @xmath201\}$ ] in the complex plane , by analytical continuation and continuous extension . if in addition is satisfied , @xmath75 can be extended to @xmath202\}$ ] . the laplace exponent @xmath203 $ ] of @xmath11 is finite on the maximal domain @xmath204 , which contains the intervals @xmath205 $ ] and @xmath206 $ ] under conditions and respectively . restricted to the interior @xmath207 , the map @xmath208 is convex and differentiable , with derivative @xmath209 . , @xmath210 is understood to be @xmath211 . ] by the strict convexity of @xmath48 , it follows that @xmath212 is strictly increasing on @xmath207 and we denote by @xmath213 its right - inverse function . under [ ] the wiener hopf factorisation remains valid for @xmath214 in the strip @xmath215 [ @xmath216 , respectively ] . [ lem : wh ] ( i ) if assumption [ a2 ] is satisfied , it holds @xmath217 = { \mathbb{e}}[e^{-\gamma d_{\mathbf e_q}}]^{-1}$ ] and @xmath218 < \infty.\ ] ] \(ii ) if assumption [ a4 ] is satisfied , it holds @xmath219 < \infty.\ ] ] \(iii ) furthermore , and remain valid when @xmath220 is replaced by any @xmath221 . it follows from the wiener hopf factorisation that @xmath222 = q ( q - \psi(\theta))^{-1}{\mathbb{e}}[e^{-i\theta d_{\mathbf e_q}}]^{-1}\ ] ] for all @xmath214 in the strip @xmath223 . the continuity of @xmath224 $ ] in the half - plane @xmath225 , the convexity of @xmath226 restricted to @xmath227\}$ ] and the fact that @xmath228 then yield . the proof of the remaining assertions is similar and omitted . in @xcite it was shown that under assumptions [ a2 ] and [ a3 ] , cramr s estimate holds for the lvy process @xmath11 , i.e. , @xmath229_{|\theta=-\gamma}}>0 , \qquad \text{as $ y\to\infty$},\ ] ] where we write @xmath230 as @xmath12 if @xmath231 . cramr s estimate can be extended to the decay of the finite time probability @xmath232 when @xmath233 jointly tend to infinity in fixed proportion . for a given proportion @xmath234 the rate of decay is either equal to @xmath220 or to @xmath235 which is given in terms of the convex conjugate @xmath236 of @xmath48 given by @xmath237 here , the proportions @xmath234 are to be positive and lie in the range of @xmath212 . this leads to the following definition . a proportion @xmath238 is _ feasible _ if there exists a @xmath239 such that @xmath240 . more specifically , in @xcite it was shown that the hglund s estimates hold for @xmath11 , _ i.e. _ , if assumptions [ a2 ] and [ a3 ] are satisfied and the proportion @xmath234 is feasible , then for @xmath43 and @xmath3 tending to infinity such that @xmath241 we have @xmath242}{\eta_v \mathbb{e}[\te{\xi_v h_1-\eta_v l^{-1}_1}h_1\mbf 1_{(h_1<\infty ) } ] } \times \frac{1}{\xi_v \sqrt{2\pi\psi''(\xi_v)}},\nonumber\end{aligned}\ ] ] and with @xmath243 , where we write @xmath244 if @xmath245 and we used that @xmath246 for @xmath247 . in particular , note that , if @xmath234 is feasible with @xmath248 , then @xmath249 so that @xmath250 . the exact asymptotic decay of the tail probabilities of @xmath39 and @xmath38 is given as follows : [ cramer ] suppose that assumptions [ a2 ] and [ a3 ] hold , and let @xmath251 . \(i ) then the following limit hold true : @xmath252\ , e^{-\gamma x},\quad x\to\infty.\ ] ] if furthermore assumption [ a4 ] holds , then we have @xmath253\ , e^{-\gamma x}\quad x\to\infty.\ ] ] \(ii ) if @xmath43 and @xmath3 tend to infinity such that @xmath241 for some feasible proportion @xmath234 then we have the following limits : @xmath254\ , e^{-\gamma x } , & \text{if $ 0<v<\psi'(\gamma)$},\\ \wt c_v{\mathbb{e}}[e^{\gamma(v ) d_t } ] x^{-1/2}e^{-\gamma(v ) x } & \text{if $ v>\psi'(\gamma)$ } , \end{cases}\\ { \mathbb{p}}(\overline u^*_{t , s}>x ) & \sim & \begin{cases } c_\gamma{\mathbb{e}}[e^{\gamma u_t}]\ , e^{-\gamma x } & \text{if $ 0<v<\psi'(\gamma)$},\\ \wt c_v{\mathbb{e}}[e^{\gamma(v ) u_t}]x^{-1/2 } e^{-\gamma(v ) x } & \text{if $ v>\psi'(\gamma)$}. \end{cases}\label{clim}\end{aligned}\ ] ] in specific cases the wiener hopf factors are known , so that the constants in can be identified . * if @xmath11 is spectrally negative , then @xmath255 and @xmath256 = \frac{\phi(q)}{\phi(q)-\gamma},\q q>0,\ ] ] where @xmath257 , with @xmath258 , @xmath259 , the largest root of the equation @xmath260 where @xmath261 $ ] is the laplace exponent . these expressions hold since @xmath262 follows an exponential distribution with parameter @xmath258 . by inverting the laplace transforms in @xmath263 we find the following explicit expression in terms of the one - dimensional distributions of @xmath11 : @xmath264 = 1+\gamma\int_0^t \mathbb{e}[e^{\gamma x_s } x_s^+]s^{-1}{\mathrm{d}}s.\ ] ] here we used that , on account of kendall s identity @xmath265 for the first passage time @xmath266 , it follows that @xmath267t^{-1}{\mathrm{d}}t = \frac{1}{\phi(q)-\gamma}.\label{invltmain}\ ] ] moreover , if @xmath268 , @xmath269 = \frac{q}{q - \psi(-\gamma ) } \left[1 + \frac{\gamma}{\phi(q)}\right].\ ] ] hence , we have @xmath270 = e^{t\psi(-\gamma ) } + \int_0^t e^{(t - s)\psi(-\gamma)}{\mathbb{e}}[e^{\gamma x_s}x_s^+]s^{-1}{\mathrm{d}}s.\ ] ] * if @xmath11 is spectrally positive , then @xmath271 and @xmath272 = { \mathbb{e}}[e^{-\gamma d_{\mathbf e_q}}]^{-1 } = \frac{\phi(q)+\gamma}{\phi(q)},\ ] ] where @xmath220 and @xmath258 , @xmath259 , are the largest roots of @xmath273 and @xmath274 for @xmath275 $ ] , and we used that @xmath276 . hence @xmath277 = 1+\gamma\int_0^t \mathbb{e}[{x}_s^-]s^{-1}{\mathrm{d}}s,\qquad { \mathbb{e}}[e^{\gamma d_t}]=1-\gamma\int_0^t{\mathbb{e}}[e^{-\gamma x_s}x^-_t]s^{-1}\,{\mathrm{d}}s.\ ] ] * the wiener hopf factors may also be identified for the meromorphic lvy processes ( * ? ? ? 1 ) : @xmath278=\prod_{n\geq 1}\frac{1-\frac{\gamma}{\rho_n}}{1-\frac{\gamma}{\zeta_n(q ) } } , \qquad{\mathbb{e}}[e^{\gamma d_{{\mathbf{e}_q}}}]= \prod_{n\geq 1}\frac{1-\frac{\gamma}{\hat{\rho}_n}}{1-\frac{\gamma}{\hat{\zeta}_n(q)}},\ ] ] where @xmath279 are the poles of @xmath75 ( which is meromorphic ) and @xmath280 are the roots of @xmath281 . the above laplace transforms in @xmath263 can be numerically inverted giving @xmath282 $ ] and @xmath283 $ ] ( see for details ( * ? ? ? \(i ) from proposition [ repr ] it follows that for @xmath284 @xmath285}{\mathbb{p}}(u_s\leq x - z){\mathbb{p}}(u_t\in{\mathrm{d}}z,\overline{u}_t\leq x ) \leftrightarrow \\ { \mathbb{p}}(\overline u^*_{t , s } > x ) = { \mathbb{p}}(\overline u_t > x ) + \int_{[0,x]}{\mathbb{p}}(u_s > x - z){\mathbb{p}}(u_t\in{\mathrm{d}}z,\overline{u}_t\leq x)\label{ustarr}\end{gathered}\ ] ] by letting @xmath286 in and using @xmath287 ( see ) we arrive at the identity @xmath288}{\mathbb{p}}(u^*_0>x - z){\mathbb{p}}(u_t\in{\mathrm{d}}z,\overline{u}_t\leq x).\ ] ] denote by @xmath289 the cramr measure that is defined on @xmath290 by @xmath291 $ ] , @xmath292 . the cramr asymptotic decay implies that @xmath293 \simeq c_\gamma , \quad \text{as $ x\to\infty$},\ ] ] where @xmath294 is the first - passage time of @xmath11 over the level @xmath43 . in view of the facts that @xmath295 is non - increasing and @xmath296 for @xmath297 and any @xmath298 , for @xmath12 if @xmath299 as @xmath12 . ] we find @xmath300\\ \nn & = & e^{-\gamma x}{\mathbb{e}}^{(\gamma)}[e^{-\gamma(u_{t^u_x}+\underline x_{t^u_x}-x)}\i_{\{t^u_x < t\}}]\\ & \leq & e^{-\gamma x}{\mathbb{e}}^{(\gamma)}[e^{-\gamma \underline x_{t}}\i_{\{t^u_x < t\ } } ] = { \rm o}(e^{-\gamma x}),\quad\mbox{as } x\rightarrow\infty , \label{oegx}\end{aligned}\ ] ] where the expectation in converges to zero by virtue of the dominated convergence theorem and the facts that @xmath301 = { \mathbb{e}}[e^{\gamma u_t}]<\infty$ ] ( by lemma [ lem : wh ] ) and @xmath302 @xmath289-a.s . as @xmath303 ( as @xmath11 drifts to infinity @xmath289-a.s . ) . combining with , the cramr asymptotics and the dominated convergence theorem yield @xmath304 , \q t\in\mbb r_+.\ ] ] as far as @xmath134 is concerned , we deduce from proposition [ repr ] , the cramr asymptotics , lemma [ lem : wh](ii ) and the dominated convergence theorem that @xmath305.\end{aligned}\ ] ] \(ii ) let @xmath234 be a feasible proportion . the proof follows by a line of reasoning that is analogous to the one given in part ( i ) , deploying hglund s estimate instead of cramr s estimate . in particular , combining , , and the dominated convergence theorem shows that when @xmath306 @xmath307.\end{aligned}\ ] ] turning to the case @xmath248 , we note that a similar reasoning as was used to prove and the fact that @xmath234 is feasible shows that , for some @xmath308 sufficiently small , @xmath309 and @xmath310 are @xmath311 as @xmath12 . hence , for any @xmath312 and @xmath313 we have @xmath314 } ( x - z)^{1/2}e^{\gamma(v ) ( x - z)}{\mathbb{p}}(u_s > x - z)e^{\gamma(v)z}{\mathbb{p}}(u_t\in\td z ) + i_\epsilon(x ) \\ & & \text{with}\ i_\epsilon(x ) = x^{1/2}e^{\gamma(v)x}\int_{[x-\epsilon , x]}{\mathbb{p}}(u_s > x - z){\mathbb{p}}(u_t\in\td z ) . \nonumber\end{aligned}\ ] ] we have that @xmath315 in view of the bounds @xmath316,\ ] ] the upper bound of which converges to zero when @xmath43 tends to infinity . hence , using and the dominated convergence theorem yields @xmath317\end{aligned}\ ] ] for any @xmath318 , where the expectation is finite by lemma [ lem : wh](iii ) and the fact @xmath250 . the proof of the asymptotics of the tail - probability @xmath319 is similar and omitted . conditional on @xmath320 being large , or on @xmath321 being large @xmath56 admits a limit in distribution , as is shown next . these limits are given by the `` drawup - measures '' @xmath322 and the `` drawdown measures '' @xmath323 , @xmath324 , that are defined as follows on the measurable space @xmath290 : @xmath325 } \mathbf 1_a\ri],\qquad \underline{\mathbb{p}}^{(s)}(a ) = { \mathbb{e}}\le[\frac{e^{s d_t}}{{\mathbb{e}}[e^{s d_t } ] } \mathbf 1_a\ri ] , \qquad a\in\mathcal f_t , s\in\theta.\ ] ] [ thm : asymp ] suppose assumptions [ a2 ] and [ a3 ] hold , and let @xmath251 . \(i ) then , conditional on @xmath326 and on @xmath327 , @xmath56 converges in distribution as @xmath12 : @xmath328 & \simeq & \overline { \mathbb{p}}^{(\gamma)}[x_t\leq x],\\ { \mathbb{p}}[x_t \leq x| \underline u^*_t > x ] & \simeq & \underline { \mathbb{p}}^{(\gamma)}[x_t \leq x].\label{clim2}\end{aligned}\ ] ] \(ii ) if @xmath43 and @xmath3 tend to infinity such that @xmath241 where @xmath234 is feasible then the following limits hold true : @xmath329 & \sim & \begin{cases } \overline { \mathbb{p}}^{(\gamma)}[x_s\leq x ] & \text{if $ 0<v<\psi'(\gamma)$,}\\ \overline { \mathbb{p}}^{(\gamma(v))}[x_t\leq x ] & \text{if $ \psi'(\gamma)>v$ , } \end{cases}\\ { \mathbb{p}}[x_t \leq x| \underline u^*_{t , s } > x ] & \sim & \begin{cases } \underline { \mathbb{p}}^{(\gamma)}[x_t\leq x ] & \text{if $ 0<v<\psi'(\gamma)$,}\\ \underline { \mathbb{p}}^{(\gamma(v))}[x_t \leq x ] & \text{if $ \psi'(\gamma)>v$}.\label{clim22 } \end{cases}\end{aligned}\ ] ] \(i ) by following a similar line of reasoning as the proof of theorem [ cramer ] it is straightforward to show that for @xmath330 $ ] , as @xmath12 , @xmath331 & \simeq & c_\gamma e^{-\gamma x}{\mathbb{e}}[e^{u x_t + \gamma u_t}],\\ { \mathbb{e}}[\te{u x_t}\mathbf 1_{\{\underline u^*_t > x\ } } ] & \simeq & c_\gamma e^{-\gamma x}{\mathbb{e}}[e^{u x_t + \gamma d_t}].\end{aligned}\ ] ] bayes lemma then yields the stated identities . the proof of ( ii ) is similar and is therefore omitted . we continue the study of the asymptotic behaviour of the tail probabilities of @xmath133 and @xmath134 in the case that the lvy measure @xmath64 of @xmath11 belongs to the class @xmath332 of convolution - equivalent measures which , we recall , is a subset of the following class @xmath333 . [ def3](class @xmath333 ) for a parameter @xmath334 we say that measure @xmath335 with tail @xmath336 belongs to class @xmath333 if * @xmath337 for each @xmath338 , * @xmath339 , * @xmath340 ( then assumed of span @xmath341 ) . ( class @xmath332 ) we say that @xmath335 belongs to class @xmath332 if * @xmath342 ; * for some @xmath343 , we have @xmath344 where @xmath345 and @xmath346 denotes convolution . the asymptotics are derived under conditions on the lvy measure @xmath347 of the ladder height process @xmath348 , which according to the vigon @xcite identity is related to the lvy measures @xmath64 of @xmath11 by @xmath349 for the renewal measure @xmath350 throughout this section we assume that for some fixed @xmath351 the following three conditions hold true : @xmath352 [ heavy ] assume that @xmath96<0 $ ] and let @xmath251 . under conditions we have : @xmath353 where @xmath354 and @xmath355 are given by @xmath356 + \int_0^t { \mathbb{e}}\left[e^{\alpha \underline{x}_{t - s}}\right]^{-1 } \mu(\td s ) , \q { \rm const}^-= { \mathbb{e}}[e^{-\alpha \underline{x}_{t}}],\ ] ] with the borel measure @xmath357 on @xmath358 given by @xmath359{\mathrm{d}}u.\ ] ] [ remmq ] * by straightforward calculations it can be verified that @xmath360 where @xmath361 denotes the laplace stieltjes transform of the measure @xmath357 . * if @xmath362 for @xmath363 then holds and @xmath364 see ( * ? ? ? * if @xmath11 is spectrally positive , then from : @xmath365 = 1+\alpha\int_0^t \mathbb{e}[{x}_s^-]s^{-1}{\mathrm{d}}s,\qquad { \mathbb{e}}[e^{-\alpha \underline{x}_{t } } ] = 1+\alpha\int_0^t{\mathbb{e}}[e^{\alpha x_s}x^-_s]s^{-1}\,{\mathrm{d}}s.\ ] ] moreover , since @xmath366 and @xmath367 , we have @xmath368 we first prove the statement concerning @xmath164 . the starting point of the proof is to take the identity noted earlier in and replace the fixed time @xmath122 by an independent exponential random variable @xmath369 with parameter @xmath263 , which yields @xmath370}{\mathbb{p}}(u^*_0>x - z){\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z,\overline{u}_{{\mathbf{e}_q}}\leq x).\ ] ] we show that both terms on the right - hand side of are asymptotically equivalent to the tail - measure @xmath371 of the ladder process @xmath348 as @xmath12 and identify the constant . throughout the proof we denote the first upward and downward passage times of @xmath11 across the level @xmath43 by @xmath372 and @xmath373 . to establish this result it suffices to show asymptotic equivalence of the two terms on the right - hand side of to the probability @xmath374 , since it is known from ( * ? ? ? 4.1 ) and ( * ? ? ? ( 5.6 ) ) that under the conditions stated in the theorem @xmath375 with the interpretation @xmath376 for @xmath377 . note that the constant in is strictly positive for all @xmath378 by the condition and @xmath379 ( as @xmath96 $ ] is strictly negative ) . we treat both terms separately , starting with the first term . we first derive upper and lower bounds for the ratio @xmath380 . by an application of the strong markov property and the definition of @xmath381 we have @xmath382 where in the last line we used that @xmath383 and @xmath384 are independent ( by the wiener hopf factorisation ) and @xmath384 and @xmath385 have the same distribution . hence we find from and that @xmath386 the first terms on the right - hand sides of and may be simplified by using that , by the markov property , we have @xmath387.\end{aligned}\ ] ] furthermore , since @xmath388 we note that @xmath389 from the dominated convergence theorem and definition [ def3](ii)(iii ) it then follows that @xmath390}{{\mathbb{p}}(\tau_x^+<{\mathbf{e}_q } ) } = { \mathbb{e}}\left[e^{\alpha x_{\tau_{-\epsilon}^-}}\i_{\{\tau_{-\epsilon}^-<{\mathbf{e}_q}\}}\right],\ ] ] and an application of the markov property yields @xmath391=\frac{{\mathbb{e}}\left[e^{\alpha \underline{x}_{{\mathbf{e}_q}}}\i_{\{\tau_{-\epsilon}^-<{\mathbf{e}_q}\}}\right ] } { { \mathbb{e}}\left[e^{\alpha \underline{x}_{{\mathbf{e}_q}}}\right]}.\ ] ] taking first @xmath12 in and and using , ( [ posr ] ) , and ( [ levyconference ] ) and that @xmath392=0 $ ] we find @xmath393}{{\mathbb{e}}\left[e^{\alpha \underline{x}_{{\mathbf{e}_q}}}\right ] } \leq \liminf_{x\to\infty } \frac{{\mathbb{p}}(\overline{u}_{{\mathbf{e}_q}}>x)}{{\mathbb{p}}(\tau_x^+<{\mathbf{e}_q } ) } \leq \limsup_{x\to\infty } \frac{{\mathbb{p}}(\overline{u}_{{\mathbf{e}_q}}>x)}{{\mathbb{p}}(\tau_x^+<{\mathbf{e}_q } ) } \leq \frac{{\mathbb{e}}\left[e^{\alpha \underline{x}_{{\mathbf{e}_q}}}\big| \tau_{-\epsilon}^- > { \mathbf{e}_q}\right]}{{\mathbb{e}}\left[e^{\alpha \underline{x}_{{\mathbf{e}_q}}}\right ] } + 1 - e^{-\alpha\epsilon}.\ ] ] letting subsequently @xmath394 and using @xmath395 = 1,\ ] ] which in turn holds as the conditional expectation is bounded above by @xmath341 and bounded below by @xmath396 , we get the following asymptotics : @xmath397}.\nonumber\end{aligned}\ ] ] next , we turn to the proof of the asymptotic decay of the second term on the right - hand side of . note that @xmath398}{\mathbb{p}}({\overline{x}}_\infty > x - z){\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z,\overline{u}_{{\mathbf{e}_q}}\leq x)}\\&&=\left(\int_{[0,y^\prime]}+\int_{(y^\prime , x - y^\prime]}+\int_{(x - y^\prime , x]}\right){\mathbb{p}}({\overline{x}}_\infty > x - z){\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z,\overline{u}_{{\mathbf{e}_q}}\leq x).\label{sumint}\end{aligned}\ ] ] we next show that the second and third inegral of the right - hand side of tend to zero as we let first @xmath43 and then @xmath399 tend to infinity . indeed , concerning the second integral we use , definition [ def3](ii)(iii ) and to show that @xmath400}{\mathbb{p}}({\overline{x}}_\infty > x - z){\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z,\overline{u}_{{\mathbf{e}_q}}\leq x)}{{\mathbb{p}}(\tau_x^+<\infty)}=\int_{(y^\prime,\infty ) } e^{\alpha z } { \mathbb{p}}(\overline{x}_{{\mathbf{e}_q } } \in{\mathrm{d}}z),\ ] ] which tends to @xmath196 as @xmath401 . for the third integral , we obtain the bound @xmath402}{\mathbb{p}}({\overline{x}}_\infty > x - z){\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z,\overline{u}_{{\mathbf{e}_q}}\leq x)&\leq & { \mathbb{p}}({\overline{x}}_\infty > y^\prime){\mathbb{p}}(\overline{x}_{{\mathbf{e}_q } } > x - y^\prime)\\ & \leq & { \mathbb{p}}(\tau^+_{y^\prime } < \infty){\mathbb{p}}(\tau^+_{x - y^\prime}<\infty).\end{aligned}\ ] ] after dividing the integral in the display by @xmath403 and letting first @xmath12 and then @xmath401 , it tends to zero . finally , the first integral on the right - hand side of is asymptotically of the same order as the left - hand side . indeed , using and definition [ def3](ii)(iii ) , and the dominated convergence theorem we find @xmath404}{\mathbb{p}}({\overline{x}}_\infty > x - z){\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z,\overline{u}_{{\mathbf{e}_q}}\leq x)}{{\mathbb{p}}(\tau_x^+<\infty ) } = \int_{[0,y^\prime]}e^{\alpha z}{\mathbb{p}}(u_{{\mathbf{e}_q}}\in{\mathrm{d}}z),\ ] ] which converges to @xmath405:=\wt b_q$ ] as @xmath401 . by combining the previous estimates we have the following asymptotics of the tail probability @xmath406 : @xmath407 noting that the right - hand side of is a pointwise limit of laplace transforms of measures and is itself such a laplace transform , it follows from ( an extension of ) the continuity theorem ( see ( * ? ? ? 15.5.2 ) ) that the corresponding measures also converge to the limiting measure with laplace transform given by @xmath408 . hence the first assertion of the theorem follows by inverting the laplace transform @xmath408 ( see remark [ remmq ] ) . concerning @xmath134 , note that by we have @xmath409}{\mathbb{p}}(\tau_{x+z}^+<\infty){\mathbb{p}}(\underline{x}_t\in{\mathrm{d}}z).\ ] ] asymptotics , the dominated convergence theorem and part ( ii ) and ( iii ) of definition [ def3 ] establish that the asymptotic decay of @xmath410 is as stated . from proposition [ repr ] it follows that the distributions of @xmath411 and @xmath412 can be identified if one is able to identify the laws of the finite time supremum and the resolvent of the lvy process reflected at its infimum . in the case of a spectrally one - sided lvy process @xmath11 such explicit expressions are provided by existing fluctuation theory . in this section we suppose that @xmath11 is spectrally negative ( the case of spectrally positive lvy process follows from considering the dual of @xmath11 ; see remark [ dualextrema](i ) ) . many fluctuation results for @xmath11 can be conveniently formulated in terms of its scale function @xmath413 that is defined as the unique continuous increasing function on @xmath414 with laplace transform @xmath415 let @xmath416 be an exponentially distributed random variable with parameter @xmath417 ( independent of @xmath369 and @xmath11 ) . [ prop : dumfinite ] ( i ) if @xmath96 \in[-\infty,0)$ ] then @xmath418\q\mbox{and}\\ \mathbb{p}(\underline u^*_{{\mathbf{e}_q},{\mathbf{e}_\beta } } > x)&=&\frac{q}{q-\beta}e^{-\phi(\beta)x}\frac{\phi(\beta)-\phi(q)}{\phi(q)}.\end{aligned}\ ] ] \(ii ) if @xmath96 \in(0,\infty)$ ] then @xmath419}(w^{(\beta)}(x - z)-\beta z^{(\beta)}(x - z ) ) w^{(q)}(z){\mathrm{d}}z \\&&- \frac{\beta}{\phi(\beta ) } \frac{w^{(q)}(x)}{w^{(q)\prime}_+(x)}\int_{[0,x ] } ( w^{(\beta)}(x - z)-\beta z^{(\beta)}(x - z))w^{(q)}({\mathrm{d}}z)\\&&\q+z^{(q)}(x)-q\frac{w^{(q)}(x)^2}{w^{(q)\prime}_+(x)},\end{aligned}\ ] ] where @xmath420 denotes the right - derivative of @xmath413 at @xmath43 . the proof of proposition [ prop : dumfinite ] is based on the representations derived in proposition [ repr ] and the form of the @xmath263-resolvent measures @xmath421 and @xmath422 of @xmath423 and @xmath424 killed upon crossing the level @xmath425 , which are defined by @xmath426 where @xmath427 and @xmath428 are the first - passage times of @xmath423 and @xmath429 over @xmath43 , @xmath430 @xmath431 in ( * ? ? ? 1 ) it was shown that these resolvent measures have a density a version of which is given by @xmath432,\\ \label{rd } r^d_x({\mathrm{d}}y ) & = & w^{(q)}(x)\frac{w^{(q)}({\mathrm{d}}y)}{w^{(q)\prime}_+(x ) } -w^{(q)}(y){\mathrm{d}}y , \qquad y \in [ 0 , x].\end{aligned}\ ] ] recall that by proposition [ repr ] we have @xmath433 + q\int_{[0,x]}{\mathbb{p}}(\overline{x}_{{\mathbf{e}_\beta}}>x - z)r^u_x({\mathrm{d}}z),\ ] ] where by ( * ? ? ? 2 ) , @xmath434=\frac{1}{z^{(q)}(x)}\ ] ] and @xmath435 similarly , @xmath436 where by @xcite : @xmath437 straightforward calculations complete the proof of ( i ) . the proof of ( ii ) follows by a similar reasoning using the identity @xmath438= z^{(q)}(x)-q\frac{w^{(q)}(x)^2}{w^{(q)\prime}_+(x)};\ ] ] see ( * ? ? ? [ prop : dum ] ( i ) if @xmath96 \in[-\infty,0)$ ] @xmath439\q\mbox{and } \\ \mathbb{p}(\underline u^*_{{\mathbf{e}_q } } > x)&= & e^{-\phi(0)x}\frac{\phi(q)-\phi(0)}{\phi(q)},\end{aligned}\ ] ] where @xmath440 . \(ii ) if @xmath96 \in(0,\infty)$ ] then @xmath441}w(x - z ) w^{(q)}(z){\mathrm{d}}z\\ & & - \psi'(0 ) \frac{w^{(q)}(x)}{w^{(q)\prime}_+(x)}\int_{[0,x ] } w(x - z)w^{(q)}({\mathrm{d}}z).\end{aligned}\ ] ] since @xmath98 follows an exponential distribution with parameter @xmath442 , we have for any @xmath443 @xmath444}\int_0^y\phi(0)e^{-\phi(0)z}{\mathrm{d}}z\mathbb{p}(z+u_{{\mathbf{e}_q}}\in { \mathrm{d}}y , { \mathbf{e}_q}<t_x^u ) \\ & = & \frac{q\phi(0)}{z^{(q)}(x)}\int_0^x\int_0^y e^{-\phi(0)z}w^{(q)}(x - y+z){\mathrm{d}}z\,{\mathrm{d}}y\\ & = & \frac{1}{z^{(q)}(x)}\le[q\int_0^x(1 - e^{-\phi(0)y})w^{(q)}(y)\,{\mathrm{d}}y\ri].\end{aligned}\ ] ] furthermore , from , @xmath445 the proof of ( ii ) follows by a similar reasoning , using the fact @xmath446 = \psi'(0)^{-1 } w(x)$ ] and the form of the resolvent . * by inverting the laplace transform we find that @xmath447\right).\ ] ] * straightforward calculations show that the double laplace transforms @xmath448 and @xmath449 of @xmath450 and @xmath451 in @xmath122 and @xmath20 are given by : @xmath452 this agrees with the form of @xmath453 obtained in @xcite . slightly differs from the one stated in ( * ? ? 2 ) , which is incorrect due to a small error in the calculation in the proof of ( * ? ? ? 2 ) . ] * from the proofs of the propositions above it is clear that we can identify the bivariate laplace transform of @xmath38 , @xmath39 , @xmath9 and @xmath10 with respect of @xmath34 and @xmath3 as long as the laws of @xmath454 , @xmath383 and resolvents of reflected process @xmath455 , @xmath456 are known . this could be done not only for one - sided lvy processes . for example one can consider the kou model , where the log - price @xmath0 is modelled by a jump - diffusion with constant drift @xmath357 and volatility @xmath457 , with the upward and downward jumps arriving at rate @xmath458 and @xmath459 with sizes following exponential distributions with mean @xmath460 and @xmath461 , @xmath462 where @xmath463 are independent standard poisson processes with rates @xmath464 , independent of a brownian motion @xmath465 , and @xmath466 are independent . then the important ingredients are identified in ( * ? ? ? * lem . 1 and prop . 3 ) ( also applied for the dual process ) . consider a risky asset whose price process @xmath467 is given as follows : @xmath468 where @xmath56 is a lvy process . in the case of the black scholes samuelson model , @xmath469 is a geometric brownian motion , with rate of appreciation @xmath470 and the volatility @xmath471 , and @xmath0 is given by the linear brownian motion @xmath472 let @xmath473 . this model is widely used in practice as a benchmark for other models . for this model we have @xmath474 , @xmath475 with @xmath476 and @xmath477 and @xmath478,\ ] ] @xmath479.\ ] ] hence from corollary [ prop : dum ] we have @xmath480 and @xmath481 hence we find for @xmath318 @xmath482e^{-2\omega x}\right),\end{aligned}\ ] ] where by and ( * ? ? ? * ( 1.1.3 ) , p. 250 ) we have @xmath483&=&{\mathbb{e}}[e^{-2\omega \overline{x}_t}]=1 - 2\omega\int_0^t \mathbb{e}[e^{-2\omega x_s } x_s^+]s^{-1}{\mathrm{d}}s\\&=&\frac{2-\sigma^2}{2 - 2\sigma^2}e^{\frac{\omega^2(1-\sigma^2)}{\sigma^2}t}{\rm erfc}\left(\frac{\omega ( 2-\sigma^2)}{\sqrt{2}\sigma}\sqrt{t}\right ) -\frac{\sigma^2}{2 - 2\sigma^2}{\rm erfc}\left(\frac{\omega \sigma}{\sqrt{2}}\sqrt{t}\right ) . \label{supremummgf}\end{aligned}\ ] ] moreover , @xmath484 & = & p_0 \frac{{\mathbb{e}}\left[e^{\gamma u_t + x_t}\right]}{{\mathbb{e}}\left[e^{\gamma u_t}\right ] } = p_0 e^{\psi(1)t}\frac{{\mathbb{e}}^{(1)}\left[e^{\gamma u_t}\right]}{{\mathbb{e}}\left[e^{\gamma u_t}\right ] } , \\ \underline{\mathbb{e}}[p_t ] & = & p_0 \frac{e^{\psi(1)t}{\mathbb{e}}^{(1)}\left[e^{\gamma d_t}\right]}{{\mathbb{e}}\left[e^{\gamma d_t}\right]},\end{aligned}\ ] ] where the measure @xmath485 is defined via @xmath486 $ ] for @xmath292 and @xmath487 . under @xmath485 we have @xmath488 straightforward calculations based on ( * ? ? ? * ( 1.1.3 ) , ( 1.2.3 ) p. 250 - 251 ) give : @xmath489&=&\frac{2+\sigma^2}{2 + 2\sigma^2}e^{\frac{\omega^2(1+\sigma^2)}{\sigma^2}t}{\rm erfc}\left(-\frac{\omega ( 2+\sigma^2)}{\sqrt{2}\sigma}\sqrt{t}\right ) \\ & & \qquad + \frac{\sigma^2}{2 + 2\sigma^2}{\rm erfc}\left(\frac{\omega \sigma}{\sqrt{2}}\sqrt{t}\right),\\ { \mathbb{e}}[e^{\gamma d_t}]&=&\frac{2+\sigma^2}{2 + 2\sigma^2}e^{\frac{\omega^2(1+\sigma^2)}{\sigma^2}t}{\rm erfc}\left(\frac{\omega ( 2+\sigma^2)}{\sqrt{2}\sigma}\sqrt{t}\right ) \\ & & \qquad + \frac{\sigma^2}{2 + 2\sigma^2}{\rm erfc}\left(-\frac{\omega \sigma}{\sqrt{2}}\sqrt{t}\right),\\ { \mathbb{e}}^{(1 ) } [ e^{\gamma u_t } ] & = & \frac{1+\sigma^2}{1 + 2\sigma^2}e^{\frac{\omega[(2\mu-\sigma^2)(\sigma^2 + 1)-\sigma^4]}{\sigma^4}t}{\rm erfc}\left ( -\frac{(2\mu-\sigma^2)(\sigma^2 + 2)-2\sigma^4}{2\sqrt{2}\sigma^3}\sqrt{t}\right ) \\ & & \qquad + \frac{\sigma^2(2\mu-3\sigma^2)}{(4\mu-2\sigma^2)(1+\sigma^2)-4\sigma^4}{\rm erfc}\left(\frac{2\mu-3\sigma^2}{2\sqrt{2}\sigma}\sqrt{t}\right),\\ { \mathbb{e}}^{(1 ) } [ e^{\gamma d_t}]&= & \frac{1+\sigma^2}{1 + 2\sigma^2}e^{\frac{\omega[(2\mu-\sigma^2)(\sigma^2 + 1)-\sigma^4]}{\sigma^4}t}{\rm erfc}\left ( \frac{(2\mu-\sigma^2)(\sigma^2 + 2)-2\sigma^4}{2\sqrt{2}\sigma^3}\sqrt{t}\right ) \\ & & \qquad + \frac{\sigma^2(2\mu-3\sigma^2)}{(4\mu-2\sigma^2)(1+\sigma^2)-4\sigma^4}{\rm erfc}\left(-\frac{2\mu-3\sigma^2}{2\sqrt{2}\sigma}\sqrt{t}\right).\end{aligned}\ ] ] this work is partially supported by the ministry of science and higher education of poland under the grant dec-2013/09/b / hs4/01496 ( 2014 - 2016 ) . the second author author also kindly acknowledges partial support by the project rare -318984 , a marie curie irses fellowship within the 7th european community framework programme .

elliptical rotation is the motion of a point on an ellipse through some angle about a vector . the purpose of this paper is to examine the generation of elliptical rotations and to interpret the motion of a point on an elipsoid using elliptic inner product and elliptic vector product . to generate an elliptical rotation matrix , first we define an elliptical ortogonal matrix and an elliptical skew symmetric matrix using the associated inner product . then we use elliptic versions of the famous rodrigues , cayley , and householder methods to construct an elliptical rotation matrix . finally , we define elliptic quaternions and generate an elliptical rotation matrix using those quaternions . each method is proven and is provided with several numerical examples . * keywords :* elliptic quaternion , rotation matrices , rodrigues formula , cayley transformation , householder transformation . * msc 2010 : * 15a63 , 15a66 , 70b05 , 70b10 , 70e17 , 53a17 .

1503.07895

a rotation is an example of an isometry , a map that moves points without changing the distances between them . a rotation is a linear transformation that describes the motion of a rigid body around a fixed point or an axis and can be expressed with an orthonormal matrix which is called a rotation matrix . @xmath0 rotation matrices form a special orthogonal group , denoted by @xmath1 , which , for @xmath2 is non - abelian . the group of @xmath0 rotation matrices is isomorphic to the group of rotations in an @xmath3 dimensional space . this means that multiplication of rotation matrices corresponds to composition of rotations . rotation matrices are used extensively for computations in geometry , kinematics , physics , computer graphics , animations , and optimization problems involving the estimation of rigid body transformations . for this reason , the generation of a rotation matrix is considered to be an important problem in mathematics . in the two dimensional euclidean space , a rotation matrix can easily be generated using basic linear algebra or complex numbers . similarly , in the lorentzian plane , a rotation matrix can be generated by double ( hyperbolic ) numbers . in higher dimensional spaces , obtaining a rotation matrix using the inner product is impractical since each column and row of a rotation matrix must be a unit vector perpendicular to all other columns and rows , respectively . these constraints make it difficult to construct a rotation matrix using the inner product . instead , in higher dimensional spaces , rotation matrices can be generated using various other methods such as unit quaternions , the rodrigues formula , the cayley formula , and the householder transformation . we will give a brief review of these methods and use elliptical versions of these methods later in the paper . * 1 . a unit quaternion :* each unit quaternion represents a rotation in the euclidean 3-space . that is , only four numbers are enough to construct a rotation matrix , the only constraint being that the norm of the quaternion is equal to 1 . also , in this method , the rotation angle and the rotation axis can be determined easily . however , this method is only valid in the three dimensional spaces ( @xcite , @xcite ) . in the lorentzian space , timelike split quaternions are used instead of ordinary usual quaternions ( @xcite , @xcite ) . * rodrigues formula :* an orthonormal matrix can be obtained using the matrix exponential @xmath4 where @xmath5 is a skew symmetric matrix and @xmath6 is the rotation angle . in this method , only three numbers are needed to construct a rotation matrix in the euclidean 3-space ( @xcite , @xcite , @xcite and , @xcite).the vector set up with these three numbers gives the rotation axis . this method can be extended to the @xmath3 dimensional euclidean and lorentzian spaces ( @xcite , @xcite , @xcite , @xcite and , @xcite ) . * cayley formula :* the formula @xmath7 gives a rotation matrix , where @xmath5 is a skew symmetric matrix . rotation matrices can be given by the cayley formula without using trigonometric functions . the cayley formula is an easy method but it does nt give the rotation angle directly ( @xcite , @xcite , @xcite , @xcite and , @xcite ) . householder transformation :* the householder transformation gives us a reflection matrix . we can obtain a rotation matrix using two householder transformations . this method is an elegant method but it can be long and tedious . also , the rotation angle has to be calculated separately . this transformation can be used in several scalar product spaces ( @xcite , @xcite , @xcite , @xcite and , @xcite ) . details about generating rotation matrices , particularly in the euclidean and lorentzian spaces , using these methods can be found in various papers , some of which are given in the reference section . those authors mostly studied the rotation matrices in the positive definite scalar product space whose associated matrices are diag@xmath8 and interpreted the results geometrically . for example , quaternions and timelike split quaternions were used to generate rotation matrices in the three dimensional euclidean and lorentzian spaces where the associated matrices were @xmath9 and @xmath10 respectively . in these spaces , rotations occur on the sphere @xmath11 or the hyperboloids @xmath12 . that is , euclidean and lorentzian rotation matrices help us to understand spherical and hyperbolic rotations . in the euclidean space , a rotation matrix rotates a point or a rigid body through a circular angle about an axis . that is , the motion happens on a circle . similarly , in the lorentzian space , a rotation matrix rotates a point through an angle about an axis circularly or hyperbolically depending on whether the rotation axis is timelike or spacelike , respectively . in this paper , we investigate elliptical rotation matrices , which are orthogonal matrices in the scalar product space , whose associated matrix is diag@xmath13 with @xmath14 . first , we choose a proper scalar product to the given ellipse ( or ellipsoid ) such that this ellipse ( or ellipsoid ) is equivalent to a circle ( or sphere ) for the scalar product space . that is , the scalar product does nt change the distance between any point on the ellipse ( or ellipsoid ) and origin . interpreting a motion on an ellipsoid is an important concept since planets usually have ellipsoidal shapes and elliptical orbits . the geometry of ellipsoid can be examined using affine transformations , because of an ellipsoid can be considered as an affine map of the unit sphere . for example , for the ellipsoid @xmath15 and the unit sphere @xmath16 we can write @xmath17 using the affine transformation @xmath18 , @xmath19 then we have , @xmath20 where @xmath21 the aim of this study is to explain the motion on the ellipsoid @xmath22 as a rotation , using the proper inner product , vector product and elliptical orthogonal matrices . in this method , the elliptical inner product , the vector product and the angles are compatible with the parameters @xmath6 and @xmath23 of the parametrization @xmath24 . we use the classical methods to generate elliptical rotation matrices . in the preliminaries section , first we explain how to define a suitable scalar product and a vector product for a given ellipsoid . then we introduce the symmetric , skew symmetric and orthogonal matrices in this elliptical scalar product space . finally , we examine the motion on an ellipsoid using elliptical rotation matrices . in section 3 , we generate the elliptical rotation matrices using various classical methods ( such as , cayley formula , rodrigues formula and householder transformation ) compatible with the defined scalar product . furthermore , we defined the elliptic quaternions and generate elliptical rotations using unit elliptic quaternions . we begin with a brief review of scalar products . more informations can be found in ( @xcite , @xcite and , @xcite ) . consider the map @xmath25 @xmath26 for @xmath27 if such a map is linear in each argument , that is,@xmath28 where , @xmath29 and @xmath30 then it is called a bilinear form . given a bilinear form on @xmath31 there exists a unique @xmath32 square matrix such that for all @xmath33 @xmath34 @xmath35 is called `` _ _ the matrix associated with the form _ _ '' with respect to the standard basis and we will denote @xmath36 as @xmath37 as needed . a bilinear form is said to be symmetric or skew symmetric if @xmath38 or @xmath39 , respectively . hence , the matrix associated with a symmetric bilinear form is symmetric , and similarly , the associated matrix of a skew symmetric bilinear form is skew symmetric . also , a bilinear form is nondegenerate if its associated matrix is non - singular . that is , for all @xmath40 , there exists @xmath41 , such that @xmath42 a real scalar product is a non - degenerate bilinear form . the space @xmath43 equipped with a fixed scalar product is said to be a real scalar product space . also , some scalar products , like the dot product , have positive definitely property . that is , @xmath44 and @xmath45 if and only if @xmath46 now , we will define a positive definite scalar product , which we call the @xmath47-inner product or elliptical inner product . let @xmath48 , @xmath49 and @xmath50 . then the map @xmath51 is a positive definite scalar product . we call it elliptical inner product or @xmath47-inner product . the real vector space @xmath43 equipped with the elliptical inner product will be represented by @xmath52 note that the scalar product @xmath36 can be written as @xmath53 where associated matrix is@xmath54{cccc}a_{1 } & 0 & \cdots & 0\\ 0 & a_{2 } & \cdots & 0\\ \vdots & \vdots & \ddots & 0\\ 0 & 0 & \cdots & a_{n}\end{array } \right ] . \label{omega}\ ] ] the number @xmath55 will be called `` _ _ constant of the scalar product _ _ '' and denoted by @xmath56 in the rest of the paper . the norm of a vector associated with the scalar product @xmath47 is defined as @xmath57 . two vectors @xmath58 and @xmath59 are called @xmath47-orthogonal or elliptically orthogonal vectors if @xmath60 . in addition , if their norms are 1 , then they are called @xmath47-orthonormal vectors . if @xmath61 is an @xmath47-orthonormal base of @xmath62 , then @xmath63 the cosine of the angle between two vectors @xmath58 and @xmath59 is defined as,@xmath64 where @xmath6 is compatible with the parameters of the angular parametric equations of ellipse or ellipsoid . let @xmath47 be a non degenerate scalar product , @xmath35 the associated matrix of @xmath47 , and @xmath65 is any matrix . * i ) * if @xmath66 for all vectors @xmath67 , then @xmath68 is called a @xmath47-orthogonal matrix . it means that orthogonal matrices preserve the norm of vectors and satisfy the matrix equality @xmath69 . also , all rows ( or columns ) are @xmath47-orthogonal to each other . we denote the set of @xmath47-orthogonal matrices by @xmath70 . that is , @xmath71 @xmath70 is a subgroup of * * gl**@xmath72 it is sometimes called the isometry group of @xmath43 associated with scalar product @xmath47 . the determinant of a @xmath47-orthogonal matrix can be either @xmath73 or @xmath74 if @xmath75 then we call it a @xmath47-rotation matrix or an elliptical rotation matrix . if @xmath76 we call it an elliptical reflection matrix . although the set @xmath70 is not a linear subspace of @xmath77 it is a lie group . the isometry group for the bilinear or sesquilinear forms can be found in @xcite . the set of the @xmath47-rotation matrices of @xmath78 can be expressed as follows:@xmath79 @xmath80 is a subgroup of @xmath81 * ii ) * if @xmath82 for all vectors @xmath67 , then @xmath83 is called a @xmath47-symmetric matrix . it satisfies @xmath84 . the set of @xmath47-symmetric matrices , defined by @xmath85 is a jordan algebra @xcite . it is a subspace of the vector space of real @xmath0 matrices , with dimension @xmath86 . any @xmath47-symmetric matrix in @xmath62 can be defined as @xmath87 _ { n\times n } \label{duz}\ ] ] where @xmath88 and @xmath89 * iii ) * if @xmath90 for all vectors @xmath67 , then @xmath91 is called a @xmath47-skew - symmetric matrix . also , @xmath92 . the set of @xmath47-skew symmetric matrices , defined by @xmath93 is a lie algebra @xcite . it is a subspace of the vector space of real @xmath0 matrices , with dimension @xmath94 , as well . any @xmath47-skew - symmetric matrix in @xmath62 can be defined as,@xmath95 _ { n\times n}\text { \ \ with \ \ } t_{ij}=\left\ { \begin{array } [ c]{cc}\dfrac{\delta a_{ij}}{a_{i}}\smallskip & i > j\\ \dfrac{-\delta a_{ij}}{a_{i}}\smallskip & i < j\\ 0 & i = j \end{array } \right . \label{ters}\ ] ] where @xmath88 and @xmath89 for example , in the scalar product space @xmath96 the symmetric and skew symmetric matrices are @xmath97{ccc}a_{11}/a_{1 } & x / a_{1 } & y / a_{1}\\ x / a_{2 } & a_{22}/a_{2 } & z / a_{2}\\ y / a_{3 } & z / a_{3 } & a_{33}/a_{3}\end{array } \right ] $ ] and @xmath98{ccc}0 & x / a_{1 } & y / a_{1}\\ -x / a_{2 } & 0 & z / a_{2}\\ -y / a_{3 } & -z / a_{3 } & 0 \end{array } \right ] .$ ] note that , even if we omit the scalar product constant @xmath56 in @xmath83 or @xmath99 they will still be symmetric or skew symmetric matrix , respectively . but then , we can not generate elliptical rotation matrices using the rodrigues and cayley formulas . so , we will keep the constant @xmath56 . now , we define the elliptical vector product , which is related to elliptical inner product . let * * @xmath100**@xmath101 for @xmath102 and @xmath103 be standard unit vectors for @xmath104 then , the elliptical vector product in @xmath62 is defined as,@xmath105@xmath106{ccccc}\mathbf{e}_{1}/a_{1 } & \mathbf{e}_{2}/a_{2 } & \mathbf{e}_{3}/a_{3 } & \cdots & \mathbf{e}_{n}/a_{n}\\ u_{11 } & u_{12 } & u_{13 } & \cdots & u_{1n}\\ u_{21 } & u_{22 } & u_{23 } & \cdots & u_{2n}\\ \vdots & \vdots & \vdots & \vdots & \vdots\\ u_{\left ( n-1\right ) 1 } & u_{\left ( n-1\right ) 2 } & u_{\left ( n-1\right ) 3 } & \cdots & u_{\left ( n-1\right ) n}\end{array } \right ] \label{vect}\ ] ] the vector @xmath107 is @xmath47-orthogonal to each of the vectors @xmath108 in this section , we will generate elliptical rotation matrices using the elliptical versions of the classical methods . for a given ellipse in the form @xmath109 we will use the scalar product @xmath110 that is , our scalar product space is @xmath111 an elliptical rotation matrix represents a rotation on @xmath112 or any ellipse similar to @xmath113 recall that ellipses with the same eccentricity are called similar . since the shape of an ellipse depends only on the ratio @xmath114 @xmath115 in equation ( [ ellipse ] ) , does not affect the rotation matrix . @xmath116the rodrigues rotation formula is a useful method for generating rotation matrices . given a rotation axis and an angle , we can readily generate a rotation matrix using this method . @xmath1 is a lie group and the space of skew - symmetric matrices of dimension @xmath3 is the lie algebra of @xmath1 . we denote this lie algebra by @xmath117@xmath118 the exponential map defined by the standard matrix exponential series @xmath119 connects it to the lie group . for any skew - symmetric matrix a , the matrix exponential @xmath119 always gives a rotation matrix . this method is known as the rodrigues formula . let @xmath91 be a @xmath47-skew symmetric matrix . then , the matrix exponential @xmath124{cc}\cos\theta & -\dfrac{\sqrt{a_{2}}}{\sqrt{a_{1}}}\sin\theta\\ \dfrac{\sqrt{a_{1}}}{\sqrt{a_{2}}}\sin\theta & \cos\theta \end{array } \right]\ ] ] gives an elliptical rotation along the ellipse @xmath125 @xmath126 that is , @xmath127 is an elliptical rotation matrix in the space @xmath128 let s consider the ellipse @xmath129 with the parametrization @xmath130 . let s take the points @xmath131 that is , if we rotate the point @xmath5 elliptically through the angle @xmath132 we get @xmath133 now , let s find elliptical rotation matrix for these ellipses using the above theorem and get same results . to calculate the elliptical rotation matrix , first , we choose the elliptical inner product @xmath134 in accordance with the ellipse @xmath135 such that @xmath136 and @xmath137 . that is , our space is @xmath138 , @xmath139 and the @xmath47-skew symmetric matrix has the form@xmath121{cc}0 & -3/2\\ 2/3 & 0 \end{array } \right ] .\ ] ] note that , @xmath140 we can obtain elliptical rotation matrix as , @xmath141{cc}\cos\theta & -\dfrac{3\sin\theta}{2}\\ \dfrac{2\sin\theta}{3 } & \cos\theta \end{array } \right ] .\ ] ] where @xmath127 is a @xmath47-orthogonal matrix in @xmath142 namely , the equalities @xmath143 and @xmath144 are satisfied . for @xmath145 we get @xmath146{cc}1/2 & -3\sqrt{3}/4\\ \sqrt{3}/3 & 1/2 \end{array } \right ] .\ ] ] so , if we rotate the point @xmath5 elliptically , we get @xmath147 thus , we get same result using elliptical rotation matrix for @xmath148 note that , @xmath149 and the angle between @xmath150 and @xmath151 is @xmath152 it can be seen that the elliptical rotation matrix @xmath127 can be also used to interpret the motion on a similar ellipse @xmath153 to @xmath148 @xmath159=@xmath160{ccc}\mathbf{e}_{1}/a_{1 } & \mathbf{e}_{2}/a_{2 } & \mathbf{e}_{3}/a_{3}\\ u_{1 } & u_{2 } & u_{3}\\ v_{1 } & v_{2 } & v_{3}\end{array } \right ] $ ] = @xmath161{ccc}0 & -u_{3}/a_{1 } & u_{2}/a_{1}\\ u_{3}/a_{2 } & 0 & -u_{1}/a_{2}\\ -u_{2}/a_{3 } & u_{1}/a_{3 } & 0 \end{array } \right ] \left [ \begin{array } [ c]{c}v_{1}\\ v_{2}\\ v_{3}\end{array } \right ] $ ] = @xmath162 where @xmath163 the matrix@xmath164{ccc}0 & -u_{3}/a_{1 } & u_{2}/a_{1}\\ u_{3}/a_{2 } & 0 & -u_{1}/a_{2}\\ -u_{2}/a_{3 } & u_{1}/a_{3 } & 0 \end{array } \right ] \label{3ters}\ ] ] is skew symmetric in @xmath165 that is , @xmath92 . so , the vector product in @xmath166 can be viewed as a linear transformation , which corresponds to multiplication by a skew symmetric matrix . the characteristic polynomial of @xmath91 is , @xmath167 whose eigenvalues are @xmath168 and @xmath169 according to characteristic polynomial @xmath170 so , if we take a unit vector @xmath171 we get @xmath172 and we can use rodrigues and cayley formulas . let @xmath91 be a skew symmetric matrix in the form @xmath173 such that @xmath174 is a unit vector . then , the matrix exponential @xmath175 gives an elliptical rotation on the ellipsoid @xmath155 furthermore , the matrix @xmath176 can be expressed as @xmath177{ccc}a_{1}u_{1}^{2}+\left ( 1-a_{1}u_{1}^{2}\right ) \cos\theta & -\tfrac{\delta u_{3}\sin\theta}{a_{1}}-a_{2}u_{1}u_{2}\left ( \cos\theta-1\right ) & \tfrac{\delta u_{2}\sin\theta}{a_{1}}-a_{3}u_{1}u_{3}\left ( \cos \theta-1\right ) \\ \tfrac{\delta u_{3}\sin\theta}{a_{2}}-a_{1}u_{1}u_{2}\left ( \cos \theta-1\right ) & a_{2}u_{2}^{2}+\left ( 1-a_{2}u_{2}^{2}\right ) \cos\theta & -\tfrac{\delta u_{1}\sin\theta}{a_{2}}-a_{3}u_{2}u_{3}\left ( \cos \theta-1\right ) \\ -\tfrac{\delta u_{2}\sin\theta}{a_{3}}-a_{1}u_{1}u_{3}\left ( \cos \theta-1\right ) & \tfrac{\delta u_{1}\sin\theta}{a_{3}}-a_{2}u_{2}u_{3}\left ( \cos\theta-1\right ) & a_{3}u_{3}^{2}+\left ( 1-a_{3}u_{3}^{2}\right ) \cos\theta \end{array } \right ] \label{rodmat}\ ] ] where @xmath157 is the rotation axis and @xmath6 is the elliptical rotation angle . a parametrization of the ellipsoid @xmath180 is @xmath181 where @xmath182 and @xmath183 let s take the points @xmath184 on the ellipsoid . let s find the rotation matrix which is rotate the point @xmath5 to @xmath185 elliptically . we have @xmath186 and @xmath187 so , @xmath188 first , using the vector product of @xmath189 and @xmath190 in @xmath191 we find the rotation axis @xmath192 @xmath193{ccc}4i & 4j & 9k\\ 3/2 & \sqrt{3}/2 & 3/2\\ -\sqrt{3}/2 & -\dfrac{1}{2 } & 3\sqrt{3}/2 \end{array } \right\vert = \left ( 1,-\sqrt{3},0\right ) .\ ] ] since @xmath194 is unit vector in @xmath191 we get @xmath195 thus , we obtain the elliptical rotation matrix@xmath196{ccc}9\cos\theta+3 & 3\sqrt{3}\left ( \cos\theta-1\right ) & -4\sqrt{3}\sin\theta\\ 3\sqrt{3}\left ( \cos\theta-1\right ) & 3\cos\theta+9 & -4\sin\theta\\ 9\sqrt{3}\sin\theta & 9\sin\theta & 12\cos\theta \end{array } \allowbreak\right]\ ] ] by using ( [ rodmat ] ) . this matrix describes an elliptical rotation on a great ellipse such that it is intersection of the ellipsoid and the plane passing through the origin and @xmath197-orthogonal to @xmath198 it can be easily found that equation of the plane is @xmath199 so , @xmath200 represents an elliptical rotation over the the great ellipse is @xmath201 @xmath202 . also , the elliptical rotation angle is @xmath203 since @xmath204 ( figure 2b ) . thus , we find @xmath205{ccc}3 & -3\sqrt{3 } & -4\sqrt{3}\\ -3\sqrt{3 } & 9 & -4\\ 9\sqrt{3 } & 9 & 0 \end{array } \right ] . \label{examp2}\ ] ] the matrix ( [ examp2 ] ) , rotates the point @xmath5 to the point @xmath185 elliptically over the great ellipse @xmath206 @xmath207 in 1846 , arthur cayley discovered a formula to express special orthogonal matrices using skew - symmetric matrices . it is called cayley rotation matrix . in this section , we will describe the cayley rotation matrix for any ellipsoid whose scalar product is @xmath47 . we call it @xmath47-cayley rotation matrix . the cayley rotation matrix is a useful tool that gives a parameterization for rotation matrices without the need to use trigonometric functions . let @xmath5 be an @xmath0 skew symmetric matrix where @xmath212 is invertible . cayley formula transforms the matrix @xmath5 into @xmath213 * * so**@xmath214 is isomorphic to @xmath117@xmath214 via the cayley formula where @xmath117@xmath214 denotes the space of skew symmetric matrices , usually associated with the lie algebra of the transformation group defined by rotations in * * so**@xmath214 . that is , the cayley map is defined as@xmath215 @xmath216 note that , if a matrix @xmath217 has the eigenvalue @xmath218 then @xmath219 is not invertible . but , since @xmath5 is skew symmetric , all its eigenvalues are purely imaginary in the euclidean space and @xmath212 is invertible . that is , the cayley formula is well - defined for all skew symmetric matrices . the inverse of the cayley map is given by @xmath220 in the minkowski space , an eigenvalue of a skew symmetric matrix can be @xmath221 in this case , the cayley formula is not valid . for detailed information on the cayley formula in minkowski space see @xcite . let @xmath91 be a skew symmetric matrix in the form ( [ 3ters ] ) such that @xmath174 . then , @xmath222 is an elliptical rotation matrix on the ellipsoid @xmath223 where @xmath58 is the rotation axis . furthermore , the matrix @xmath224 can be written in the form @xmath225{ccc}a_{1}u_{1}^{2}{\small -}u_{2}^{2}a_{2}{\small -}u_{3}^{2}a_{3}{\small + } 1 & a_{2}u_{1}u_{2}{\small -}\tfrac{2\delta u_{3}}{a_{1}}{\small + } u_{1}u_{2}a_{2 } & \tfrac{2\delta u_{2}}{a_{1}}{\small + } a_{3}u_{1}u_{3}{\small + } u_{1}u_{3}a_{3}\\ \tfrac{2\delta u_{3}}{a_{2}}{\small + } a_{1}u_{1}u_{2}{\small + } u_{1}u_{2}a_{1 } & a_{2}u_{2}^{2}{\small -}u_{1}^{2}a_{1}{\small -}u_{3}^{2}a_{3}{\small + } 1 & a_{3}u_{2}u_{3}{\small -}\tfrac{2\delta u_{1}}{a_{2}}{\small + } u_{2}u_{3}a_{3}\\ a_{1}u_{1}u_{3}{\small -}\tfrac{2\delta u_{2}}{a_{3}}{\small + } u_{1}u_{3}a_{1 } & \tfrac{2\delta u_{1}}{a_{3}}{\small + } a_{2}u_{2}u_{3}{\small + } u_{2}u_{3}a_{2 } & a_{3}u_{3}^{2}{\small -}u_{1}^{2}a_{1}{\small -}u_{2}^{2}a_{2}{\small + } 1 \end{array } \right ] . \label{cayleyrot}\ ] ] using these equalities , it can be seen that @xmath228 also , since @xmath229 and @xmath230 we have @xmath231 that is , @xmath224 is an elliptical rotation matrix . the matrix ( [ cayleyrot ] ) can be obtained after some tedious computations . the eigenvalues of the matrix ( [ cayleyrot ] ) are @xmath232 also , the eigenvector corresponding to @xmath211 is @xmath233 which is the rotation axis . the matrix ( [ cayleyrot ] ) rotates a vector elliptically on the ellipsoid @xmath223 about the axis @xmath157 through the elliptical angle @xmath6 where @xmath234 let s find the elliptical rotation matrix representing a elliptical rotation on the ellipsoid @xmath235 about the axis @xmath236 using the matrix ( [ cayleyrot ] ) , we find @xmath177{ccc}0 & 0 & 2\\ 3/2 & 0 & 0\\ 0 & 1/3 & 0 \end{array } \right ] .\ ] ] the elliptical rotation angle corresponding to this matrix , can be found to be @xmath237 from the formula ( [ tantetat ] ) . where @xmath239 is a nonzero vector . this transformation describes a reflection about a plane or hyperplane passing through the origin and orthogonal to @xmath240 householder transformations on spaces with a non - degenerate bilinear or sesquilinear forms are studied in @xcite . every orthogonal transformation is the combination of reflections with respect to hyperplanes . this is known as the cartan dieudonn theorem . a constructive proof of the cartan dieudonn theorem for the case of generalized scalar product spaces is given by uhlig @xcite and fuller @xcite . an alternative proof of the cartan dieudonn theorem for generalized real scalar product spaces is given by rodrguez - andrade and etc . they used the clifford algebras to compute the factorization of a given orthogonal transformation as a product of reflections . householder transformations are widely used in tridiagonalization of symmetric matrices and to perform qr decompositions in numerical linear algebra @xcite . householder transformation is cited in the top 10 algorithms of the 20th century @xcite . generalized householder matrices are the simplest generalized orthogonal matrices . a generalized householder , or @xmath47-householder , matrix has the form in this paper , we use the elliptical version of the householder transformation to generate an elliptical rotation matrix in a real elliptical scalar product space @xmath52 let @xmath243 _ { n\times n}$ ] be a householder matrix . then where @xmath245 is the kronecker delta . * * the householder matrix @xmath246 is a @xmath47-symmetric and @xmath47-orthogonal matrix . to prove these , first we will show that @xmath247 : @xmath248 is a positive real number@xmath249 so , using the above transformations , we can write the elliptical reflection matrix as@xmath253{ccc}\left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{1}v_{1}^{2 } & -2a_{2}v_{1}v_{2 } & -2a_{3}v_{1}v_{3}\\ -2a_{1}v_{2}v_{1 } & \left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{2}v_{2}^{2 } & -2a_{3}v_{2}v_{3}\\ -2a_{1}v_{3}v_{1 } & -2a_{2}v_{3}v_{2 } & \left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{3}v_{3}^{2}\end{array } \right ] \label{house1}\ ] ] for the ellipsoid @xmath223 . the householder transformation does nt change the length of a vector in @xmath254 . it describes an elliptical reflection about a plane passing through the origin @xmath197-orthogonal to @xmath240 using the matrix ( [ house1 ] ) , we can express elliptical reflections on any ellipsoid . let s find the elliptical reflection matrix which reflects a point elliptically on the ellipsoid @xmath255 about the plane which is passes through the origin and is @xmath47-orthogonal to @xmath256 our elliptical inner product is @xmath257 . so , @xmath258 and we get @xmath259{ccc}15/19 & -8/19 & -6/19\\ -8/19 & 3/19 & -12/19\\ -12/19 & -24/19 & 1/19 \end{array } \right ] .\ ] ] the determinant of this matrix is @xmath73 and the equality @xmath260 holds . we find the reflection of @xmath261 to be @xmath262 let s check it : the middle point of @xmath263 $ ] is @xmath264 the equation of the plane is @xmath265 where @xmath266 in our case , it is @xmath267 so @xmath268 lies on this plane . also , it can be easily seen that @xmath269 is @xmath47-orthogonal to the plane . for the @xmath3 dimensional ellipsoid associated with the scalar product @xmath47 , the elliptical reflection matrix @xmath246 is in the form @xmath270{ccccc}\left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{1}v_{1}^{2 } & -2a_{2}v_{1}v_{2 } & -2a_{3}v_{1}v_{3 } & \cdots & -2a_{n}v_{1}v_{n}\\ -2a_{1}v_{2}v_{1 } & \left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{2}v_{2}^{2 } & -2a_{3}v_{2}v_{3 } & \cdots & -2a_{n}v_{2}v_{n}\\ -2a_{1}v_{3}v_{1 } & -2a_{2}v_{3}v_{2 } & \left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{3}v_{3}^{2 } & \cdots & -2a_{n}v_{3}v_{n}\\ \vdots & \vdots & \vdots & \ddots & \vdots\\ -2a_{1}v_{n}v_{1 } & -2a_{2}v_{n}v_{2 } & -2v_{n}a_{3}v_{3 } & \cdots & \left\vert \mathbf{v}\right\vert _ { \mathcal{b}}^{2}-2a_{n}v_{n}^{2}\end{array } \right]\ ] ] where @xmath271 by direct computation , we see that@xmath281 and@xmath282 thus , we get@xmath283 moreover , we have @xmath284@xmath285 which means that @xmath286 hence , we obtain the elliptical rotation matrix@xmath287 which rotates @xmath288 to @xmath289 elliptically . the rotation axis is the vector @xmath194 and the elliptic rotation angle can be found using elliptical inner product . let s find the elliptic rotation matrix that rotates @xmath290 to @xmath291 on the ellipsoid @xmath292 using the matrix ( [ house1 ] ) , we get @xmath293{ccc}4 & -1 & -2\\ -1 & 4 & -2\\ -4 & -4 & -3 \end{array } \right ] \text { \ and \ } \mathcal{h}_{\boldsymbol{y}}=\dfrac{1}{25}\left [ \begin{array } [ c]{ccc}9 & -16 & -12\\ -16 & 9 & -12\\ -24 & -24 & 7 \end{array } \right ] .\ ] ] thus , the rotation matrix is @xmath294{ccc}4 & -1 & 2\\ -1 & 4 & 2\\ -4 & -4 & 3 \end{array } \right]\ ] ] which satisfies @xmath295 , @xmath75 and @xmath69 where @xmath296 note that , we can obtain same matrix using ( [ rodmat ] ) . quaternions were discovered by sir william r. hamilton in 1843 and the theory of quaternions was expanded to include applications such as rotations in the early 20th century . the most important property of the quaternions is that every unit quaternion represents a rotation and this plays an important role in the study of rotations in 3-dimensional vector spaces . using unit quaternions is a useful , natural , and elegant way to perceive rotations . quaternions are used especially in computer vision , computer graphics , animation , and kinematics . quaternion algebra @xmath297 is an associative , non - commutative division ring with four basic elements @xmath298 satisfying the equalities @xmath299 we can express any quaternion @xmath300 as @xmath301 or @xmath302 where the symbols @xmath303 and @xmath304 denote the scalar and vector parts of @xmath305 respectively@xmath249 if @xmath306 then @xmath300 is called a pure quaternion . the conjugate of @xmath300 is denoted by @xmath307 and defined as @xmath308 . the norm of a quaternion @xmath309 is defined by @xmath310 and is denoted by @xmath311 . we say that @xmath312 is unit quaternion if @xmath313 the set of unit quaternions are denoted by @xmath314 every unit quaternion can be written in the form @xmath315 where @xmath316 is a unit vector satisfying the equality @xmath317it is called the axis of the quaternion @xcite , @xcite . let @xmath300 and @xmath318 be two quaternions . then , the linear transformation @xmath319 defined by @xmath320 is a quaternion that has the same norm and scalar as @xmath318 . since the scalar part of the quaternion @xmath318 does nt change under @xmath321 we will only examine how its vector part @xmath322 changes under the transformation @xmath323 we can interpret the rotation of a vector in the euclidean 3-space using the quaternion product @xmath324 . if @xmath325 is a unit quaternion , then , using the linear transformation @xmath326 , the corresponding rotation matrix can be found as @xmath327{ccc}q_{0}^{2}+q_{1}^{2}-q_{2}^{2}-q_{3}^{2 } & -2q_{0}q_{3}+2q_{1}q_{2 } & 2q_{0}q_{2}+2q_{1}q_{3}\\ 2q_{1}q_{2}+2q_{3}q_{0 } & q_{0}^{2}-q_{1}^{2}+q_{2}^{2}-q_{3}^{2 } & 2q_{2}q_{3}-2q_{1}q_{0}\\ 2q_{1}q_{3}-2q_{2}q_{0 } & 2q_{1}q_{0}+2q_{2}q_{3 } & q_{0}^{2}-q_{1}^{2}-q_{2}^{2}+q_{3}^{2}\end{array } \right ] . \tag{4}\ ] ] this rotation matrix represents a rotation through angle @xmath328 about the axis @xmath329 in the lorentzian space , the rotation matrix corresponding to a timelike quaternion @xmath309 is , @xmath327{ccc}q_{1}^{2}+q_{2}^{2}+q_{3}^{2}+q_{4}^{2 } & 2q_{1}q_{4}-2q_{2}q_{3 } & -2q_{1}q_{3}-2q_{2}q_{4}\\ 2q_{2}q_{3}+2q_{4}q_{1 } & q_{1}^{2}-q_{2}^{2}-q_{3}^{2}+q_{4}^{2 } & -2q_{3}q_{4}-2q_{2}q_{1}\\ 2q_{2}q_{4}-2q_{3}q_{1 } & 2q_{2}q_{1}-2q_{3}q_{4 } & q_{1}^{2}-q_{2}^{2}+q_{3}^{2}-q_{4}^{2}\end{array } \right]\ ] ] where the set of unit timelike quaternions satisfies the properties @xmath330 @xmath331 @xmath332 ( see @xmath333 the set of elliptic quaternions will be denoted by @xmath341 this set is an associative , non - commutative division ring with our basic elements @xmath342 if we take @xmath343 we get the usual quaternion algebra . the elliptic quaternion product table is given below . [ c]|c||c|c|c|c| & @xmath344 & @xmath345 & @xmath346 & @xmath347 + @xmath344 & @xmath344 & @xmath345 & @xmath346 & @xmath347 + @xmath345 & @xmath345 & @xmath348 & @xmath349 & @xmath350 + @xmath346 & @xmath346 & @xmath351 & @xmath352 & @xmath353 + @xmath347 & @xmath347 & @xmath354 & @xmath355 & @xmath356 + the elliptic quaternion product of two quaternions @xmath357 and @xmath309 is defined as@xmath358 where @xmath359 and @xmath360 are the elliptical scalar product and the elliptical vector product , respectively . if @xmath361 and @xmath300 are pure , then@xmath362{ccc}i / a_{1 } & j / a_{2 } & k / a_{3}\\ p_{1 } & p_{2 } & p_{3}\\ q_{1 } & q_{2 } & q_{3}\end{array } \right\vert .\end{aligned}\ ] ] the elliptic quaternion product for @xmath363 can be expressed as @xmath364{cccc}p_{0 } & -a_{1}p_{1 } & -a_{2}p_{2 } & -a_{3}p_{3}\\ p_{1 } & p_{0 } & -\dfrac{p_{3}\delta}{a_{1 } } & \dfrac{p_{2}\delta}{a_{1}}\\ p_{2 } & \dfrac{p_{3}\delta}{a_{2 } } & p_{0 } & -\dfrac{p_{1}\delta}{a_{3}}\\ p_{3 } & -\dfrac{p_{2}\delta}{a_{3 } } & \dfrac{p_{1}\delta}{a_{3 } } & p_{0}\end{array } \right ] \left [ \begin{array } [ c]{c}q_{0}\\ q_{1}\\ q_{2}\\ q_{3}\end{array } \right ] .\allowbreak$ ] @xmath364{cccc}p_{0 } & -2p_{1 } & -2p_{2 } & -p_{3}\\ p_{1 } & p_{0 } & -p_{3 } & p_{2}\\ p_{2 } & p_{3 } & p_{0 } & -p_{1}\\ p_{3 } & -2p_{2 } & 2p_{1 } & p_{0}\end{array } \right ] \allowbreak\allowbreak\left [ \begin{array } [ c]{c}q_{0}\\ q_{1}\\ q_{2}\\ q_{3}\end{array } \right ] .$ ] [ c]|c||c|c|c|c| & @xmath344 & @xmath345 & @xmath346 & @xmath347 + @xmath344 & @xmath344 & @xmath345 & @xmath346 & @xmath347 + @xmath345 & @xmath345 & @xmath369 & @xmath370 & @xmath371 + @xmath346 & @xmath346 & @xmath372 & @xmath369 & @xmath345 + @xmath347 & @xmath347 & @xmath346 & @xmath373 & @xmath73 + remember that the algebra is formed by a vector space @xmath374 equipped with a quadratic form @xmath375 with the following equalities@xmath376 is called a _ clifford algebra _ and is denoted by @xmath377 if @xmath378 is a base for an n - dimensional vector space @xmath379 then @xmath380 is formed by the multivectors @xmath381 with @xmath382 since the clifford product of two even multivectors is an even multivector , they define an even subalgebra of @xmath383 . the even subalgebra of an @xmath3-dimensional clifford algebra is isomorphic to a clifford algebra of @xmath384 dimensions and it is denoted by @xmath385 the hamiltonian quaternion algebra @xmath297 is isomorphic with the even subalgebra @xmath386 by @xmath387 and the split quaternion algebra @xmath388 is isomorphic with the even subalgebra @xmath389 by @xmath390 @xcite@xmath249 similarly , the elliptic quaternion algebra is an even subalgebra of the clifford algebra for the quadratic form @xmath397 recall that the elliptical inner product can be obtained by using the equality @xmath398 .\ ] ] so , we get @xmath399 for @xmath400 and @xmath401 thus , we can construct an elliptic quaternion algebra for any elliptical inner product space . conjugate , norm and inverse of an elliptic quaternion @xmath309 can be defined similar to usual quaternions : @xmath402 also , each elliptic quaternion @xmath309 can be written in the form @xmath403 where here , @xmath406 is a unit vector in the scalar product space @xmath166 satisfying the equality @xmath407 . it is called the axis of the rotation . for example , if @xmath408 then @xmath409 and we can write each a unit elliptic quaternion represents an elliptical rotation on the an ellipsoid . if @xmath414 is a unit elliptic quaternion , then the linear map @xmath415 gives an elliptical rotation through the elliptical angle @xmath416 about the axis @xmath417 where @xmath418 the elliptical rotation matrix to corresponding to the quaternion @xmath300 is @xmath419{ccc}q_{0}^{2}{\small + } a_{1}q_{1}^{2}-a_{2}q_{2}^{2}-a_{3}q_{3}^{2 } & 2a_{2}q_{1}q_{2}-2\dfrac{q_{0}q_{3}\delta}{a_{1 } } & 2a_{3}q_{1}q_{3}+2\dfrac { q_{0}q_{2}\delta}{a_{1}}\\ 2a_{1}q_{1}q_{2}+2\dfrac{q_{0}q_{3}\delta}{a_{2 } } & q_{0}^{2}-a_{1}q_{1}^{2}{\small + } a_{2}q_{2}^{2}-a_{3}q_{3}^{2 } & 2a_{3}q_{2}q_{3}-2\dfrac { q_{0}q_{1}\delta}{a_{2}}\\ 2a_{1}q_{1}q_{3}-2\dfrac{q_{0}q_{2}\delta}{a_{3 } } & 2a_{2}q_{2}q_{3}+2\dfrac{q_{0}q_{1}\delta}{a_{3 } } & q_{0}^{2}-a_{1}q_{1}^{2}-a_{2}q_{2}^{2}{\small + } a_{3}q_{3}^{2}\end{array } \right ] . \label{quatrot}\ ] ] it can be seen that @xmath420 is a linear transformation and preserves the norm . using the equalities,@xmath421 we can obtain ( [ quatrot ] ) . so , the rotation matrix ( [ quatrot ] ) is an elliptical rotation matrix on the ellipsoid @xmath223 . that is , the equalities @xmath422 and @xmath423 are satisfied . also , note that , if we take @xmath424 , the standard rotation matrix is obtained . now , let s choose an orthonormal set @xmath425 satisfying the equalities @xmath426 if @xmath427 is a vector in the plane of the @xmath316 and @xmath428 , we can write it as @xmath429 to compute @xmath430 , let s find how @xmath431 and @xmath428 change under the transformation @xmath432 since @xmath433 is parallel to @xmath417 we have @xmath434 by ( [ vect ] ) and @xmath435 . so , @xmath316 is not changed under the transformation @xmath436 . it means that @xmath316 is the rotation axis . on the other hand,@xmath437 since we know that @xmath438 for orthogonal , pure quaternions , we obtain@xmath439 that is , @xmath427 is rotated through the elliptical angle @xmath328 about @xmath316 by the transformation @xmath440 let s find the general elliptical rotation matrix for the ellipsoid @xmath292 using ( [ quatrot])@xmath441 we obtain,@xmath419{ccc}q_{0}^{2}+2q_{1}^{2}-2q_{2}^{2}-q_{3}^{2 } & 4q_{1}q_{2}-2q_{0}q_{3 } & 2q_{0}q_{2}+2q_{1}q_{3}\\ 2q_{0}q_{3}+4q_{1}q_{2 } & q_{0}^{2}-2q_{1}^{2}+2q_{2}^{2}-q_{3}^{2 } & 2q_{2}q_{3}-2q_{0}q_{1}\\ 4q_{1}q_{3}-4q_{0}q_{2 } & 4q_{0}q_{1}+4q_{2}q_{3 } & q_{0}^{2}-2q_{1}^{2}-2q_{2}^{2}+q_{3}^{2}\end{array } \right ] .\ ] ] here , @xmath442 and @xmath443 where @xmath444 for example , the unit quaternion @xmath445 represents an elliptical rotation on the ellipsoid @xmath446 through the elliptical angle @xmath447 about the axis @xmath448 and the elliptical rotation matrix is * step 8 . * find the elliptical rotation matrix that rotates @xmath455 to @xmath476 elliptically on the ellipsoid @xmath223 using rodrigues matrix@xmath477{ccc}a_{1}u_{1}^{2}{\small + } \left ( 1{\small -}a_{1}u_{1}^{2}\right ) c & { \small -}\dfrac{\delta u_{3}s}{a_{1}}{\small -}a_{2}u_{1}u_{2}\left ( c{\small -}1\right ) & \dfrac{\delta u_{2}s}{a_{1}}{\small -}a_{3}u_{1}u_{3}\left ( c{\small -}1\right ) \\ \dfrac{\delta u_{3}s}{a_{2}}{\small -}a_{1}u_{1}u_{2}\left ( c{\small -}1\right ) & a_{2}u_{2}^{2}{\small + } \left ( 1{\small -}a_{2}u_{2}^{2}\right ) c & { \small -}\dfrac{\delta u_{1}s}{a_{2}}{\small -}a_{3}u_{2}u_{3}\left ( c{\small -}1\right ) \\ { \small -}\dfrac{\delta u_{2}s}{a_{3}}{\small -}a_{1}u_{1}u_{3}\left ( c{\small -}1\right ) & \dfrac{\delta u_{1}s}{a_{3}}{\small -}a_{2}u_{2}u_{3}\left ( c{\small -}1\right ) & a_{3}u_{3}^{2}{\small + } \left ( 1{\small -}a_{3}u_{3}^{2}\right ) c \end{array } \right]\ ] ] where @xmath478 @xmath479{ccc}\mathcal{n}\left ( \mathbf{v,}\boldsymbol{a}\right ) -2a_{1}v_{1}^{2 } & -2a_{2}v_{1}v_{2 } & -2a_{3}v_{1}v_{3}\\ -2a_{1}v_{2}v_{1 } & \mathcal{n}\left ( \mathbf{v,}\boldsymbol{a}\right ) -2a_{2}v_{2}^{2 } & -2a_{3}v_{2}v_{3}\\ -2a_{1}v_{3}v_{1 } & -2a_{2}v_{3}v_{2 } & \mathcal{n}\left ( \mathbf{v,}\boldsymbol{a}\right ) -2a_{3}v_{3}^{2}\end{array } \right ] $ ] @xmath489{ccc}q_{0}^{2}{\small + } a_{1}q_{1}^{2}-a_{2}q_{2}^{2}-a_{3}q_{3}^{2 } & 2a_{2}q_{1}q_{2}-2\dfrac{q_{0}q_{3}\delta}{a_{1 } } & 2a_{3}q_{1}q_{3}+2\dfrac { q_{0}q_{2}\delta}{a_{1}}\\ 2a_{1}q_{1}q_{2}+2\dfrac{q_{0}q_{3}\delta}{a_{2 } } & q_{0}^{2}-a_{1}q_{1}^{2}{\small + } a_{2}q_{2}^{2}-a_{3}q_{3}^{2 } & 2a_{3}q_{2}q_{3}-2\dfrac { q_{0}q_{1}\delta}{a_{2}}\\ 2a_{1}q_{1}q_{3}-2\dfrac{q_{0}q_{2}\delta}{a_{3 } } & 2a_{2}q_{2}q_{3}+2\dfrac{q_{0}q_{1}\delta}{a_{3 } } & q_{0}^{2}-a_{1}q_{1}^{2}-a_{2}q_{2}^{2}{\small + } a_{3}q_{3}^{2}\end{array } \right ] $ ] in the @xmath491 dimensional spaces , rotations can be classified such as simple , composite , and isoclinic , depending on plane of rotation . simple rotation is a rotation with only one plane of rotation . in a simple rotation , there is a fixed plane . the rotation is said to take place about this plane . so points do not change their distance from this plane as they rotate . orthogonal to this fixed plane is the plane of rotation . the rotation is said to take place in this plane . on the other hand , a rotation with two or more planes of rotation is called a composite rotation . the rotation can take place in each plane of rotation . these planes are orthogonal . in @xmath492 it is called a double rotation . a double rotation has two angles of rotation , one for each plane of rotation . the rotation is specified by giving the two planes and two non - zero angles @xmath23 and @xmath6 ( if either angle is zero , then the rotation is simple ) . finally , the isoclinic rotation is a special case of the composite rotation , when the two angles are equal @xcite . in the @xmath493 dimensional euclidean and lorentzian spaces , a skew symmetric matrix is decomposed as @xmath494 using two skew - symmetric matrices @xmath495 and @xmath496 satisfying the properties @xmath497 @xmath498 and @xmath499 hence , the rodrigues and cayley rotation formulas can be used to generate 4 dimensional rotation matrices ( @xcite , @xcite , @xcite , @xcite , and @xcite ) . rodrguez - andrade , , g. aragn - gonzlez , j.l . aragn , l. verde - star , an algorithm for the cartan - dieudonn theorem on generalized scalar product spaces , linear algebra and its applications , vol . 434 , issue 5 , ( 2011 ) , 1238 - 1254 . r. w. brackett : robotic manipulators and the product of exponentials formula . mathematical theory and networks and systems . proceeding of international symposium . berlin , springer - verlag ( 1984 ) : 120 - 127 .

we point out that the flavor problem in theories with dynamical electroweak symmetry breaking can be effectively decoupled if the physics above the tev scale is strongly conformal , and the electroweak order parameter has a scaling dimension with . there are many restrictions on small values of : for , electroweak symmetry breaking requires a fine - tuning similar to that of the standard model ; large- conformal field theories ( including those obtained from the ads / cft correspondence ) require fine - tuning for ; ` walking technicolor ' theories can not have , according to gap equation analyses . however , strong small- conformal field theories with avoid all these constraints , and can give rise to natural dynamical electroweak symmetry breaking with a top quark flavor scale of order , large enough to decouple flavor . small- theories also have an acceptably small peskin - takeuchi parameter . this class of theories provides a new direction for dynamical electroweak symmetry breaking without problems from flavor or electroweak precision tests . a possible signal for these theories is a prominent scalar resonance below the tev scale with couplings similar to a heavy standard model higgs .

hep-ph0409274

how is electroweak symmetry broken ? the most important theoretical clue we have is the hierarchy problem , the problem of understanding the smallness of the weak scale compared to much higher scales in physics such as the planck scale . perhaps the most elegant solution of the hierarchy problem is dynamical electroweak symmetry breaking @xcite . this is the idea that the scale of electroweak symmetry breaking is determined by a new strong interaction scale . this naturally explains the smallness of the electroweak scale , since the strong interaction scale is given in terms of uv quantities by _ ew~_uv e^-g_c^2/g_uv^2 , where @xmath8 is the strength of the coupling in the uv and @xmath9 is the critical value where electroweak symmetry is broken . for @xmath10 , the electroweak scale is naturally exponentially small compared to @xmath11 . this mechanism is already realized in nature in the strong interaction sector , explaining why the qcd scale is naturally small compared to higher scales . this paradigm for electroweak symmetry breaking makes the general prediction that the electroweak symmetry breaking sector is strongly coupled at the tev scale . within a few years , the lhc will definitively settle the fundamental question of whether electroweak symmetry breaking sector is weakly or strongly coupled . until the lhc turns on , we must rely on indirect constraints . dynamical electroweak symmetry breaking faces a number of potential difficulties . first , strong interactions at the tev scale can ruin the agreement of the standard model with precision electroweak data . however , if the physics that breaks electroweak symmetry is a strongly coupled theory with no large or small parameters , ` dimensional analysis ' ( nda ) gives an estimate for the peskin - takeuchi @xmath7 and @xmath12 parameters s_nda ~ , t_nda ~. for comparison , the value of the @xmath7 parameter from scaled - up qcd is @xcite s_qcd ~0.3 . these are rough estimates , and are comparable to the size of the current 95% confidence level bounds @xcite . these do not rule out models of dynamical electroweak symmetry breaking . the models that _ are _ ruled out ( without fine tuning ) are those containing a large number @xmath4 of degrees of freedom , in which @xmath13 . these include large ` technicolor ' or ` walking technicolor ' theories @xcite , and randall - sundrum ( rs ) models @xcite with gauge fields in the bulk @xcite , which are related to large-@xmath4 conformal theories ( cft s ) by the ads / cft correspondence @xcite . another general problem with models of dynamical electroweak symmetry breaking is that flavor is generally not decoupled from the tev scale . in technicolor models , this is because the order parameter that breaks electroweak symmetry is a techni - fermion bilinear @xmath14 with mass dimension @xmath15 . the standard - model fermion masses therefore arise from 4-fermion operators connecting the standard model fermions with the technifermions @xcite . these operators have dimension 6 , and therefore become strong at low scales . in particular the top coupling becomes strong at a scale _ t ~_ew()^1/2 ~5 , where @xmath16 is the scale where the electroweak symmetry breaking sector becomes strongly coupled . @xmath17 is the scale where flavor must be addressed in these models . the flavor problem is less severe in models of ` walking ' technicolor , in which it is assumed that the electroweak order parameter @xmath14 has a large anomalous dimension , and scales as an operator with dimension @xmath18 @xcite . walking technicolor theories are similar to a cft with a nearly marginal ( slightly relevant ) operator that runs slowly and becomes strong and breaks electroweak symmetry . analyses based on the truncated schwinger - dyson equations show that in asymptotically free theories , @xmath19 @xcite . the scale where the top coupling becomes strong is then raised for @xmath20 to _ t _ ew ~10 . attempts to make realistic models based on strong top dynamics can be found in . in this paper , we will instead attempt to avoid strong flavor - dependent dynamics at low scales . a simple way to avoid the restriction @xmath19 is to assume that the theory is at an interacting conformal fixed point above the tev scale . this class of theories offers a solution of the hierarchy problem that is identical to asymptotically free theories such as technicolor . if the cft is coupled to a gauge theory that is asymptotically free , this gauge theory will become strong in the ir , causing the cft to flow away from its fixed point . , and we do not consider it here . ] the resulting non - perturbative dynamics can give rise to electroweak symmetry breaking . another possibility is that the cft contains a nearly marginal operator that becomes strong in the ir . these mechanisms are attractive because it generates an exponentially large hierarchy . another possible mechanism exists if the cft has a relevant operator that transforms nontrivially under a global symmetry , a discrete symmetry . the coefficient of this operator can then be naturally small , and can set the scale for the breaking of conformal and electroweak symmetry . in this mechanism , the large hierarchy is put in by hand in the form of a small coefficient , but it is technically natural . in a strong cft , flavor arises from couplings of the form @xmath21 , where @xmath22 is a standard - model fermion and @xmath23 is a cft operator with quantum numbers of the higgs . in order to decouple flavor , we would like to have the scaling dimension @xmath24 of the operator @xmath23 as small as possible . in cft s , bound on the scaling dimension of a scalar operator is @xmath25 @xcite . in the limit @xmath26 , the scalar operator behaves as a weakly - coupled scalar field , which is just the standard - model higgs . the theory is therefore fine - tuned and does not solve the hierarchy problem . however , for @xmath0 , with @xmath27 , the top quark becomes strongly coupled at the scale _ t ~_ew ( ) ^1/. this scale is _ exponentially _ large for small @xmath2 , and therefore we can plausibly have sufficiently large @xmath2 to avoid fine - tuning , while decoupling the flavor to high scales . how @xmath17 must be to avoid flavor - changing neutral currents depends on the nature of flavor violation at this scale . the most pessimistic case imaginable is that there are unsuppressed strong contributions to operators that contribute to @xmath28@xmath29 mixing at the scale @xmath17 . this requires @xmath30 , which is obtained for @xmath31 . if we assume some suppression of flavor violation for the lightest generation , we expect that the flavor scale can be significantly lower . for example , a single yukawa suppression of four - fermion operators contributing to @xmath28@xmath29 mixing lowers the flavor scale to @xmath32 , which requires @xmath33 . such values of @xmath2 are definitely plausible . for example , in f - theory constructions of ads@xmath34 duals , one finds scalar operators with dimension @xmath35 and @xmath36 @xcite . the possible application of non - supersymmetric cft s with low - dimension scalar operators to the electroweak hierarchy problem was also discussed in . the randall - sundrum model gives an explicit example of a 4d cft , and has been extensively discussed as a solution of the hierarchy problem . in this model , the higgs is usually localized in the the ir brane to obtain a large hierarchy . in this case the higgs field can be thought of as a bulk field with a large mass , and in the corresponding 4d interpretation the electroweak order parameter has a large ( @xmath37 ) scaling dimension . to obtain sufficiently large fermion masses , the fermions are put on the ir brane or in the bulk @xcite . in 4d language , this corresponds to generating fermion masses by making them mix with composite fermions so that they can feel the symmetry breaking in the strong sector . the mixing of the standard - model fermions with composite fermions was considered previously in the context of qcd - like technicolor @xcite . theories of this type are interesting alternatives to our scenario . as we discuss in the appendix , these theories generally have a potentially viable region of the parameter space where @xmath38 is just at the current experimental bound while corrections to @xmath39 require fine - tuning at the @xmath40 level @xcite . however , we will pursue scenarios where the standard - model fermions are completely elementary , just like in conventinal technicolor theories . it is a simple matter to modify the rs model to give the electroweak order parameter a smaller dimension : one simply puts a higgs scalar in the bulk , and leaves the fermions on the uv brane . electroweak symmetry is broken by a higgs potential localized on the ir brane ( ensuring that this is an ir effect ) and the bulk higgs field communicates electroweak symmetry breaking to the fermions on the uv brane . taking the higgs bulk mass parameter to be negative makes the dimension of the higgs operator in the 4d cft description smaller . however , we will show that scalar operators with @xmath5 are fine - tuned in rs . this can be traced to the fact that rs is a large-@xmath4 theory , and this fine - tuning is common to all large-@xmath4 theories . we are therefore led to a rather dark corner of theory space : non - supersymmetric 4d strongly - coupled conformal field theories with small @xmath4 . these can have a scalar operator with dimension @xmath41 with @xmath42 , and can dynamically break electroweak symmetry at the tev scale while giving large fermion masses without flavor - changing neutral currents . small-@xmath4 theories also have an acceptably small @xmath7 parameter . not much is known about the dynamics of such theories , and so our discussion of these theories is necessarily speculative . above the tev scale , the theory becomes conformally invariant , and the new strong conformal dynamics can be directly tested in direct analogy with the way qcd is tested at a high - energy @xmath43 collider . however , even the lhc will be limited to exploring the lightest ` hadrons ' of the cft , and it is not possible to make rigorous predictions for this regime . in the case where @xmath2 arises from a ( moderately ) small parameter in the fundamental theory , we argue that the theory contains a prominent scalar resonance near ( but somewhat below ) the tev scale , with couplings similar to those of a heavy standard - model higgs , but deviating from the standard - model couplings by order @xmath2 . this provides an interesting and well - motivated signal to look for at the lhc , whose observation would clearly motivate going to even higher energies in the future . this paper is organized as follows . in section 2 , we review the constraints on the operator dimension @xmath24 in various types of known models and argue that small-@xmath4 theories avoid all constraints and can have @xmath5 without fine tuning . in section 3 , we study the phenomenology of these theories , focusing mainly on the possibility of a higgs - like scalar resonance . section 4 contains our conclusions . as discussed in the introduction , an important question for models of dynamical electroweak symmetry breaking based on conformal field theory is the scaling dimension @xmath24 of the ` higgs ' operator @xmath23 that acts as the electroweak order parameter . in order to decouple flavor , we would like to have @xmath24 as small as possible , while avoiding fine - tuning . in this section we review what is known about low - dimension scalar operators , and argue that theories with strong coupling and small @xmath4 can give operators with @xmath5 without fine tuning . general theorems of conformal field theory tell us that a scalar operator @xmath23 must have dimension @xmath25 @xcite . furthermore , an operator with dimension @xmath46 is a free field ( meaning that correlation functions of @xmath23 are the same as those of a free field ) . it is therefore clear that for @xmath24 sufficiently close to 1 , the theory is equivalent to a cft weakly coupled to a higgs field , which clearly does not solve the hierarchy problem . in cft language , this is because the operator @xmath47 has dimension close to @xmath48 , which is relevant for @xmath5 . the existence of a relevant operator that can not be forbidden by symmetries means that the fixed point is not reached unless the coefficient of the relevant operator is tuned . this argument seems to suggest that the theory is fine - tuned for any @xmath5 , but it is limited to weak coupling because we have assumed that the dimension of @xmath47 is approximately @xmath48 . since the anomalous dimension @xmath49 is of order a one - loop factor , we expect the anomalous dimension of @xmath47 to also be of order @xmath50 for a weakly - coupled theory with @xmath51 . but for strong coupling the operator @xmath47 will have a large anomalous dimension , and its dimension will not have any simple relation to the dimension of @xmath23 . '' has no _ a priori _ meaning . the remarks above apply if we define @xmath47 to be the operator of lowest scaling dimension in the operator product expansion of @xmath52 other than the unit operator . ] exceptions to this are large-@xmath4 theories , as we will discuss below . how small can we take @xmath53 ? in a nearly free theory with an elementary scalar , @xmath2 is an anomalous dimension , which is of order a loop - counting factor in the theory . in other words , perturbation theory is an expansion in powers of @xmath2 , and we expect it to break down when @xmath54 . for values like @xmath55 , perturbation theory is no longer a reliable guide . in these theories , it is possible that the operator @xmath47 has a sufficiently large anomalous dimension to make it an irrelevant operator . we conclude that the general theorem does not imply that theories with @xmath56 are fine - tuned . we now consider large-@xmath4 cft s . in large-@xmath4 theories , the matrix elements of operators factorize , and we can conclude that the operator @xmath47 has dimension near @xmath48 ( up to @xmath57 corrections ) even if the theory is strongly coupled . this is true even for theories with a large t hooft parameter , such as the 4d dual of the randall - sundrum ( rs ) model , as we will discuss below . the operator @xmath47 is therefore relevant for all @xmath5 , and we conclude that large-@xmath4 cft s with scalar operators with @xmath5 are fine - tuned . large-@xmath4 theories are also disfavored as a dynamical electroweak symmetry breaking sector because they give contributions to the @xmath7 parameter that grow with @xmath4 . in theories with a small t hooft parameter , this is simply because @xmath7 arises from a vacuum polarization effects that count the number of microscopic states . we will see that rs models , which have a large t hooft parameter , also predict large @xmath7 in the absence of fine - tuning . for these reasons , we are led to consider theories that do not have large @xmath4 , and which can not be obtained from rs setup . @xmath58 gauge theories with @xmath59 flavors of dirac fermions loses asymptotic freedom for @xmath60 , and for @xmath61 with @xmath62 the theory has a weakly coupled ` banks - zaks ' fixed point @xcite . this allows us to infer the existence of qcd - like theories that flow to strongly coupled cft s in the ir . the loop expansion parameter at the banks - zaks fixed point is @xmath63 , so we know there is a conformal window for a range of @xmath63 . the conformal window ends at some value @xmath64 , and the cft s at this end of the conformal window are necessarily strongly coupled . although this argument is strictly speaking limited to large @xmath4 ( where @xmath63 can be thought of as a continuous parameter ) , it is very reasonable to assume that there are also small-@xmath4 asymptotically free gauge theories that flow to strong conformal fixed points . it is therefore natural to ask whether theories of this kind can give rise to a conformal sector with scalar operators with @xmath5 . to obtain a tractable approximation to the non - perturbative dynamics of qcd - like theories , it is traditional to truncate the schwinger - dyson equations by replacing the full gauge propagator and gauge - fermion vertices by their tree - level values in the landau gauge , giving rise to the so - called ` gap equation . ' however , this limits the parameterization of the non - perturbative effects to spontaneous chiral symmetry breaking via a dynamically generated ` constituent quark mass . ' in particular , it is not able to describe the conformal fixed point dynamics of the banks - zaks fixed point . the conformal window was studied using the gap equation in , which concluded that the conformal window in qcd ends at @xmath65 . it would be very interesting to further explore non - perturbative approximations to qcd - like theories that can capture the physics of the perturbative end of the conformal window and model the non - perturbative dynamics at the strong end of the conformal window . this is beyond the scope of the present work . we now consider the randall - sundrum ( rs ) model , which gives us a concrete , calculable cft . to make the connection to conformal field theory as transparent as possible , it is convenient to write the 5d ads metric as ds^2 = ^2 _ dx^dx^- , where @xmath66 is the ads curvature , and @xmath67 parameterizes the extra dimension . the fact that this metric is invariant under the transformation , @xmath68 and @xmath69 , naturally leads to the interpretation that the parameter @xmath70 is proportional to the energy scale in the corresponding 4d cft . we assume that all dimensionful quantities in the 5d lagrangian are @xmath71 in units of @xmath66 . the physical size of dimensionful couplings at a position @xmath70 in the bulk is then given by @xmath72 . for simplicity , we will use units where @xmath73 . we consider a 5d complex scalar doublet @xmath74 with bulk mass @xmath75 . we will be interested in the situation where @xmath74 gets a vev , breaking an @xmath76 gauge symmetry . in this section for simplicity . we will also ignore an @xmath77 gauge symmetry we need to put in the bulk for any realistic model to guarantee a custodial @xmath78 symmetry in the 4d cft . adding these complications is straightforward and does not alter our conclusions . ] it is therefore convenient to parameterize the field by h = e^i_a _ a ( c 0 + ) , where @xmath79 and @xmath80 ( @xmath81 ) are real fields . when @xmath79 gets a vev , the zero mode of @xmath80 parameterizes the goldstone degrees of freedom . in this section , we are interested in the effects of the vev , and so we will mainly concentrate on the field @xmath79 . solving the equation of motion in the bulk , we obtain the most general 4d poincare invariant solution : = a ^n - 4 + b ^-n where @xmath82 . we restrict attention to masses satisfying the breitenlohner - freedman bound @xmath83 @xcite , so that @xmath84 . for the special case @xmath85 , the general solution is = ^-2 + ^-2 . we will focus on the generic case below . we now discuss the cft interpretation of the solution for @xmath86 . according the the ads / cft correspondence , the field @xmath79 is associated with an operator @xmath23 of the 4d cft . as we will review below , in the conventional case the operator @xmath23 has dimension @xmath87 , which implies @xmath19 . to obtain a 4d cft with @xmath5 , we must change the uv boundary condition @xcite . specifically , we must choose the uv boundary condition to set @xmath88 , so that the ir boundary condition determines @xmath89 . this can be done by adding a uv boundary lagrangian _ uv = -m |h|^2 = -m ^2 . the uv boundary condition is then _ uv & = 0 , + . |_uv & = 0 , which gives ( n - 4 + m ) a _ uv^n - 4 = ( n - m ) b _ uv^-n . the generic solution is @xmath90 , but we can tune @xmath88 by taking m = 4 - n. note that we are tuning a relevant operator that can not be forbidden by symmetries . we now specify the ir boundary conditions . we assume that the ir boundary conditions set . |_ir & = , + . |_ir & = 0 . this can be viewed as the result of adding an @xmath78 invariant ir potential of the form v_ir = ( |h|^2 - ^2)^2 and taking the limit @xmath91 so that the radial component of the field is frozen . ( note that the boundary conditions and ensure that the fields @xmath80 each have a massless goldstone zero mode . ) the results for the coefficients @xmath92 and @xmath89 are then b = _ ir^n & ` generic ' 0 & tuned and a = b _ uv^4 - 2n & ` generic ' _ ir^4 - n & tuned now suppose that the vev for the field @xmath79 spontaneously breaks an @xmath78 gauge symmetry in the bulk and generates a mass for a fermion localized on the uv brane . this is the rs description of breaking the gauge symmetry by the vev of the operator @xmath23 and giving mass to an elementary fermion coupling to @xmath23 . the 4d gauge boson mass is m_w^2 & = _ _ ir^_uv d ^2 + & = _ ir^2 & ` generic ' _ ir^2 & tuned where @xmath93 is the gauge coupling of the 4d gauge zero mode . in the tuned case with @xmath94 , the gauge boson mass grows large as @xmath95 increases , so @xmath96 is sensitive to the uv scale . the physical interpretation of this case is that the gauge symmetry is broken in the uv , not in the ir . this means that this case does _ not _ describe dynamical symmetry breaking . therefore , we will restrict to @xmath97 in the rest of the paper . however , note that even if @xmath97 , taking the @xmath98 limit in the tuned case leads to m_w^2 12 g_4 ^ 2 ^ 2_ir^2 ( _ uv / _ ir ) , so the gauge boson mass is logarithmically sensitive to the uv . we will discuss the 4d interpretation of the logarithm shortly . now consider fermions , a doublet @xmath99 and a singlet @xmath100 localized on the uv brane . we add the uv boundary term _ uv = _ uv^-1 - , where the factor @xmath101 multiplying the kinetic term arises from the conformal factor in the metric . the fermion mass is therefore m_t & = y_t _ uv _ uv + & = y_t _ ir ( _ ir / _ uv ) ^n - 1 & ` generic ' y_t _ ir ( _ ir / _ uv ) ^3 - n & tuned we therefore obtain ~ ( ) ^n - 1 & ` generic ' ( ) ^3 - n & tuned in the 4d cft , the uv boundary term corresponds to coupling an elementary fermion to the cft operator @xmath23 : _ cft = c_t q t^c + the results for the fermion and gauge boson masses are therefore consistent with the 4d interpretation that @xmath23 has scaling dimension d = n & ` generic ' 4 - n & tuned in the tuned case , recall that we require @xmath102 to avoid breaking the gauge symmetry in the uv . at the critical value @xmath103 shows that we have logarithmic uv sensitivity . this is easy to understand . the effective lagrangian at the ir scale has the form _ eff = z |d_h|^2 - v(h ) + , where @xmath104 is the lightest kk mode of the scalar . the logarithm in reflects the fact that the wavefunction factor @xmath105 depends logarithmically on the uv scale , because in the @xmath106 ( @xmath107 ) limit @xmath104 is just a weakly - coupled scalar . ( note that @xmath108 and @xmath109 are quantities defined at the ir scale . ) there is no logarithm in @xmath110 as @xmath98 because the fermion yukawa coupling is not renormalized ( apart from the wavefunction renormalization in @xmath105 ) within our approximation of treating @xmath111 as a background field . we therefore conclude that the tuned case can indeed describe operators with dimension @xmath44 . now we study how much fine tuning is actually necessary to reach the tuned case . if we deviate from the fine - tuning condition by @xmath112 , the coefficient @xmath92 now has to be non - zero to satisfy the uv boundary condition : b ~ m _ ir^n ( ) ^2n-4 . the condition that this is a small correction at the ir brane requires @xmath113 , ( ) ^4 - 2d 1 , where we have used @xmath114 . this quantifies the amount of the fine - tuning . note that for @xmath115 we have the same tuning as in the standard model . of course , we need only something like @xmath116 to push the flavor scale up to @xmath117 tev , but even in this case the fine - tuning is @xmath118 , which is unacceptable . the 4d cft interpretation of this fine - tuning is easy to understand . since the cft contains an operator @xmath119 , it also contains the operator @xmath120 which is invariant under all symmetries . the ads / cft correspondence relates operator products to multi - particle states , and the dimensions of composite operators factorize essentially as a consequence of the factorization properties of multi - particle states in a weakly - coupled 5d field theory . the dimension of the operator @xmath120 is therefore @xmath48 , which means it is a _ relevant _ operator . note that is exactly the amount of fine - tuning required to suppress the effects of a relevant operator of dimension @xmath48 . a convincing check of this interpretation of the fine - tuning can be obtained by computing the vacuum energy in the presence of the vev as a function of the tuning parameter @xmath121 . the bulk action integral vanishes thanks to the bulk equation of motion , leaving only the boundary terms . since the uv boundary condition is chosen to cancel the boundary term at the uv brane , we only get the contribution from the ir boundary : v(_ir ) & ~^5 . |_ir + & ~_ir^4 . on the other hand , in the 4d cft language , we are adding the operator _ uv = _ uv ^ in the uv . if the dimension of @xmath120 is @xmath122 , then the coupling @xmath123 should scale with energy as ( ) = _ uv ( ) ^d-4 . the vacuum energy associated with @xmath124 therefore has the form v & ~_ir^4 + & ~_ir^4 . comparing this with , we see that @xmath122 is indeed equal to @xmath48 and that the tuning @xmath125 precisely corresponds to setting @xmath126 . note that this is a tree - level calculation on the ads side , which corresponds to a calculation at the leading order in the @xmath57 expansion on the cft side . loop corrections correspond to @xmath57 corrections , and are expected to give corrections to the relation @xmath127 . these corrections are suppressed by at least a loop factor , and are small whenever perturbation theory is under control . they therefore can not be used to make the rs model with @xmath5 natural , but it does illustrate that the relation @xmath127 is not exact in general . we will have more to say about the 4d interpretation of the rs model with @xmath5 in subsection [ subsec : lightscalarrs ] below . we showed above that @xmath5 requires tuning in the rs setup , but argued that this is because these are large-@xmath4 cft s . but there is another problem with large-@xmath4 theories . the fact that rs theories are large-@xmath4 theories means that we also expect them to give large contributions to the @xmath7 parameter . however , because the 4d cft corresponding to an rs model also has large t hooft parameter , we can not use nda to estimate @xmath7 . we therefore briefly review the size of @xmath7 in these theories . complete results with numerical coefficients can be found in . contributions to @xmath7 can arise in various ways , but here let us focus on the bulk contribution to @xmath7 arising from the mixing between the unperturbed gauge zero mode and the bulk kk modes via the higgs kinetic term @xcite . this is not necessarily the largest contribution to @xmath7 , but it can be easily estimated , and we will see below that it is already too large . the leading effect of this kind comes from tree - level mixing between the zero - mode gauge bosons and the excited kk modes : s_bulk ~16 g_5 ^ 2 ( ) ^4 , where @xmath128 is the 5d gauge coupling , @xmath129 is the 4d vev that breaks electroweak symmetry , and @xmath130 is the mass of the lightest kk mode . ( recall we are using units where @xmath73 . ) the 5d gauge coupling parameterizes the cft contribution to the running of the 4d gauge coupling . for example , if there is a gauge kinetic term localized on the uv brane with coefficient @xmath131 , we have = + . in order to avoid a landau pole below the uv scale , we require g_5 ^ 2 g_4 ^ 2 ( _ uv / _ ir ) . in order to have @xmath132 ( roughly the current experimental limit ) , we must have m_kk^2 4 v^2 ^1/2 . this shows that small values of @xmath7 can only arise from a hierarchy between @xmath133 and @xmath134 , which however requires fine tuning . the amount of fine tuning is given by ~ ( ) ^2 where @xmath135 is the cutoff in ir units and @xmath136 counts the number of kk modes below the cutoff . we see that we can make @xmath7 small only at the price of either fine - tuning , or taking @xmath137 to be small , meaning that the cutoff is so low that it describes only a small number of kk modes . in this case , the extra dimension is not really buying any predictive power relative to a general effective theory with a cutoff near the tev scale . said another way , since we must allow arbitrary higher - dimension operators suppressed by powers of @xmath135 , the expansion parameter of the effective theory is @xmath138 . in this section , we discuss the general phenomenology of a electroweak symmetry breaking sector consisting of a small-@xmath4 strongly coupled cft with an electroweak order parameter with dimension @xmath0 , with @xmath56 . in the case where the ( moderately ) small value of @xmath2 arises from a small parameter in the fundamental theory , we argue that the theory contains a prominent sub - tev scalar resonance which can be thought of as a higgs that is ` partially ' composite at the tev scale . the compositeness is ` partial ' in the sense that the couplings of the scalar to the strong tev scale dynamics is parametrically suppressed by powers of @xmath2 . although we are interested in 4d cft s that do not have a simple higher - dimensional interpretation , we begin by determining the properties of the lightest scalar kk modes in the rs model . we find the results illuminating , and will argue that they have a simple physical interpretation that extends to the case of a partially composite higgs . to make a fully realistic rs model , we need a model with custodial symmetry . as explained in this can be done by gauging a @xmath139 in the bulk and breaking this down to @xmath140 on the uv brane . in the bulk , which is necessary to properly embed @xmath141 for bulk fermions . if fermions are elementary , as assumed here , there is no need for the @xmath142 factor in the bulk gauge group . ] since we are using rs only as a guide , we will ignore the issue of custodial symmetry and consider the model with only @xmath140 gauged in the bulk . however , it is straightforward to construct a fully realistic rs model at the price of fine - tuning . we now consider the kk decomposition of the bulk scalar doublet @xmath74 . we are mainly interested in the kk modes of @xmath79 , which contains the light higgs - like boson ( see ) . the eigenvalue condition for kk masses of the @xmath79 with the fine - tuned boundary condition in the uv and the shifted boundary condition @xmath143 in the ir which corresponds to after subtracting the vev is given by = - , where x = , y = . expanding this for small @xmath144 , @xmath145 , and @xmath2 , we obtain the mass of the lightest mode : m_0 ^ 2 & = 4 _ ir^2 ^-1 + & 4 _ ir^2 & for @xmath146 + & for @xmath147 + the 4d cft interpretation of these results is the following . the scalar would be exactly massless in the limit of unbroken conformal symmetry , but the conformal symmetry is broken both in the uv and the ir . for _ finite _ @xmath2 in the limit @xmath148 , the ir breaking of conformal invariance dominates . hence , if the scalar couples to the cft with full strength , the conformal breaking should give @xmath149 . however , in our case we expect from the general cft theorem that the scalar should decouple from the cft for @xmath150 . we exactly see this behavior in the first case in , where we have @xmath151 , and the interpretation of this is that @xmath2 controls the couplings of the scalar to the part of the cft in which conformal symmetry is spontaneously broken ( by the ir brane ) . on the other hand , the uv breaking of conformal symmetry dominates for finite @xmath95 and @xmath150 . in this limit the scalar decouples from the cft at low energies ( as required by the general theorem ) but only logarithmically , just like an elementary scalar . for realistic models , the question of whether the breaking of conformal invariance is dominantly in the uv or the ir depends on the scale @xmath95 where the theory approaches the conformal fixed point . this scale could be as high as the planck scale even if the flavor scale is much lower , since the top flavor interactions may be a weak perturbation on the strong cft dynamics . in this case @xmath152 , and since we are interested in @xmath56 , we expect that the breaking of conformal symmetry is dominated in the ir . we will therefore mainly focus on this case in the following . we have taken @xmath3 in the above discussion in order to focus on the parametrics , but we are really interested in @xmath56 . we can easily compute the mass of the scalar kk modes in this simple model to see what we might expect for larger values of @xmath2 . in order to focus on the contribution to the scalar mass from the ir breaking of conformal symmetry we take @xmath148 , in which case the eigenvalue equation for @xmath153 becomes j_-1 + ( x ) = 0 , while the eigenvalue equation for the ` goldstone ' modes @xmath80 is j_1(x ) = 0 . these can be solved numerically , and the mass of the lightest scalar resonances as a function of @xmath2 are plotted in fig . [ fig : kkmasses ] . we see that there is a light ` higgs ' even for moderate values of @xmath2 . note that for @xmath154 we recover a tower of completely degenerate doublets .. ] in the rs version of the cft.,width=453 ] to get more evidences for our claim that @xmath2 controls the coupling of the scalar to the cft , we now compute the couplings of this scalar resonance to gauge bosons and fermions . writing the scalar field as ( x , ) = _ n = 0^f_n ( ) _ n(x ) , the kk wavefunctions @xmath155 are given by f_n ( ) = . the bessel functions are linearly independent for @xmath156 , which is sufficient for our purposes . the boundary conditions determine the ratio @xmath157 and the mass @xmath158 . the uv boundary conditions then give = ( ) ^2 . where @xmath159 . for simplicity we again take @xmath148 to focus on the ir effects , in which case this simply gives @xmath160 . for the lightest kk mode we have @xmath161 for all @xmath70 , and we can expand the bessel function @xmath162 to obtain f_0 ( ) . the two terms are parametrically the same size at the ir brane because @xmath163 ( see ) . we determine @xmath164 by demanding that the real scalar field @xmath165 is canonically normalized : 12 = _ _ ir^_uv d f^2_0 ( ) . the integral over the first term in involves @xmath166 , and the additional factor of @xmath167 makes this term dominate in the determination of @xmath164 . a similar integral appears when computing the @xmath168 coupling , and we therefore find that the @xmath168 coupling is equal to the standard model value up to @xmath169 corrections . more specifically , we find g_w w = g_w w^(sm ) . similarly , when computing the coupling of @xmath153 to the top quark , we note that the first term in dominates at large @xmath70 , so the top coupling is also approximately equal to the standard model , but this time with corrections suppressed by @xmath170 . this is so small that dominant corrections in fact come from perturbative top loops . therefore , we see that in the rs model the couplings of the light scalar to top quarks is very nearly equal to the standard model value , while the couplings to gauge bosons has a larger , order @xmath171 deviation from the standard model value . we now describe the 4d cft interpretation of these results . the kk modes get massive due to the breaking of conformal symmetry , which occurs in both the uv and the ir . the uv breaking can be decoupled by sending @xmath148 . as @xmath150 a full scalar doublet is becoming massless , with deviations from standard model couplings suppressed ( at least ) by powers of @xmath2 . therefore , this supports our interpretation that the scalar doublet is coupled to the strong cft with a coupling that vanishes as a power of @xmath2 . in rs setup fine - tuning is required to decouple the uv breaking of conformal invariance , so this is not a viable solution to the hierarchy problem , but as explained in section [ sec : scalarop ] this fine - tuning need not be present in small-@xmath4 models , which we turn to next . we now consider a strongly coupled small-@xmath4 cft with an electroweak order parameter ( ` higgs operator ' ) @xmath23 with dimension @xmath0 . the key feature of this theory is that @xmath2 is small enough to generate a large hierarchy between the flavor scale and the electroweak scale ( see ) , but at the same time large enough that the operator @xmath47 ( ` higgs mass term ' ) is irrelevant . we argued above that for @xmath3 , the dimension of @xmath47 is @xmath172 , which is therefore relevant . we are assuming that @xmath56 , but the @xmath169 corrections to the dimension of @xmath47 are large enough to make the operator irrelevant . there are at least two ways one can imagine this coming about . the first possibility is that there is no large or small parameter in the theory , and the fact that @xmath24 is close to @xmath173 is simply a numerical accident . in this case , we expect the low - energy effective theory below the scale @xmath174 to be a general strongly coupled nda theory with no small parameters . another possibility is that there is in fact a moderately small parameter in the fundamental theory that controls the size of @xmath2 . ( for example , in the rs model this parameter is the mass of the bulk scalar . ) for @xmath56 it is expected that not all quantities will be numerically small even if parametrically suppressed by a power of @xmath2 . we assume one such numerical accident that the anomalous dimension of @xmath47 is large . , and therefore is expected to be more robust than the smallness of power corrections . ] in this subsection , we will describe the effective field theory below the tev scale for this theory . we will start with the limit @xmath3 , and then extrapolate to @xmath56 . as argued above , for @xmath3 the correlation functions of the operator @xmath23 are approximated by the correlation function of an elementary scalar doublet @xmath104 . in the range of energies where the theory is approximately conformal , the theory can therefore be written in terms of a lagrangian where @xmath104 is coupled to a strong sector : = | d h|^2 - m^2 |h|^2 - 14 |h|^4 + c_1 ( _ 1 h^+ ) + , where @xmath175 is an operator in the strongly - coupled sector of the theory with the same quantum numbers as @xmath104 . we are interested in the case where the coupling to the strong sector is sufficiently strong so that the operators @xmath176 and @xmath177 have large anomalous dimensions , and are irrelevant . this allows us to neglect the couplings @xmath178 and @xmath179 in the conformal regime without fine tuning , and is the basic reason that this class of theories solves the hierarchy problem . in general , large anomalous dimensions for the operators @xmath176 and @xmath177 imply that @xmath180 , where @xmath2 is defined as the anomalous dimension of @xmath104 itself . however , it is an assumption of our analysis that @xmath56 is still small enough that an expansion in powers of @xmath2 gives at least qualitatively correct results for observables . in order for the coupling @xmath181 to be important near the fixed point of the theory , the scaling dimension of the operator must be sufficiently small ( 3 in the limit @xmath3 ) . we must assume that such an operator exists in order for @xmath104 to have sufficiently strong couplings to the strong sector of the cft . in order to make a contribution of order @xmath2 to the anomalous dimension of the scalar field , we must have c_1 ~4 . the powers of @xmath182 are counted using nda , which assumes that the theory has no large or small parameters other than @xmath2 . this is valid because our theory is by assumption a small-@xmath4 theory . does apply . ] we can also consider couplings such as = c_2 _ 2 |h|^2 + c_3 ( _ 3 h^2 + ) + however , since by assumption @xmath176 has scaling dimension greater than 4 , @xmath183 is irrelevant near the fixed point . the coupling @xmath184 is only important at the fixed point if the @xmath185 triplet operator @xmath186 has a sufficiently low dimension ( 2 in the limit @xmath3 ) , but there is no reason to expect that an operator of such low dimension exists in the strong sector of the cft . this obviously generalizes to coupling involving higher powers of @xmath104 , and we conclude that the coupling of a single power of @xmath104 to the strong sector of the cft will dominate at the fixed point . now suppose that the conformal symmetry is broken at an ir scale @xmath174 . ( we assume that the conformal fixed point is reached at very high energies , and therefore neglect any uv breaking of conformal symmetry . ) for example , we can have a new non - abelian gauge group that gauges a global symmetry of the cft . if the new gauge coupling is asymptotically free , it will get strong in the ir and break the conformal symmetry , as discussed in the introduction . the point of this is that the breaking of conformal symmetry occurs in the strong sector of the cft , since the new gauge fields do not couple directly to @xmath104 . therefore , @xmath104 learns the conformal breaking only via coupling to the cft , and we would like to understand how @xmath2 controls this communication . below the scale @xmath174 , the coupling @xmath181 in will generate all possible interactions of the scalar field @xmath104 . in particular , it will generate the @xmath176 and @xmath177 terms , which are no longer rendered irrelevant by the strong conformal dynamics . the effective lagrangian below the scale @xmath174 is then _ eff = |d h|^2 + ( , ) , where @xmath187 is an order-1 function that parameterizes the effects of the strong sector , and @xmath188 is the gauge covariant derivative of the standard model . the factors of @xmath182 are put in according to nda , which assumes that the theory is strongly coupled at the scale @xmath174 with no large or small parameters other than @xmath2 . here we are again using the assumption that this is a small-@xmath4 theory . expanding out , we obtain _ eff = |d h|^2 - m_h^2 |h|^2 - 14 |h|^4 + , with m_h^2 ~^2 , ~16 ^ 2 ^2 . note that the result for the mass agrees with the rs calculation . if @xmath189 , electroweak symmetry will be spontaneously broken by a vev for @xmath104 , and we obtain m_w^2 = 12 g_2 ^ 2 v^2 , m_z^2 = 12 ( g_1 ^ 2 + g_2 ^ 2 ) v^2 from the @xmath104 kinetic term , where = , with @xmath190 as in the standard model . we therefore obtain ~4 v. the scale of new strong dynamics is parametrically below the scale @xmath191 , but for @xmath56 there are large uncertainties of the nda estimates . to be smaller than the gauge loop contribution to the anomalous dimension , @xmath192 . in fact , at this critical value we have @xmath193 . ] we now discuss the couplings of @xmath104 to the standard model gauge bosons . because @xmath194 , the function @xmath187 in the effective lagrangian contains all possible electroweak breaking couplings of @xmath104 and gauge bosons with @xmath195 coefficients in terms of @xmath196 and @xmath174 . for example , the leading correction to the @xmath197 coupling is _ eff ~ ( ) ^2 ~g_2 ^ 2 v w_^2 h + . this gives an @xmath169 correction to the coupling of the higgs to gauge bosons , as anticipated from the rs computation . we now consider the couplings to the top quark . in the conformal regime , we add a coupling of the form = c_t ( h q t^c + ) between the scalar doublet @xmath104 and the elementary top quark . the field @xmath104 has an anomalous dimension @xmath2 that suppresses the coupling @xmath198 at low energies , so that the top quark yukawa coupling at the weak scale is y_t = c_t(_t ) ( ) ^ , where @xmath17 is the scale where the top quark coupling is generated . as discussed above , we assume that @xmath199 is sufficiently large that @xmath200 at the weak scale . the leading coupling of @xmath104 to the top quark is therefore the same as in the standard model . the leading corrections to this come from @xmath104 loop effects such as the one shown in fig . [ fig : htt ] , which gives y_t ~ ( ) ^3 ~ , where the factor of @xmath201 arises because the diagram is ir dominated . this diagram is the same as in the standard model , except that the 3 point higgs interaction does not have the standard model value . this can be seen by expanding the function @xmath187 in , where we see that the vev and the 2- and 3-point functions of the higgs expanded about the vev are independent . this is expected to give a deviation from the standard model prediction of between @xmath202 and @xmath40 . coupling in the low energy effective theory below @xmath174 . the bullet represents a local operator that includes cft contributions to the 3-point function as well.,width=188 ] we now turn to the phenomenology of this class of models . one important question in these models is the size of precision electroweak corrections . the effective operator that contributes to @xmath7 is obtained from the effective lagrangian : _ eff ~ h^w^ h b _ ~ w^_3 b _ + . this gives a contribution to @xmath7 with nda strength , @xmath203 , similar to small-@xmath4 technicolor . the @xmath12 parameter requires breaking of custodial symmetry , and therefore requires an additional hypercharge loop . the effective coupling that gives rise to @xmath12 is therefore _ eff ~ boson field . this gives a contribution to @xmath12 that is parametrically smaller than the nda value : t ~ ~ t_nda . taken at face value , this is not so great for precision electroweak fits , since the best fit for nonzero electroweak corrections has both @xmath7 and @xmath12 positive and comparable . however , given the uncertainties in these estimates these theories are still viable . the phenomenology at high energy colliders depends on the structure of strong resonances at the tev scale . we can not make any rigorous prediction about these resonances other than the fact that they must exist to unitarize ww scattering ( the ` no lose theorem ' ) . in the case where the small dimension of the electroweak order parameter is due to a moderately small parameter in the fundamental theory , we argued above that there will be a prominent scalar resonance below the tev scale whose couplings to gauge bosons and the top quark are parametrically close to the standard - model values . the deviation from the standard - model value for the coupling to the top quark is quite small , between @xmath202 and @xmath40 , while the deviation from the standard model value for the @xmath168 coupling is of order @xmath2 , which is expected to be at least @xmath40 . the possibility of studying a ` non - standard ' heavy higgs - like scalar at the lhc has been discussed by a number of authors @xcite . the discovery of such a particle and the measurement of deviations from the standard - model predictions for its couplings may well be the first indication of strong conformal dynamics as the origin of electroweak symmetry breaking . we have described a new paradigm for dynamical electroweak symmetry breaking in which the electroweak symmetry breaking sector is a conformal field theory ( cft ) above the tev scale . conformal symmetry and electroweak symmetry are broken at the tev scale , triggered by an asymptotically - free gauge group or a slightly relevant operator becoming strong , or by a relevant operator with a coefficient made small naturally by symmetries . any of these mechanisms stabilizes the weak scale against quantum corrections and gives a solution of the hierarchy problem . the flavor problem of dynamical electroweak symmetry breaking is solved if the dimension of the cft operator that acts as the order parameter for electroweak symmetry breaking has a dimension @xmath24 close to 1 , the dimension of an elementary higgs scalar field . for @xmath0 , the scale where the top quark yukawa coupling becomes strong is raised to @xmath204 , where @xmath205 . for @xmath206 , this is large enough to effectively decouple flavor . finding a cft with the required properties is highly nontrivial . weakly coupled cft s can certainly have operators with dimension near 1 in the form of an elementary scalar higgs field @xmath104 , but this clearly does not solve the hierarchy problem because of the existence of the relevant operator @xmath207 with dimension near 2 . what is required is a strongly - coupled cft with a scalar operator @xmath23 with dimension @xmath0 , where strong cft dynamics renders the operator @xmath47 irrelevant by giving it a large anomalous dimension . we have shown , however , that strong cft s with large @xmath4 , including those obtained from the ads / cft correspondence , have the property that the dimension of @xmath208 is close to @xmath48 , and are therefore fine - tuned for @xmath5 . we are therefore led to consider strongly - coupled , small-@xmath4 cft s . these theories also naturally have small electroweak precision corrections , addressing another strong constraint on models of dynamical electroweak symmetry breaking . the difficulty is that there are no reliable theoretical tools for studying such theories , and in fact no explicit examples are known . in the case where the smallness of @xmath2 is due to a small parameter in the fundamental theory , we argued that there will be a prominent scalar resonance with couplings to gauge bosons and the top quark that are comparable to that of a heavy standard - model higgs , but with @xmath169 deviations that can be measured at lhc . we believe that this gives strong motivation to experimental studies of a heavy higgs - like particle , and look forward to a decisive test of these ideas . we would like to thank n. arkani - hamed and a. cohen for discussions , and j. terning and r. sundrum for comments on the manuscript . finally , we express our gratitude to k. agashe for pointing out an error in the appendix . m. a. l. is supported by nsf grant phy-009954 . t. o. is supported by doe contract de - fg03 - 91er-40676 . in this appendix , we consider the possibility that fermion masses are generated by mixing with fermionic operators of the cft . that is , we suppose that the cft contains operators with quantum numbers conjugate to the standard model fermion fields , and we include interaction terms = z_q q ^c + z_u u^c + z_d d^c + z_l l ^c + z_e e^c , where @xmath209 are standard model fermions , and @xmath210 are fermionic cft operators . the unitarity limit on the dimension of the cft operators is @xmath211 , the dimension of a free fermion , so the couplings @xmath212 can be marginal or even relevant without approaching the unitarity limit for cft operators . this means that the flavor scale where these operators are generated can be decoupled completely , just as in the standard model . we will show that this mechanism generally require a mild @xmath213% tuning to accomodate constraints on @xmath214 and the @xmath12 parameter together with the heavy top mass . this mechanism for generating fermion masses was first considered in the context of technicolor theories in . it has been revived recently in the context of rs theories , where it corresponds to putting standard model fermions in the bulk @xcite . we consider this framework here using our cft language , which makes it clear that our analysis is model - independent . in , the couplings @xmath212 are @xmath215 matrices in flavor space . these @xmath216 s act as spurions that violate the @xmath217 flavor symmetry that is otherwise present , and their transformation properties under the flavor symmetry determine the structure of flavor violation in this model . we will normalize the cft operators @xmath210 so that @xmath218 corresponds to strong coupling at the scale of conformal and electroweak symmetry breaking @xmath174 . if the conformal and electroweak symmetry breaking is strong with no large or small parameters , nda tells us that the quark mass matrices are m_u = c z_q^z_u , m_d = c z_q^z_d , where @xmath219 . are the same for up- and down - type fermions . ] to get the top quark mass , we must therefore have ( z_q^z_u)_33 ~ ~110 . because the @xmath216 s violate flavor and custodial symmetry , they give rise to corrections to the @xmath220 parameter and the @xmath221 vertex . note that the leading corrections to the @xmath220 parameter do not involve internal gauge bosons , so we can ignore the custodial symmetry breaking from @xmath141 . in this limit the standard model and the cft each have a separate @xmath139 symmetry , which are broken to diagonal subgroups by @xmath222 and @xmath223 . specifically , the @xmath216 s transform like z_l & l_sm z_l l_cft^ + z_r & r_sm z_r r_cft^ , where @xmath224 and @xmath225 . electroweak symmetry is broken with nda strength in the strong sector in the pattern @xmath226_{\rm cft } \to su(2)_{\rm cust , cft}$ ] , so custodial symmetry breaking in the standard model sector depends on the spurion @xmath227 . this spurion has @xmath228 , while a custodial symmetry violating contribution to the @xmath105 mass has @xmath229 . therefore . we thank kaustubh agashe for pointing out our mistake . ] m_z^2 ~ ( z_u z_u^- z_d z_d^)^2 ~ ( z_u z_u^)^2 in order to satisfy the constraint on the @xmath220 ( or @xmath12 ) parameter , we require ( z_u z_u^)^2 1100 . on the other hand , the correction to the @xmath230 vertex is g_z b|b ~4 ( z_q^z_q)_33 . satisfying the experimental constraint requires ( z_q^z_q)_33 1100 . we see that generically there is a tension in satisfying all three constraints : the @xmath220 parameter constraint , the @xmath39 constraint and the top mass condition . for example , with @xmath231 , forces @xmath232 , which makes the @xmath220 constraint completely safe while requiring an additional contribution to the @xmath214 coupling that cancels the non - standard one to @xmath202 accuracy . with @xmath233 , the @xmath39 constraint is just on the edge while the @xmath220 constraint needs @xmath202 tuning . however , we can ` compromise ' and choose @xmath234 , which puts us just on the edge of the @xmath220 constraint , while we must find additional contributions to @xmath214 that cancel the unwanted cft contribution to @xmath40 accuracy . therefore , in this part of the parameter space , this framework is viable with mild tuning . b. holdom , phys . b * 150 * , 301 ( 1985 ) ; t. akiba and t. yanagida , phys . b * 169 * , 432 ( 1986 ) ; t. w. appelquist , d. karabali and l. c. r. wijewardhana , phys . lett . * 57 * , 957 ( 1986 ) ; k. yamawaki , m. bando and k. i. matumoto , phys . * 56 * , 1335 ( 1986 ) ; t. appelquist and l. c. r. wijewardhana , phys . d * 35 * , 774 ( 1987 ) ; phys . d * 36 * , 568 ( 1987 ) . j. m. maldacena , adv . * 2 * , 231 ( 1998 ) [ int . j. theor . phys . * 38 * , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] ; e. witten , adv . * 2 * , 253 ( 1998 ) [ arxiv : hep - th/9802150 ] ; h. verlinde , nucl . b * 580 * , 264 ( 2000 ) [ arxiv : hep - th/9906182 ] ; j. maldacena , unpublished remarks ; e. witten , 1999 itp santa barbara conference ` new dimensions in field theory and string theory , ' , http://www.itp.ucsb.edu/online/susy c99/discussion ; e. verlinde and h. verlinde , jhep * 0005 * , 034 ( 2000 ) [ arxiv : hep - th/9912018 ] ; n. arkani - hamed , m. porrati and l. randall , jhep * 0108 * , 017 ( 2001 ) [ arxiv : hep - th/0012148 ] ; r. rattazzi and a. zaffaroni , jhep * 0104 * , 021 ( 2001 ) [ arxiv : hep - th/0012248 ] . t. appelquist , m. einhorn , t. takeuchi and l. c. r. wijewardhana , phys . b * 220 * , 223 ( 1989 ) ; v. a. miransky and k. yamawaki , mod . lett . a * 4 * , 129 ( 1989 ) ; k. matumoto , prog . phys . * 81 * , 277 ( 1989 ) ; v. a. miransky , m. tanabashi and k. yamawaki , phys . lett . b * 221 * , 177 ( 1989 ) ; mod . lett . a * 4 * , 1043 ( 1989 ) . t. appelquist , j. terning and l. c. r. wijewardhana , phys . lett . * 79 * , 2767 ( 1997 ) [ arxiv : hep - ph/9706238 ] . y. grossman and m. neubert , phys . b * 474 * , 361 ( 2000 ) [ arxiv : hep - ph/9912408 ] ; t. gherghetta and a. pomarol , nucl . b * 586 * , 141 ( 2000 ) [ arxiv : hep - ph/0003129 ] ; s. j. huber and q. shafi , phys . d * 63 * , 045010 ( 2001 ) [ arxiv : hep - ph/0005286 ] ; phys . b * 498 * , 256 ( 2001 ) [ arxiv : hep - ph/0010195 ] . c. csaki , j. erlich and j. terning , phys . d * 66 * , 064021 ( 2002 ) [ arxiv : hep - ph/0203034 ] ; k. agashe , a. delgado , m. j. may and r. sundrum , jhep * 0308 * , 050 ( 2003 ) [ arxiv : hep - ph/0308036 ] ; g. cacciapaglia , c. csaki , c. grojean and j. terning , [ arxiv : hep - ph/0401160 ] . j. bagger , s. dawson and g. valencia , nucl . b * 399 * , 364 ( 1993 ) [ arxiv : hep - ph/9204211 ] ; j. bagger , v. d. barger , k. cheung , j. f. gunion , t. han , g. a. ladinsky , r. rosenfeld , c. p. yuan , phys . d * 52 * , 3878 ( 1995 ) [ arxiv : hep - ph/9504426 ] ; k. iordanidis and d. zeppenfeld , phys . d * 57 * , 3072 ( 1998 ) [ arxiv : hep - ph/9709506 ] ; h. j. he , y. p. kuang , c. p. yuan and b. zhang , phys . b * 554 * , 64 ( 2003 ) [ arxiv : hep - ph/0211229 ] ; b. zhang , y. p. kuang , h. j. he and c. p. yuan , phys . d * 67 * , 114024 ( 2003 ) [ arxiv : hep - ph/0303048 ] .

the dominant constituents of the universe s matter are believed to be collisionless in nature and thus their modelling in any self - consistent simulation is extremely important . for simulations that deal only with dark matter or stellar systems , the conventional n - body technique is fast , memory efficient , and relatively simple to implement . however when extending simulations to include the effects of gas physics , mesh codes are at a distinct disadvantage compared to sph codes . whereas implementing the n - body approach into sph codes is fairly trivial , the particle - mesh technique used in mesh codes to couple collisionless stars and dark matter to the gas on the mesh , has a series of significant scientific and technical limitations . these include spurious entropy generation resulting from discreteness effects , poor load balancing and increased communication overhead which spoil the excellent scaling in massively parallel grid codes . in this paper we propose the use of the collisionless boltzmann moment equations as a means to model the collisionless material as a fluid on the mesh , implementing it into the massively parallel flash amr code . this approach which we term `` collisionless stellar hydrodynamics '' enables us to do away with the particle - mesh approach and since the parallelisation scheme is identical to that used for the hydrodynamics , it preserves the excellent scaling of the flash code already demonstrated on peta - flop machines . we find that the classic hydrodynamic equations and the boltzmann moment equations can be reconciled under specific conditions , allowing us to generate analytic solutions for collisionless systems using conventional test problems . we confirm the validity of our approach using a suite of demanding test problems , including the use of a modified sod shock test . by deriving the relevant eigenvalues and eigenvectors of the boltzmann moment equations , we are able to use high order accurate characteristic tracing methods with riemann solvers to generate numerical solutions which show excellent agreement with our analytic solutions . we conclude by demonstrating the ability of our code to model complex phenomena by simulating the evolution of a two armed spiral galaxy whose properties agree with those predicted by the swing amplification theory . methods : numerical hydrodynamics

1210.5246

galaxies are complex systems consisting of a multitude of components that often require different approaches for both analytical and numerical study . the gaseous component can be described with the classic hydrodynamics equations ( che ) thanks to frequent collisions between constituent particles that act to isotropise the local pressure . on the other hand , stellar and dark matter particles lack collisions which effectively means the local pressure is anisotropic and should be described by a stress tensor . although this modification is rather straightforward to implement in the che ( see e.g. * ? ? ? * ; * ? ? ? * ) , this approach has mostly been applied to analytical and semi - analytical studies of growing instabilities in stellar systems ( e.g. * ? ? ? when it comes to numerical simulations of collisionless galactic components , n - body methods have often been the method of choice ( e.g. * ? ? ? one owes this to the relative ease with which they can be implemented and one can start modelling three - dimensional collisionless systems - often with rather small numbers of particles . in particular , large dark - matter - only n - body simulations ( e.g. the _ millennium simulation _ * ) have led to significant advances in our understanding of cosmological structure formation , allowing some of the founding tenets of the current paradigm to be tested . as simulations have grown in complexity , incorporating the effects of baryonic matter , it has become necessary for theorists to develop a wider range of techniques to model the gaseous component . for this , there exist two dominant approaches ; that of smooth particle hydrodynamics ( sph ) and alternatively , grid codes . whilst both approaches agree for simple tests where analytic solutions are available , discrepancies are still often found when modelling more complex phenomena . sph is very robust and memory efficient , allowing large cosmological runs to self - consistently model the effects of gas physics and collisionless matter ( see for example the _ owls _ and _ gimic _ simulations ; * ? ? ? * ; * ? ? ? * ) . however , due to the higher shock capturing powers and more flexible refinement criteria of mesh codes , along with the tendency of sph to suppress the formation of hydrodynamic instabilities and turbulent mixing @xcite , grid codes present a powerful alternative . in particular , grid based schemes are better suited to modelling small - scale physical processes such as heat conduction and radiative transfer . unfortunately , when coupling the collisionless matter to the collisionally dominated baryonic matter , it begins to become clear that the n - body approach is not always ideally suited to grid codes . in sph the gas is discretised into particles of a given mass , allowing the effects of gravity to be calculated using an n - body method in a very convenient self - consistent manner for all components . grid codes however require the mapping of the collisionless particle masses to the mesh , from which the global density field can be used to obtain the gravitational potential . this can be done using the multigrid technique @xcite with either a nested grid or adaptive mesh refinement ( amr ) . alternatively a fourier transform or a hybrid of the two can accelerate this process . once the gravitational acceleration is computed on the grid , it is then interpolated back to the particles . these are updated using a leapfrog method . this particle - mesh approach , a mapping of discrete particle properties to and from the mesh , represents a series of scientific and technical limitations to grid codes which degrade the physical reliability of their results and hampers the scalability of grid based simulations . these include : * _ spurious generation of entropy _ - the particle - mesh technique allows numerical noise to be introduced if insufficient particles are present in a given resolution element or `` cell . '' it has been argued by @xcite that in regions where the mesh is over - sampled relative to the number density of particles , then some cells may receive mass mapped from particles whilst their neighbours may not . this can lead to strong variations in the local density field when realistically the field should be smooth and continuous . this can drive spurious weak shocks and turbulence on small scales , generating an unphysical entropy excess . @xcite cites this as the origin of the higher entropy cores seen in galaxy cluster simulations when run using grid codes , compared to those run with sph codes which have cuspy low entropy cores @xcite . + as actually these particles should represent a continuous mass field , this effect can be limited by mapping the discrete mass of a given particle over a larger region using some pre - defined kernel function ( for example the cloud - in - cell technique ; * ? ? ? * ) , this however leads to more technical problems . * _ increased communication overhead _ - in massively parallel programs , a large amount of effort is often put into minimising communication between different nodes over the network . such communication is relatively time consuming , with latencies ( the time taken to establish and terminate a communication ) many orders of magnitude greater than the clock speed of the processor . as such , many calculations could be performed whilst waiting for a given group of processes to communicate , and if there is a list of different processes to communicate with , the time adds up . therefore any common communication scheme and grouping of messages , along with a reduction in the number of different nodes to which we must communicate with , presents a significant speed up in the code . + when particle masses are mapped using a kernel , there will inevitably be some overlap with neighbouring cells which may or may not reside on different nodes . although communication over the network is natural for any parallel simulation , it is lower when dealing with a system which advects only fluxes through cell surfaces instead of a system which can map mass to any neighbouring cell , whether it is directly adjacent to a face or merely in contact with an edge or vertex of the cell . this is demonstrated in figure [ fig : pmvsflux ] in which a schematic layout of two dimensional ( 2d ) block of cells ( as used in the flash code ) is shown along with the dashed outline of the neighbouring blocks which surround it . the internal cells in the block are shown in red and a surrounding layer four guardcells deep contains boundary data copied from neighbouring blocks . grey cells are guardcells which do not need to be updated or communicated for the given scheme , whilst green cells are guardcells which need to be updated or exchanged with neighbours . a 2d hydrodynamic simulation ( top panel ) requires the communication of boundary data to compute the four fluxes through its cell faces with the grey corner guardcells not being used . the bottom panel however , shows how mapping of particle mass over a kernel of finite size can lead to particle properties being mapped to any of the surrounding guardcells . this data then needs to be communicated to the internal cells of the blocks to which they correspond . thus in 2d the communication overhead is doubled , whilst in 3d it can increase by a factor of 3.3 ( 6 face fluxes / 20 edge and corner cells to which mass may be mapped ) . sophisticated space filling curves can help optimise the distribution of blocks ( or individual cells ) across nodes so that as many adjacent neighbours as possible reside on the same processor , minimising inter - node communication . however increased overhead remains unavoidable , especially as we always remain bottle - necked by the slowest processor if the simulation is to proceed in lock - step . * _ poor load balancing _ - the use of particles in conjunction with a mesh also presents a load - balancing issue as many forms of astrophysical structure formation result in the natural concentration of the majority of mass in dense compact regions . this leads to too many particles accumulating in one or two cells whilst the remaining cells have few or none , even if the initial particle distribution is fairly uniform . ultimately , the mass field ends up poorly sampled in the under - dense regions of the simulation whilst the mapping of properties to and from the mesh is performed by a limited number of cells within the densest regions . as already mentioned , the hydrodynamics requires that as many neighbouring cells as possible be stored on the same processor . therefore these heavily occupied dense cells are all stored on the same few processors leading to extremely poor load balancing . although distributing these dense blocks more evenly would improve the load balancing for the particle - mesh algorithm , it would ruin the load balancing for the hydrodynamics , creating an impossible system to optimise . fortunately the use of the collisionless boltzmann moment equations allows us to overcome all of these limitations since it allows the stellar and dark matter components to be represented as a collisionless fluid in a near identical manner to the standard baryonic gas . this removes the need to develop a separate parallelisation scheme for the particle - mesh approach and allows us to use the same well parallelised hydrodynamics scheme . instead of communicating guardcell data for the gas and stellar material separately , the data can now be grouped and communicated all at once , minimising the latency overhead and maximising bandwidth usage . ideal load - balancing is also achieved as the time taken to calculate the inter - cell fluxes is independent of the amount of collisionless mass within a cell . thus processors will not be left idle whilst others calculate , provided that the cells are evenly distributed across processors . given that the hydrodynamics in the flash amr code we use has been shown to scale well for tens of thousands of processors @xcite , then by using the same parallelisation scheme for the collisionless material , we naturally preserve this scaling . in an era of increasingly large and complex simulations , this scalability is vital in keeping performance in - line with scientific requirements . most importantly , as the density of the collisionless material is now represented in each cell continuously , this removes the potential for spurious generation of entropy through discreteness effects introduced when mapping particle properties to the mesh . it also removes limitations involved when converting gas into stars of discrete masses - problematic if cells contain insufficient mass but yet have densities that satisfy star formation criteria . we present in this paper our new implementation of the boltzmann moment equations which allows collisionless material to be modelled as a collisionless fluid on the mesh . a major application of this technique will be in modelling the behaviour of large collections of stars , such as those in a galaxy , as a collisionless fluid in much the same way that we model gas hydrodynamics in grid codes . we will therefore refer to it from here on as `` collisionless stellar hydrodynamics . '' we begin in [ sec : boltzmanneqns ] with a derivation of the zeroth , first and second order moment equations of the collisionless boltzmann equation and then proceed in [ sec : numericalschemes ] to outline the flash hydrodynamic code ( [ sec : flashcode ] ) and the different numerical schemes we use to numerically integrate the equations . these include both a simpler riemann solver free technique ( [ sec : ktscheme ] ) and a more sophisticated riemann solver scheme with characteristic tracing ( [ sec : riemannsolver ] ) . in [ sec : tests ] we outline our discovery of how standard hydrodynamic test problems for which analytic solutions are known , can be modified under special conditions to reproduce the results of a collisionless fluid . this enables us to use the sod - shock - tube test problem to test our results , the findings of which are given in [ sec : sodshock ] . we continue our tests with both a pressure - free spherical collapse test ( [ sec : sphercollapse ] ) and an anisotropic collapse with non - isotropic dispersion tensor in [ sec : anisocollapse ] . as a more complex test of our collisionless stellar hydrodynamics , in [ sec : diskgal ] we track the growth of a two armed spiral in a stellar disc , contrasting the resulting structure with the predictions of the swing amplification theory . we then conclude our findings with an overview of our results , a discussion of the runtime performance of our code and the future implications of our work in [ sec : summary ] the dynamics of collisionless particles are fundamentally described by the collisionless boltzmann equation for the distribution function of particles in the six dimensional position - velocity phase space @xmath0 @xmath1 where @xmath2 is the acceleration due to gravity and curvilinear terms ( in non - cartesian coordinate systems ) . in this paper , we use cartesian coordinates and hence @xmath3 . equation ( [ boltzmann ] ) is a six - dimensional , time - dependent partial differential equation and although its direct numerical solution is possible ( see , e.g. * ? ? ? * ) high resolution simulations are prohibitively expensive even for the most powerful supercomputers . n - body methods circumvent this problem by solving a system of ordinary differential equations for the dynamics of individual particles , which on galactic scales often represent a cluster of objects rather than individual physical entities . in contrast to this _ discrete _ mechanics approach , a _ continuous _ mechanics approach is based on taking the first three velocity moments of equation ( [ boltzmann ] ) . more specifically , equation ( [ boltzmann ] ) is multiplied by ten quantities @xmath4 , where @xmath5 , and integrated over the velocity space @xmath6 to obtain partial differential equations for the volume density @xmath7 , the mean velocity @xmath8 , and velocity dispersions @xmath9 of collisionless particles . the resulting continuity , momentum , and velocity dispersion equations , named boltzmann moment equations ( bme ) , are as follows , + @xmath10 here , @xmath11 and @xmath12 is the symmetric velocity dispersion tensor . all indices obey the usual einstein summation rule . equations ( [ contin])-([dispers ] ) are closed by using the usual zero - heat - flux approximation and setting the third - order moments to zero , @xmath13 . we discuss the applicability of this approximation in appendix b. equations ( [ contin ] ) and ( [ moment ] ) are in fact the jeans equations ( e.g. * ? ? ? * ) , from which the classic hydrodynamics equation are derived assuming a diagonal and isotropic form of the stress tensor @xmath14 and introducing the gas pressure @xmath15 and internal energy per unit volume @xmath16 , where @xmath17 and @xmath18 are the number of translational and total ( translational plus internal ) degrees of freedom . in the boltzmann moment equations ( bme ) , no simplifying assumptions regarding the form of the stress tensor are made . due to this reason , every component of @xmath14 needs to be evolved in time separately . equation ( [ dispers ] ) does that for the quantity @xmath19 , which can be regarded as an analogue to the total energy @xmath20 in the che . in total , one has 10 equations for ten unknown quantities : @xmath21 , @xmath22 , and @xmath23 . we note that equations ( [ contin])-([dispers ] ) can be convolved to the usual che if an isotropic and diagonal form of @xmath14 is assumed ( see appendix a ) . the fundamental similarity of the bme with the che makes it straightforward to couple collisionless ( stars , dark matter ) and collisional ( gas , dust ) galactic subsystems in one grid - based code . below , we provide a detailed implementation of two numerical solvers which can accurately follow the behaviour of a collisionless fluid in the flash amr code . a full solution of the collisionless boltzmann equation requires a detailed knowledge of the velocity phase space distribution function for a collisionless fluid . unfortunately no known solution is currently available and therefore the first scheme we choose to implement is one which neither requires any detailed knowledge of the systems equation of state , nor any consideration of the behaviour of the different entropy and pressure waves around cell interfaces as would be used by more sophisticated riemann solver schemes . for this purpose we adopt the unsplit solver of @xcite for hyperbolic partial differential equations ( kt scheme hereafter ) . the scheme is second order accurate in both space and time using a simplistic runge - kutta midpoint technique to ensure that the solution is second order accurate in time . this removes the need to deconvolve the moment equations into their separate eigenvectors within the riemann interaction fan . despite being riemann solver free , leading to an averaging over interactions within the riemann fan , when modelling a conventional ideal gas @xcite have shown that it possess very low numerical diffusion and can accurately capture sharp shocks and contact discontinuities . this makes it an ideal means to test the reliability of higher order more sophisticated schemes and as it is also highly extensible , higher order moments of the boltzmann equation can be readily incorporated . the second scheme we implement involves a more accurate modified state reconstruction using a new set of eigenvectors so that riemann solvers can be directly applied . unlike the kt scheme , the riemann solver approach achieves second order accuracy in time by reconstructing the time averaged fluid properties ( primitive or characteristic ) using more detailed knowledge of the eigenvector structure of the fluid equations . this grants the shock superior resolving powers , allowing sharper changes in the fluid variables to be tracked with less numerical diffusion . the scheme is also found to be relatively efficient compared to more dissipative schemes . as far as the authors are aware , this represents the first time that direct solution of the riemann problem for an anisotropic collisionless fluid has been achieved , with previous efforts focussing on either the use of less complex schemes such as the aforementioned kt scheme or alternatively more complete direct integration of the collisionless boltzmann equation which represents a more significant computational overhead . flash was developed at the alliance centre for astrophysical thermonuclear flashes as part of a doe grant . originally intended for the study of x - ray bursts and supernovae , it has since been adapted for many astrophysical conditions and now includes modules for relativistic hydrodynamics , thermal conduction , radiative cooling , magneto - hydrodynamics , thermonuclear burning , self - gravity and particle dynamics via a particle - mesh approach . in this study we use flash version 3.2 . the code is massively parallel , scaling well over more than tens of thousands of processors @xcite , and is highly modular enabling several different solvers for the hydrodynamics and stellar dynamics to be introduced . the code features an oct - tree structure with the simulation being divided into `` blocks '' , each of which contains a fixed number of cells ; @xmath24 in each dimension for our simulations . in addition to these cells , an additional four `` guardcells '' either side of the internal cells are required , which contain a copy of the properties of the neighbouring blocks in - case they are stored on a different node . the code is capable of adaptive mesh refinement , increasing the resolution in regions where a user defined refinement criteria is satisfied and conversely removing refinement in regions where excessive refinement is both unnecessary and wasteful . refinement is achieved by sub - dividing each `` parent '' block into @xmath25 `` child '' blocks . those at the highest level of refinement for their region of space are termed `` leaf blocks . '' the code enforces a maximum jump of two between levels of refinement . at refinement level @xmath26 , a fully refined amr grid will have a total of @xmath27 cells on a side . to identify regions of rapid flow change , flash s refinement and de - refinement criteria can incorporate the adapted @xcite error estimator . this calculates the modified second derivative of the desired variable , normalised by the average of its gradient over one cell . with this applied to the density as is common place , we impose maximum and minimum levels of refinement ; @xmath28 and @xmath29 . we implement into the flash amr code the semi - implicit numerical advection scheme of @xcite . we briefly outline the scheme below but refer readers to their paper for a more in - depth discussion . in conjunction with the semi - implicit scheme , we use a runge - kutta integration scheme for the temporal integration using the following time derivative for the fluid variables , @xmath30 ( eqn 4.2 in * ? ? ? * ) , @xmath31 where the numerical fluxes are defined as , @xmath32 here @xmath33 and @xmath34 represent the fluid variable values at the left and right side of cell @xmath17 with the indices @xmath35 and @xmath36 representing the interpolated values to the left and right sides of the respective cell interfaces . the fluxes @xmath37 are those taken from the derived bme ( [ contin ] , [ moment ] and [ dispers ] ) . the quantity @xmath38 represents the `` spectral frequency '' of the system ; @xmath39 as this is the speed with which information is propagated , it is unsurprising that this equates to the effective sound speed of the collisionless fluid . this can be computed component wise with no loss of numerical accuracy , or alternatively as we show in [ sec : tests ] that it is possible to reconcile the bme with the che , we can use the conventional sound speed formula ( [ eqn : soundspeed ] ) in a directionally dependant manner . this makes use of a value for @xmath40 and the dispersion tensor @xmath41 which can be treated as a directionally dependant pressure @xmath42 . @xmath43 the scheme can easily be extended to arbitrary orders of temporal and spatial accuracy ( see * ? ? ? * ) , however we choose to implement a second order accurate runge - kutta solver , providing a good compromise between accuracy and computational cost . higher order schemes would require additional guardcell updates at the many midpoints , increasing communication overhead , whereas the second order accurate scheme can perform the update using just the information in locally stored guardcells . the only modification which needs to be made in order for the scheme to function well with the adaptive mesh , is that flux conservation be checked at refinement boundaries . fluxes calculated for higher resolution cells are considered to be more accurate than those at lower levels of refinement . therefore the mean flux from the four higher resolution cells on a refinement boundary is calculated and passed to the neighbouring block at the coarser level of refinement to be used in place of the locally calculated flux . we implement our riemann solver based collisionless stellar hydrodynamics module into flash using the muscl - hancock dimensionally unsplit scheme which has already been successfully implemented for the che including magneto - hydrodynamics @xcite . we adopt an unsplit approach over a less memory intensive split version ( e.g. ppm of * ? ? ? * ) given its superior ability to preserve flow symmetries . this is necessary given the highly non - linear nature of the bme which would otherwise quickly exaggerate any anisotropies . as an aside , we also note that we have significantly reduced the memory consumption of the default muscl - hancock scheme provided with flash through more efficient memory management , ensuring that the increased memory overhead of using the unsplit scheme is equivalent to only a single extra amr block . the results for the reduced memory version are of course bit - wise identical to the default release . our collisionless scheme differs from the standard hydrodynamics scheme primarily through our determination of the new eigenvalues and eigenvectors which are required for the characteristic tracing step . this allows for the calculation of the time averaged fluid properties which can then be used to calculate the second order accurate fluxes using a conventional riemann solver such as the approximate hll family of schemes including the hllc solver ( see , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) or more accurate roe scheme @xcite . we begin by expressing the zeroth , first and second order moments of the boltzmann equation ( see equations [ contin ] , [ moment ] and [ dispers ] ) in compact matrix notation form ( neglecting gravity for the moment ) , @xmath44 with the conserved advection quantities @xmath45 and their corresponding fluxes * f*(*u * ) , * g*(*u * ) and * h*(*u * ) in the respective @xmath46 , @xmath47 and @xmath48-directions being , @xmath49 , & { \bf f}({\bf u } ) = \left [ \begin{array}{l } \rho v_x \\ \rho v_x v_x + \rho\sigma_{xx}^2\\ \rho v_y v_x \\ \rho v_z v_x \\ e_{xx } v_x + 2\rho\sigma_{xx}^2 v_x\\ e_{yy } v_x \\ e_{zz } v_x \\ \end{array } \right ] , \\ { \bf g}({\bf u } ) = \left [ \begin{array}{l } \rho v_y \\ \rho v_x v_y \\ \rho v_y v_y + \rho\sigma_{yy}^2\\ \rho v_z v_y \\ e_{xx } v_y \\ e_{yy } v_y + 2\rho\sigma_{yy}^2 v_y\\ e_{zz } v_y \\ \end{array } \right ] , & { \bf h}({\bf u } ) = \left [ \begin{array}{l } \rho v_z \\ \rho v_x v_z \\ \rho v_y v_z \\ \rho v_z v_z + \rho\sigma_{zz}^2 \\ e_{xx } v_z \\ e_{yy } v_z \\ e_{zz } v_z + 2\rho\sigma_{zz}^2 v_z\\ \end{array } \right ] , \end{array } \label{conservedfluxes}\ ] ] where @xmath50 . for simplicity , in this paper we assume a diagonalised velocity dispersion tensor in the above flux vectors , neglecting off diagonal terms , ie . @xmath51 . as the dimensions of the matrix and therefore complexity of the problem grow quickly with the increasing order of moments that we consider , we leave the tracking of the off diagonal terms and higher order moments to our next paper . by taking the conservative equations and expanding the derivatives , it is possible to express equivalent advection equations for the primitive quantities ; density , bulk velocity and velocity dispersion : @xmath52 , & \tilde{{\bf f}}({\bf w } ) = \left [ \begin{array}{l } \rho v_x \\ v_x v_x + \sigma_{xx}^2\\ v_y v_x \\ v_z v_x \\ \rho\sigma_{xx}^2 v_x + \gamma \rho\sigma_{xx}^2 v_x\\ \rho\sigma_{yy}^2 v_x \\ \rho\sigma_{zz}^2 v_x \\ \end{array } \right ] , \\ \tilde{{\bf g}}({\bf w } ) = \left [ \begin{array}{l } \rho v_y \\ v_x v_y \\ v_y v_y + \sigma_{yy}^2\\ v_z v_y \\ \rho\sigma_{xx}^2 v_y \\ \rho\sigma_{yy}^2 v_y + \gamma \rho\sigma_{yy}^2 v_y\\ \rho\sigma_{zz}^2 v_y \\ \end{array } \right ] , & \tilde{{\bf h}}({\bf w } ) = \left [ \begin{array}{l } \rho v_z \\ v_x v_z \\ v_y v_z \\ v_z v_z + \sigma_{zz}^2 \\ \rho\sigma_{xx}^2 v_z \\ \rho\sigma_{yy}^2 v_z \\ \rho\sigma_{zz}^2 v_z + \gamma \rho\sigma_{zz}^2 v_z\\ \end{array } \right ] , \end{array } \label{primitivefluxes}\ ] ] where @xmath53 ( see [ sec : tests ] ) . for brevity we now consider just the @xmath46-directional flux vectors in the conserved quantities , @xmath54 and primitive variables , @xmath55 . as the components of the primitive flux vectors @xmath56 are each functions of the individual primitive quantities @xmath57 , the jacobian of the system in the x - direction , @xmath58 , can be expressed in terms of the partial derivatives of @xmath55 with respect to @xmath59 : @xmath60 \ ; .\ ] ] the eigenvalues @xmath61 of the system are then obtained by solving for the roots of the polynomial , @xmath62 where @xmath63 is the identity matrix . physically , the eigenvalues represent the speeds of signal propagation within the system and are thus important for determining the extent to which fluid bounding a cell interface affects the flux through it over the time step . unlike a collisionally dominated fluid which has five eigenvalues , the ( diagonalised ) collisionless fluid has seven to account for the anisotropy in the dispersion tensor ( for a non - diagonalised version , this increases to ten ) . we determine the eigenvalues for the @xmath46-direction to be @xmath64^t \ ; , \label{eigenvalues}\ ] ] with @xmath65 being the sound speed in the @xmath46-direction . the directionally dependant sound speed , @xmath66 , is in close analogy to that of a standard collisionally dominated gas , @xmath67 the right eigenvectors of the jacobian , @xmath68 , are those that obey the relation @xmath69 whilst the left eigenvectors of the jacobian , @xmath70 , are those which satisfy @xmath71 . using the jacobian for the primitive variable matrix equations , @xmath72 \ ; , \label{jacobian}\ ] ] and the equivalent eigenvalues ( [ eigenvalues ] ) , the eigenvector matrices are found to be : @xmath73 \ ; , \label{righteigvec}\ ] ] @xmath74 \ ; . \label{lefteigvec}\ ] ] with the general form of the eigenvalues and corresponding eigenvectors calculated , we can implement them into the muscl - hancock scheme in - line with the approach taken by @xcite which we outline below . first we compute the left , right and central differences of the primitive variables using the cell centred values within each cell , where @xmath17 is the cell index . @xmath75 we then map them onto the characteristic variables using the left eigenvectors calculated using ( [ lefteigvec ] ) , where @xmath76 is the matrix corresponding to the @xmath77 row of the left eigenvector matrix , and the primitive variables @xmath78 are defined at the cell centres , @xmath79 as the scheme is higher than first order , the reconstruction of the time averaged variables at the cell interfaces , can lead to the formation of new non - physical local maxima or decreasing minima . to prevent their formation which otherwise leads to spurious post shock oscillations , we apply monotonicity constraints to the characteristic differences , ensuring that the time averaged variables are total variation diminishing ( tvd , see * ? ? ? * ) , @xmath80 the monotonised differences are then projected back onto the primitive variables , where @xmath81 is the matrix corresponding to the @xmath77 column of the right eigenvector matrix ( [ righteigvec ] ) , @xmath82 we now calculate the values of the primitives at the left , @xmath83 , and right , @xmath84 , cell interfaces using the primitive variable monotonised differences , @xmath85 \delta { \bf w}_{i}^m \ ; , \label{stoneeqn40 } \\ \tilde{{\bf w}}_{r , i-1/2 } = { \bf w}_{i } - \left [ \frac{1}{2 } - max(\lambda_{i}^{0 } , 0 ) \frac{\delta t}{2\delta x } \right ] \delta { \bf w}_{i}^m \ ; , \label{stoneeqn41}\end{aligned}\ ] ] where @xmath86 and @xmath87 are the largest and smallest eigenvalues based upon the cell centred data . we can now perform the characteristic tracing step , removing from the quantities calculated in equations ( [ stoneeqn40 ] ) and ( [ stoneeqn41 ] ) , contributions from each set of waves that travelled towards the interface but did not reach it during the forward half - time - step @xmath88 using @xcite , @xmath89 { \tilde { \bf r}}_i^{\alpha } \ ; , \label{stoneeqn42 } \\ { \bf w}_{r , i-1/2 } = \tilde{{\bf w}}_{r , i-1/2 } + \frac{\delta t}{2\delta x}\sum_{{\lambda}^{\alpha } < 0 } \left [ ( \lambda_i^0 - \lambda_i^{\alpha}){\tilde { \bf l}}^{\alpha } \cdot \delta { \bf w}_i^m \right ] { \tilde { \bf r}}_i^{\alpha } \ ; . \label{stoneeqn43}\end{aligned}\ ] ] with time averaged primitive values calculated for each cell face , godunov s fluxes can now be computed and applied using a conventional riemann solver technique such as the accurate roe scheme . if using an approximate riemann solver such as the hll scheme , then due to the averaging of the intermediate eigenstates between the fastest left and right waves , it is necessary to add the additional terms ( [ stoneeqn44 ] ) and ( [ stoneeqn45 ] ) shown below , to equations ( [ stoneeqn42 ] ) and ( [ stoneeqn43 ] ) respectively . @xmath90 { \tilde { \bf r}}_i^{\alpha } \ ; , \label{stoneeqn44 } \\ { \bf w}_{r , i-1/2 } = - \frac{\delta t}{2\delta x}\sum_{{\lambda}^{\alpha } > 0 } \left [ ( \lambda_i^{\alpha } - \lambda_i^{0}){\tilde { \bf l}}^{\alpha } \cdot \delta { \bf w}_i^m \right ] { \tilde { \bf r}}_i^{\alpha } \ ; . \label{stoneeqn45}\end{aligned}\ ] ] these remove the effect of waves which propagate away from the cell interface which would otherwise be included , allowing for the scheme to achieve second order accuracy . through the use of operator splitting , we can deal with the addition of extra source and sink terms , including gravity , separately after the collisionless stellar hydrodynamics step . to ensure that the gravitational acceleration term is kept second order accurate in time , we store the acceleration from the previous time step and interpolate forward to account for temporal variation in the gravitational field . additional source and sink terms can include star formation and supernova , although we only consider gravity in this paper . notwithstanding a formal similarity of the che and bme , testing the latter is nontrivial as it requires knowing analytical solutions for the dynamics of _ collisionless _ systems . fortunately , some of the test problems that are used to benchmark the performance of hydro codes can be adapted to test the collisionless systems . this is because equations ( [ contin])-([dispers ] ) can be made formally identical to those of classical hydrodynamics in special cases . indeed , for a one - dimensional flow of a non - gravitating collisionless fluid along the @xmath46-direction , the bme become @xmath91 = 0 \ ; .\end{aligned}\ ] ] if we now assume that @xmath92 is equal to the gas pressure @xmath93 in classic hydrodynamics and @xmath94 then as @xmath95 , i.e. the ratio of specific heats @xmath96 is equal to 3 , then equations ( [ contin1d])-([dispers1d ] ) turn into the following form @xmath97 = 0 \ ; , \end{aligned}\ ] ] which are exactly one - dimensional classic hydrodynamics equations . the fact that the bme for a one - dimensional flow turns into the che for @xmath53 is not a coincidence but is related to the fact that a one - dimensional flow of a collisionless fluid has only _ one _ translational degree of freedom @xmath98 , unlike a collisional fluid which can still be characterised by @xmath99 due to frequent collisions that equipartition the kinetic energy of particles between the three translational degrees of freedom . this means that in order to properly describe the one - dimensional flow of a collisionless fluid using the classical hydrodynamics approach , one needs to assume that the fluid has only one translational degree of freedom , i.e. @xmath100 . it is straightforward to show that for a two - dimensional flow of a collisionless fluid , one needs to consider a specific case with @xmath101 and @xmath102 and also @xmath103 , i.e. a flow of a collisional fluid with two degrees of freedom . the above analysis allows us to use the standard sod shock tube problem , often employed to benchmark collisional hydro codes . to generate an analytic solution to a one - dimensional discontinuous state along the @xmath46-coordinate , we choose @xmath53 and use the standard solution procedure @xcite . for a two - dimensional discontinuous state in the @xmath104 plane at an arbitrary angle @xmath105 to the @xmath46-direction , an analytic solution is obtained by choosing @xmath103 . in the 1d case , we run the test along the @xmath46-coordinate and set @xmath92 and @xmath21 at @xmath106 $ ] to 1.0 , while at @xmath107 $ ] the x - component of the stress tensor is set to 0.1 and the stellar density is 0.125 . the numerical resolution is 200 grid zones and the cells are equidistantly spaced . in the 2d case , we align the initial fluid discontinuity along the [ 1,1,0 ] plane and use the same densities and velocity dispersions as were set for the 1d setup setting @xmath108 in order to yield agreement with the results for an isotropic collisionally dominated gas . this yields a shock wave which propagates at @xmath109 to the @xmath46 and @xmath47 axes . again we use 200 cells in the @xmath46 and @xmath47 directions and note that the setup can easily be extended to three dimensions although for brevity we do not show these here . figures [ fig1 ] and [ fig2 ] show the results of our tests for the 1d and 2d setups , respectively , at @xmath110=0.1 using the hllc riemann solver . in particular , the top - left , top - right , bottom - left , and bottom - right images show the density , velocity , the x - component of the stress tensor @xmath111 , and @xmath112 . the latter two quantities may be regarded as analogues to the gas pressure @xmath93 and internal energy density @xmath113 in classical hydrodynamics . in the 1d case , the plotted velocity is @xmath114 , whereas in the 2d case we plot the absolute value of velocity @xmath115 along the diagonal @xmath116 , i.e. , along the direction of the shock wave propagation . it is evident that the riemann - solver - based numerical scheme can trace the shock waves ( propagating to the right ) very well , usually within one to two grid zones . contact discontinuities ( propagating to the left ) are smeared to a greater extent , usually by 5 - 6 grid zones . nevertheless , the position of the shock waves and contact discontinuities along with the values of the flow variables are reproduced quite accurately . in a conventional collisional shock , the energy of the macroscopic motion of the gas is transfered through collisions into random isotropic kinetic motions of the particles ( thermal energy ) . in a collisionless fluid , the energy of macroscopic motion is transfered into an effective increase in the velocity dispersion as stellar streams with different velocities mix at the shock front . open circles in figure [ fig3 ] present the results of the 1d sod shock tube test using the kt scheme . it is evident that this scheme does not provide as good an agreement with the analytic solution as the hllc riemann solver , particularly for the velocity dispersion ( bottom - right panel ) . in the kt scheme , the shock fronts and contact discontinuities are smeared out over roughly twice as many grid zones as in the hllc scheme . however encouragingly , the positions of the shocks and discontinuities and also the values of the flow variables are still accurately reproduced . all in all , the hllc scheme shows an undoubtedly superior performance though at a higher cpu cost , with the kt scheme typically taking around 22% less time to run and a near identical amount of memory . the gravitational collapse of a pressure - free sphere is used to assess the code s ability to accurately treat converging spherical flows on the cartesian grid . this test is also useful for estimating the performance of the gravitational potential solver on dynamical problems for which one has to use a finite - difference form of the gravitational potential _ gradient_. since the test setup involves no stress tensor , the resulting equations are identical in the collisionless and collisional fluid dynamics cases and one can use an analytic solution describing the collapse of every mass shell in the limit of an infinite sphere radius @xcite . to run this test , we set a cold homogeneous sphere of unit radius and density ( for convenience , the gravitational constant is also set to unity ) and let it collapse under its own gravity . we use a block size of @xmath117 cells and six levels of refinement leading to an effective grid size of @xmath118 cells . we adopt isolated gravitational boundary conditions . unfortunately , we must consider a cloud of finite radius in cartesian geometry with a sharp outer boundary to preserve the cloud sphericity . as a result , a rarefaction wave develops after the onset of the collapse , propagating towards the coordinate origin and necessitating complicated corrections to the analytic solution . figure [ fig4 ] compares the results of our numerical simulation with the `` uncorrected '' analytic solution of @xcite ( solid line ) at 0.985 free fall timescales , when the initial density has increased by nearly three orders of magnitude . the majority of cells within the sphere lie upon the analytic solution indicating a good homologous collapse with only a small peak above the analytic solution forming within the very central few cells . in addition to this small peak , the initially sharp boundary of the cloud is smeared out over several cells as a result of the rarefaction developing . the analytic radii of the cloud however shows reasonable agreement with the radii at which the half peak density is reached . in general , our results are found to be in strong agreement with those of other authors applying this test problem to the classic hydrodynamics ( e.g. , * ? ? ? * ) . as a third test , to demonstrate that the code can model the formation of anisotropies , we initialise two copies of the same sphere as in [ sec : sphercollapse ] with a non - negligible dispersion tensor . in one we initially set an isotropic dispersion tensor whilst in the second we initially weight the different diagonal terms @xmath119 with the ratio of @xmath120 . we set the system up so that the initial maximum effective dispersion energy is only 25% of that needed to support the sphere against gravitational collapse . this allows the sphere to collapse down from its initially symmetric configuration and deform into an ellipsoid . the velocity dispersion is then free to change as the simulation evolves . we plot in figure [ fig5 ] the density contour for @xmath121 at regular time intervals for both the cloud with an isotropic dispersion tensor ( left hand panel ) and an anisotropic dispersion tensor ( right hand panel ) . the isotropic dispersion tensor leads to a collapse which is isotropic as expected whilst in the anisotropic case , the collapse proceeds quickest in the direction with the lowest dispersion term . this leads to a flattening of the sphere into an ellipse . in figure [ fig6 ] we plot the ellipticity of the collapsing anisotropic cloud as a function of time . the ellipticity is determined to be the ratio between the minor and major axes of the ellipsoid over a specified iso - density surface . the solid line corresponds to the ellipticity of the iso - density surface of @xmath122 and the dashed line is the ellipticity for @xmath123 . initially , since the velocity dispersion is lowest in the x - direction , material along that axes collapses fastest , with the ellipticity increasing as the density builds . this leads to a greater effective pressure , @xmath124 , from the velocity dispersion . eventually the core collapses to a density and pressure sufficient to stop further collapse . at this point the sphere begins to isotropise as material from the other two axial directions catches up and falls onto the core . this collapse of material onto the core , drives up the pressure and material is forced out around @xmath125 along the x - direction where least resistance is offered . this again increases the ellipticity and results in the major and minor axes of the ellipse switching places as the x - axis becomes the major instead of the minor axis and vice versa for the z - axis . the system then oscillates between a high degree of ellipticity and near spherical symmetry as the different axes of the ellipsoid collapse and rebound at different times and speeds . these oscillations repeat over many free fall timescales . material at a lower density follows the same cyclical pattern of varying ellipticity but with less pronounced maxima and minima . as a result of the non - linear nature of the collapse and its relative separation from the violent oscillations within the core , transitions at lower density are also smoother . the final test problem describes the growth of a spiral structure in a gravitationally unstable stellar disc . the stability analysis of a thin stellar disc to a local axisymmetric perturbation states that the disc is gravitationally unstable if @xcite @xmath126 where @xmath127 is the radial stellar velocity dispersion , @xmath128 is the epicycle frequency , and @xmath129 is the stellar surface density . a finite disc thickness and non - axisymmetric perturbations may increase the critical toomre parameter @xmath130 by a factor of unity . the initial setup consists of a self - gravitating , rotating stellar disc submerged in a fixed dark matter ( dm ) halo . the latter is described by the quasi - isothermal density profile of the form @xmath131 where @xmath132 is the galactocentric distance . the central density @xmath133 pc@xmath134 and the characteristic scale length of the quasi - isothermal halo @xmath135 kpc are calculated based on the assumed dm halo mass @xmath136 and halo radius , @xmath137 kpc , using relations provided in @xcite . the computational box spans 30 kpc in each coordinate direction @xmath138 . the numerical procedure for generating a rotationally supported stellar disc in the combined gravitational potential of stars and dm halo is described in detail in @xcite . the rotation curve , the radial gas surface density profile and the radial profile of the toomre parameter @xmath139 of the initial stellar disc are plotted in fig . [ figsp1 ] with the solid , dashed , and dot - dashed lines , respectively . evidently , the disc is initially gravitationally unstable . figure [ figsp2 ] presents a series of disc images at several successive times since the beginning of numerical simulations . usually , an initial seed perturbation is required to drive the system out of equilibrium and initiate the growth of a spiral structure . however , in our case , the initial perturbation is introduced due to a re - map of the initial equilibrium configuration from the cylindrical coordinates @xcite onto the cartesian coordinates ( the flash code ) . the growth of a two - armed spiral mode is apparent in the figure . to quantify the growth rate of spiral modes in our model , we employ global fourier amplitudes defined as @xmath140 where @xmath141 is the spiral mode , @xmath142 is the surface density of the stellar disc , @xmath143 is the polar angle , and @xmath144 is the total mass of the stellar disc . the amplitudes are calculated on a square region with size @xmath145 kpc , which is centred on the coordinate origin and encompasses the growing spiral . the global fourier amplitudes can be regarded as the amplitude of a spiral density perturbation relative to the surface density of the axisymmetric disc . figure [ figsp3 ] shows the temporal evolution of the first six global fourier amplitudes ( in log units ) in our model stellar disc . initially all amplitudes are negligibly small except for that corresponding to the @xmath146 mode . this is a numerical artifact and arises due to the finite resolution within the core . similar to the results of @xcite in which they model a static king - profile that is initially in hydrostatic equilibrium , the finite resolution within the core leads to an under - resolved gravitational potential . this results in a slight relaxation and expansion of the very central region . an outward flow of material results through the faces of the central most cell and due to the fact that our disk is constructed so as to promote the growth of gravitational instabilities , this generates a small yet non - negligible perturbation . although similar behaviour may occur in all mesh codes , those using cartesian geometries are particularly susceptible given their strong natural bias for flows along the axes of the mesh . despite being present from the very onset of the simulation , the @xmath146 mode slowly decays and does not lead to the growth of a spurious four armed spiral given the dominance of the @xmath147 mode at late times by around two orders of magnitude . the @xmath147 mode demonstrates the fastest growth rate . it saturates at around @xmath148 myr and declines slowly in the subsequent evolution due to continuing disc heating . the maximum amplitude of the @xmath147 mode is about @xmath149 , which suggests a non - axisymmetric density perturbation of about 2% relative to the underlying axisymmetric disc . an appealing physical interpretation for the growth of a spiral structure in our model disc is swing amplification . amplification occurs when any leading spiral disturbance , e.g. , introduced by the initial seed perturbation , unwinds into a trailing one due to differential rotation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in addition , swing amplification needs a feedback mechanism that can constantly feed a disc with leading spiral disturbances . trailing short - wavelength disturbances propagating through the disc centre and emerging on the other side as leading ones provide a feedback for the swing amplifier @xcite . we can check if the growth mechanism of our spiral structure is consistent with the predictions of the swing amplification theory . for our spiral pattern to be triggered by the swing amplifier , there should be no inner lindblad resonance ( ilr ) for the @xmath147 mode . otherwise , the corresponding trailing disturbances will damp at the ilr , thus failing to pass through the disc centre and promoting the growth of the @xmath147 mode . figure [ figsp4 ] shows the radial profiles of @xmath150 and @xmath151 at @xmath152 myr , where @xmath150 is the angular velocity of the stellar disc and @xmath128 is the epicycle frequency . both quantities are the azimuth averages . in particular , the solid and dashed lines show the corresponding values for the @xmath147 and @xmath153 modes , respectively . the dash - dot - dot - dotted and dash - dotted lines show the angular velocity @xmath154 of the global spiral pattern measured at two distinct positions in the disc : 2.0 kpc and 4.0 kpc . in theory , @xmath154 should be independent of radius , though in practice it is always slightly faster in the inner regions than in the outer ones due to gradual winding of a spiral pattern . the radial position of lindblad resonances are determined as @xmath155 , i.e. , as the radial distance where the forcing frequency of the spiral pattern coincides with the epicycle frequency of the stars . in particular , the inner and outer lindblad resonances correspond to the plus and minus signs , respectively . figure [ figsp4 ] demonstrates that there is clearly no ilr for the @xmath147 mode , whereas there is one marginally for the @xmath153 mode ( and thus for any higher - order mode ) . indeed , the dash - dotted and dash - dot - dot - dotted lines never intersect the @xmath156 curve , whereas they do intersect the @xmath157 curve at @xmath158 kpc , thus damping @xmath153 disturbances that try to pass through the disc center . although slight growth over time of the @xmath153 mode is observed in figure [ figsp3 ] , it is damped relative to the @xmath147 mode by over two orders of magnitude . higher order modes for which a clearer ilr will be present are damped to a much greater extent . this explains why the @xmath147 mode ( and not any higher - order mode ) ultimately dominates the spiral pattern . another consistency check on our model is the position of the outer lindblad resonance ( olr ) . spiral disturbances can not propagate through the olr , which effectively limits the radial extent of a spiral pattern . figure [ figsp4 ] indicates that the olr for the @xmath147 mode is located approximately at 4 kpc and this value is in agreement with the radial extent of the spiral pattern in fig . [ figsp2 ] . we conclude that our numerical results are in general agreement with the predictions of the swing amplification theory . in this paper we have outlined a series of important limitations that are encountered in mesh codes when using the particle - mesh technique . used to consistently integrate collisionless stellar or dark matter components into simulations which model gas physics on an eulerian mesh , the particle - mesh technique has the advantage of being relatively straight forward to implement but is unfortunately subject to several key limitations . from a technical perspective these include poor load balancing and increased communication overhead . from a physical point of view , discreteness effects incurred when discrete particle properties are mapped to the mesh , are believed to result in spurious entropy generation . in galaxy formation simulations , this can have a significant influence on the reservoir of cold gas available to form stars , potentially changing the entire dynamics of the system . to overcome these limitations we have proposed the use of the collisionless boltzmann moment equations as a powerful alternative . such an approach allows us to model the collective properties of collisionless objects such as stars as a fluid , instead of relying upon the traditional n - body approach . in this paper we refer to this approach as `` collisionless stellar hydrodynamics '' given its primary use in the modelling of the properties of large collections of stars , such as those in a galaxy , in much the same way as we model gases in eulerian grid codes . although direct integration of the collisionless boltzmann equation has been undertaken by other authors , it is often prohibitively expensive to perform . this arises as it is necessary to track both the spatial domain and its corresponding velocity phase space . however @xcite demonstrates that the ability to resolve small scale velocity fluctuations is much greater in these codes than for conventional n - body simulations . instead of a costly direct integration of the boltzmann equation , we opt to derive the zeroth , first and second order moments under the usual zero heat flux approximation and taking the third order moments to be negligible . although ultimately an approximation , we show that this yields results in - line with a series of demanding test problems , requiring a factor of @xmath159 less memory than is needed using direct integration over the complete velocity phase space . we implement these new collisionless stellar hydrodynamic equations into the massively parallel flash amr code using two different numerical solvers for hyperbolic partial differential equations . the first scheme by @xcite , requires no information on the equation of state of a collisionless fluid or any assumptions about the behaviour of fluid interactions within the riemann fan . this riemann solver free prescription allows us to track sharp discontinuities with relatively low numerical diffusion and provides a very useful comparison with which to compare our more sophisticated second scheme . in future , its extensibility is hoped to be used to explore the impact of including the higher order moments which we exclude here . by assuming a diagonalised dispersion tensor , we have derived the eigenvalues and eigenvectors for the first three moments of the collisionless boltzmann equations . these have been implemented into a characteristic tracing method based on the muscl - hancock unsplit advection scheme . this allows time averaged primitive states to be calculated at the cell interfaces using characteristic tracing , from which time averaged fluxes can be computed using a range of riemann solvers . we validate our code using a suite of tests with analytic solutions , including some commonly used to benchmark conventional hydrodynamic codes . by realising that the classic hydrodynamic equations and boltzmann moment equations can be reconciled under certain conditions , we are able to generate analytic solutions for the sod shock test which are applicable for collisionless systems . of particular note , we find that since a collisionless fluid has only one degree of freedom in a given direction and is unable to equipartition the kinetic energy of particles between the three dimensions through collisions , it has an effective ratio of specific heats , @xmath160 . thus in 1d , a sod shock solution for a collisionally dominated gas with @xmath160 , matches that for a collisionless fluid . using this insight , we confirm that our numerical schemes ; both the kt scheme and characteristic tracing method , accurately reproduce analytic solutions . although the kt scheme runs 22% faster with near identical memory consumption , our riemann solver based scheme shows notably better resolving power of sharp shocks and discontinuities , spreading them over roughly half the number of cells . through the use of a spherical pressure - free collapse problem , we confirm that the code can maintain good flow symmetry in convergent flows in the absence of any anisotropy in the dispersion tensor . agreement with the peak density within the collapsed sphere also confirms the validity of the poisson solver used in flash . through the inclusion of a non - negligible anisotropic dispersion tensor , we extend the spherical collapse to observe the behaviour of an initially homologous spherical cloud as it collapses and deforms into an elliptical profile . although initial behaviour of the system occurs in - line with expectations , the long term behaviour is more complicated with repeated transitions between an elliptical and spherical profile , along with switching of the major and minor axes after the initial core implosion . although the inclusion of gas physics may help to isotropise the internal structure of the cloud and damp these oscillations , they will nevertheless remain important to the dynamics outside of the core . we will explore in more detail interactions between the collisionally dominated gas physics and the stellar material in future papers . in particular , we will extend our scheme to include off - diagonal terms and a more thorough exploration of the effects of higher order terms . although powerful in its current form , the addition of off - diagonal terms in the velocity dispersion tensor will allow us to directly measure the vertex deviation , a key observable used for probing galactic structure . to date observers have made extensive measurements of our own milky way s vertex deviations which can hide detailed information about the kinematic properties of the bar and spiral arms . however noise in particle based simulations has made accurate theoretical predictions difficult , something which our new collisionless stellar hydrodynamics code can allow us to overcome . having already confirmed the applicability of the collisionless stellar hydrodynamics code to the formation of spiral structure , we have found excellent agreement between our numerical simulations of the growing spiral pattern and the predictions of the swing amplification theory , with the correct relation between the disc size and the outer lindblad resonances being observed . as a final note , we find the time taken in flash for the riemann solver based collisionless stellar hydrodynamics routine is only 7% greater than that for the original classic hydrodynamic scheme , allowing the collisionless stellar hydrodynamics to scale linearly with the classic hydrodynamics . we also find that instead of doubling the overall amount of time spent communicating when both gas and collisionless stellar hydrodynamics are included , our grouping of communications results in a net increase of only 1.2% to the overall communication time compared to that of just a single phase hydrodynamic scheme . thus we conclude by confirming that our new collisionless stellar hydrodynamic approach using the boltzmann moment equations both preserves the excellent scaling previously demonstrated in flash as well as enhancing the level of detail we can expect to be able to extract from our results , with none of the discreteness effects observed for particle - mesh techniques . the authors acknowledge the generous support of the austrian fwf for project numbers p21097-n16 ( nigel l. mitchell ) and lise meitner fellowship m1255-n16 ( eduard i. vorobyov ) . the software used in this work was in part developed by the doe nnsa - asc oascr flash center at the university of chicago . the computations were performed at the vienna scientific cluster ( vsc ) and the astrocluster of the institute for astrophysics , funded by the university of vienna . on general grounds one can argue that the boltzmann moment equations ( [ contin])-([dispers ] ) should convolve to the classic hydrodynamics equations ( che ) if the stress tensor @xmath14 is diagonal and isotropic . in this section we demonstrate this by assuming that @xmath161 for @xmath162 and @xmath163 , implying that @xmath164 . the equation of continuity ( [ contin ] ) is identical to that of the che and the momenta equations ( [ moment ] ) trivially turn into their collisional counterparts with a substitution @xmath165 ( and identical relations for other components ) , where @xmath93 is the gas pressure . it takes a bit more math to show that the dispersion equations ( [ dispers ] ) can be convolved to the equation for the total energy per unit volume @xmath166 . let us expand equation ( [ dispers ] ) into the component form in cartesian coordinates . @xmath167 after summing up equations ( [ eqn1])-([eqn3 ] ) we obtain @xmath168 \nonumber \\ \label{eqn4 } & - & \rho u_i a_i = 0 \ ; , \end{aligned}\ ] ] where we have divided the resulting equation by a factor of 2 and have taken into account that @xmath164 and @xmath169 . in the classical hydrodynamics the gas pressure relates to the internal energy density via the relation @xmath170 , where @xmath171 for a flow with three translational degrees of freedom , meaning that @xmath172 in the above equation . the final step is to note extract the diagonal part from the symmetric tensor @xmath173 by introducing the viscous stress tensor @xmath174 . @xmath175 with this final transformation , equation ( [ eqn4 ] ) turns into the usual hydrodynamics equation for the total energy per unit volume @xmath176 of a viscous fluid . @xmath177 - \rho u_i a_i = 0 \ ; .\ ] ] when deriving the boltzmann moment equations , each equation for the @xmath178-th moment inevitably requires the knowledge of the @xmath179 moment . this chain is conventionally closed by setting the third - order moments to zero , i.e. , @xmath13 . in this section , we discuss the validity of this approximation . for this purpose it is useful to refer to a similar problem in the collisional hydrodynamics , for which the closure problem exists as well . indeed , the che are derived from the _ collisional _ boltzmann equation and if the zero - heat - flux approximation were relaxed , then equation ( [ eqn5 ] ) would have an additional term @xmath180 a contemporary swindle is to use the empirical fourier law for thermal conduction and assume that @xmath181 where @xmath182 is the gas temperature and @xmath183 is the heat conduction coefficient . unfortunately , such a trick can not be applied to a collisionless fluid as it , strictly speaking , has no temperature due to the lack of local thermal equilibrium . however , this analysis can give us an insight as to when the zero - heat - flux approximation can not be neglected . in the che , such a situation arises on a quasi - stationary contact discontinuity between two gases with vastly different temperatures when the thermal diffusion time scale @xmath184 can be much shorter than dynamical one @xmath185 , where @xmath186 is the characteristic size of the system . in most astrophysical environments , however , the bulk motions of gas ensure that @xmath187 and the third - order moments can be neglected . in the case of collisionless systems , the diffusion time scale is played by the typical relaxation time @xmath188 , which should be compared against @xmath189 . it is apparent that the zero - heat - flux approximation is expected to be valid if @xmath190 this condition is usually met in any rotationally supported system like a galactic stellar disc , with @xcite highlighting stellar dynamic models which show the third order moments decay faster than the relaxation timescales . this means that the zero - heat - flux approximation is approached faster than isotropisation . however , this condition may break in globular clusters that are thought to be supported against gravitational collapse by random motions of stars ( i.e. , by high velocity dispersions ) . this does not invalidate the whole boltzmann moment approach approach , but simply means that an important piece of physics is missing and needs to be taken into account . we work on developing useful approximations that can treat such cases in collisionless fluids . antypas , k.b . , calder , a. c. , dubey , a. , gallagher , j. b. joshi , j. , lamb , d. q. , linde , t. , lusk , e. et al . , `` flash : applications and future . '' , parallel computational fluid dynamics 2005 : theory and applications , 2006 , 325 ( elsevier science )

we study the quantum melting of stripe phases in models with competing short range and long range interactions decaying with distance as in two space dimensions . at zero temperature we find a two step disordering of the stripe phases with the growth of quantum fluctuations . a quantum critical point separating a phase with long range positional order from a phase with long range orientational order is found when , which includes the coulomb interaction case . for the transition is first order , which includes the dipolar case . another quantum critical point separates the orientationally ordered ( nematic ) phase from a quantum disordered phase for any value of . critical exponents as a function of are computed at one loop order in an expansion and , whenever available , compared with known results . for finite temperatures it is found that for orientational order decays algebraically with distance until a critical kosterlitz - thouless line . nevertheless , for it is found that long range orientational order can exist at finite temperatures until a critical line which terminates at the quantum critical point at . the temperature dependence of the critical line near the quantum critical point is determined as a function of .

1612.03098

quantum systems with fluctuating stripe order have received an increasing attention of the condensed matter community in the last years@xcite . different quasi - two - dimensional systems such as quantum hall systems@xcite , cuprates and iron - based superconductors@xcite as well as heavy fermion compounds@xcite present inhomogeneous and/or anisotropic electronic / spin structures at very low temperatures . a common feature of these systems is the possible existence of a quantum critical point ( qcp ) separating phases that break translational and/or rotational symmetries@xcite . with growing temperature stripe patterns melt , typically by the proliferation of dislocations and disclinations . the theory of thermal melting of stripes is reasonable well understood@xcite . when quantum flutuations play a role , as in the electronic systems cited above , the phenomenology of anisotropic phases may be much more complex than in the corresponding thermal , or classical counterpart . on general grounds it is expected that upon lowering the temperature a crossover must occur between thermal and quantum dominated fluctuations . eventually , quantum fluctuations may be responsible for a true @xmath9 quantum critical point ( qcp ) . the behavior of thermodynamic and electronic transport properties must be strongly influenced by the proximity of the qcp , and then it is important to characterize the different possible universality classes of quantum phase transitions@xcite . a quantum theory of stripe melting which can account for this rich phenomenology is still under development . a model for continuous stripe quantum melting in metals , driven by a particular type of topological defects , double dislocations , was recently proposed in connection with the physics of cuprate superconductors @xcite . this model considers the simultaneous presence of charge and spin stripes . as a consequence , it is argued that single charge stripe dislocations ( which would lead to frustration in the spin stripes ) remain gapped , justifying that the physics of charge stripe melting should be mainly governed by the proliferation of double dislocations . upon disordering , this leads to a `` stripe loop metal '' phase , different from the usual fermi liquid phase . when spin order is not considered , the model leads to similar results as we reach in the present study , which only deals with one type of degrees of freedom . in the context of quantum hall phases , the condensation of lattice defects was recently studied@xcite . a very general framework of a gauge field theory of quantum liquid crystals at @xmath9 in continuum space has also been proposed and recently reviewed in [ ] . another context in which the results of the present work can be relevant is the field atomic gases@xcite and in particular in ultra - cold dipolar fermi gases @xcite . the experimental realization of ultra - cold atom systems is progressing rapidly and promises to be an ideal testing ground for the physics of quantum strongly correlated systems . several electronic liquid crystalline phases are predicted to be present in these systems @xcite . recently the thermal melting of two dimensional dipolar fermi gases was addressed in [ ] . when all the dipoles are at a generic tilting angle @xmath10 with respect to the @xmath11 plane of the system , the thermal melting of stripes is found to be driven by the proliferation of dislocations , leading to an effective anisotropic @xmath12 model , which then allows to predict the elastic and critical properties straightforwardly . nevertheless , for tilting angle @xmath13 , i.e. when all dipoles are oriented perpendicular to the @xmath11 plane , the system recovers rotation invariance on the plane , and then long range stripe order is forbiden at finite @xmath14 @xcite . while longe range positional order is forbidden , orientational order of stripes survives to dislocations proliferation , leading to a nematic - like phase @xcite . in the present work we show that , for dipolar interactions in a rotationally symmetric system , only quasi - long - range nematic order in the orientation of stripes is possible at finite temperatures . it is worth mention that while most properties of a quantum phase transition are dictated by symmetry , dynamical properties depend essentially on microscopy . in the context of fermi liquids , landau damping could produce over - damped modes with dynamical exponent @xmath15@xcite near the isotropic - nematic phase transition . in the presence of a tetragonal lattice , ising - nematic fluctuations could renormalize the dynamical exponent as well@xcite , which adds to the complexity of the observed phenomenology . the emergent phenomenology related with striped phases is supposed to come from the competition between short range attractive interactions and long range repulsive ones@xcite at a microscopic level . in this perspective , the relevance of residual long range interactions is an interesting and important problem and , to the best of our knowledge , an open one . for instance , in quantum hall systems , it is known that long - range coulomb interactions change the nature of the smectic phase@xcite . moreover , in the context of charge cold atoms , coulomb as well as dipole interactions play a relevant role@xcite . recently , the effects of long ranged interactions in classical two - dimensional stripe melting was considered@xcite . it was shown that , consistent with common assumptions , for sufficiently short - ranged interactions , positional correlations are short ranged while a kosterlitz - thouless transition from an isotropic to a quasi - long - range nematic phase takes place . interestingly , it was found that for sufficiently long - ranged repulsive interactions a second order phase transition occurs between a disordered isotropic phase and a truly long - range orientationally ordered nematic phase . this result implies that , for classical models , the lower critical dimension depends on the range of the interaction : the larger the range of the repulsive interaction , the smaller the lower critical dimension . in this work we extend the approach to orientational phases in systems with long ranged competing interactions to the realm of quantum melting of stripe patterns . the starting point is a coarse - grain effective hamiltonian for isotropic , competing short range and long range interactions which decay with distance as @xmath0 . this action is a well known continuum limit of a large family of truly microscopic hamiltonians , and in this sense the connection of our results with specific microscopic models is straightforward . then , starting from the well known stripe ground states of the effective model , we promote the displacement field of the stripe modulation to a quantum time dependent operator , introduce dislocations and obtain the relevant dispersion relations for the quantum elasticity theory with long range competing interactions . at finite temperatures the mean squared fluctuations of the displacement field diverge with the size of the system independently of the range of interactions , and then positional order can be strictly short ranged , a result well known in models with short range interactions . a more interesting scenario emerges for the orientational degrees of freedom which are stable to gaussian fluctuations . then , after introducing dislocations , the relevant topological defects in stripe forming systems , we consider the interaction between two far apart portions of the striped pattern through a multipolar expansion of the relevant density ( electronic or magnetic density ) . this leads to a model of quantum rotors in the plane of the system with generalized dipolar interactions @xcite . analysis of this model shows that , when temperature fluctuations dominate over quantum ones , the well known kosterlitz - thouless type phase transition driven by disclination unbinding is restricted to sufficiently short ranged competing interactions , @xmath7 . if the interactions are long ranged enough , @xmath16 , a genuine second order phase transition to a phase with long range orientational order may exist at finite @xmath14 . at @xmath9 the conclusions from the quantum elastic theory are different : for sufficiently weak quantum fluctuations the mean square fluctuations of the displacement field may be finite , while they diverge for strong enough fluctuations . this precludes the existence of a quantum phase transition from a smectic - like phase to a nematic - like one at a critical value of the strength of quantum fluctuations . in the context of a mcmillan - degennes theory this phase transtion turns to be of first order for @xmath17 and second order for sufficiently long range repulsive interactions @xmath1 . subsequently , at higher values of the quantum fluctuations , orientational order also breaks down leading to a fully disordered phase . for short range interactions , @xmath7 , the quantum nematic - isotropic transition in @xmath18 is in the universality class of the thermal xy model in @xmath19 . for sufficiently long ranged interactions , @xmath16 , the critical exponents depend continuously on the interaction range . finally , the critical line emerging from the quantum critical point at the @xmath9 isotropic - nematic transition is obtained for sufficiently low temperatures . no critical line emerges from the nematic - smectic transition point because positional order is completely absent at finite temperatures . the organization of the paper is as follows : in section [ sec : generaltheory ] we briefly introduce the model for competing interactions at different scales and proceed in subsection [ sec : melting ] to a presentation of the quantum elastic theory of stripe melting in a pedagogical way . in subsection [ sec : rotors ] we introduce the model of quantum rotors in the plane , from which most of the results will be extracted . then , in section [ sec : criticality ] we describe the main results for the isotropic - nematic phase transitons at @xmath9 . two subsections describe , respectively , the behavior for @xmath7 and @xmath16 . complementing the discussion of the @xmath9 phase diagram , in section [ sec : smectic ] the nematic - smectic phase transition is described . in section [ sec : thermal ] the results for the finite @xmath14 phase diagrams are presented for both @xmath7 and @xmath16 . finally , a discussion of the results and some conclusions are presented in section [ sec : conclusions ] . the simplest model to describe the effect of competing interactions at different scales in two spatial dimensions can be cast in the following coarse - grained hamiltonian @xcite : @xmath20&=&\frac{1}{2}\int d^2 x\,\left(\vec{\nabla}\phi(\vec{x } ) \right)^2 \nonumber \\ \nonumber & + & \frac{1}{2}\int d^2x\int d^2x'\ \phi(\vec{x}){j}(\vec{x}-\vec{x'})\phi(\vec{x'})\\ & + & \frac{1}{2\beta}\int d^2x\ v(\phi(\vec{x } ) ) \ ; . \label{eq : ham}\end{aligned}\ ] ] @xmath21 represents a density field or scalar order parameter and , in the absence of external fields , the hamiltonian has the ising symmetry @xmath22 . @xmath23 is a repulsive long ranged isotropic interaction which decays as a power law of the distance in the form @xmath24 . although it is convenient to work with an arbitrary exponent @xmath5 , physically relevant examples are the coulomb interaction in charged systems ( @xmath2 ) and the dipolar interaction between out - of - plane magnetic moments ( @xmath4 ) in magnetic films . additionally , entropic contributions generate a local potential @xmath25 which , for simplicity , we consider to be a degenerate double well potential of the form @xmath26 , with @xmath27 and @xmath28 . finally , @xmath29 is the inverse temperature . in the absence of long - range interactions , the system tends to form condensate states , with @xmath30 . due to the ising symmetry , both condensates are equally probable and the system tends to phase separate . the long range repulsion @xmath31 frustrates this tendency and the ground state ends being inhomogeneous and/or anisotropic . the simplest and most commonly found configuration of this type is a unidirectional modulation characterized by a single wave vector @xmath32 @xcite . in this case , the order parameter can be written in the form @xmath33 , where the fourier coefficients @xmath34 determine the profile of the modulation . long wave - length fluctuations can be parametrized by a displacement field @xmath35 in the form @xmath36)$ ] , where @xmath37 is the average direction of the modulation and @xmath38 stands for the modulus of @xmath39 . a typical configuration is illustrated in figure [ fig : stripes ] . the shaded area represents regions with @xmath40 while in the white regions @xmath41 . the fourier spectrum @xmath34 can be computed , in principle , using a mean - field approximation . when expressed in terms of @xmath42 the effective hamiltonian ( [ eq : ham ] ) reads @xcite : @xmath43 where @xmath44 represents the energy of the unperturbed stripes configuration , @xmath45 is the fourier transform of @xmath46 and the @xmath47 s are stiffness coefficients . @xmath48 measures the local elastic response of the stripes to compression or elongation of the pattern , while @xmath49 is the local elastic response to bending of the stripes . @xmath50 comes from the long range repulsive interaction , which generates a non local interaction between far apart pairs of layers , as illustrated in fig.[fig : stripes ] . it has been shown @xcite that the non local term results irrelevant in the long wave - length limit for @xmath51 , but becomes important when @xmath52 . thus , it is equivalent to use the simpler model hamiltonian : @xmath53 which interpolates between the different possible cases . for @xmath52 , the long range tail of the repulsive interaction dominates at long wave - lengths and consequently the sigma depending term must be kept . on the other hand , for fast enough decaying repulsive interaction ( @xmath51 ) the appropriate @xmath54 contribution is recovered by just fixing @xmath55 . building upon the classical effective hamiltonian ( [ heff1 ] ) , which has already been studied in , our goal is to build a theory of quantum stripe melting , where quantum fluctuations are considered in the same foot as thermal fluctuations , and long range interactions are appropriately taken into account . the well known classical theory of stripe melting @xcite has to be recovered as the high temperature limit of the present formalism when @xmath7 . since @xmath56 represents an elastic displacement field , it is natural to see the modes @xmath45 of this field as phonons . this context justifies the use of a path integral approach to treat the quantum fluctuations of the field . accordingly , we promote the displacement field adding an imaginary time coordinate @xmath57 @xcite . the imaginary time slab extends from zero to @xmath58 . intertwined with the assumption that the normal modes are phonons it is the assumptions that the quantum field @xmath59 posses a kinetic energy quadratic in the `` velocity '' @xmath60 . consequently , the action for the quantum problem reads : @xmath61 , \label{qs1}\end{aligned}\ ] ] where @xmath62 measures the strength of quantum fluctuations . the equality below the time integral indicates that the quantum action is constrained to periodic realizations of the field @xmath42 along the imaginary time direction . now , defining the fourier transform of the field @xmath63 in the imaginary time direction in the form @xmath64 with the frequencies given by @xmath65 satisfying the periodicity condition in the interval @xmath66 $ ] , the action ( [ qs1 ] ) can be recast as : @xmath67 this is a gaussian action for which the correlator is : @xmath68 from this we can obtain the mean square fluctuations ( msf ) of translation and orientation degrees of freedom at a gaussian level , which inform us on the stability of the possible phases in a system of stripes , depending on the range of the repulsive interaction @xmath5 , the strength of quantum fluctuactions and temperature . the msf of the displacement field @xmath42 is given by : @xmath69 additionally , considering that in the small deviation limit the angular orientation of the stripe pattern is given by @xmath70 , we get for the msf of the angular orientation : @xmath71 for finite temperatures @xmath72 one can conclude that positional order is always short ranged . even at zero quantum fluctuations ( @xmath73 ) , the msf of the displacement field diverges with the linear size of the system for any @xmath5 . on the other hand , orientational order is always stable within the gaussian approximation , for weak enough quantum fluctuation ( @xmath73 ) and low enough temperature ( @xmath74 ) while , for high enough temperature or strong enough quantum fluctuations the orientational fluctuations diverge . under such conditions some orientational phase transition is expected at intermediate values of @xmath14 and @xmath62 . this picture may change ( and indeed changes ) under the effects of topological excitations @xcite . at zero temperature the main difference is that the msf of the displacement field becomes finite for weak enough quantum fluctuations , independently of the value of @xmath5 . on the other hand , for strong enough quantum fluctuations ( @xmath75 ) the msf of the displacement field diverges , which suggest the existence of a quantum phase transition at some intermediate value of the strength of the quantum fluctuations . we will discuss the nature of the quantum smectic - nematic phase transition in section [ sec : smectic ] . the picture obtained at a gaussian level is at least incomplete since no phase transitions are detected , even when they are expected . in the following , we will complete the description by inclusion of relevant topological defects , specially dislocations , which are essential to obtain qualitatively correct phase diagrams . in a stripe system , a dislocation can be seen as a pair of stripes of opposite densities coming to an end in the middle of the stripe pattern , see fig.[fig : dislocation ] . when the integration path encloses the dislocation core . the length of the dashed line corresponds to the value of the line integral . the parameter `` @xmath76 '' corresponds to the modulation length of the stripe pattern @xmath77 . ] the strength of a dislocation is characterized by the line integral over a closed counterclockwise path around the dislocation core , as shown in fig . [ fig : dislocation ] . the double arrowhead signals the difference in the number of stripes at the top and the bottom of the drawn contour . this condition is expressed in the form : @xmath78 where @xmath79 is an integer number and @xmath76 stands for the modulation length . it is important to realize that , differently from the general case , in the case under study an odd number of stripes ( layers ) can not form dislocations , otherwise the natural tendency of the system to create a modulated symmetrical pattern would be violated and consequently this will cost too much energy . this means that although usually the parameter @xmath76 represents the width of a single layer , in our case as we are forced to add pairs of layers it represents the whole modulation length . we will see in what follows that this particular feature has important implications for the dynamical properties of the system . also , it is known that normally dislocations can not move in an arbitrary way inside a crystal . specifically , due to the so called `` glide constraint '' , dislocations are forbidden to climb inside the crystal , which would imply that they can not move along the stripe direction . such movement inside a crystal would imply the existence of points in which the particles are destroyed or created , which obviously violates mass conservation . however in the present case dislocations are formed by a certain number of layer pairs with opposite densities , which means that the translation of a dislocation core along the stripe pattern maintains a constant local average of the order parameter , implying that the total mass or charge of our system is not affected by this kind of motion . the euler equation for the initial action ( [ eq : qs2 ] ) is : @xmath80 this equation has to be solved subject to the constraint given by eq . ( [ eq : cond1 ] ) , considering that at any time @xmath81 $ ] the integration contour is chosen encircling the dislocation . this is not immediate since there is nt a smooth and single valued solution for this problem . the way out is to introduce cuts across which @xmath82 jumps discontinuously . following usual conventions we take this cut going from the dislocation center to infinity along the _ y _ axis ( fig.[fig : dislocation ] ) . the displacement @xmath83 should present a discontinuity of value @xmath84 when crossing the cut . considering this and the form in which the operators in the euler equation act near the cut , we realize that a source term of the form @xmath85 should be added at the right hand of equation ( [ eq : eu1 ] ) in order to obtain a single valued solution satisfying condition ( [ eq : cond1 ] ) . this source term considers a moving dislocation with trajectory given by @xmath86 . after this the displacement field is given by : @xmath87 from which one can read the green function for a single dislocation : @xmath88 the next step is to generalize the previous result to a system of interacting dislocations . promoting the delta density of a single dislocation to the case of a field of dislocations , the displacement can be written as @xmath89 , where @xmath90 is given by ( [ eq : gk1 ] ) . now , the complete elastic action can be written in the symmetric form : @xmath91 where the component fields are given by : @xmath92 introducing the cut into the function @xmath93 in order to make it single valued leads to the appearance of spurious infrared divergences in @xmath94 and @xmath95 when @xmath96 . this issue is a known fact in the classical theory of stripe melting @xcite . the solution comes from the observation that , while the displacement field @xmath82 posseses a discontinuity of magnitude @xmath97 for all @xmath98 , the order parameter , which is proportional to @xmath99 , is a smooth function . then , it is safe enough to neglect these divergent terms . after this , @xmath100 and @xmath101 are given by : @xmath102 finally , to complete the construction of the effective action for the interacting dislocations field @xmath103 , it is important to include the cost of an isolated dislocation @xmath104 , which is not taken into account in the long wave - length contribution of the displacement field . this yields : @xmath105 where @xmath106 represents the dislocation core diameter , and serves as a short distance cut - off for the effective action . now , remembering that for small layer deviations the orientation angle of the stripes is @xmath107 , the orientational correlator is @xmath108 which , to leading order in the long wavelength ( @xmath109 ) and low temperature ( @xmath110 ) limits , leads to : @xmath111 this is a generalization of the corresponding results for classical smectic systems @xcite when one includes the effects of long range repulsive interactions and quantum fluctuations . from here it is straightforward to infer the form of the corresponding orientational action : @xmath112 in order to study critical properties of the stripe system , we would like to build the most general effective action that preserves local rotational symmetry , having the action ( [ eq : so ] ) as the one obtained in the spin wave approximation . to this end , instead of tracking back the effect of the higher order contributions in the original action , which could be rather intricate , one can rebuild the full orientational hamiltonian considering the symmetries of the stripe pattern . to identify the real symmetries of the system we should notice that the basic cell of any stripe system is composed by a double layer of opposite densities with a small longitudinal width , as shown in fig . [ fig : stripes ] . in principle , any low energy configuration of the system of stripes can be built from these basic cells . in reference a model which takes into account the interactions between near and distant striped cells was introduced . in the next section we will discuss the qualitative aspects of the model in more detail , stressing the particular choice of observables for the identification of the different phase transitions . the orientation of each basic cell can be defined by means of a local director vector @xmath113 . in general , the director @xmath114 points perpendicular to the iso - density curves @xmath115 , and in the case of sharp interfaces it is peaked at the domain walls as shown in fig . [ fig : stripes ] . however , it is simple to realize that stripe configurations have no vectorial order since , integrating over the whole sample , @xmath116 . notice that in two adjacent domain walls the directors always point in opposite directions . moreover , if @xmath114 is globally rotated by @xmath117 , the final state is exactly de same as the original one . therefore , the orientational order is generally characterized by a quadratic function of @xmath114 , say a symmetric traceless tensor @xmath118 , where the index @xmath119 refers to two orthogonal directions . this is a nematic order parameter , which by construction is invariant under global rotations by @xmath117 . an orientationally ordered phase is characterized by @xmath120 , while for a completely disordered ( isotropic ) phase , @xmath121 . to completely characterize possible phase transitions it is necessary to compute fluctuations : @xmath122 . from a technical point of view , this is a difficult calculation since it is a four point correlation function , @xmath123 . from a physical point of view , this means that the nematic order parameter is not obtained as a linear response to an homogeneous external field but instead it is necessary to compute a quadratic response . within this approach , the minimum approximation to compute the nematic order parameter is the `` self consistent screening approximation '' , where infinite sets of two - loop diagrams can be computed self consistently @xcite . here we propose an alternative observable to characterize the isotropic - nematic phase transition @xcite . note that in the stripe phase , the directors are ordered in an `` antiferromagnetic '' structure . each block of two adjacent domain walls are characterized by two directors pointing in opposite directions and separated by half a stripe period . thus , the system is structured in two interpenetrated sub - lattices , both of them ferromagnetically ordered with opposite directions . we can define a _ sub - lattice director _ @xmath124 , where @xmath125 is the sub - lattice index . this is illustrated in fig . [ fig : sublattice ] . then , similarly to what is done to describe antiferromagnetic order , it is useful to work with two linear cobinations of the sub - lattice directors @xmath126 where @xmath127 is the stripe period . in the stripe phase @xmath128 , reflecting the fact that there is no dipolar or vector order . in turn , the staggered order parameter , eq . ( [ eq : stagger ] ) , reflects the vectorial order of one of the sub - lattices , @xmath129 . it can be noted that if @xmath130 the system is orientationally ordered . for the sake of characterizing the isotropic - nematic phase transition one can work either with the tensor order parameter @xmath131 or with the vector one , @xmath132 . both of them characterize the same phase transition . on the other hand , the critical exponents can be quite different because @xmath131 and @xmath132 are different observables . while @xmath131 is a quadratic response to an external uniform field , @xmath132 is a linear response to a staggered conjugate field . in this sense , the latter is easier to compute since it is a linear function of the density gradient @xmath133 . from the point of view of the symmetry , the nematic order is characterized by its invariance under rotations by @xmath117 . this is obvious in the tensor order parameter since @xmath134 is a quadratic function of @xmath114 . on the other hand , @xmath132 changes sign under rotation by @xmath117 . however , form eq . ( [ eq : stagger ] ) and fig . [ fig : stripes ] it is clear that a change of sign in @xmath132 represents a change from high density ( shaded areas ) to low density ( white areas ) regions , and vice - versa . then , @xmath132 and @xmath135 represent the same state in the thermodynamic limit , as it should be . to describe the nematic transition using the staggered director we will assume that one sub - lattice is slaved to the oder . thus , we are not considering compression fluctuations , which can be treated perturbatively . in the following , orientational order will be characterized by a space - time dependent unit director , given by @xmath136 the hamiltonian for this order parameter can be expressed as the sum of two contributions : @xmath137=h_{sr}[\vec n]+h_{\ell r}[\vec n ] , \label{eq : h}\ ] ] where @xmath138 is a local function describing the effect of short - ranged interactions between the unit cells and @xmath139 codify the long - ranged interactions and it is in general non - local . from symmetry considerations the effective local hamiltonian should have the form @xcite : @xmath140\;,\ ] ] since this is the more general quadratic form that is local and rotational invariant . of course , it could contain non - quadratic local terms that we will treat perturbatively in a renormalization group approach . the coupling constants @xmath141 and @xmath142 will be defined later in terms of the stripe melting parameters such as the stripe stiffness and dislocation s energy . to evaluate the contribution of the long ranged interactions we start by coarse graining the system , covering the plane with small rectangles of area @xmath143 with @xmath144 , centered at positions @xmath145 corresponding to a domain - wall sub - lattice as indicated in fig . [ fig : stripes ] . considering two well separated rectangles of areas @xmath146 and @xmath147 , located at a distance @xmath148 , the interaction between them can be expressed in the form : @xmath149 performing a multipolar expansion of the interaction @xmath150 , summing over all possible pairs of rectangles and retaining only the leading dipole contributions , the long range part of the hamiltonian can be written in the form : @xmath151 where @xmath152 , the dipolar moment is given by @xmath153 and @xmath154 is a short distance cutoff . since we are interested in the classical as well as in the quantum behavior of this system , it is convenient to work in the euclidean efective action formalism , in which we write the partition function as a funcional integral of the form @xmath155=\int { \cal d } \hat n \ e^{-s_g[\vec n(\vec x,\tau)]+\int d^2 x d\tau \ , \vec h(\vec x)\cdot \vec n(\vec x)}\ ] ] where the unitary vector @xmath156 is a function of position and euclidean time , satisfying the periodic boundary conditions @xmath157 . @xmath158 is the conjugate field of the order parameter . for static configurations , @xmath159 , @xmath160=\beta h[\vec n]$ ] , where @xmath161 is given by eq . ( [ eq : h ] ) . as described above , @xmath161 is strongly constrained by symmetry . conversely , the order parameter dynamics can not be deduced by means of symmetry properties only . as discussed in the previous section , for the present case we consider a conservative local dynamics that preserves time reversal . then , the euclidean action can be written as : @xmath162+\frac{g}{2}\int_0^\beta d\tau\int d^2x\int d^2x ' \\ \nonumber & \times&\,\omega(\vert\vec{x}-\vec{x}'\vert)\left(\frac{\vec{n}(\vec{x},\tau)\cdot\vec{n}(\vec{x}',\tau ) } { \vert\vec{x}-\vec{x}'\vert^{\sigma+2}}\right . -(\sigma+2)\\ & \times & \left.\frac{\vec{n}(\vec{x},\tau)\cdot(\vec{x}-\vec{x}')\vec{n}(\vec{x}',\tau)\cdot(\vec{x}-\vec{x}')}{\vert\vec{x}-\vec{x}'\vert^{\sigma+4}}\right ) . \label{eq : sg}\end{aligned}\ ] ] @xmath160 $ ] describes the orientational order properties of a stripe forming system due to competing interactions at different scales . @xmath163 represents the director of one sub - lattice , as described above , and @xmath164 is the linear response to a staggered conjugate field with a periodicity @xmath76 . in eq . ( [ eq : sg ] ) there are three local terms , given by the coupling constants @xmath165 . @xmath166 measures the intensity of the quantum fluctuations , while @xmath141 and @xmath142 are stiffness coefficients . the effects of long - ranged interactions are contained in the last two non - local terms , both proportional to the coupling constant @xmath167 . notice that these terms are a generalization of a dipolar interaction , with an isotropic component as well as an anisotropic one . both components decay as @xmath168 . the euclidean action ( [ eq : sg ] ) matches the one obtained in the spin wave limit , eq . ( [ eq : so ] ) , when the coupling constants are given by : @xmath169 it is expected that at length scales much longer than the modulation length the anisotropy in the spatial stiffnesses vanishes resulting in a simplification of the form of the effective action @xcite . as a result of this renormalization for long wave - lengths the simpler general form is obtained : @xmath170 \\ \nonumber & + & \frac{g}{2}\int_0^\beta d\tau\int d^2x\int d^2x'\,\omega(\vert\vec{x}-\vec{x}'\vert)\left(\frac{\vec{n}(\vec{x},\tau)\cdot\vec{n}(\vec{x}',\tau ) } { \vert\vec{x}-\vec{x}'\vert^{\sigma+2}}\right . \\ & - & \left.(\sigma+2)\frac{\vec{n}(\vec{x},\tau)\cdot(\vec{x}-\vec{x}')\vec{n}(\vec{x}',\tau)\cdot(\vec{x}-\vec{x}')}{\vert\vec{x}-\vec{x}'\vert^{\sigma+4}}\right ) . \label{eq : soq}\end{aligned}\ ] ] in the next section we will explore the ground state ( @xmath9 ) properties of this action . it has been shown @xcite that the universality class does not change if the anisotropic dipolar interaction is replaced by an attractive isotropic term , decaying with the same power law as the original interaction . thus , the action can be further simplified to read : @xmath171 } \\ \nonumber & -&\frac{g}{2}\int_0^\beta d\tau\int d^2x\int d^2x'\,\omega(\vert\vec{x}-\vec{x}'\vert ) \frac{\vec{n}(\vec{x},\tau)\cdot\vec{n}(\vec{x}',\tau ) } { \vert\vec{x}-\vec{x}'\vert^{\sigma+2}}\ ; , \end{aligned}\ ] ] which is a non - linear sigma model with long range interactions . moreover , the non - linear and the linear sigma model belong to the same universality class , provided the order parameters share the same symmetry properties and the interaction of both models have the same long wave - length behavior . then , it is convenient to work with ginzburg - landau type expansion : @xmath172 \\ \nonumber & -&\frac{g}{2}\int_0^\beta d\tau\int d^2x\int d^2x'\,\omega(\vert\vec{x}-\vec{x}'\vert ) \frac{\vec{n}(\vec{x},\tau)\cdot\vec{n}(\vec{x}',\tau)}{\vert\vec{x}-\vec{x}'\vert^{\sigma+2}},\end{aligned}\ ] ] where the restriction on the modulus of the vector @xmath163 has been lifted . the degrees of freedom of the non - linear sigma model are the goldstone modes of the linear model in the broken symmetry phase where @xmath173 . the longitudinal fluctuations are gapped , therefore not contributing for the critical properties . we found convenient to re - write eq . ( [ eq : sglin ] ) for an @xmath174-dimensional vector @xmath175 in fourier space for general spatial dimension @xmath176 and a long ranged interaction decaying as @xmath177 : @xmath178 where @xmath179 are the matsubara frequencies , the stiffnesses @xmath166 and @xmath141 have been set to one by means of a rescaling of the space - time variables and a sum over the vector component indexes @xmath180 is understood . by fixing @xmath181 , we recover the original model . the critical properties of the model ( [ eq : sogl1 ] ) at @xmath9 have been studied by means of the epsilon expansion technique in reference . in the following subsections we summarize the results , offering an alternative way to compute critical exponents , specially useful at finite temperature which we present in the next section . as we have already discussed , the behavior of positional and orientational fluctuations allow us to classify the models according to the range of the interaction , whether @xmath7 or @xmath16 . when @xmath7 the interactions are effectively short ranged and the non - local terms in ( [ eq : sglin ] ) are irrelevant in the renormalization group sense . then , ones can simply fix @xmath55 in eq . ( [ eq : sogl1 ] ) , which at @xmath9 then reads : @xmath182 this is the well known classical @xmath183 model in effective dimension @xmath184 @xcite . its critical behavior is very well established and very good approximations for the critical exponents are known . the upper critical dimension is @xmath185 , or equivalently the spatial upper critical dimension is @xmath186 . while for @xmath187 the gaussian fix point is stable and mean field gives the correct critical behavior , for @xmath18 it is necessary to make an @xmath188 expansion . it turns out that there is a non - trivial fix point at @xmath189 and @xmath190 of order @xmath6 . the susceptibility and correlation length exponents are , respectively : @xmath191 in the special case of interest in this work , @xmath192 , @xmath18 at @xmath9 is equivalent to a classical xy model in three dimensional space at finite temperature . the temperature in the classical system plays the role of the inverse coupling constant in the quantum system at @xmath9@xcite . it is known that the classical model has a critical point driven by spin wave fluctuations at a finite @xmath193 and several critical exponents for the @xmath19 xy universality class have been computed numerically with great precision @xcite . these results are compatible with more recent ones on the quantum @xmath19 @xmath194 model @xcite . this quantum critical point ( qcp ) is depicted in figure [ fig : phasediag1 ] . in @xmath18 . @xmath195 marks a @xmath9 first order phase transition between a phase with long range positional order and a long range orientational order one . a quantum critical point at @xmath196 separates the long range ordered phase from a quantum disordered one . at finite temperature only quasi long range orientational order is present ending at a kt critical line . ] in this case the generalized dipolar interaction , last term of eq . ( [ eq : sglin ] ) , is generally a relevant interaction depending on the space dimension @xmath176 . starting from the action ( [ eq : sogl1 ] ) , the susceptibility at one loop order reads : @xmath197 where @xmath198 is determined self - consistently from @xmath199 by performing the change of variables : @xmath200 where @xmath201 and @xmath202 is the area of @xmath176-dimensional sphere , we find that @xmath203 where @xmath204 is the susceptibility of the short range interacting case ( @xmath55 ) , computed from eq . ( [ eq : slocalt0 ] ) , with renormalized values of @xmath205 , @xmath42 and @xmath176 . this relation allows us to compute the critical exponent of the long - ranged interacting models from the knowledge of the local models , at least at the one loop approximation . for instance , the critical exponent of the susceptibility should satisfy @xmath206 . then , using eq . ( [ eq : gamma - local ] ) we immediately find , @xmath207 which coincides with the result obtained by a direct calculation within the perturbative renormalization group at linear order in an @xmath6-expansion@xcite . the upper critical dimension @xmath208 for @xmath52 is obtained from the well known value @xmath209 , for @xmath55 , and the scaling relations of eq . ( [ eq : relations ] ) , giving @xmath210 . we note that the upper critical dimension depends continuously on @xmath5 . this is a direct consequence of the fact that the dynamical exponent @xmath211 also depends on @xmath5 , since the dispersion relation of the lowest energy modes is @xmath212 . of course , @xmath213 when we take the limit @xmath214 . a second critical exponent for @xmath52 can be obtained by noticing that the scaling of the wave vectors @xmath215 implies that any length scale @xmath216 in the original system is related to the length scale @xmath217 in the local @xmath55 system by @xmath218 . in particular , the correlation length should satisfy @xmath219 , implying that the associated critical exponent will be given by @xmath220 . ( [ eq : nu - local ] ) and the scaling relations of eq . ( [ eq : relations ] ) , we find @xmath221,\ ] ] which coincides with the expression computed by means of a linear expansion in @xmath222 @xcite . it is interesting to note that , the largest the interaction range , the smaller the critical dimension . as a consequence , for @xmath223 , the exponents coincide with the mean - field ones @xmath224 , @xmath225 . in particular , these exponents are exact for the coulomb interaction ( @xmath2 ) in @xmath18 . once two critical exponents are known , other ones like @xmath226 can be immediately obtained by scaling and hyperscaling relations@xcite : @xmath227 taking into account that , at one loop approximation , @xmath211 . in table [ tb : exponents ] we summarize the results obtained for different values of the interaction range parameter @xmath5 . @xmath228 & \frac{1}{2}\left[1-\frac{1}{2}\left(3-d\right)\right ] \\ \cline{2 - 5 } & \alpha & 0 & \frac{1}{\sigma}\left(\frac{3\sigma}{2}-d\right ) \left [ 1-\frac{1}{\sigma}\left(\frac{n+2}{n+8}\right ) \left(d+\frac{\sigma}{2}\right)\right ] & \frac{\left(3-d\right)}{2 } \left [ 1-\left(\frac{n+2}{n+8}\right ) \frac{\left(d+1\right)}{2}\right ] \\ \hline\hline & \xi(t , r = r_c ) & t^{-\frac{1}{\sigma}\left(\frac{2d}{\sigma}-1\right ) } & t^{-\frac{2}{\sigma } } & t ^{-1 } \\ \cline{2 - 5 } t\neq 0 & \chi(t , r = r_c ) & t^{-\left(\frac{2d}{\sigma}-1\right ) } & t^{-2 } & t^{-2 } \\ \cline{2 - 5 } & \delta r(t)=r_c - r & -t^{\left(\frac{2d}{\sigma}-1\right ) } & -t^{2 } & -t^{2 } \\ \hline\hline \end{array}\ ] ] in @xmath18 . @xmath195 marks a @xmath9 phase transition between a phase with long range positional order ( red line ) and a long range orientationally ordered one . for @xmath229 the transition is of first order induced by quantum fluctuations , while for @xmath230 it is of second order . a quantum critical point at @xmath196 separates the long range orientationally ordered phase from a quantum disordered one . the finite temperature critical line ( red ) separates a long range ordered phase from a disordered one , ending at the qcp . ] although positional order is destroyed at any finite temperature already at the level of gaussian fluctuations , as shown in section [ sec : melting ] , it can actually exist at @xmath9 . from eqs . ( [ eq : theta - u ] ) and ( [ eq : theta2 ] ) it can be inferred that , at @xmath9 , the proliferation of unbounded dislocations destroys positional order in a region where orientational order is still possible . then , besides the isotropic - nematic transition described in section [ sec : criticality ] , the system also displays a nematic - smectic quantum phase transition . a theory of the thermal nematic - smectic phase transition was developed by mcmillan and degennes@xcite . some generalizations to quantum systems were presented in refs . . here , we follow similar procedures to understand the nature of the smectic - nematic quantum phase transition as a function of the interaction range parameter @xmath5 . the theory starts from an orientationally ordered state characterized by the director @xmath231 , pointing for instance in the @xmath37 direction : @xmath232 . on top of this we consider a positional order parameter given by a complex function @xmath233 . @xmath62 is the smectic order parameter , indicating the presence of a density modulation . the phase is proportional to the displacement field @xmath59 . to build a rotationally invariant action , the smectic order parameter and the nematic fluctuations should be coupled by a `` covariant derivative '' @xmath234 and @xmath235 ( @xmath37 is the coordinate along the director @xmath163 , while the coordinate @xmath236 is transversal ) . then , the action for the smectic order parameter coupled to nematic fluctuations is given by : @xmath237&=&\int d^2x\ d\tau\left\ { \frac{a}{2}\mid\psi\mid^2 + \frac{b}{4}\mid\psi\mid^4 \right . \label{eq : mdg } \\ & + & \left . \frac{c_x}{2}\left|\frac{\partial \psi}{\partial x}\right|^2+\frac{c_y}{2}\left|\left(\frac{\partial } { \partial y}-ik_0n_y\right)\psi\right|^2\right\ } \nonumber\\ & + & \int\frac{d^2k}{(2\pi)^3}d\omega \left(g_0\omega^2 + g k^\sigma\right ) \hat{n}_y(\vec{k},i\omega)\hat{n}_y(-\vec{k},-i\omega)\ ; . \nonumber\end{aligned}\ ] ] in this equation , @xmath106 controls the phase transition , while @xmath238 . @xmath239 and @xmath240 are smectic elastic constants . since the system is anisotropic , they are generally different . the last term of eq . ( [ eq : mdg ] ) represents the nematic dynamics , where @xmath166 and @xmath167 are couplings associated to the small angular fluctuations of the director , which can be traced back to the action ( [ eq : solr ] ) . ( [ eq : mdg ] ) resembles the effective free energy for smectic - nematic transition in 3d liquid crystals and the metal - superconductor transition in type 1 superconductors @xcite . from the first line of eq . ( [ eq : mdg ] ) , it is simple to conclude that , in the absence of nematic fluctuations , the phase transition is second order , driven by the sign of the parameter @xmath106 . however , nematic fluctuations can change the nature of the transition depending on the range of the interactions . to see this , we can integrate out the gaussian transverse nematic fluctuations in @xmath241 . this leads us to an effective action @xmath242 $ ] , such that : @xmath243)=\int d[n_y]\exp(-\mathcal{s}\left[\psi , n_y\right]).\ ] ] the derivative of @xmath242 $ ] with respect to @xmath244 , lead us to : @xmath245}{\delta \psi}=a\psi+b\psi^3+c_yk_0 ^ 2\langle n_y^2\rangle\psi , \label{eq : ders}\ ] ] where @xmath246 computing the integral in eq . ( [ eq : ny2 ] ) at leading order in @xmath244 and replacing the result in eq . ( [ eq : ders ] ) , we obtain : @xmath245}{\delta \psi } = \left\ { \begin{array}{lcl } \tilde{a}\psi-\tilde c_y k_0 ^ 2 \psi^{\frac{4}{\sigma}}+\tilde b\psi^3 & \mbox{~,~ } & \frac{4}{3}\leq\sigma\leq2\\ & & \\ \tilde{a}\psi+\tilde{b}\psi^3 & \mbox{~,~ } & \sigma\leq\frac{4}{3}\\ \end{array}\right . , \label{eq : sderivative}\ ] ] where @xmath247 , @xmath248 and @xmath249 are renormalized couplings . the first line of eq . ( [ eq : sderivative ] ) implies that the system has a first order smectic - nematic transition induced by fluctuations for @xmath250 . this result includes the short - range interaction case @xmath55 . moreover , it is clear from the second line of eq . ( [ eq : sderivative ] ) that for @xmath251 the transition is second order , provided the nematic fluctuations are weak enough to keep the renormalization of the quartic coupling @xmath248 positive . if this condition is not satisfied , it is necessary to include higher order powers of @xmath244 in the calculation . summarizing , we expect that the two - dimensional smectic - nematic quantum phase transition is discontinuous for short range as well as long range interaction models , provided @xmath250 . the dipolar interaction is included in this regime . on the other hand , for extremely long ranged interactions @xmath251 ( including the coulomb case ) , the transition is of second order . the transitions are shown in figures [ fig : phasediag1 ] and [ fig : phasediag1 ] . it is interesting to note that there is no critical line ending at the quantum transition point since any finite temperature completely destroys positional order . this fact could change in the presence of a substrate where the translation invariance in broken to a discrete group . in this case , the smectic phase could be stabilized at finite temperature . in order to understand the @xmath252 phase diagram , we analize in this section thermal fluctuations . similarly to the @xmath9 case , there are two very distinct behaviors depending on the range of interactions , which we proceed to describe separately . in this case , the model of eq . ( [ eq : sogl1 ] ) coincides with the quantum rotor model in @xmath176 dimensions and finite @xmath14 , extensively discussed in the literature@xcite . @xmath18 is a special case , because at finite @xmath14 , it is the lower critical dimension . thus , for negligible quantum fluctuations a kosterlitz - thoules ( kt ) classical phase transition takes place at a critical @xmath254 . near the critical line , we expect that thermal fluctuations dominate ( except at the qcp ) . so , there should be a kt line that connects the classical and the quantum critical points as shown in figure [ fig : phasediag1 ] . the long range orientational order at @xmath9 is destroyed by temperature fluctuations , producing a quasi - long - range ordered ( qlro ) phase , characterized by power low decaying of the order parameter correlation functions . this expectations are supported by numerical simulations of the @xmath12 model in three dimensions , with a finite third dimension @xcite . althought we do not expect that an @xmath188 expansion could produce sensible results for @xmath18 , they will be useful for the long range interaction regime ( @xmath52 ) where the lower critical dimension is @xmath255 . in general , the inverse susceptibility very near the qcp is expected to be of the form : @xmath256 where @xmath257 is the gap opened when one moves away from the qcp . @xmath258 is the distance to the critical point . this gap can be computed in the limits @xmath259 and @xmath260 , leading to the following results@xcite : @xmath261 where @xmath262 is a non - universal constant . this long ranged interacting regime is very different from the previous one . the lower critical dimension is @xmath255 , then for @xmath18 we have a classical thermal second order phase transition for relatively small quantum fluctuations@xcite . consequently , a second order line connects the classical and the quantum critical point , producing a whole region of the phase diagram of truly long ranged order , depicted in figure [ fig : phasediag2 ] . to compute the low temperature behavior near the qcp we closely follow the methods of ref . . first , by summing up over the high frequency modes , we obtain an effective action for the zero frequency mode the result is an effective `` classical '' theory whose parameters are renormalized by quantum fluctuations . starting from the model of eq . ( [ eq : sogl1 ] ) we obtain : @xmath263 where the couplings @xmath264 and @xmath265 are given by : @xmath266 from eq . ( [ eq : st ] ) it is straightforward to compute the inverse susceptibility at one loop approximation : @xmath267 where @xmath268 in these equations @xmath264 and @xmath265 are given by eqs . ( [ eq : rbeta ] ) and ( [ eq : ubeta ] ) respectively . before proceeding with the computation of the integrals it is interesting to investigate how the susceptibility scales with the transformation ( [ eq : relations ] ) . ( [ eq : rbeta ] ) and ( [ eq : relations ] ) the gap parameter at @xmath52 is related with the one computed at @xmath55 as : @xmath269 moreover , performing the same transformation on eq . ( [ eq : ubeta ] ) we find , @xmath270 using these results we can show that , in the context of one loop approximation , the susceptibility for @xmath52 , given by eqs . ( [ eq : chibeta ] ) and ( [ eq : nbeta ] ) , is related with the one computed with the short - ranged interaction model by the expression @xmath271 where the scaling relations are given by eq . ( [ eq : relations ] ) . interestingly , the same scaling properties satisfied by the susceptibility at @xmath9 are satisfied at finite , albeit small , temperatures . equation ( [ eq : chibetascaling ] ) can be used to compute several critical properties of the long - ranged interaction model at finite temperature . the inverse susceptibility for @xmath52 at finite temperature has a similar expression as in the local case , @xmath272 however , the gap parameter @xmath264 is now computed using the local model result ( [ eq : r2 ] ) and the scaling relations ( [ eq : relations ] ) . we find , @xmath273 the condition @xmath274 determines the critical line near the qcp : @xmath275 moreover , the correlation length at the qcp ( @xmath276 ) as @xmath277 can be obtained from @xmath278 , giving : @xmath279 these results are summarized in table [ tb : exponents ] and figure [ fig : phasediag2 ] . let us conclude this section by discussing the validity of the approximations made , for the particular case @xmath18 with @xmath280 . in the usual @xmath183 model in @xmath184 the upper critical dimension is @xmath186 ( @xmath185 ) , while the lower critical dimension is @xmath281 . for this reason , figure [ fig : phasediag1 ] displays a kt line , there is no symmetry breaking across the critical line , and the low temperature phase is actually a qlro phase . evidently , an @xmath6 expansion around the upper critical dimension fails in this limit . however , when @xmath52 , the upper as well as the lower critical dimensions depends on @xmath5 and they are actually given by @xmath282 , @xmath210 . consequently , the two - dimensional system is always above the lower critical dimension . for this reason , the critical line depicted in figure [ fig : phasediag2 ] is truly second order and an @xmath6 expansion produce qualitatively good results . of course , as usual , the numerical values computed at order @xmath6 have growing errors when we move away form the upper critical dimension @xmath210 . it is interesting to note that for @xmath18 with coulomb interactions ( @xmath2 ) , @xmath283 and then the critical exponents are dominated by the gaussian fixed point and therefore are exact . the perturbative , finite @xmath14 calculations , provide the correlation length as well as the susceptibility diverging as @xmath284 at the qcp , while the critical line behaves as @xmath285 while @xmath277 . in this work we developed a theory for the melting of stripe phases in two dimensional quantum systems with competing interactions of variable range , considering both quantum and thermal fluctuations . our main conclusion is that the nature of the phase transitions as a consequence of the melting process can be very different depending on the range of the competing repulsive interactions , @xmath5 , extending considerably the known results which , almost exclusively , were restricted to short range interactions . a mapping of the problem to a model of quantum rotors in the plane with generalized dipolar interactions allowed us to obtain several interesting properties of the phase transitions and universality classes of the models . at @xmath9 we showed that the melting of stripes proceeds through a two step process , which can produce two quantum critical points for sufficiently long ranged repulsive interactions , @xmath230 , while when @xmath3 the smectic - nematic transition turns out to be of first order . at finite temperatures only some kind of orientational order is possible . when @xmath7 the well known critical phase with algebraic orientational correlations is present , ending at a kt line . but for sufficiently long range repulsive interactions , @xmath16 , a phase with long range nematic order is possible , ending at a second order critical line . at @xmath9 , critical exponents can be computed due the equivalence of the quantum @xmath18 problem at @xmath9 with the classical model of rotors in @xmath286 dimensions at finite temperature , which properties for short range interactions are well known . an approximate treatment for finite temperatures allowed us to compute the behavior of thermodynamic quantities near the qcp , specially the temperature dependence of the uniform susceptibility , the correlation length and the critical line , summarized in table [ tb : exponents ] . in this work the melting of stripes defined by a scalar density order parameter was studied . our results could be tested , e.g. in ultra - cold dipolar fermi gases in the case where the dipoles point perpendicular to the plane of the system , in which case the system of dipoles recovers space rotational invariance . important extensions for future work are the consideration of a vector order parameter , e.g. inclusion of different spin components , and also an interaction among different degrees of freedom , which is important for the physics of high @xmath193 superconductors and other strongly correlated electronic systems in which electron and spin density waves are intertwinned , as considered , e.g. in [ ] . lattice anisotropies can be also important perturbations in real situations which deserve to be studied in future work . the brazilian agencies _ conselho nacional de desenvolvimento cientfico e tecnolgico _ ( cnpq ) , _ fundao carlos chagas filho de amparo pesquisa do estado do rio de janeiro _ ( faperj ) , and _ coordenao de aperfeioamento de pessoal de nvel superior _ ( capes ) are acknowledged for partial financial support . d.g.b also acknowledges partial financial support by the associate program of the abdus salam international centre for theoretical physics , ictp , trieste , italy . 56ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1103/revmodphys.75.1201 [ * * , ( ) ] link:\doibase 10.1080/00018730903122242 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.87.457 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.82.394 [ * * , ( ) ] link:\doibase 10.1103/physrevb.59.8065 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.076803 [ * * , ( ) ] http://dx.doi.org/10.1038/nature09597 [ * * , ( ) ] http://dx.doi.org/10.1038/nature08716 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.117.127001 [ * * , ( ) ] link:\doibase 10.1126/science.1063539 [ * * , ( ) ] , link:\doibase 10.1126/science.1134796 [ * * , ( ) ] , link:\doibase 10.1146/annurev - conmatphys-070909 - 104117 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.23.316 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.64.115312 [ * * , ( ) ] link:\doibase 10.1103/physrevb.55.142 [ * * , ( ) ] http://gen.lib.rus.ec/book/index.php?md5=d95c8ba24f70b2291c7a02cdf2039e91[__ ] , ed . , world scientific lecture notes in physics 67 ( , ) @noop _ _ ( , ) link:\doibase 10.1103/physrevb.86.115138 [ * * , ( ) ] link:\doibase 10.1103/physrevb.91.035122 [ * * , ( ) ] @noop * * ( ) \doibase http://dx.doi.org/10.1016/j.physc.2012.04.014 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.215301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.114.205302 [ * * , ( ) ] link:\doibase 10.1103/physreva.82.013643 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.115301 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.235124 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.145304 [ * * , ( ) ] @noop * * , link:\doibase 10.1103/physrevb.79.075437 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.114.116101 [ * * , ( ) ] link:\doibase 10.1103/physrevb.73.085101 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.075127 [ * * , ( ) ] link:\doibase 10.1016/0921 - 4534(93)90581-a [ * * , ( ) ] @noop * * , ( ) @noop * * , @noop * * , ( ) link:\doibase 10.1103/physrevb.89.094112 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.98.200604 [ * * , ( ) ] link:\doibase 10.1103/physreve.88.062140 [ * * , ( ) ] link:\doibase 10.1103/physrevb.70.134430 [ * * , ( ) ] link:\doibase 10.1103/physrevb.14.3110 [ * * , ( ) ] @noop _ _ , vol . ( , ) link:\doibase 10.1103/physrevb.64.184106 [ * * , ( ) ] @noop _ _ ( , , ) \doibase http://dx.doi.org/10.7566/jpscp.3.016018 [ * * ( ) , http://dx.doi.org/10.7566/jpscp.3.016018 ] link:\doibase 10.1103/physrevb.63.214503 [ * * , ( ) ] link:\doibase 10.1103/physrevd.87.114504 [ * * , ( ) ] link:\doibase 10.1103/physreva.4.1238 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1103/physrevb.78.085124 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.039901 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.100509 [ * * , ( ) ] link:\doibase 10.1103/physreva.6.936 [ * * , ( ) ] link:\doibase 10.1103/physreva.7.1673 [ * * , ( ) ]

we study the existence of monotone heteroclinic traveling waves for a general fisher - burgers equation with nonlinear and possibly density - dependent diffusion . such a model arises , for instance , in physical phenomena where a saturation effect appears for large values of the gradient . we give an estimate for the critical speed ( namely , the first speed for which a monotone heteroclinic traveling wave exists ) for some different shapes of the reaction term , and we analyze its dependence on a small real parameter when this brakes the diffusion , complementing our study with some numerical simulations . maurizio garrione , marta strani + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ saturating diffusion , fisher - burgers equation , density - dependent diffusion , traveling waves , admissible speeds . .15 cm _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 34c37 , 35k55 , 35k57 . .15 cm _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

1702.03782

this paper concerns one - dimensional quasilinear reaction - diffusion pdes . we study the existence of monotone heteroclinic connections of traveling - wave type for reaction - diffusion equations with saturating and density - dependent diffusion , in presence of a convective term ; precisely , we deal with the equation @xmath0 ) - { \partial}_x h(u ) + f(u ) , \quad ( u = u(t , x)),\ ] ] under a suitable set of assumptions ( see section [ sez2 ] ) . not to overload the notation and the readability of the paper , our main results will focus on the special case when @xmath1 is of mean curvature type , namely @xmath2 ( see , for instance , @xcite ) . however , the presented results may hold true for a larger class of operators , as it will be pointed out later on in the paper . reaction - convection - diffusion type models like , with a special focus on traveling waves , are very popular and can be used to describe different mathematical and physical phenomena : the main areas of application , for which we quote a few references , are , for instance , neurophysics and biophysics @xcite , chemical physics @xcite , population genetics @xcite , tumor growth @xcite and mathematical ecology @xcite . as describing quite complicated realistic models , most of the above applications concern systems of equations in higher dimensional spaces . inside this vast class , the fisher - kolmogorov - petrovsky - piskounov ( fkpp ) equation @xmath3 introduced by fisher @xcite to model the spread of an advantageous gene inside a @xmath4-dimensional population , and the viscous burgers equation @xcite @xmath5 finding application , for instance , in problems related to gas dynamics and traffic flows , are probably among the most popular . in the recent years , an increasing interest has been devoted , as well , to a `` combination '' of and , called _ fisher - burgers _ ( or _ burgers - fisher _ ) equation , which is obtained by setting @xmath6 , @xmath7 , @xmath8 and @xmath9 in equation ; this is perhaps the easiest example of pde taking into account both the reaction - diffusion and the convection effects . such an equation has been widely studied by means of both theoretical and numerical methods ( see , e.g. , @xcite ) , and represents the starting point of our investigation . in order to better set up our scheme and introduce our results for generalized fisher - burgers equations displaying saturating effects , we thus now start briefly reviewing the results which are available , in our perspective , for equations and . after its introduction @xcite ( as pointed out , for instance , in @xcite , actually the investigation and analysis of traveling waves in chemical reactions is probably first due to luther @xcite ) , the fisher equation was widely studied , from a mathematical point of view , starting with the paper @xcite . as mentioned above , it was meant to describe the evolution of the relative concentration of a gene inside a population ; under the natural assumption @xmath10 ( meaning that no reaction can be present if the gene is completely spread or not spread at all ) , it was already observed in these first articles the natural appearance of heteroclinic solutions connecting the equilibria @xmath11 and @xmath4 - we will call them _ @xmath12connections _ - of traveling wave type , embodying the transition from the situation where the gene is not present to the one where every individual possesses it . by standard phase - plane arguments , it is immediate to see ( see , for instance , @xcite ) that such solutions are necessarily monotone ; it was then proved in @xcite that they also have some stability properties for a certain class of initial data . for a better comprehension of the model , it is thus natural to search first for solutions having the form @xmath13 , with @xmath14 an increasing function such that @xmath15 , @xmath16 . the values of @xmath17 yielding such solutions are called _ admissible speeds _ , and it turns out that the set of the admissible speeds for equation is an unbounded interval @xmath18 ; @xmath19 takes the name of _ critical speed_. for example , in the fisher case it is @xmath20 . later on , reaction terms other than the fisher one were studied : the most popular are the so called _ type b _ and _ type c _ ( according to the terminology in @xcite ) . the former has been considered in some combustion problems , while the latter appears naturally in biology ( see @xcite ) , in relation with the death rates of populations presenting different genotypes with respect to a particular gene , or in aggregative - diffusive models for population dynamics ( see , e.g. , @xcite for the case of a sign - changing diffusion ) . by some monotonicity arguments , for both of these situations ( see section [ sez4 ] below ) it may be seen that the admissible speed exists and is unique . on the other hand , the viscous burgers equation often appears as a simplification of more complicated models : a well known example is the one given by the navier - stokes equations , describing the dynamics of compressible or incompressible fluids . if we consider without the second order viscous term and with general flux function describing the convection phenomenon , it is well known that solutions with discontinuities ( called internal shocks ) are allowed in the class of the so called _ entropy solutions _ ; in particular , in order to provide uniqueness , it is common to assume the flux function to be convex ( for more details , see @xcite ) . on the other hand , when the viscosity is switched on " , all the discontinuities that appear at the hyperbolic level turn out into smooth transition layers . in this setting , smooth traveling wave solutions to the cauchy problem associated with have been extensively studied over the past years ; indeed , when looking for a solution of the form @xmath21 , equation can be explicitly solved by integration , leading to @xmath22 where @xmath23 ( @xmath24 being an integration constant ) , and @xmath25 . in particular , @xmath26 by definition , so that @xmath27 and represents a decreasing function connecting the values @xmath28 and @xmath29 . finally , it holds that @xmath30 meaning that , once the values @xmath31 and @xmath32 are chosen , there exists a unique @xmath33connection of traveling wave type , having speed @xmath17 . what we have described so far holds true when the diffusion is _ linear_. however , assuming that a diffusive process in nature takes place in a homogeneous way , disregarding local biological and physical conditions ( both environmental - like light , humidity , temperature - and intrinsic in the behavior of the described population - e.g. , its density or concentration , its gradient , cf . , for instance , @xcite ) , seems to be quite a strong restriction . for instance , it may be more likely to imagine that individuals move from richly to scarcely populated areas , or , on the contrary , that under some conditions they tend to aggregate ( for instance under the risk of extinction ) , in opposition to a diffusion - type phenomenon . similarly , it may be expected that the diffusion relents once the gradient of the density becomes large , in view of a kind of blockage effect at too high gradients . of course , there are many ways of introducing such effects in a model , leading to different pictures for the considered problem ; many authors worked in this direction , part of the associated bibliography being available inside the references we mention at the end of the paper . our choice in the present article is to take into account a usual diffusion process ( towards less `` populated '' areas ) which depends on the population density and displays a saturation effect for large gradients ; explicitly , motivated , among the others , by the papers @xcite and @xcite , we consider a diffusion depending on @xmath34 through the mean - curvature type operator ( as the one considered in @xcite ) and possibly density - dependent . in any case , we warn the reader that modifying the assumptions in the diffusion makes possible , in principle , the appearance of other kinds of solutions which are not regular any more , since the functional space to appropriately set the problem into may include discontinuous functions ( for instance , in the case one is led to work in the space of bounded variation functions ) . however , we here disregard this issue , focusing since the beginning on regular traveling waves . as for the admissible speeds , and referring to the previous discussion about the fisher equation , introducing a density - dependent and saturating diffusion may lead to some different outcomes with respect to the linear case . if , on one hand , the case of fisher - type nonlinearities like @xmath8 essentially displays the same features , on the other hand it may be seen that the situation changes when the reaction term is of type c : in the recent paper @xcite this was highlighted when dealing with the model @xmath35 namely equation with @xmath36 and @xmath37 . let us also notice that , wishing to deal instead with a similar issue for the quasilinear counterpart of the burgers equation @xmath38 the situation is slightly different from the one for , since equation admits all the constants as equilibria . fixed two equilibria @xmath39 to be connected , in order to find monotone @xmath33-connections of traveling wave type @xmath21 one thus has to solve the boundary value problem @xmath40 asking @xmath41 for every @xmath42 ( or @xmath43 for every @xmath42 ) . integrating the differential equation on the real line and noticing that @xmath44 in view of the monotonicity , it is immediate to see that it has to hold @xmath45 so that there exists at most a unique @xmath46 giving a monotone @xmath33connection ; we stress that , in the burgers case @xmath9 , the previous condition is exactly ( and in this case @xmath43 for every @xmath42 ) . hence , here the picture is a bit different - actually , as we will see in section 3 , the situation is much more restrictive - and we can explicitly solve the problem , due to the particular form of the equation . it is thus natural to wonder which effect on the admissible speeds prevails when combining a nonlinear advection @xmath47 and a nonlinear forcing term @xmath48 in equation , together with a ( realistic ) saturating and density - dependent diffusion . this will be our main concern ; precisely , we study the problem @xmath49 ) ' - ( h'(v)+c)v ' + f(v ) = 0 \\ v(-\infty)=u_- , \ ; v(+\infty)=u_+ , \quad v'(s ) > 0 , \end{array } \right.\ ] ] derived from the search for monotone heteroclinic traveling waves for , and we determine the set of the admissible speeds for different choices of @xmath48 , providing some numerical evidence , as well . complementarily , we also start investigating the admissible speeds in the small viscosity limit , analyzing the dependence of the critical speed on a small viscosity parameter braking the diffusion ( see section [ sez6 ] ) . in this respect , let us mention that the case without reaction was taken into account , e.g. , in @xcite ; on the other hand , the linear diffusion case with both a reaction and a convective term was analyzed , for instance , in @xcite , where it was shown that one could in general provide only an estimate of @xmath19 . in this framework , the nonlinear ( but not saturating ) diffusion case has been studied in @xcite , as well as in @xcite for the case of porous media - type equations . finally , the specific case of the linear fisher - burgers equation , with particular emphasis on traveling waves , has attracted the interest of several authors in these last years ( we refer , as examples , to the previously mentioned papers @xcite ) . for completeness , let us also cite the paper @xcite , the monography @xcite , as well as the references @xcite for reviews and detailed explanation about density - dependent processes , and , as an example , the papers @xcite for the study of different boundary value problems for the mean - curvature operator . on the other hand , traveling waves for the fkpp equation have been largely studied , respectively , in association with porous media - type diffusions or when the diffusion process takes place on a riemannian manifold . far from being complete in the references , we quote , respectively , @xcite and @xcite , referring the reader to the references in these papers , as well . we close this introduction with a plan of the paper . in section [ sez2 ] we present the general procedure to reduce equation to the first order ; we here adapt to the presence of the nonlinear term @xmath50 the technique already exploited , for instance , in the papers @xcite . essentially , it consists in a suitable change of variables heavily relying on the fact that we search for _ monotone _ solutions . for our purposes , it is here crucial that the shape of the nonlinearity @xmath48 is invariant under `` rescalement '' by @xmath50 . even more , this may happen in presence of a wider class of operators ( see assumption @xmath51 in section [ sez2 ] ) . section [ sez3 ] is the core of the paper and it is entirely devoted to the study of the admissible speeds for equation , assuming the reaction term to be positive on @xmath52 . in particular , the main result of the section is the following . [ ilprincipale ] let @xmath53 be a continuous and positive function and let @xmath1 be as in . moreover , let @xmath54 be a @xmath55-function such that @xmath56 , and @xmath57 be a continuous function such that @xmath58 exists , @xmath59 for every @xmath60 and @xmath61 , @xmath62 for every @xmath63 $ ] , for suitable constants @xmath64 ( notice that this implies @xmath10 ) . then , there exists @xmath65 , with @xmath66 } \frac{d(u)f(u)}{u } } - \min_{u \in [ 0 , 1 ] } h'(u),\ ] ] such that has a solution if and only if @xmath67 . the proof is carried out by means of a shooting technique and lower - upper solutions , after having performed the change of variables described in section [ sez2 ] . we underline that theorem [ ilprincipale ] gives the same bound for the critical speed as in the linear case , and finds immediate application to the fisher - burgers equation . of course , such a bound may be negative , in principle , but for sure we have an unbounded set of positive admissible speeds . incidentally , we observe that the analysis of section [ sez3 ] can be easily extended to the case of suitable general convective terms of the form @xmath68 ( see , for instance , @xcite ) . on the other hand , in section [ sez4 ] we start taking into account reaction terms @xmath48 which are not positive in a right neighborhood of zero . in propositions [ possibileb ] and [ possibilec ] , we first give some sufficient conditions for the existence of a _ positive _ admissible speed . due to the presence of the convection , we observe as well the possible appearance of _ a unique negative _ admissible speed ( cf . @xcite ) , differently from @xcite ; at the same time , we are able to show examples where no admissible speeds exist . the numerical description of these two situations is the concern of section [ sez5 ] . finally , section [ sez6 ] is devoted to the study of equation in the small viscosity limit , meaning that we consider our problem with a small parameter @xmath69 in front of the diffusive term . using the results obtained in the previous sections , we show that the positive critical speeds are proportional to @xmath69 . this result is meaningful in the spirit of the possible appearance of a phenomenon know as _ metastable dynamics _ , as explained in details at the end of the section . in particular , we expect the speed rate of convergence of the time dependent solutions towards their asymptotic limits to be influenced by the size " of the viscosity , as already proved for the linear - diffusion case ( see , for instance , the pioneering article @xcite ) . these features are currently the object of our study , and may represent a possibly interesting follow - up of the present work . in this section , we introduce the main assumptions on the functions appearing in and we describe the change of variables needed in order to reduce such an equation to a first order ode , when looking for monotone traveling waves . we will underline the differences with respect to the case when @xmath36 , referring the reader , e.g. , to the presentation given in @xcite for the general procedure . we focus on the partial differential equation @xmath70 ) - \partial_x h(u ) + f(u),\ ] ] and we assume the following hypotheses : * @xmath71 is a @xmath55-function such that @xmath72 , @xmath73 , @xmath74 for every @xmath75 , and @xmath76 , for suitable constants @xmath77 ; moreover , @xmath78 * @xmath79 \to \mathbb{r}$ ] is a @xmath55-function such that @xmath80 and there exist @xmath81 with @xmath82;\ ] ] * @xmath83 \to \mathbb{r}$ ] is a @xmath55-function such that @xmath56 ; * @xmath84 \to \mathbb{r}$ ] is a @xmath55-function such that @xmath10 and there exists @xmath85 with @xmath86 a couple of comments are in order . first , hypothesis @xmath87 expresses the concept of saturating diffusion ; the monotonicity of @xmath1 ensures the possibility of performing the change of variables below , while @xmath73 will be useful for the estimate below , as we explain in remark [ ilcurvatura ] . finally , assumption ensures that @xmath88 defined in below presents a singularity , as is the case for the curvature operator . concerning the last assumption , we have required for simplicity that @xmath89 , but many of the results we state ( as we have already seen in the statement of theorem [ ilprincipale ] ) can be given under weaker assumptions , with minor changes . we also notice that @xmath90 states that @xmath11 and @xmath4 are equilibria ( in general , the only ones ) for , so that it makes sense to search for @xmath12connections ; in order for such connections to be _ increasing _ , the positive sign of @xmath91 in a neighborhood of @xmath4 is necessary , as an elementary argument in the phase - plane shows . moreover , depending on the shape of the reaction term in a neighborhood of @xmath11 , we can distinguish between three different cases which have by now become classical in the literature : * @xmath92 . in this case , @xmath59 for every @xmath60 ; a prototype model is the so - called _ fisher - type _ nonlinearity @xmath8 . in this case , we say that @xmath91 _ is of type a _ and we write @xmath93 . * @xmath94 and @xmath95 on @xmath96 $ ] . in this case , we say that @xmath91 _ is of type b _ and we write @xmath97 . * @xmath94 and @xmath98 for every @xmath99 ; a prototype model is given by the cubic nonlinearity @xmath100 . in this case , we say that @xmath91 _ is of type c _ and we write @xmath101 . when looking for monotone heteroclinic traveling waves @xmath21 solving , we are thus led to study the boundary value problem @xmath102 ) ' - ( c+h'(v ) ) v ' + f(v ) = 0 \\ v(-\infty)=0 , \ ; v(+\infty)=1 , \ ; v ' > 0 . \end{array } \right.\ ] ] we observe that @xmath79 \to d([0 , 1])=:[0 , d_1]$ ] is invertible ( since @xmath103 implies that @xmath50 is strictly increasing ) ; thus , setting @xmath104 , can be rewritten as @xmath105 since @xmath50 and @xmath106 are strictly monotone , the map @xmath107 is invertible , so that we can treat @xmath108 as the new independent variable and write @xmath109 as a function of @xmath108 . setting @xmath110 , we can proceed similarly as in @xcite , obtaining @xmath111 where @xmath112 , @xmath113 , @xmath114 , and @xmath115 is the primitive of @xmath116 satisfying @xmath117 . writing @xmath118 and denoting by @xmath119 the functional inverse of @xmath120 , is thus equivalent to @xmath121 we observe that , in view of the invertibility of @xmath50 , the shape of @xmath122 is preserved , namely @xmath123 ( resp . @xmath124 ) if and only if @xmath125 ( resp . @xmath124 ) . + for the sake of simplicity , from now on we consider the particular case where @xmath126 so that @xmath127 and becomes @xmath128 where we explicitly note that @xmath129 ^ 2}}.\ ] ] henceforth , problem @xmath130 will be our main object of interest . in this section , we study the admissible speeds for the partial differential equation @xmath131 - { \partial}_x h(u ) + f(u),\ ] ] with @xmath132 assuming that the reaction term @xmath48 satisfies hypothesis @xmath90 with @xmath133 , namely @xmath93 . moreover , we require conditions @xmath134 and @xmath135 to be satisfied . in view of the previous section , we directly focus our attention on problem @xmath130 ; however , as stated in theorem [ ilprincipale ] , in order for a solution of such a problem to give rise to a `` proper '' heteroclinic traveling wave ( namely , a function defined on the whole @xmath136 reaching both equilibria in infinite time ) it is necessary to assume in addition that @xmath137.\ ] ] otherwise , as highlighted in @xcite , coming back to the original variables may give rise to a `` sharp '' traveling wave , i.e. , a traveling wave reaching the equilibria in a finite time and then extended to the whole @xmath136 with constant value . we refer to ( * ? ? ? * remarks 2.1 and 2.2 ) for further comments . with this in mind , from now on we will always study the solvability of the first order two - point problem @xmath130 . we first give a lower bound on the critical speed @xmath19 . [ necessariabf ] let @xmath93 and set @xmath138 . let @xmath139 be a solution to @xmath130 , with @xmath140 in a right neighborhood of @xmath11 . then , @xmath141 the proof is similar to the one of ( * ? ? ? * lemma 3.1 ) . in particular , since @xmath139 has to be positive in a right neighborhood of @xmath11 , we have @xmath142 . moreover , from the differential equation in @xmath130 we deduce @xmath143,\ ] ] and , taking the upper limit for @xmath144 on both sides , we have @xmath145.\ ] ] noticing that @xmath146 , @xmath147 and @xmath148 , we have @xmath149 whence , since @xmath150 , it has to be @xmath151 and @xmath141 we now state and prove our main result concerning the critical speed for equation ; we point out that it is a generalization of the result previously obtained in ( * ? ? ? * theorem 1.4 ) , concerning a convection - reaction - diffusion equation with linear diffusion ( i.e. , @xmath6 ) . [ teorema ] let @xmath93 . then , there exists @xmath65 , with @xmath152 } \frac{d(u)f(u)}{u } } - \min_{u \in [ 0 , 1 ] } h'(u),\ ] ] such that problem has a solution if and only if @xmath67 . the main idea of the proof is to make the ( unique ) solution to @xmath153 denoted by @xmath154 , to intersect , for ( possibly more than ) a suitable value of @xmath17 , with the corresponding solution to @xmath155 which we denote by @xmath156 . in order to prove theorem [ teorema ] , we need to state the following additional lemma concerning the solution to . [ valoriback ] let @xmath91 satisfy @xmath157 . then , for every @xmath158 } \frac{d(u)f(u)}{u } } - \min_{u \in [ 0 , 1 ] } h'(u),\ ] ] it holds @xmath159.\ ] ] moreover , @xmath160.\ ] ] we start proving . let us fix @xmath161 } \frac{d(u)f(u)}{u } } - \min_{u \in [ 0 , 1 ] } h'(u)$ ] ; we notice that , with this choice , it holds @xmath162 } h'(u ) > 0.\ ] ] assume by contradiction that there exists @xmath163 such that @xmath164 since @xmath165 is continuous and @xmath166 , we can assume , without loss of generality , that @xmath167 then , using the differential equation in , thanks to - which ensures that the coefficient in front of @xmath88 therein is always positive - we are able to construct a sequence @xmath168 satisfying , at the same time , @xmath169 a contradiction . for the second part of the lemma , let us now observe that @xmath170 is a lower solution for in @xmath171 $ ] , so that @xmath172 for every @xmath173 $ ] . on the other hand , if there existed @xmath174 $ ] such that @xmath175 , then the differential equation in would lead to @xmath176 , implying that @xmath165 is decreasing in a neighborhood of @xmath177 , a contradiction . hence , is proved . lemma [ necessariabf ] ensures the first inequality in ; our aim is now to show that every @xmath46 such that @xmath178 } \frac{d(u)f(u)}{u } } - \min_{u \in [ 0 , 1 ] } h'(u)\ ] ] is admissible . in view of lemma [ valoriback ] , the solution @xmath154 is defined for every @xmath179 $ ] . on the other hand , fixed @xmath17 satisfying we want to construct a positive lower solution to . to this end , we use the solutions to @xmath180 in dependence of @xmath181 . indeed , noticing that @xmath182 , problem has the solution @xmath183 that is defined in a suitable right neighborhood @xmath184 of @xmath185 , for some @xmath186 . we want to choose @xmath187 in such a way that @xmath188 solves the differential inequality @xmath189 in a right neighborhood of @xmath185 . since @xmath190 we thus have to choose @xmath181 so as to guarantee that , for every @xmath179 $ ] , it holds @xmath191 whence , for every @xmath63 $ ] , @xmath192 dividing by @xmath193 , taking the maximum for @xmath63 $ ] and solving for @xmath187 , we end up with @xmath194 } h'(u ) \pm \sqrt{(c+\min_{u \in [ 0 , 1]}h'(u))^2 - 4\sup_{u \in [ 0 , 1 ] } \frac{d(u ) f(u)}{u}}\;\right],\ ] ] which yields a positive value of @xmath187 thanks to ( which , as already observed , implies that @xmath195 } h'(u ) > 0 $ ] ) . thus , for this choice of @xmath187 , @xmath188 is a positive lower solution to in the interval @xmath196 . as a consequence , there exists a solution @xmath156 to satisfying @xmath197 the argument is now a standard uniqueness one ( see @xcite for further details ) : by continuity and connectedness ( in view of the previous considerations about @xmath154 ) , the two solutions @xmath198 and @xmath154 have to intersect , hence they have to be the same . thus , we have found the desired solution to problem @xmath130 . [ ilcurvatura ] as already anticipated , our explicit choice of @xmath1 is mainly due to the importance of the curvature operator in literature and to the need for a certain readability of the main statement ; however , it can be seen that hypothesis @xmath87 ( together with the consequent properties of @xmath88 ) is sufficient in order to perform the above arguments . indeed , @xmath73 guarantees , through lhpital rule , that @xmath199 for @xmath200 , so that lemma [ necessariabf ] holds true also under assumption @xmath87 ( otherwise , using the same reasonings we would obtain a different bound on @xmath19 ) ; on the other hand , yields a singularity for @xmath88 ( which in the case is obtained for @xmath201 ) . we now examine some statements which can be obtained through the first - order approach - possibly directly from theorem [ teorema ] - for some particular cases of equation , among which , the fisher and the burgers ones . setting @xmath202 , and assuming @xmath93 and @xmath1 as in , equation turns into @xmath203 in the case @xmath204 , the admissible speeds for were determined in ( * ? ? ? * proposition 3.2 ) . we can now obtain such a result as a consequence of the following corollary . [ ilcorollario ] let @xmath93 . then , there exists @xmath205 , with @xmath206 } \frac{d(u)f(u)}{u}},\ ] ] such that has a heteroclinic traveling wave connecting monotonically @xmath11 and @xmath4 if and only if @xmath67 . indeed , whenever @xmath207 on @xmath208 $ ] , the case @xmath209 yields in turns @xmath210 , as in @xcite ; actually , an estimate of @xmath19 was therein given under the weaker bound @xmath211 in terms of @xmath212 . however , this is the case also for us , by slightly modifying the end of the proof of theorem [ teorema ] . the nonlinear fisher equation , obtained by choosing @xmath8 , fits into this framework . with reference to equation , we now consider the case when @xmath1 is given by and @xmath95 , and we analyze the associated first - order model . explicitly , we thus study the viscous conservation law @xmath213 which reduces to the burgers equation with mean - curvature type diffusion when choosing @xmath9 . by proceeding with the change of variables described in section [ sez2 ] , we have to solve the first order two - point problem @xmath214 we observe that , if @xmath215 , then the solution shot forward from @xmath11 is identically zero ; hence , we restrict our attention to the speeds @xmath17 such that @xmath216 ( observe that this bound is also given by lemma [ necessariabf ] ) . it is now immediate to see that theorem [ teorema ] can not be applied to this equation : since assumption @xmath157 is not satisfied , lemma [ valoriback ] fails . the difficulty lies in the fact that it is not ensured that both the forward solution @xmath217 and the backward one @xmath165 are both positive and everywhere defined . thus , an intersection between @xmath156 and @xmath154 may not appear anymore . finally , in this case we may also lose uniqueness from @xmath218 . hence , at first , we need to understand the behavior of the solutions shot backward from @xmath218 : denoted by @xmath139 one of these solutions , the particular structure of the differential equation in allows us to directly integrate by separation of variables , obtaining @xmath219 thus , the right - hand side in such an equality has to be negative for @xmath220 , so that it has to be @xmath221 since it is required that @xmath222 for every @xmath220 , it now follows that @xmath223 and @xmath224 if and hold , we now impose @xmath225 and obtain a unique admissible value of @xmath17 for the existence of an increasing \{0 , 1}-connection : @xmath226 this is compatible with and only if @xmath227 such a condition imposes a concavity shape on @xmath47 , and is indeed not fulfilled by the burgers flux @xmath9 , as is expected in view of the portrait for the burgers equation highlighted in the introduction . on the other hand , a concave flux like @xmath228 enters this setting and we observe that in this case @xmath229 . of course , this argument works whatever the limits the heteroclinic solution has to approach ( we recall that , for equation , all the constants are equilibria ) ; in particular , fixed two equilibria @xmath31 and @xmath32 to be connected , direct computations show that , under the validity of conditions analogous to and , it has to be @xmath230 on the other hand , according to ( * ? ? ? * remark 2.3 ) , _ decreasing _ \{0 , 1}-connections for can be obtained by solving @xmath231 and a similar argument can be performed , showing that a decreasing connection exists , taking @xmath17 as in , under the validity of @xmath232 we can thus state the following proposition . [ propb ] let @xmath47 satisfy either condition or . then , a monotone ( increasing or decreasing , respectively ) @xmath12connection of traveling wave type @xmath21 solving exists if and only if @xmath233 . such a proposition can be easily adapted to @xmath33-connections ( with @xmath31 , @xmath32 arbitrary distinct equilibria ) , suitably changing conditions and , and can be applied , for instance , to the burgers case @xmath9 recalled in the introduction , recovering the mentioned picture . to make our discussion complete , we here consider both the effects of a reaction and a convection term , choosing for simplicity @xmath1 as in , @xmath234 , for some @xmath235 , and @xmath93 such that @xmath236 for all @xmath237 $ ] ( the quasilinear fisher - burgers equation is recovered for the choices @xmath238 and @xmath8 ) . from , it follows immediately that it holds @xmath239 in particular , in the fisher - burgers case with @xmath209 , it holds @xmath240 and the critical speed @xmath19 is exactly given by @xmath241 . on the other hand , if we apply formula in the case @xmath242 for @xmath243 , only leads , in general , to an estimate of @xmath19 . explicitly , we have @xmath244 and it appears more complicated , in view of the nonlinear diffusion , to obtain the precise value of @xmath19 as in ( * ? ? ? * section 5 ) . however , the case of convex fluxes ( corresponding to the choice @xmath245 ) appears to be more significant in literature . we are now interested in studying the admissible speeds for problem @xmath130 when @xmath91 may vanish inside the interval @xmath52 . we will see that , in general , the presence of a convective term changes the picture with respect to the case @xmath246 , and the situation is more complicated than in the previous section : even if @xmath247 , the occurrence of a unique _ negative _ admissible speed ( compare , for instance , with @xcite ) , as well as the nonexistence of admissible speeds , are here possible . we refer the reader to section [ sez5 ] for more details . to begin with , in this section we thus first deal with the problem of the existence of _ positive _ admissible speeds . one possible motivation for the search for positive speeds , among the others , is that they possess some interesting features in dependence on a small parameter braking the diffusion , as we will see in section [ sez6 ] . moreover , there are cases where we have no chance for the existence of negative speeds : for instance , if @xmath248 and @xmath249 ( as in the burgers case ) , then @xmath250 implies that the only solution to the differential equation in @xmath130 with @xmath251 is the zero one . as in section [ sez3 ] , we restrict ourselves to the case when @xmath1 is of mean curvature type , defined in . moreover , we assume the following condition on the convective term @xmath252 : * @xmath83 \to \mathbb{r}$ ] is a @xmath55-function such that @xmath56 and @xmath253 for every @xmath254 . this is the case , for instance , of a convex convective term like the one appearing in the classical burgers equation . we first state a preliminary lemma regarding again some monotonicity properties of the solutions with respect to @xmath255 . as before , we denote by @xmath165 the solution to , which is unique thanks to the fact that @xmath88 is increasing . on the other hand , this time we also have uniqueness ( and continuous dependence on @xmath255 ) of the positive solution to , which will be denoted by @xmath217 ; this follows from direct integration if @xmath97 , and from ( * ? ? ? * theorem 3.9 ) if @xmath101 . [ lemma1 ] let @xmath256 or @xmath101 and let @xmath257 . then , the following inequalities hold : @xmath258 being @xmath259 the unique ( positive ) solution to . similarly , @xmath260 being @xmath261 the unique solution to . moreover , if @xmath262 , then @xmath263 in view of the uniqueness , the statement follows for instance from ( * ? ? ? * exercise 4.1 ) . as a direct consequence of lemma [ lemma1 ] , in the case @xmath264 or @xmath265 , if a positive admissible speed exists then it is necessarily unique . in the figures in the next sections , we will depict the connection corresponding to such a unique speed in some different cases ( see figures [ figb1 ] , [ figb2 ] , [ figc1 ] and [ figc2 ] ) . in this respect , we explicitly mention that we will always show the solution to the first order problem ( @xmath266 ) under study ; indeed , a direct numerical search for the heteroclinic solution - that is the same , for the admissible speed - for problem is much more difficult , since the boundary conditions @xmath15 , @xmath16 require a too high degree of subjectivity in choosing the initial data for the numerical simulations . we thus see another advantage of dealing with the equivalent first - order two - point problem on [ 0 , 1 ] . we also observe that the simulations have been realized through the use of a numerical euler method ( using the software mathematica@xmath267 ) , with a sufficiently small step ( 0.0005 ) and an accuracy of at least @xmath268 , so that the situations depicted appear sufficiently reliable . we recall that @xmath256 means that there exists @xmath269 such that @xmath270 for every @xmath271 $ ] and @xmath272 in @xmath273 . to start with , we observe that the unique positive solution to for @xmath274 , denoted as before by @xmath275 , can be obtained by direct integration in the interval @xmath276 $ ] , leading to @xmath277 in particular , in view of assumption @xmath278 we immediately see that @xmath279 exists and @xmath280 if and only if @xmath252 satisfies @xmath281 thus , this turns out to be a necessary condition for the existence of a positive admissible speed ( otherwise , the solution shot from the left blows - up ) . clearly , such a condition is anyway not sufficient : by the monotonicity properties previously highlighted , the existence of a positive admissible speed will indeed be possible only if @xmath282 so that some estimate on the backward solution @xmath283 is required , as well . with the following proposition , we aim at showing that it is possible to find explicit situations when is fulfilled , being aware of the nonoptimality of the statement ; actually , a precise estimate of the solutions of the differential equation in ( @xmath266 ) is far from being straight . in order to simplify the statement , we here assume that @xmath204 , so that @xmath209 ; the extension of the statement to density - dependent diffusions is straight , as will be clear from the proof . [ possibileb ] let @xmath256 and let @xmath47 fulfill . fix @xmath284 such that @xmath285 where @xmath286 . then , sufficient conditions for the existence of a _ positive _ admissible speed for problem @xmath130 are the following : * if @xmath287 , @xmath288 * if @xmath289 , fixed two other constants @xmath290 with @xmath291 and @xmath292 such that @xmath293 , @xmath294 we preliminarily recall that the unique positive solution @xmath295 to for @xmath274 is given by @xmath296 ( recall that @xmath209 ) . we first consider case 1 ) . in view of @xmath297 , formula immediately gives that @xmath298 on the other hand , @xmath299 is a supersolution for which always stays bounded below from @xmath4 . consequently , in view of the monotonicity of @xmath88 , we have @xmath300 where @xmath283 denotes the solution to with @xmath274 . in view of @xmath301 , this yields @xmath302 so that is fulfilled and the conclusion follows from the usual monotonicity arguments . on the contrary , if we are in case 2 ) , we use @xmath303 to infer , through formula , that @xmath304 reasoning similarly as in case 1 ) in the interval @xmath305 , we also have @xmath306 so that , under the validity of @xmath307 , we deduce @xmath308 now we can apply the same argument as in ( * ? ? ? * proposition 3.9 , ( 3 ) , case 2 ) : in particular , the solution to @xmath309 is a subsolution for @xmath310 in the interval @xmath311 $ ] . since , for every @xmath312 , @xmath313 in view of @xmath314 we obtain that @xmath315 , so that @xmath316 and the conclusion follows as before . in the previous statement , the constants @xmath317 and @xmath290 are meant to control the behavior of the solutions @xmath217 and @xmath318 and may be chosen quite freely . however , as expectable , it is immediate to see that the controls @xmath297 and @xmath303 are more restrictive as long as @xmath317 grows and , symmetrically , the more @xmath290 approaches @xmath317 , the stronger the controls @xmath314 and @xmath307 become , so that a certain care is required . the only substantial change in presence of a density - dependent diffusion is represented by the appearance of @xmath319 under the integral involving @xmath48 , so that the above existence conditions should be slightly modified in order to obtain the same statement . when no convection effects are present in the considered model , namely @xmath37 , assumptions and are trivially satisfied , and proposition [ possibileb ] may thus be adapted so to represent the natural generalization of ( * ? ? ? * proposition 3.9 ) to the case of density - dependent diffusions . we conclude this subsection with some numerical simulations showing an application of proposition [ possibileb ] when @xmath320 is either less or greater than one ( figures [ figb1 ] and [ figb2 ] respectively ) . also , in figure [ figb3 ] we show the necessity of condition for the existence of increasing @xmath321-connections . we now consider @xmath101 , namely there exists @xmath269 for which @xmath98 for @xmath322 and @xmath59 for @xmath323 . as a notation , we set @xmath324 , @xmath325 and @xmath326 in this case , it is immediate to see that the necessary condition may not be sufficient to ensure that the forward solution @xmath295 to is such that @xmath327 exists and is bounded away from @xmath4 , since the function defined in is here only a subsolution . the existence of an admissible speed ( either positive or negative ) is thus even more influenced , in this case , by the interplay between the growths of @xmath328 and @xmath329 . as before , we are going to examine some conditions related to the existence of positive admissible speeds , in the easier case @xmath204 ( similarly as before , slight changes allow to cover also the general situation ) . first , a necessary condition can be deduced from the case @xmath330 ( cf . * proposition 3.9 ) ) . [ ftipoc ] let @xmath101 and let @xmath252 fulfill @xmath331 . then , if @xmath255 is an admissible speed for @xmath130 , necessarily it has to be @xmath332 and has to hold . we just sketch the argument ( which is similar to the one in ( * ? ? ? * proposition 3.9 ) ) . on one hand , if @xmath333 is violated , it follows that @xmath334 , namely @xmath335 and the nonexistence of a positive admissible speed follows from the usual monotonicity argument ( see lemma [ lemma1 ] ) . + on the other hand , if @xmath336 is not true , we have that @xmath337 is a lower solution from the left for @xmath338 ( recall that @xmath253 ) . thus , for @xmath255 we have that @xmath339 as long as @xmath340 , meaning that @xmath341 reaches the value @xmath4 in finite time . hence , no positive admissible speeds exists . finally , the necessity of the condition on @xmath328 comes from the above discussion , comparing with the case @xmath256 . we explicitly notice that the necessary condition , together with , is not sufficient for the existence of a positive admissible speed , as it instead was for @xmath342 ( compare with @xcite ) . actually , the forward solution @xmath343 may blow - up even if and are satisfied . however , if @xmath344 has a very slow growth , we can recover the sufficiency of without further assumptions , as shown in the following ( nonoptimal ) statement . [ possibilec ] let @xmath252 fulfill and @xmath101 satisfy . fix @xmath317 such that @xmath345 then , sufficient conditions for the existence of a _ positive _ admissible speed for problem @xmath130 ( with @xmath209 ) are the following : @xmath346 and * if @xmath287 , @xmath347 * if @xmath289 , fixed other constants @xmath348 and @xmath292 such that @xmath349 , @xmath350 the idea is to proceed similarly as in proposition [ possibileb ] in order to ensure that @xmath351 ( noticing preliminarily that is automatically satisfied in view of ) . we preliminarily observe that @xmath352 immediately gives @xmath353 indeed , if we assume that there exists @xmath354 such that @xmath355 ( we recall that this last quantity is strictly below @xmath4 in view of the choice of @xmath317 ) , with @xmath356 for every @xmath357 , by the monotonicity of @xmath88 we infer that @xmath358 yielding , in view of @xmath359 , the contradiction @xmath360 we now focus on @xmath283 . in case 1 ) , we have @xmath361 and one can proceed similarly as in proposition [ possibileb ] , exploiting the fact that @xmath299 is a supersolution for . in view of the monotonicity of @xmath88 , it follows that @xmath362 so that @xmath363 in view of @xmath364 and is fulfilled . in case 2 ) , we exploit again the same reasoning as in proposition [ possibileb ] , inferring first that @xmath365 then , the backward solution to @xmath366 is again a subsolution for @xmath310 in the interval @xmath311 $ ] . since , for every @xmath367 , @xmath368 in view of @xmath369 we obtain that @xmath370 , so that @xmath371 and the conclusion follows as before . similarly as in the case @xmath256 , if @xmath37 then all the assumptions on @xmath47 are trivially satisfied , so that proposition [ possibilec ] may be easily adapted to generalize ( * ? ? ? * proposition 3.9 ) to the case of density - dependent diffusions . we also observe that we may have taken into account more general reaction terms , as in ( * ? ? ? * proposition 3.11 ) , obtaining similar pictures . we omit further details for briefness . in figures [ figc1 ] and [ figc2 ] , similarly as before , we depict two possible applications of proposition [ possibilec ] . in this section , we aim at showing some situations when there may be a unique negative admissible speed or even no admissible speeds , when either @xmath256 or @xmath101 ; as before , we will assume that the convection term satisfies assumption @xmath278 . we start by observing that , both in the case @xmath256 and @xmath101 , the forward cauchy problem ( i.e. , the problem with initial condition @xmath251 ) has a unique positive solution @xmath217 . if @xmath256 , this follows again by integrating the equation , obtaining formula . if @xmath101 , on the other hand , the solution to itself is unique and positive in a neighborhood of @xmath11 as a consequence of the monotonicity of @xmath88 , regardless of the sign of @xmath17 . we have already recalled this fact @xcite for @xmath151 ( actually , the same motivation works for @xmath372 ) ; if @xmath373 , setting @xmath374 we obtain the equivalent backward cauchy problem @xmath375 which has a unique solution as @xmath376 is increasing and @xmath377 ( in a left neighborhood of @xmath218 ) . consequently , @xmath139 is indeed unique . + this allows to deduce some monotonicity properties ( see again @xcite ) of @xmath217 : since @xmath188 decreases on increasing of @xmath17 , @xmath156 increases with @xmath17 also if @xmath250 . in particular , @xmath378 depends monotonically on @xmath17 ( notice indeed that in this case @xmath156 can not vanish for @xmath379 , in view of the sign of its derivative , so that @xmath378 is well defined ) . on the other hand , the backward solution @xmath165 is always unique and monotone with respect to @xmath17 ( but may possibly blow - up ) . + we can thus make a first observation based on a comparison with the problem without convection @xmath380 assuming the validity of condition . we recall that problem possesses a unique positive admissible speed @xmath381 , as it is possible to see proceeding as in the proof of ( * ? ? ? * proposition 3.9 ) , with very minor changes due to the presence of @xmath382 . the following proposition relates @xmath381 with the admissible speeds ( if any ) for problem @xmath130 through a first estimate . we denote by @xmath383 ( resp . @xmath384 ) the solution to the forward ( resp . backward ) cauchy problem associated with the differential equation in , with initial condition @xmath385 ( resp . @xmath386 ) . let @xmath256 or @xmath101 satisfy and let @xmath252 fulfill hypothesis @xmath278 . if @xmath46 is an admissible speed for problem @xmath130 , then it has to hold @xmath387 } h'(u ) \leq c \leq c_f^ * - \min_{u \in [ 0,1 ] } h'(u).\ ] ] in view of the sign of @xmath388 and the monotonicity properties of the solutions with respect to @xmath17 , we immediately have that @xmath17 can be admissible only if @xmath389 } h'(u ) > 0 $ ] ; moreover , it has to be @xmath390 . now , if there exists @xmath391 such that @xmath392 } h'(u ) -\sigma$ ] , then the solution @xmath393 to @xmath394 is a subsolution for the forward cauchy problem @xmath395 since for @xmath396 we have @xmath397 , by monotonicity the appearance of a @xmath398-connection is then impossible . analogously , if @xmath399 } h'(u ) + \sigma$ ] , then @xmath400 and we have @xmath401 , so that a @xmath398-connection is again impossible . the control gives an effective estimate on the admissible speed for @xmath130 only once this is already known to exist and the value of @xmath381 is known ; however , since it can be easily shown ( cf . @xcite ) that @xmath381 is subject to the bound @xmath402 estimate implies , for instance , that @xmath403 } h'(u ) \leq c \leq 2\sqrt{d_1 f'(u_0 ) } - \min_{u \in [ 0,1 ] } h'(u).\ ] ] thus , we have a further evidence that there may not be positive admissible speeds if the convection is too large . in the following , we aim at describing some possible situations we could encounter ; since a general statement appears quite delicate to be given , we limit ourselves to some heuristic " descriptions and numerical simulations . as remarked before , if @xmath249 then the admissible speed ( if any ) has to be positive , since for @xmath250 the forward solution @xmath156 is necessarily the zero one ( for the same reason , any @xmath404 can not be admissible ) . thus , in this framework nonexistence holds every time that fails ; in figure [ figb0 ] we illustrate this situation . now , assuming that @xmath405 , there are essentially two situations in which the admissible speed may be negative : * holds but fails , so that an intersection between @xmath217 and @xmath165 may be guaranteed only for a negative value of @xmath17 ; * does not hold ( so that the forward solution @xmath406 is not defined in @xmath407 ) , but , since @xmath408 there exists a suitable _ negative _ @xmath409 such that @xmath410 so that @xmath411 exists and is less than @xmath4 . the obstacle to the existence of an admissible speed , in both cases , may be the possibility that @xmath412 is not defined for @xmath413 $ ] , since @xmath154 has already blown up to @xmath4 for some @xmath414 $ ] ( since @xmath415 is negative , the blow - up is here possible ) . in figures [ figb4 ] and [ figb5 ] , we depict two situations of existence in presence of a convex convective term , referring the reader to the considerations in the captions therein . in figure [ figb7 ] , we present a numerical simulation for a non - convex convective term , while in figure [ figb8 ] we show a nonexistence situation . similarly as in the above discussion , also for @xmath101 an obstacle to the existence of an admissible speed may be represented by the possibility that @xmath327 is not defined ( namely , @xmath416 is large ) , so that @xmath17 has to be chosen sufficiently negative and blow - up from the right is then possible . if this is not the case , the validity of , roughly speaking , will be related with the sign of the admissible speed . intuitively , we may thus expect an `` informal '' picture similar to the following , which takes into account the interplay between the growths of @xmath328 and @xmath329 : there exist two constants @xmath417 with @xmath418 such that * @xmath419 there exists @xmath19 positive ; * @xmath420 there exists @xmath19 negative ; * @xmath421 there may not exist an admissible speed @xmath19 . as already mentioned , one should also control the backward solution @xmath165 , which may blow up for @xmath422 . since a rigorous proof and an estimate of @xmath423 and @xmath424 in the general case do not appear easy from a theoretical point of view , we just provide some numerical simulations complementing the cases previously taken into account ( see figures [ figc3]-[figc4 ] below ) . we finally turn our attention to the behavior of the admissible speeds when a small parameter brakes the diffusion , namely considering the pde @xmath425 ) -{\partial}_x h(u)+ f(u),\ ] ] for some @xmath426 . first , we deal with the case @xmath427 and @xmath246 . setting @xmath21 , we observe that , if @xmath428 solves , then @xmath14 solves @xmath429 so that , dividing both sides by @xmath430 and setting @xmath431 , we have @xmath432 where @xmath433 as already remarked , such a problem has a critical speed @xmath434 such that we will not consider explicitly the case @xmath101 , since it is more delicate in view of assumption . if @xmath93 , we know that @xmath437 ; consequently , the critical speed for the original problem is given by @xmath438 where @xmath19 is the critical speed for @xmath439 . on the other hand , in order to estimate @xmath440 if @xmath256 , we notice that the internal zeros of @xmath256 possibly represent equilibria to be asymptotically reached by a heteroclinic solution . if the uniqueness for the backward cauchy problems is fulfilled ( as is the case here ) , we thus have to rule out the speeds yielding other than @xmath12connections . in particular , concerning increasing connections between @xmath441 and @xmath4 , thanks to lemma [ necessariabf ] it is immediate to deduce that it has necessarily to be @xmath442 ; indeed , for values larger than this threshold , increasing @xmath443connections appear . a completely symmetric argument can provide a bound from below , so that @xmath444 to be compared , in the linear case , with the estimate in @xcite . this ensures that , also in these situations , the admissible speed has order at most equal to @xmath445 . let us now turn our attention to the general equation @xmath446 ) - h'(u ) { \partial}_x u + f(u)\ ] ] studied in the present paper , with @xmath91 and @xmath252 satisfying hypotheses @xmath157 and @xmath331 . for the sake of simplicity , we only focus on _ positive _ admissible speeds ; by using the same procedure as before and denoting by @xmath440 the critical speed for such equation , on one hand theorem [ teorema ] ensures that thanks to the previous computations , we observe that if the convective term is convex and satisfies @xmath249 , as for instance in the model case of a burgers flux , then the critical speed for reaction terms of type @xmath450 is of order @xmath69 ; similar considerations can be done when @xmath264 , when a positive admissible speed exists in dependence on @xmath451 ( cf . propositions [ possibileb ] and [ possibilec ] ) . the previous computations suggest that , if we add a small parameter in front of the diffusive term , the speed rate of the heteroclinic traveling waves for under the additional assumption @xmath249 is proportional to @xmath69 , hence it is smaller as @xmath69 goes to zero . this is meaningful in light of the possible appearance of a phenomenon known as _ , whereby the time dependent solutions of an evolutive problem approach their stable steady state in an exponentially long time interval as the viscosity coefficient goes to zero . such behavior has been extensively studied in the linear diffusion case , both for reaction - diffusion and advection - diffusion type equations , to be considered in a _ bounded _ interval of the real line . to mention some results concerning this issue , we recall here @xcite and @xcite for viscous conservation laws and @xcite for burgers type equations , as well as the fundamental contributions @xcite for phase transition problems described by the allen - cahn equation . the number of references is very large and underlines how such a problem has arisen a big deal of interest in the mathematical community over the past years . it may thus be a natural and interesting issue to understand what happens , in this perspective , when considering a saturating diffusion ; one could hopefully expect that , once we reduce our study to a bounded interval @xmath452 and we complement the equation with appropriate boundary conditions , a metastable behavior for the time dependent solution appears . of course , the first issue to be addressed in this direction is the study of the stability properties of the steady states of the initial - boundary value problem associated with . precisely , given @xmath453 , with @xmath454 , one has to consider the problem @xmath455 ) + \alpha \partial_x h(u)+ \beta f(u ) , & \qquad & x\in i , t > 0,\\ & \gamma u(t , \pm \ell ) + \delta{\partial}_x u ( t , \pm \ell)=0 , & \qquad & t > 0,\\ & u(0 , x ) = u_0(x ) , & \qquad & x\in i , \end{aligned}\right.\ ] ] where @xmath456 and @xmath457 , while the functions @xmath458 and @xmath91 satisfy suitable assumptions . once the existence of a steady state @xmath459 for such a problem is proved , there is a broad range of techniques to investigate its stability / instability . as usual , one could linearize the original system around the steady state , and subsequently perform a spectral analysis in order to study the exact location of the eigenvalues . another possible way could be to try to find a lyapunov functional for the original equation ; here the main difficulty is to deal with the sign of the solution at the boundary when computing the time derivative of the lyapunov functional . in possible association with these issues , the stability properties of the traveling waves found in the previous sections may be investigated , as well . to the best of our knowledge , for this problem such directions have not been rigorously studied yet in literature , and this will be the object of a future investigation . d. bonheure and l. sanchez , _ heteroclinic orbits for some classes of second and fourth order differential equations _ , in : a. canada , p. drbek , a. fonda eds . , handbook of differential equations : ordinary differential equations , vol . 3 , elsevier , amsterdam , 2006 . p. p. n de groen and g.e . karadzhov , _ slow travelling waves on a finite interval for burgers-type equations _ , advanced numerical methods for mathematical modelling . j. comput . appl . math . * 132 * ( 2001 ) , 155189 . b. h. gilding and r. kersner , travelling waves in nonlinear diffusion - convection reaction equations , progress in nonlinear differential equations and their applications , 60 , birkhuser verlag , basel , 2004 . k. p. hadeler , _ travelling fronts and free boundary value problems _ , in : j. albretch , l. collatz and k. h. hoffman eds . , numerical treatment of free boundary value problems , birkhuser verlag , basel , 1981 .

we study the effect of a massive central singularity on the structure of a triaxial galaxy using-body simulations . starting from a single initial model , we grow black holes with various final masses and at various rates , ranging from impulsive to adiabatic . in all cases , the galaxy achieves a final shape that is nearly spherical at the center and close to axisymmetric throughout . however , the rate of change of the galaxy s shape depends strongly on the ratio of black hole mass to galaxy mass . when , the galaxy evolves in shape on a timescale that exceeds orbital periods , or roughly a galaxy lifetime . when , the galaxy becomes axisymmetric in little more than a crossing time . we propose that the rapid evolution toward axisymmetric shapes that occurs when provides a negative feedback mechanism which limits the mass of central black holes by cutting off their supply of fuel . # 1#23.6pt rutgers astrophysics preprint series no .

astro-ph9709106

massive compact objects appear to be ubiquitous components of galactic nuclei ( @xcite ) . while these mass concentrations need not be black holes current observations fail by orders of magnitude to resolve the schwarzschild radius there is a growing concensus that the black hole model is reasonable and , in many ways , simpler and more compelling than its alternatives . particularly attractive is the possibility that we are observing the black holes that once powered quasars and that still provide the energy source of active galactic nuclei . this hypothesis is at least crudely consistent with the observed properties of quasars and agn ; in particular , quasar luminosities imply accumulated masses of order @xmath6 or more per galaxy ( @xcite ) , in approximate agreement with the masses inferred from the nuclear kinematics of nearby galaxies . the strongest of the current black hole candidates have masses ranging from @xmath7 in the case of the milky way to @xmath8 in the case of m87 ; expressed as a fraction @xmath2 of the stellar mass of their host galaxies , black hole masses range from @xmath9 to @xmath10 ( @xcite ) . central singularities containing such a large fraction of a galaxy s mass should have a substantial influence on the dynamics and evolution of the surrounding stars and gas . peebles ( 1972 ) and young ( 1980 ) noted that the growth of a black hole in a pre - existing stellar core would pull the surrounding stars into a power - law density cusp . power - law cusps are now known to be universal components of early - type galaxies ( @xcite ) , although they are often found to be less steep than the @xmath11 dependence that follows from adiabatic growth of a black hole growth in an isothermal sphere ( @xcite ) . another consequence of the slow growth of a black hole is a polarization of the stellar velocity field : orbits of stars near the black hole become slightly more circular as a consequence of its growth ( @xcite ; @xcite ) . such a polarization may have been observed in m87 ( @xcite ) . both of these dynamical processes were first studied in the context of spherical galaxy models , and both are predicted to strongly influence the stellar distribution only within the radius of influence of the black hole , @xmath12 , where @xmath13 is the velocity dispersion of stars in the nucleus . however , the influence of a massive central singularity in a non - axisymmetric galaxy can extend far beyond @xmath14 . many of the orbits in triaxial potentials are boxes , orbits that pass arbitrarily close to the center after a sufficiently long time . a number of authors have suggested that the scattering and redistribution of stars on box orbits by a central black hole would cause at least the inner parts of a triaxial galaxy to become rounder or more axisymmetric ( @xcite ; @xcite ; @xcite ) . this prediction was first tested in a self - consistent way by norman , may & van albada ( 1985 ) , who performed @xmath0-body integrations of triaxial galaxy models in which a central point mass was gradually grown . they found a modest evolution toward sphericity at the centers of their models . however , the small number of particles used by norman , may & van albada in their simulations , coupled with the relatively large softening lengths that they were forced to adopt , made it difficult for them to interpret their results in an unambiguous way . more recent @xmath0-body studies of triaxial systems by udry ( 1993 ) , dubinski ( 1994 ) and others while based on improved algorithms and larger numbers of particles have likewise lacked the temporal and spatial resolution needed to accurately reproduce the motion of stars very near to the central point mass . the geometry of barred spiral galaxies is slightly easier to treat than that of elliptical galaxies , both because bars are essentially two - dimensional objects and because their densities are not very strongly peaked toward the center . a number of @xmath0-body studies of barred galaxies ( @xcite ; @xcite ; @xcite ) have shown that a bar can be essentially destroyed that is , converted into a nearly axisymmetric , bulge - like component by a central mass concentration containing a few percent of the total disk mass . test - particle integrations in barred or triaxial potentials ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) indicate that the primary physical mechanism responsible for bar dissolution is chaos induced in the stellar motions by the central mass . attempts to construct self - consistent nonaxisymmetric systems by orbital superposition often fail if only regular , i.e. non - chaotic , orbits are included ( @xcite ; @xcite ; @xcite ; @xcite ) . in this paper , we present an @xmath0-body study of the influence of a growing black hole on a pre - existing triaxial galaxy . our study benefits to a great degree from advances in the art of @xmath0-body simulation since the time of norman , may & van albada s ( 1985 ) pioneering work . our computer code , described in section 2 , employs individual time steps for the particles . in this way we are able to accurately follow the stellar trajectories over a great range of spatial scales , to well within the radius of influence of the black hole . we also assign different masses to particles with different binding energies in order to increase the effective resolution near the black hole . our integrations employ enough particles , @xmath15 , that collisional relaxation effects are negligible . our results ( sections 3,4 ) are in qualitative agreement with the predictions of gerhard & binney ( 1985 ) and others : the black hole causes the initially triaxial galaxy to become rounder near the center and more axisymmetric throughout . but we find that the evolution toward axisymmetry is nearly complete ; i.e. the model reaches an equilibrium state that differs only negligibly from ( oblate ) axisymmetry over a wide range in radii . furthermore we find that the timescale over which the galaxy s shape evolves is strongly dependent on the mass of the black hole . when @xmath2 increases above @xmath10 , the evolution to an axisymmetric state takes place in a time not much longer than the crossing time ; while for @xmath16 , the evolution requires of order @xmath4 crossing times or longer . we argue , based on test - particle integrations , that the sudden decrease in the galaxy response time when @xmath2 exceeds @xmath17 is a consequence of a transition to global stochasticity in the phase space of the boxlike orbits , orbits that are necessary for maintaining triaxiality . our results suggest that there is a characteristic mass fraction , of order @xmath18 , above which a central black hole will cause its host galaxy to evolve rapidly to an axisymmetric state . this mass fraction is close to the maximum value of @xmath2 observed in nearby galaxies : several galaxies have @xmath19 , and the current record - holders , ngc 4342 and ngc 3115 , both have @xmath20 . we suggest that this agreement is more than a coincidence . the fueling of massive black holes in agn and quasars requires matter to be channeled into the nucleus from large distances , and gravitational torques resulting from a bar - shaped or triaxial mass distribution are an efficient way to accomplish this . the masses of nuclear black holes may be regulated by a negative - feedback mechanism that shuts off the supply of fuel to the nucleus once @xmath2 exceeds roughly the critical value found here ( section 5 ) . the @xmath0-body code used here was a hybrid , combining elements from the `` mean field '' code of hernquist & ostriker ( 1992 ) and the nbody programs of aarseth ( 1994 ) . the equation of motion for the stars was @xmath21 the black hole , of mass @xmath22 , remained fixed at the origin . the first term of equation 1 describes the interaction of the stars with the black hole , while the second represents the star - star interactions . the latter were computed via a basis - function expansion of the potential , with coefficients @xmath23 that were periodically updated . the indices @xmath24 and @xmath25 run from @xmath26 to @xmath27 and @xmath26 to @xmath28 ; we used @xmath29 radial basis functions and @xmath30 angular basis functions . some test experiments with larger @xmath27 and @xmath28 produced very similar results . stars were advanced with individual timesteps ( sometimes differing by as much as a factor of @xmath31 ) using aarseth s ( 1994 ) fourth - order nbody1 integrator . the timesteps were updated after each step by a function involving the force and the first three time derivatives of the force , multiplied by the square root of an accuracy parameter @xmath32 . we used @xmath33 , smaller than the value recommended by aarseth @xmath34 . the coefficients @xmath23 for the potential expansion were updated at fixed time intervals @xmath35 , after the coordinates of the stars were predicted to a common time . the same initial model was used for all of the integrations described here . we generated a steady - state , triaxial , nonrotating @xmath0-body galaxy containing @xmath36 particles as follows . \1 . a random , spherical distribution of @xmath37 particles with density profile @xmath38 , @xmath39 was set up . the particles were initially at rest . the combined mass of these particles was set equal to one ; fixing @xmath40 thus determined the model units . this configuration was evolved forward with the @xmath0-body code , with @xmath41 , until a time @xmath42 . during this time , the model collapsed and re - expanded , forming in the process a nearly steady - state , triaxial bar as in aguilar and merritt ( 1990 ) and cannizzo & hollister ( 1992 ) . the particles in the equilibrium triaxial model were sorted by binding energy into 10 groups . each particle in the most - bound group was replaced by 10 particles , whose coordinates were computed by integrating the triaxial model for 2 additional times units and storing the positions and velocities at 10 equally - spaced times . in the same way , the particles in the next binding - energy group were replaced by 9 particles each , then by 8 , etc . ; particles in the least - bound group were not subdivided . the total mass of the model was left unchanged . this scheme for subdividing particles was designed to increase the resolution of the @xmath0-body code very near the center , where the black hole produces a cusp in the density profile . each of the @xmath43 particles so generated was then replaced by two particles , one at the original position @xmath44 and one at @xmath45 . this procedure guaranteed that the center of mass of the model was initially at the origin and that the model had no net momentum . the resulting set of @xmath36 particles was evolved for another 20 time units to ensure that the model had reached equilibrium . the configuration obtained at the end of this integration was defined as the initial model . the initial model is a triaxial body with a short - to - long axis ratio @xmath46 of approximately 0.5 , and a triaxiality index " @xmath47 . the model is strongly peanut - shaped as seen from the intermediate axis ( figure [ fig5 ] ) , a common feature of @xmath0-body bars . the spherically - symmetrized properties of the model are shown in figure [ fig3 ] . the density profile exhibits a constant - density core of radius @xmath48 , and a half - mass radius of 0.48 . the particle motions are approximately isotropic in the core , and become increasingly radial in the envelope ( figure [ fig3]c ) a relic of the radial collapse . table 1 lists some of the global properties of the initial model . roughly 89% of the mass in the original spherical cloud remained bound after the collapse . ( thus the bound mass of the initial triaxial model is 0.89 in the units adopted here . ) approximating the model as spherically symmetric , the full period of a circular orbit at the radius containing one - half of the bound mass is @xmath49 . figure [ fig1 ] shows that the initial model is close to equilibrium . plotted there is the time dependence of the axis ratios of the most - bound 2% , 10% and 50% of the particles , defined as the axis ratios of a homogeneous spheroid with the same moment - of - inertia tensor as the particles . over an elapsed time of 50 units roughly 17 half - mass orbital periods there is no discernible change in the shape of the model . the radii containing 1% , 2% , etc . of the total mass were likewise found to remain essentially constant over this interval . the black hole was grown by increasing the mass at the center of the model according to the relation @xmath50 with @xmath51 ( figure [ fig2 ] ) . various values for @xmath1 and @xmath52 were used , as discussed below . because the mass of stars bound to the initial model is 0.89 in our units , the final black hole mass expressed as a fraction of the bound mass of the galaxy is approximately @xmath53 . the black hole softening length @xmath54 of equation ( 1 ) was computed via the relation @xmath55 in other words , the softening length was chosen such that a typical star at a distance @xmath54 from the black hole would have an orbital velocity of 8 in model units . for comparison , the central velocity dispersion of the initial model is approximately 1.2 ( figure [ fig3]b ) . the softening length which is a function of @xmath1 is given in table 1 for the three values of @xmath1 used here . our adopted softening length was always much smaller than the radius of influence @xmath14 of the black hole . by choosing @xmath56 , we were able to accurately reproduce the formation of a steep stellar cusp as the black hole grew . the @xmath0-body code conserved the total energy of the model to @xmath57 over 50 time units with no black hole . once the black hole had grown , energy was typically conserved to @xmath58 over the same time interval . the growth of the black hole caused the initially triaxial model to become nearly spherical near the center . we therefore begin by describing the spherically - symmetrized structure of the models at late times , and we compare our results with those of earlier modelling that imposed spherical symmetry . here and below , smooth estimates of spatially - dependent properties were computed from the discrete particle coordinates via nonparametric function estimators , either adaptive kernels or smoothing splines ( @xcite ; @xcite ) . figure [ fig3]a shows the dependence of the central density profile at @xmath59 on @xmath1 , for three integrations with @xmath60 . the initial model has a small core , with radius @xmath61 . this core is replaced by a cusp following the growth of the black hole in all three models ; within the cusp , the dependence of density on radius is well approximated as a power law , @xmath62 with @xmath63 . in the runs with the smaller black holes , @xmath64 and @xmath65 , the signature of the core persists even after growth of the black hole in the form of an inflection in the density profile at @xmath66 . quinlan , hernquist & sigurdsson ( 1995 ) computed the evolution of the stellar distribution function @xmath67 of a spherical model as the mass of a central singularity was slowly increased . in models with cores , the growth of the black hole produced profiles similar to those of figure [ fig3]a ( e.g. their fig . 3 ) ; the logarithmic slope of the central density cusp was usually close to @xmath68 and only weakly dependent on the initial density profile . their results were confirmed by sigurdsson , quinlan & hernquist ( 1995 ) in spherically - symmetric @xmath0-body integrations . the velocity dispersion profile ( computed by assuming a spherical velocity ellipsoid ) also exhibits a cusp after the growth of the black hole ( figure [ fig3]b ) , with @xmath69 , as expected for particles moving in response to an inverse - square force law . the velocity dispersion in the initial model falls slightly toward the center , a common feature in models with steeper - than - isothermal central density profiles ( e.g. @xcite ) . this drop is retained following the growth of the smallest black hole , @xmath64 , and even for @xmath70 the velocity dipsersion profile exhibits a mild inflection at @xmath66 . these results are again similar to those of quinlan , hernquist & sigurdsson ( 1995 ) ( e.g. their fig . 4 ) and sigurdsson , quinlan & hernquist ( 1995 ) . adiabatic growth of a central point mass in an initially constant - density core is expected to circularize the stellar orbits slightly ( @xcite ; @xcite ) . the effect is subtle , since orbital eccentricities are left essentially unchanged by slow changes in the potential ( @xcite ) ; the velocity polarization is a second - order effect resulting from the different character of orbits in the keplerian and harmonic - oscillator limits . we nevertheless see the effect here ( figure [ fig3]c ) : the velocity anisotropy , @xmath71 , falls slightly below zero for @xmath72 in the models with @xmath73 and @xmath74 , indicating a slight bias toward circular motions . however , @xmath75 does not drop below about @xmath76 . quinlan , hernquist & sigurdsson ( 1995 ) and sigurdsson , quinlan & hernquist ( 1995 ) found similar , modest changes in @xmath75 at small radii in spherical models with comparable black hole masses . the `` observable '' velocity dispersion profile is presented in figure [ fig3]d . here the averaging was carried out over circular rings in the plane of the sky , defined as the plane perpendicular to the intermediate axis of the model . the @xmath77 velocity cusp is clearly visible in the projected profile as well . to summarize : the slow growth of a central point mass induces changes near the center of the @xmath0-body model that are similar to those predicted by young ( 1980 ) , goodman & binney ( 1984 ) , quinlan , hernquist & sigurdsson ( 1995 ) and sigurdsson , quinlan & hernquist ( 1995 ) on the basis of spherical models . the constant - density core is replaced by an @xmath78 cusp ; the velocity dispersion increases as @xmath77 near the black hole ; and the velocity ellipsoid becomes mildly elongated in directions tangential to the radius vector . all of these changes are confined to a radius @xmath72 . not all effects of the central singularity are confined to small radii , however . figure [ fig4 ] illustrates the time - dependence of the model axis ratios for three integrations with @xmath79 and @xmath60 . axis ratios were defined in terms of the principal axes @xmath80 of the homogenous ellipsoid with the same moment - of - inertia tensor as the @xmath0-body model . following the usual practice , particles were first ranked according to binding energy and the inertia tensor was computed for various subsets , e.g. the most - bound 2% , 10% etc . figure [ fig4 ] shows that the model changes shape even at large radii , out to and exceeding the half - mass radius . near the center at radii corresponding to the inner few percent of the mass in stars all of the models become nearly spherical , on a timescale that is comparable to @xmath52 . at larger radii , both axis ratios @xmath81 and @xmath46 initially increase , but the evolution appears to slow or halt once the models approach axisymmetry , i.e. when @xmath81 comes close to one this is seen most clearly in the integration with @xmath70 , where the evolution is essentially complete once the black hole has reached its full mass at @xmath82 . in the runs with the smaller black hole masses , axisymmetry within the half - mass radius is barely reached at the final time step and slow evolution continues until the end of the run . the evolution timescale is also a function of location in the model , i.e. longer at larger radii . the evolution seen here toward rounder or more axisymmetric shapes is not surprising ; similar evolution has been observed in a number of @xmath0-body studies of triaxial galaxies following an increase in the central density ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the mechanism that drives the evolution is discussed in detail below ; here we note only that the evolution appears to be dependent both on the presence of a central singularity and on departures from axisymmetry , since it slows or stops once the models are close to axisymmetric . one nevertheless worries that the evolution seen in figure [ fig4 ] might be due in part to systematic errors in the @xmath0-body code . most errors in the force computations or the time integration would tend to produce spurious evolution toward spherical symmetry . we note that the run with the largest black hole exhibits essentially no evolution once axisymmetry is reached ; the model accurately maintains its highly flattened , axisymmetric shape . on the other hand , one might not expect integration errors to have much effect in axisymmetric models since angular momentum conservation would keep most particles from approaching the black hole . the fact that the stellar cusps , once formed , persist without significant change also argues in favor of the @xmath0-body integrations being carried out correctly in the vicinity of the black hole . but here , too , a cautionary note is in order . the representation of the potential via a basis set ( eq . 1 ) is not optimal . since the basis functions used here correspond to mass components with central cusps , the @xmath0-body code might tend to produce cusps or to spuriously maintain them once formed . although the timescale for the evolution in figure [ fig4 ] is strongly dependent on @xmath1 , the equilibrium figures of the three models are very similar : all models tend toward axisymmetry , even at radii exceeding the half - mass radius , and the short - to - long axis ratio of the final configuration varies from @xmath83 at the smallest radii to @xmath84 near the half - mass radius in each of the models . figure [ fig5 ] illustrates the projected appearance of the model with the intermediate - mass black hole , @xmath73 , at the final time step ( @xmath85 ) , as seen from a point in the equatorial plane . the initial model is also shown for comparison , as seen from the intermediate axis . the elongation of the initial model is approximately constant with radius , and its isophotes are strongly peanut - shaped at intermediate radii , the isophotes of the model with @xmath73 become progressively more elongated with increasing radius , although they remain rounder than those of the initial model at least until a radius of @xmath86 , where the dynamical time is @xmath87 , roughly equal to the integration time . boxy or peanut - shaped isophotes like those in figure [ fig5 ] are a common feature of @xmath0-body galaxies ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . by contrast , real elliptical galaxies have isophotal shapes that are almost precisely elliptical : the deviations are typically around @xmath88 , rarely more than about @xmath89 , and are as often in the direction of diskiness as of boxiness ( @xcite ) . as far as we know , no convincing explanation for the lack of significant boxiness in real ellipticals has yet been given . based on figure [ fig5 ] , we propose that the precisely ellipsoidal shapes of elliptical galaxies may be a consequence of evolution induced by nuclear black holes . inspection of figure [ fig4 ] reveals one intriguing result . in the integration with the largest black hole , @xmath70 , the model reaches axisymmetry at all three radii displayed @xmath90 by roughly the time the black hole has ceased growing , i.e. by @xmath91 . in other words , the evolution of the model s figure appears to be limited only by the growth rate of the black hole . by comparison , in the integrations with smaller black holes , @xmath73 and @xmath64 , the model figure evolves at different rates at different radii , and the evolution requires much longer to reach completion roughly 100 time units ( @xmath92 orbital periods ) at the half - mass radius in the model with @xmath73 , and perhaps 200 - 300 time units ( @xmath93 orbital periods ) in the model with @xmath64 . only at the very centers of these two models where the orbital period is a small fraction of the black hole growth time does the shape of the model appear to evolve at a rate that is limited by the growth of the black hole . outside of the very center , the model figure evolves on a timescale that is approximately a constant multiple ( different for the two values of @xmath1 ) of the local orbital period . to better understand the very rapid evolution observed in the model with @xmath70 , we repeated this integration with three different values of the black hole growth time : @xmath94 and @xmath95 . the shortest of these growth times , @xmath96 , is small compared even with the orbital period at the center of the initial model ; hence the growth of the black hole in this integration is essentially impulsive as seen by the stars . figure [ fig6 ] reveals that the model responds to the @xmath70 black hole about as quickly as it can ; in the impulsive limit , the model evolves to axisymmetry at all radii on a timescale that is close to the local orbital period . for instance , the intermediate - to - long axis ratio @xmath81 determined by the 50% most - bound stars reaches unity by a time of @xmath97 when @xmath96 , compared to a circular orbital period of @xmath49 at the half - mass radius ( table 1 ) . these experiments suggest that the timescale on which the galaxy responds in shape to the appearance of the black hole is a steep function of the black hole s mass . figure [ fig7 ] plots the evolution toward axisymmetry of the most - bound 50% of the stars for the three values of @xmath1 . the curves for the two smaller black hole masses were taken from the simulations with @xmath60 , which is shorter than the apparent response time of the galaxy at the half - mass radius for these values of @xmath1 . the curve for @xmath70 is from the integration with @xmath98 , short enough that the galaxy s response at the half - mass radius is not strongly limited by the black hole growth rate . figure [ fig7 ] suggests that the response time of the galaxy to the black hole drops suddenly when @xmath1 exceeds @xmath99 . the critical mass fraction can be estimated more accurately by looking at the integration with the longer black - hole growth time , @xmath100 ( figure [ fig6]a ) . axisymmetry at all radii is reached in these integrations somewhat before the black hole has attained its final mass of @xmath74 . for instance , at the radius containing @xmath101 of the stars , @xmath102 when @xmath103 , at which time the black hole mass is @xmath104 , or @xmath105 . at the radius containing @xmath106 of the stars , approximate axisymmetry is reached when @xmath107 ; taking into account the orbital time at this radius suggests a critical mass fraction for transition to axisymmetry of @xmath108 . thus , evolution toward axisymmetry takes place very rapidly once @xmath2 exceeds @xmath109 . the evolution toward axisymmetric shapes seen here is an expected consequence of stochasticity in the stellar orbits , induced by the central singularity . the argument which is not original with us goes roughly as follows ; fuller discussion may be found in the papers of schwarzschild ( 1981 ) , norman , may & van albada ( 1985 ) , gerhard & binney ( 1986 ) , gerhard ( 1987 ) , udry & pfenniger ( 1988 ) , merritt & fridman ( 1996 ) , barnes & hernquist ( 1996 ) , merritt & valluri ( 1996 ) , and merritt ( 1997a , b ) . many of the stars in the initial , triaxial model are on regular box orbits that carry them close to the center , once per orbital period . ( a regular orbit is defined as an orbit with three isolating integrals of the motion , two in addition to the energy . ) most of the box orbits lose two of their three integrals of motion once the black hole appears , essentially because of large - angle deflections suffered during close passages to the center . ( the exceptions are boxlike orbits that lie close to a stable resonant orbit that avoids the center , e.g. the @xmath110 `` banana '' orbit . ) trajectories defined by only one integral of motion the energy densely fill the region within the equipotential surface @xmath111 . the time - averaged shape of such an orbit is approximately spherical . because box orbits with a variety of shapes are crucial for maintaining strongly triaxial figures , the model can not preserve its triaxial shape once the black hole appears . it begins to evolve toward more nearly spherical , or at least axisymmetric , shapes . \5 . as axisymmetry is approached , the boxlike orbits disappear , since axisymmetric potentials can only support tube orbits orbits that conserve angular momentum about the symmetry axis and avoid the center . the black hole is then no longer effective at inducing stochasticity in the orbital motion , and the galaxy evolves to a steady , axisymmetric state in a few crossing times . the evolution of the @xmath0-body models described above is consistent on a qualitative level with this picture . especially compelling to us is the fact that the @xmath0-body models evolve in shape only when two conditions are satisfied : the model is triaxial , i.e. departs from axial symmetry ; and @xmath112 . the initial , strongly triaxial model does not evolve ( figure [ fig1 ] ) , presumably because the box orbits which it contains are regular , or nearly so consistent with the expected behavior of boxlike orbits in triaxial models with smooth cores ( e.g. @xcite ) . and after the black hole is grown , the models cease evolving once axisymmetry is reached ( figures [ fig4 ] , [ fig6 ] ) , i.e. once the box orbits have disappeared . furthermore , the evolution of the model s shape is rapid compared to the expected rate of two - body relaxation , nor is any evidence of collisional relaxation apparent in the test integrations of the initial model without a black hole . taken together , these facts suggest that the evolution which we observe is being driven primarily by the interaction of boxlike orbits with the central singularity . further evidence in favor of this picture is provided by the timescales for evolution observed in the @xmath0-body models . once a black hole is introduced , the box orbits would be expected to respond by gradually filling the phase - space volume made available to them i.e. the volume defined by the full energy hypersurface , minus those regions occupied by regular orbits , mostly tubes . this relaxation toward a uniform population of the accessible phase space is called chaotic mixing ( @xcite ) , and it has a characteristic timescale that can be measured via test - particle integrations in fixed potentials . merritt & valluri ( 1996 ) carried out such calculations in two triaxial models , one with a central density cusp , the other with a smooth core and a central point mass containing @xmath113 of the mass of the model . those authors evolved ensembles of @xmath37 particles started from various regions on the equipotential surface and recorded their relaxation to a steady state . mixing timescales in both models were found to depend on the starting point of the ensemble . ensembles located initially in the most stochastic parts of phase space e.g. near the short axis of the figure were found to mix most rapidly , filling much of their allowed region after just @xmath114 orbital periods . other ensembles were found to mix more slowly , remaining trapped in limited regions of phase space for @xmath4 orbital periods or longer . merritt & valluri ( 1996 ) suggested an average relaxation time associated with chaotic mixing in these models of @xmath115 orbital periods . this estimate is quite consistent with the rate at which the @xmath0-body model with @xmath64 is found here to evolve the intermediate - to - long axis ratio changes from @xmath116 to @xmath117 in about 50 half - mass orbital periods ( figure [ fig7 ] ) . merritt & valluri ( 1996 ) did not extend their mixing calculations to triaxial models with larger black hole masses , nor are we aware of any other 3-d test - particle studies that could be usefully compared to our @xmath0-body models with @xmath73 and @xmath74 . for this reason , we carried out new test - particle integrations in triaxial models with central point masses of various sizes . we adopted the mass model of merritt & valluri ( 1996 ) , which has density @xmath118 the central density @xmath119 is related to the total mass @xmath120 by @xmath121 . the parameter @xmath122 is a core radius ; we chose @xmath123 . the axis ratios were taken to be @xmath124 and @xmath125 , similar to the values in the initial @xmath0-body model . to this model was added a central point mass containing various fractions @xmath1 of the total mass @xmath120 ; we used @xmath126 , and @xmath127 . test - particle integrations were then carried out for sets of orbits with starting points distributed over the half - mass equipotential surface , as described in merritt & valluri ( 1996 ) . of greatest interest here is the configuration space volume filled by boxlike orbits over a relatively short interval of time . we accordingly integrated orbits for only @xmath128 periods of the long - axis orbit . we found a striking result : over this short time interval , the behavior of the boxlike orbits changes suddenly as @xmath1 is increased ( figure [ fig8 ] ) . for @xmath129 , the boxlike orbits particularly the thin boxes with starting points near the long axis of the figure , which are crucial for maintaining a triaxial shape showed relatively little evidence of stochastic evolution . while often not regular , the orbits in these models filled regions quite similar in shape to the regions occupied by orbits in the model with @xmath41 . by contrast , when @xmath1 was increased from @xmath65 to @xmath74 , essentially none of the boxlike orbits were found to maintain well - defined boxy shapes over 10 orbital periods . instead , they quasi - randomly filled a rougly axisymmetric region , with almost no hint of `` memory '' from one oscillation to the next of the location of the previous turning point . the only exceptions were orbits with starting points near to the @xmath130 banana . we speculate that the change in the character of the boxlike orbits as @xmath1 is increased from @xmath65 to @xmath74 is indicative of a transition to global stochasticity ( e.g. @xcite ) . in many dynamical systems , one observes a sharp transition as a parameter is varied : from stochastic motion that is closely bounded by kam surfaces , i.e. regular orbits ; to motion that is strongly interconnected over large portions of the space . in the latter regime , stochastic trajectories at a given energy are nearly indistinguishable , moving rapidly in just a few oscillations from one part of stochastic phase space to another . this is just the behavior illustrated in figure [ fig8 ] as @xmath1 is increased above @xmath65 . whether or not this interpretation is correct , figure [ fig8 ] suggests that the timescale over which stochasticity induces changes in the phase - space population of a triaxial model should drop from @xmath131 orbital periods when @xmath129 , to @xmath132 orbital periods when @xmath133 . this is what we observe in the @xmath0-body models ( figure [ fig7 ] ) . we might predict an even steeper dependence of evolution rates on @xmath1 in the @xmath0-body models than in the test - particle integrations , since the @xmath0-body models develop a central density cusp with a strength that increases with @xmath1 ( figure [ fig3]a ) , and a cusp will itself induce stochasticity in the motion of boxlike orbits ( @xcite ) . one final question concerns the end states of the @xmath0-body models . their figures are remarkably close to axisymmetric . this is particularly true with regard to the integrations with the largest black hole , @xmath70 , where @xmath81 at the final time step is consistent with unity at all radii displayed in figures [ fig4 ] and [ fig6 ] . the approach to axisymmetry is not as complete in the integrations with the smaller black holes , but figure [ fig4 ] suggests that a modest increase in the integration times would probably have produced a precisely axisymmetric end state for these values of @xmath1 as well . one might have expected the @xmath0-body models to reach equilibrium without evolving so completely to axisymmetry . for instance , merritt ( 1997b ) , in a study of the triaxial self - consistency problem , found that models with @xmath134 central density cusps could exist in equilibrium for values of @xmath81 exceeding @xmath135 . but there are a number of reasons why the @xmath0-body models should prefer even more axisymmetric states than the self - consistency studies would suggest . first , the @xmath0-body models contain both steep stellar cusps and central singularities ; hence one would expect stochasticity to be more destructive of the boxlike orbits in this study than in that of merritt ( 1997b ) . second , the self - consistency studies are relatively crude in their treatment of the effects of the stochasticity ; for instance , merritt ( 1997b ) simply eliminated orbits that gave evidence of instability after 50 orbital periods . in fact , orbital instability rates scale roughly with the local orbital frequency , making them a strong function of radius . hence one should eliminate a larger fraction of the boxlike orbits near the center than at large radii , making it more difficult to reconstruct a self - consistent triaxial shape . third , the rapid evolution toward spherical symmetry at the very centers of the @xmath0-body models must influence the behavior of orbits at larger energies , perhaps accelerating their evolution . it is conceivable that modestly triaxial equilibria do exist even with central singularities , but that our scheme of growing the black hole in a pre - existing triaxial model prevents us from arriving at these end states . we have found that a central singularity containing more than @xmath136 the mass of its host galaxy induces a rapid transition from triaxiality to axisymmetry in the distribution of the surrounding stars , both near the center and at appreciable distances . less massive singularities induce a more gradual evolution , but the end state is still close to axisymmetric ; in fact , our results suggest that the shape of an initially triaxial galaxy containing a central singularity of mass @xmath1 is approximately independent of @xmath1 . ( other features of the final state , like the central density profile , clearly do depend on @xmath1 . the final state also presumably depends on the details of the initial model , which we did not vary ) . we argued that the sudden change in the rate of evolution of the galaxy s shape as @xmath2 exceeds @xmath104 is due to a transition to global stochasticity in the phase space of the boxlike orbits , orbits which are required to maintain triaxiality . while it would be dangerous to generalize from such a small set of experiments , it seems reasonable to suppose that a sudden transition to axisymmetry would occur in most triaxial models as the mass of a central singularity is increased beyond some critical value . we discuss this hypothesis in more detail below ; for the moment , we assume that such a transition generally occurs for central point masses of order @xmath137 . is there any evidence from real galaxies for such a critical mass ? it is well known that nuclear black holes often contain of order @xmath138 the mass of their host galaxies . figure [ fig9 ] shows @xmath2 for the seven galaxies with the largest , well - determined black hole mass fractions . @xmath139 is defined in figure [ fig9 ] in the usual ( and , for this study , appropriate ) way as the mass of the stellar bulge in the case of disk galaxies , and as the total stellar mass in the case of elliptical galaxies . uncertainties in estimates of @xmath1 vary considerably from galaxy to galaxy ; each of the black hole masses in figure [ fig9 ] is considered to be fairly accurate , with probable uncertainties of order @xmath106 or less . ( the galaxy ngc 4486b may have @xmath2 as great as @xmath140 but has been omitted from figure 9 since the uncertainty in the mass of its black hole is very large ; in fact @xmath1 for this galaxy is consistent with zero ( @xcite ) ) . figure [ fig9 ] confirms a tendency for the most massive black holes to contain approximately @xmath141 times the stellar mass of their host galaxies , both in very bright ( m87 ) and very faint ( m32 ) galaxies . the largest , well - determined mass fractions are associated with ngc 4342 and ngc 3115 , which have @xmath142 and @xmath143 , respectively . the remainder of the galaxies have @xmath16 . based on this modest sample , it appears that nature avoids making nuclear black holes with masses exceeding @xmath109 the mass of their host galaxy remarkably similar to the critical mass fraction that induces a transition to axisymmetry in our simulations . we suggest that this agreement is not a coincidence . the fueling of massive black holes in quasars and agn requires matter to be funneled into the nucleus from large distances , of order a kiloparsec or more ( @xcite ) . models that produce fuel locally , e.g. from a dense star cluster surrounding the black hole ( @xcite ) , generally fail to supply mass at a rate adequate to power the most luminous active nuclei . in order for infall to occur over kiloparsec scales , the angular momentum of the infalling stars or gas must somehow be removed . while many of the details of this process remain obscure , a number of lines of evidence suggest that large - scale asymmetries in the stellar distribution are a necessary ingredient ( @xcite ) . gravitational torques from non - axisymmetric perturbations triggered by galaxy interactions , internal self - gravitational instabilities , or both can remove much of the angular momentum from the interstellar medium , allowing gas to fall into the nucleus ( @xcite ) . support for this picture comes from computer simulations of galaxy interactions ( @xcite ; @xcite ; @xcite ) ; from the observed association of nuclear activity with asymmetric geometries and bars ( @xcite ; @xcite ; @xcite ) ; and from the fact that the host galaxies of nearby quasars and agn often have close , interacting companions ( @xcite ; @xcite ; @xcite ) . any mechanism that enforced large - scale axisymmetry in the stellar distribution of a galaxy would therefore be expected to strongly inhibit the fueling of a nuclear black hole and hence to limit its mass . to be effective , such a mechanism would need to operate in little more than a galaxy crossing time . this is because the black holes that powered quasars must have acquired most of their mass in @xmath144 years or less ( @xcite ) , and because the timescale for subsequent , tidally - triggered fueling is believed to be comparably short ( @xcite ) . the rapid evolution toward axisymmetry seen here when @xmath2 exceeds a critical value could be such a mechanism . in this picture , nuclear black holes might be expected to often contain of order @xmath141 times the mass of their host galaxies , and their masses should never appreciably exceed @xmath145 unless the galaxy lost a significant fraction of its stellar mass after the black hole formed . an additional prediction , of course , is that galaxies containing `` maximal '' black holes should be close to axisymmetric . some of the galaxies in figure [ fig9 ] ( ngc 4342 , ngc 3115 , ngc 4258 ) are strongly rotating , and even the slowly - rotating ellipticals like m87 may have formed through mergers of rapidly - rotating disks . if so , their nuclear black holes may have grown in the pre - existing disks a rather different environment from the one modelled here . test - particle integrations in two - dimensional barred galaxy models have shown that central mass concentrations can destroy the major families of bar - supporting orbits when @xmath2 exceeds several percent with @xmath139 here defined as the mass of the bulge plus disk ( @xcite ; @xcite ) . this prediction has been verified in self - consistent @xmath0-body simulations of barred galaxies ; for instance , norman , sellwood & hasan ( 1996 ) find that rapid evolution toward axisymmetry takes place when @xmath1 exceeds about @xmath146 in a two - dimensional barred galaxy . however , much of the orbital instability induced by central mass concentrations is associated with motion out of the principal planes ( @xcite ; @xcite ; @xcite ; @xcite ) , and there are indications ( @xcite ; @xcite ) that bars in fully 3d disk galaxy simulations are more fragile than in two dimensions , evolving rapidly in shape when the black hole mass fraction exceeds only @xmath147 . although a great deal of work remains to be done , it is plausible that the critical black hole mass fraction for transition to axisymmetry is close to @xmath148 in models with a fairly wide range of geometries and kinematics . most of the black holes in figure [ fig9 ] fall short of the maximum mass proposed here ; in fact , the mean value of @xmath2 among the roughly ten , secure black hole candidates is only @xmath149 ( @xcite ) . while this fact does not invalidate our identification of @xmath150 as an upper limit , we would nevertheless like to understand why real black holes tend to be smaller . following are a number of possible explanations . some host galaxies may have reached an axisymmetric state through processes not related to the presence of a nuclear black hole . for instance , raha et al . ( 1991 ) showed how coherent bending instabilities can convert a thin bar into an axisymmetric bulge in a few crossing times , and merritt ( 1997b ) argued that a steep central density cusp can induce axisymmetry in much the same way that a central singularity can . a gradual transition to axisymmetry takes place in our simulations even for black hole masses below the critical value ( figure [ fig7 ] ) . a sufficiently slowly - growing black hole would therefore be expected to shut off its supply of fuel before accreting the full @xmath151 of its galaxy s mass . `` slowly growing '' here means that the black hole gains mass on a timescale long compared to the orbital period at the radius which is supplying most of the fuel . however , it might be difficult to reconcile such small growth rates with the short observed lifetimes of quasars . real galaxies are not structurally identical to our @xmath0-body models ; in particular , they sometimes have stellar envelopes in which the density falls off more slowly than the @xmath152 dependence of our models . adding stars at large radii would increase @xmath139 without , presumably , strongly influencing the response of the inner regions to the black hole . galaxy mergers would not change the average ratio of black hole mass to galaxy mass , since black holes from merging galaxies would themselves eventually merge and come to rest in the nucleus ( @xcite ; @xcite ) . but galaxy mergers could greatly affect the average ratio of black hole mass to _ bulge _ mass , since mergers convert gas to stars and disks to spheroids . of course , galaxy - galaxy interactions might also cause black holes to _ grow _ at the expense of their host galaxies by triggering infall of gas into the nucleus ( @xcite ) , and there is some evidence linking quasar activity with galaxy interactions ( @xcite ) . but it seems unlikely that the masses of nuclear black holes have increased significantly since the quasar epoch at @xmath153 ( @xcite ) . hence , the net effect of mergers has probably been to reduce the average ratio of black hole mass to bulge mass , possibly by a large factor . one expected consequence is that the masses of black holes in elliptical galaxies at the current epoch should often fall below the @xmath151 upper limit . this prediction is consistent with the ( meager ) data of figure [ fig9 ] : the black holes in elliptical galaxies ( m87 , m32 , ngc 3377 ) all have @xmath16 , and the largest fractional masses are seen in disk galaxies ( ngc 4342 , 3115 , 4258 ) . however , at least one disk galaxy the milky way is believed to have a black hole mass that falls significantly below the proposed upper limit ( @xcite ) . while triaxial or barlike distortions may be necessary for the efficient growth of black holes , they are probably not sufficient . transport of stars or gas into the nucleus requires a reduction in orbital angular momentum of roughly nine orders of magnitude ; gravitational torques from large - scale bars can only account for one or two of these ( @xcite ) . the gas - dynamical processes that are responsible for removing the remaining angular momentum are poorly understood ( @xcite ) . the fact that most barred spiral galaxies do not exhibit significant nuclear activity suggests that even stellar systems with strongly non - axisymmetric distortions may sometimes fail to channel mass into the nucleus . if a black hole s mass is close to maximal , its host galaxy should be nearly axisymmetric . one can hope to falsify this prediction by looking for evidence of minor - axis rotation or isophote twists in the galaxies with the largest black holes . none of the galaxies in figure [ fig9 ] are known to show such signatures . the most extensively modelled galaxy in figure [ fig9 ] is m32 , whose kinematics can be very well reproduced by axisymmetric models ( @xcite ) . in triaxial galaxies where the black hole mass is substantially below the proposed upper limit , slow evolution of the galaxy s shape should still take place . based on figure [ fig7 ] , a spheroid containing a black hole with @xmath154 should evolve to axisymmetry in @xmath155 half - mass orbital periods , roughly a galaxy lifetime for a typical bright elliptical . for @xmath156 , the evolution time increases to @xmath157 orbital periods , which is still shorter than the lifetimes of many faint elliptical galaxies . for instance , in m32 , the circular orbital period at the half - light radius is only about @xmath158 years ( @xcite ) , or @xmath159 of the age of the universe . faint ellipticals with moderate - to - large black holes , @xmath160 , should therefore have evolved to axisymmetric shapes by now . ( faint ellipticals also tend to have steep central density cusps ( @xcite ) , a factor which would further enhance the importance of orbital stochasticity in these galaxies ( @xcite ) . ) the hubble type distribution of faint ellipticals , @xmath161 , is in fact consistent with axisymmetry , although triaxial shapes can not be ruled out ( @xcite ; @xcite ) . bright ellipticals appear to be at least mildly triaxial as a class ( @xcite ; @xcite ) consistent with their longer dynamical times if we assume that @xmath2 for these galaxies is typically no greater than for the ellipticals in figure [ fig9 ] . nuclear black holes may have acquired most of their mass in environments very different from their current ones , before the epoch of galaxy formation ( @xcite ) . if massive black holes predate galaxies , the mechanism discussed here would not have been effective at limiting their masses . the discovery of nuclear black holes containing much more than @xmath89 the mass of their host galaxies would lend support to the view that at least some black holes acquired most of their mass before galaxies formed . this work was supported by nsf grants ast 93 - 18617 and ast 96 - 17088 and by nasa grant nag 5 - 2803 . we thank j. sellwood , m. valluri , h. zapolsky and the anonymous referee for comments that improved the presentation . aarseth , s. j. 1994 , in galactic dynamics and n - body simulations , eds . g. contopoulos & l. vlahos ( berlin : springer ) , 277 aguilar , l. a. & merritt , d. 1990 , apj , 354 , 33 allen , a. j. , palmer , p. & papaloizou , j. 1992 , mnras , 256 , 695 bahcall et al . 1996 , apj , 479 , 642 barnes , j. e. 1995 , in _ the formation of galaxies _ , eds . c. muoz - tun & f. sanchez , 399 barnes , j. e. & hernquist , l. 1991 , apj , 370 , l65 barnes , j. e. & hernquist , l. 1996 , apj , 471 , 115 begelman , m. , c. , blandford , r. d. & rees , m. j. 1980 , nature , 287 , 307 bender , r. , dbereiner , s. & mllenhoff , c. 1988 , a&a suppl , 74 , 385 binney , j. j. 1980 , mnras , 190 , 873 block , d. l. & stockton , a. 1991 , aj , 102 , 1928 bower , g. a. et al . 1997 , preprint cannizzo , j. k. & hollister , t. c. 1992 , apj , 400 , 58 dahari , o. 1985 , apj suppl . , 57 , 643 dehnen , w. 1995 , mnras , 274 , 919 dubinski , j. 1992 , apj , 401 , 441 dubinski , j. 1994 , apj , 431 , 617 eckart , a. & genzel , r. 1997 , mnras , 284 , 576 fridman , t. & merritt , d. 1997 , aj , 114 , 1479 friedli , d. & benz , w. 1993 , a&a , 268 , 65 friedli , d. 1994 , in `` mass - transfer induced activity in galaxies , '' ed . e. shlosman ( cambridge : cup ) , 268 gebhardt , k. 1996 , aj , 112 , 105 gerhard , o. e. & binney , j. j. 1985 , mnras , 216 , 467 gerhard , o. e. 1987 , in structure and dynamics of elliptical galaxies , ed . t. de zeeuw ( dordrecht : reidel ) , 241 goodman , j. & binney , j. 1984 , mnras , 207 , 511 goodman , j. & schwarzschild , m. 1981 , apj , 245 , 1087 green , p. j. & silverman , b. w. 1994 , nonparametric regression and generalized linear models ( london : chapman & hall ) gunn , j. e. 1979 , in _ active galactic nuclei _ , eds . c. hazard & s. mitton ( cambridge : cup ) , 213 hasan , h. & norman , c. 1990 , apj , 361 , 69 hasan , h. , pfenniger , d. & norman , c. 1993 , apj , 409 , 91 hernquist , l. 1989 , nature , 340 , 687 hernquist , l. 1992 , apj , 400 , 460 hernquist , l. & ostriker , j. 1992 , apj , 386 , 375 hutchings , j. b. & neff , s. g. 1991 , aj , 101 , 434 kandrup , h. e. & mahon , m. e. 1994 , phys . e , 49 , 3735 kaufmann , d. e. & contopoulos , g. 1996 , a&a , 309 , 381 kormendy , j. 1992 , in testing the agn paradigm , eds . s. s. holt , s. g. neff & c. m. urry ( new york : aip ) , 23 kormendy , j. et al . 1996 , apj , 459 , l57 kormendy , j. et al . 1997 , univ . hawaii preprint ifa-97 - 20 kormendy , j. & richstone , d. o. 1995 , ara&a , 33 , 581 lake , g. & norman , c. 1983 , apj , 270 , 51 lichtenberg , a. j. & lieberman , m. a. 1983 , regular and stochastic motion ( berlin : springer ) , 213 lin , d. n. c. , pringle , j. e. & rees , m. j. 1988 , apj , 328 , 103 lynden - bell , d. 1963 , the observatory , 932 , 23 macchetto , f. , marconi , a. , axon , d. j. , capetti , a. , sparks , w. & crane , p. 1997 , space telescope science institute preprint series no . 1163 mackenty , j. w. 1990 , apj suppl . , 72 , 231 may , a. , van albada , t. s. & norman , c. a. 1985 , mnras , 214 , 131 merritt , d. 1997a , in the second stromlo symposium : the nature of elliptical galaxies , a.s.p . 116 , eds . m. arnaboldi , m. , g. s. da costa & p. saha ( provo : asp ) , 32 merritt , d. 1997b , apj , 486 , 102 merritt , d. & fridman , t. 1995 , a. s. p. conf . 86 , fresh views of elliptical galaxies , ed . a. buzzoni , a. renzini & a. serrano ( provo : astronomical society of the pacific ) , 13 merritt , d. & fridman , t. 1996 , apj , 460 , 136 merritt , d. & oh , s .- 1997 , aj , 113 , 1279 merritt , d. & valluri , m. 1996 , apj , 471 , 82 miralda - escud , j. & schwarzschild , m. 1989 , apj , 339 , 752 miyoshi , m. , moran , j. , herrnstein , j. , greenhill , l , nakai , n. , diamond , p. & inoue , m. 1995 , nature , 373 , 127 noguchi , m. 1988 , mnras , 228 , 635 norman , c. , may , a. & van albada , t. 1985 , apj , 296 , 20 norman , c. a. , sellwood , j. a. & hasan , h. 1996 , apj , 462 , 114 norman , c. a. & silk , j. 1983 , apj , 266 , 502 osterbrock , d. e. 1993 , apj , 404 , 551 peebles , p. j. e. 1972 , gen . , 3 , 63 pfenniger , d. 1984 , a&a , 134 , 373 pfenniger , d. & friedli , d. 1991 , a & a , 252 , 75 pfenniger , d. & de zeeuw 1989 , in `` dynamics of dense stellar systems , '' ed . d. merritt ( cambridge : cup ) , 81 quinlan , g. d. 1996 , newa , 1 , 35 quinlan , g. d. , hernquist , l. & sigurdsson , s. 1995 , apj , 440 , 554 raha , n. , sellwood , j. a. , james , r. a. & kahn , f. d. 1991 , nature , 352 , 411 rees , m. 1990 , science , 247 , 817 rees , m. 1992 , in physics of active galactic nuclei , eds . w. j. duschl & s. j. wagner ( berlin : springer ) , 662 ryden , b. s. , lauer , t. r. & postman , m. 1993 , apj , 410 , 515 ryden , b. s. 1996 , apj , 461 , 146 schwarzschild , m. 1981 , in the structure and evolution of normal galaxies , eds . s. m. fall & d. lyden - bell ( cambridge : cup ) , 43 schwarzschild , m. 1993 , apj , 409 , 563 sellwood , j. a. & moore , e. 1998 , in preparation shlosman , i. , begelman , m. c. & frank , j. 1990 , nature , 345 , 679 sigurdsson , s. , hernquist , l. & quinlan , g. d. 1995 , apj , 446 , 75 simkin , s. , su , h. & schwarz , m. p. 1980 , apj , 237 , 404 soltan , a. 1982 , mnras , 200 , 115 statler , t. s. 1995 , in fresh views of elliptical galaxies , a.s.p . a. buzzoni , a. renzini & a. serrano ) ( provo : asp ) , 27 tremblay , b. & merritt , d. 1996 , aj , 111 , 2243 turner , e. l. 1991 , aj , 101 , 5 udry , s. 1993 , a&a , 268 , 35 udry , s. & pfenniger , d. 1988 , a & a , 198 , 135 van den bosch , f. c. 1997 , phd thesis , leiden university van der marel , r. p. , de zeeuw , p. t. , rix , h. w. & quinlan , g. d. 1997 , nature , 385 , 610 voit , g. m. & shull , j. m. 1988 , apj , 331 , 197 wahba , g. 1990 , spline models for observational data ( philadelphia : siam ) young , p. 1980 , apj , 242 , 1232 evolution of the axis ratios of the initial model in the absence of a black hole . solid lines : @xmath81 and @xmath46 ; dashed lines : triaxiality index , @xmath162 . the numbers in the lower left corner of each frame are the fraction of particles , ranked by binding energy , that were used in computing the axis ratios . ( a)-(c ) spherically - averaged properties of the initial model ( thin lines ) and final models ( heavy lines ) in the integrations with @xmath60 , for @xmath163 and @xmath164 . the leftward extent of the curves increases with @xmath1 . ( a ) space density ; reference line has a logarithmic slope of @xmath68 . ( b ) one - dimensional velocity dispersion ; reference line has a logarithmic slope of @xmath165 . ( c ) anisotropy parameter , @xmath166 . ( d ) projected velocity dispersion , as seen from the direction of the intermediate axis ; reference line has a logarithmic slope of @xmath165 . evolution of the axis ratios of the @xmath0-body models for @xmath60 . ( a ) @xmath64 ; ( b ) @xmath73 ; ( c ) @xmath70 . solid lines : @xmath81 and @xmath46 ; dashed lines : triaxiality index , @xmath162 . the numbers in the lower left corner of each frame are the fraction of particles , ranked by binding energy , that were used in computing the axis ratios . contours of the surface mass density as seen from the direction of the intermediate axis . left column : initial model ; right column : final time step ( @xmath85 ) of the model with @xmath73 . contours are separated by 0.3 dex in the top row , and by 0.5 dex in the bottom two rows . the heavy ellipse on the left has an axis ratio of 0.45 ; the ellipse on the right has an axis ratio of 0.6 . evolution of the axis ratios of the @xmath0-body models with @xmath70 . ( a ) @xmath100 ; ( b ) @xmath98 ; ( c ) @xmath96 . solid lines : @xmath81 and @xmath46 ; dashed lines : triaxiality index , @xmath162 . the numbers in the lower left corner of each frame are the fraction of particles , ranked by binding energy , that were used in computing the axis ratios . evolution of the intermediate - to - long axis ratio @xmath81 defined by the most - bound 50% of the stars , for @xmath167 . the growth time of the black hole was @xmath60 for the two smaller values of @xmath1 , and @xmath98 for @xmath70 . test - particle integrations in a triaxial model with a smooth core ( eq . 4 ) and a central point mass containing a fraction @xmath1 of the total mass of the model . the @xmath168-axis is the short axis of the figure . each of the five orbits is defined by its starting point on the equipotential surface , which remained fixed in angular position as @xmath1 was increased . orbits were integrated for approximately 10 full oscillations . black hole masses @xmath1 , in units of the solar mass , and mass fractions @xmath2 for seven galaxies with securely - detected black holes and large black hole mass fractions ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . @xmath139 is defined as the mass of the stellar bulge in the disk galaxies ( filled circles : ngc 4342 , ngc 3115 , ngc 4258 ) , and as the total stellar mass in the elliptical galaxies ( open circles : m32 , m84 , m87 , ngc 3377 ) . the dashed line is the proposed upper limit to @xmath2 .

we present the results of _ spitzer _ irs low - resolution infrared 535 m spectroscopy of 17 nearby ulirgs at 0.2 , optically classified as non - seyferts . the presence of optically elusive , but intrinsically luminous , buried agns is investigated , based on the strengths of polycyclic aromatic hydrocarbon emission and silicate dust absorption features detected in the spectra . the signatures of luminous buried agns , whose intrinsic luminosities range up tol , are found in eight sources . we combine these results with those of our previous research to investigate the energy function of buried agns in a complete sample of optically non - seyfert ulirgs in the local universe at 0.3 ( 85 sources ) . we confirm a trend that we previously discovered : that buried agns are more common in galaxies with higher infrared luminosities . because optical seyferts also show a similar trend , we argue more generally that the energetic importance of agns is intrinsically higher in more luminous galaxies , suggesting that the agn - starburst connections are luminosity - dependent . this may be related to the stronger agn feedback scenario in currently more massive galaxy systems , as a possible origin of the galaxy downsizing phenomenon .

0912.2748

ultraluminous infrared galaxies ( ulirgs ) are characterized by a spectral energy distribution dominated by infrared emission , and an absolute infrared luminosity with l@xmath5 @xmath6 10@xmath3l@xmath4 @xcite . this means that in ulirgs , ( 1 ) very luminous energy sources with l @xmath6 10@xmath3l@xmath4 are present , ( 2 ) the energy sources are hidden by dust , which absorbs most of the primary radiation , and ( 3 ) the heated dust grains then radiate this energy as infrared dust emission . the dust - obscured energy sources are nuclear fusion occurring inside rapidly formed stars ( starbursts ) and/or the release of gravitational energy generated by spatially compact , mass - accreting , supermassive black holes ( smbhs ) ( i.e. , agn activity ) . because the ulirg population becomes very important at @xmath7 1 , in terms of cosmic infrared radiation density @xcite , understanding the hidden energy sources of ulirgs is closely coupled with clarifying the connection between star formation and smbh growth in the dust - obscured galaxy population of the early universe . because distant ulirgs are generally faint , detailed studies of nearby ( @xmath1 0.3 ) ulirgs continue to play an important role in understanding the nature of the ulirg population of the universe . nearby ulirgs have two important observational properties . first , the fraction of optical seyferts ( optically identified agns ) is known to be substantially greater in ulirgs , compared with galaxies with lower infrared luminosities ( l@xmath5 @xmath8 10@xmath3l@xmath4 ) @xcite . second , although galaxies with l@xmath5 @xmath8 10@xmath3l@xmath4 generally exhibit strong , spatially extended infrared emission originating from stars distributed inside them , the infrared dust emission of nearby ulirgs is dominated by a spatially compact feature @xcite , suggesting that a large amount of dust is concentrated in the nuclear regions of ulirgs , and heated by spatially compact energy sources . because the nuclear dust concentration increases in ulirgs @xcite , the putative agns in ulirg s cores are surrounded by a large amount of gas and dust , and ionizing uv radiation from the agn can be blocked at the inner part ( @xmath810 pc ) in virtually all lines - of - sight @xcite . such _ buried _ agns lack well - developed narrow - line - regions ( nlrs ) at 101000 pc scale , the primary sources of forbidden emission lines , and so are elusive through the conventional optical and infrared spectroscopic classification , looking for high - excitation forbidden emission lines from the nlrs @xcite . low - resolution infrared 2.535 @xmath0 m spectroscopy is an effective tool for studying such elusive @xcite buried agns , for the following reasons . first , the effects of dust extinction are sufficiently small ( @xmath80.06a@xmath9 nishiyama et al . 2008 ; 2009 ) that buried agns are detectable . second , in a normal starburst with moderate metallicity ( @xmath60.3 solar ) , consisting of uv - emitting hii regions , molecular gas and dust , and photo - dissociation regions ( pdrs ) , the pahs in the pdrs are excited by far - uv photons from stars without being destroyed , so strong pah emission is usually detected @xcite . however , pahs in close proximity to an agn are destroyed by the high - energy x - ray radiation of the agn @xcite , so that pah emission is virtually absent from a pure agn , as demonstrated from observations @xcite . in a galaxy hosting an agn , strong pah emission can be observed , if pahs survive in the regions shielded from the high - energy radiation of the agn , and pah - exciting far - uv photons from local energy sources ( i.e. , stars ) are available there . finally , in a normal starburst , stellar energy sources and gas / dust are spatially well mixed @xcite , so that the absolute optical depths of dust absorption features in the infrared spectra can not exceed certain thresholds , whereas the optical depths can be arbitrarily large in a buried agn , because the energy source ( a compact mass - accreting smbh ) is more centrally concentrated than dust / gas @xcite . thus , in principle , starbursts and buried agns are distinguishable on the basis of the strengths of pah emission and dust absorption features in the infrared spectra . from the low - resolution infrared 2.535 @xmath0 m spectra of a large number of nearby ulirgs , obtained with _ iso _ , _ spitzer _ and _ akari _ infrared satellites , the detectable buried agn fraction has indeed been found to increase in ulirgs , compared with galaxies with lower infrared luminosities @xcite . however , because the sample is not statistically complete , some bias and ambiguity could remain . observations of a complete sample will clearly lead to a better understanding of the true nature of the nearby ulirg population . in this paper , we present the _ spitzer _ irs low - resolution infrared 535 @xmath0 m spectra of previously unobserved , optically non - seyfert ulirgs at @xmath1 0.3 . by combining these data with our previous results @xcite , we can carry out a systematic investigation of the energetic importance of buried agns in nearby ulirgs , based on the complete sample . throughout this paper , h@xmath10 @xmath11 75 km s@xmath12 mpc@xmath12 , @xmath13 = 0.3 , and @xmath14 = 0.7 are adopted for the sake of consistency with our previously published papers . the ulirgs in the _ iras _ 1 jy sample @xcite are our targets . this sample consists of 118 ulirgs at @xmath1 0.3 , whose _ iras _ 60 @xmath0 m fluxes are larger than 1 jy . of these , 33 are optically classified as seyferts , and 85 as non - seyferts @xcite . the investigation of optically elusive , but intrinsically luminous , buried agns in the 85 optically non - seyfert ulirgs is our main objective . the 85 sources consist of 43 optically liner , 32 optically hii - region , and 10 optically unclassified ulirgs @xcite . @xcite investigated the buried agn fraction in the complete sample of 48 ulirgs at @xmath1 0.15 , classified optically as liners and hii - regions , in the iras 1 jy sample , on the basis of _ spitzer _ low - resolution infrared 535 @xmath0 m spectra . these authors found strong signatures of luminous buried agns in a sizable fraction ( 16/48 ; 33% ) of the observed ulirgs . four optically unclassified ulirgs at @xmath1 0.15 were not studied . @xcite extended the _ spitzer _ irs low - resolution infrared spectroscopic study to ulirgs at @xmath7 0.15 , classified optically as non - seyferts ( liners , hii - regions , and unclassified ) , in the iras 1 jy sample , by analyzing the spectra of 20 sources available at that time . as a result of this extension , a large number of ulirgs with l@xmath15 @xmath16 10@xmath17l@xmath4 are now included , so that a meaningful investigation of the buried agn fraction as a function of galaxy infrared luminosity has become possible , by dividing galaxies into different infrared luminosity classes . specifically , the classes are defined by l@xmath5 @xmath8 10@xmath3l@xmath4 , 10@xmath3l@xmath4 @xmath18 l@xmath5 @xmath8 10@xmath17l@xmath4 , and l@xmath5 @xmath16 10@xmath17l@xmath4 . @xcite found a trend that the detectable buried agn fraction systematically increases with increasing galaxy infrared luminosity . however , the ulirg sample was not statistically complete , because unobserved , optically non - seyfert ulirgs remained in the iras 1 jy sample . among 33 ulirgs at @xmath7 0.15 optically classified as non - seyferts in the iras 1 jy sample , five liner , six hii - region , and two optically unclassified ulirgs were not studied in @xcite . additionally , four optically unclassified ulirgs at @xmath1 0.15 were not investigated by @xcite and @xcite . in total , there were 17 ( 20% of the total 85 sources ) optically non - seyfert ulirgs without published _ spitzer _ irs 535 @xmath0 m low - resolution spectra in the _ iras _ 1 jy sample . because _ spitzer _ irs 535 @xmath0 m low - resolution spectroscopy of one source ( iras 08474 + 1813 ) was scheduled in cycle 3 , we observed the remaining 16 ulirgs on our own , during spitzer cycle 5 . table 1 summarizes the properties of these 17 ulirgs . observations of all 17 ulirgs were performed with the infrared spectrograph ( irs ) @xcite on board the spitzer space telescope @xcite . all four modules , short - low 2 ( sl2 ; 5.27.7 @xmath0 m ) and 1 ( sl1 ; 7.414.5 @xmath0 m ) , and long - low 2 ( ll2 ; 14.021.3 @xmath0 m ) and 1 ( ll1 ; 19.538.0 @xmath0 m ) were used to obtain the full 535 @xmath0 m low - resolution ( r @xmath2 100 ) spectra . table 2 contains details of the observation log . the slit width was 3@xmath196 or two pixels for sl2 ( 1@xmath198 pixel@xmath12 ) and 3@xmath197 or @xmath22 pixels for sl1 ( 1@xmath198 pixel@xmath12 ) . for ll2 and ll1 , the slit widths were 10@xmath195 and 10@xmath197 , respectively , corresponding to @xmath22 pixels for both ll2 ( 5@xmath191 pixel@xmath12 ) and ll1 ( 5@xmath191 pixel@xmath12 ) . the latest pipeline - processed data products available at the time of our analysis were used . frames taken at position a were subtracted from those taken at position b to remove background emission , consisting mostly of zodiacal light . the spectra were then extracted in the standard manner . apertures with 46 pixels were employed for sl and ll data . all sources were dominated by spatially compact emission whose size is similar to the psf of sl and ll . the spectra extracted for the a and b positions were then summed . wavelength and flux calibrations were made on the basis of the _ spitzer _ pipeline - processed data . for the sl1 spectra , the data at @xmath20 @xmath6 14.5 @xmath0 m in the observed frame are invalid ( infrared spectrograph data handbook , version 1.0 ) , and thus they were removed . for the ll1 spectra , the data at @xmath20 @xmath6 35 @xmath0 m were not used , because they are noisy and not necessary for our scientific discussions . for the flux calibration , we did not re - calibrate our spectra using the _ iras _ measurements at 12 @xmath0 m and 25 @xmath0 m , because the _ iras _ 12 @xmath0 m and/or 25 @xmath0 m fluxes are only upper limits in many sources ( table 1 ) . hence , the absolute flux calibration is dependent on the accuracy of the pipeline - processed data , which is taken to be better than 20% for sl and ll ( infrared spectrograph data handbook ) . this level of flux uncertainty does not significantly affect our main conclusions . for ulirgs with _ iras _ 25 @xmath0 m detection , we confirmed that the _ spitzer _ irs 25 @xmath0 m flux agrees with the _ iras _ 25 @xmath0 m data within 30% . for ulirgs with _ iras _ 25 @xmath0 m non - detection , the measured _ spizter _ irs 25 @xmath0 m flux is always smaller than the _ iras _ 25 @xmath0 m upper limits . for all ulirgs , flux discrepancies between sl1 and ll2 were discernible , ranging from 40% to a factor of @xmath22 , and the sl1 flux ( 3@xmath197 wide slit ) was always smaller than the ll2 flux ( 10@xmath195 ) . the discrepancies were generally large in double - nuclei ulirgs with separations of a few to several arcsec ( marked sources in the column 1 of table 2 ) , possibly because the ll2 spectra cover the emission from both nuclei and a diffuse component , while the sl1 spectra only probe the emission from a brighter nucleus . we adjusted the sl1 ( and sl2 ) flux to match the ll2 flux , to minimize missing pah fluxes . appropriate spectral binning with 2 , 4 , or 8 pixels was applied to reduce the scattering of data points particularly at sl2 ( 5.27.7 @xmath0 m ) and ll2 ( 14.021.3 @xmath0 m ) for faint ulirgs , and at @xmath20 @xmath2 10 @xmath0 m for ulirgs that show deep 9.7 @xmath0 m silicate dust absorption features . the infrared 535 @xmath0 m low - resolution spectra of the 17 observed ulirgs are shown in figure 1 . the full 535 @xmath0 m spectra of these ulirgs are published here for the first time . the spectra in figure 1 are useful for observing the overall 535 @xmath0 m spectral shapes and broad 9.7 @xmath0 m and 18 @xmath0 m silicate dust absorption features , but not for observing pah emission . following @xcite and @xcite , figure 2 presents enlarged spectra at @xmath20 = 5.214.5 @xmath0 m to better exhibit the pah emission properties . all ulirgs in figure 2 show clear signs of pah emission at @xmath21 = 6.2 @xmath0 m , 7.7 @xmath0 m , and 11.3 @xmath0 m in the rest frame . to estimate the strengths of these pah emission , we adopted a linear continuum , following @xcite and @xcite . data points at slightly shorter and longer wavelengths than the 6.2 @xmath0 m , 7.7 @xmath0 m , and 11.3 @xmath0 m pah emission were used to determine linear continuum levels , which are shown as solid lines in figure 2 . pah emission features above the adopted continuum levels were fitted with gaussian profiles . the estimated rest - frame equivalent widths ( ew@xmath22 ) and luminosities of the 6.2 @xmath0 m , 7.7 @xmath0 m , and 11.3 @xmath0 m pah emission are summarized in table 3 . as noted by @xcite and @xcite , we estimated the strength of the 7.7 @xmath0 m pah emission in such a way that the uncertainty caused by the strong , broad 9.7 @xmath0 m silicate dust absorption feature is minimized . this definition of the 7.7 @xmath0 m pah emission strength is significantly different from that used in previous papers ( e.g. , genzel et al . we will not use 7.7 @xmath0 m pah emission strengths in our main discussions , because of the difficulty of comparing our data with data in the literature , as well as possible large uncertainties in the 7.7 @xmath0 m pah strengths . to estimate the strengths of the silicate dust absorption features , we use @xmath23 and @xmath24 , defined by @xcite . the value of @xmath23 is the optical depth of the 9.7 @xmath0 m silicate absorption feature , relative to a power - law continuum , determined from data points at @xmath21 = 7.1 @xmath0 m and 14.2 @xmath0 m , to minimize the effects of pah emission . the value of @xmath24 is the optical depth of the 18 @xmath0 m silicate absorption feature , relative to a power - law continuum determined from data points at @xmath21 = 14.2 @xmath0 m and 24 @xmath0 m . these continua are shown as dotted lines in figure 1 . because these continuum levels are determined using data points just outside the 9.7 @xmath0 m and 18 @xmath0 m features , and close to the absorption peaks , the measured optical depths are taken as true dips caused by the silicate dust absorption feature . the @xmath23 and @xmath24 values were estimated from several data points , close to the absorption peaks . the @xmath23 values for all the ulirgs are shown in table 4 ( column 2 ) . the @xmath24 values are also shown in table 4 ( column 3 ) for ulirgs with clearly detected 18 @xmath0 m silicate absorption features . many ulirgs display dips on the shorter wavelength side of the 6.2 @xmath0 m pah emission , which we ascribe to the 6.0 @xmath0 m h@xmath25o ice absorption feature ( bending mode ) . figure 3 presents enlarged spectra at @xmath20 = 5.29 @xmath0 m for ulirgs with clearly detected ice absorption features . the spectrum of iras 14202 + 2615 is also included , as an example of non - detection . the observed optical depths ( @xmath26 ) are summarized in table 5 for detected sources . the detection rate of this 6.0 @xmath0 m ice absorption feature is substantially higher in optically liner ulirgs ( 5/5 ; 100% ) than in hii - region ulirgs ( 1/6 ; 17% ) . for the sake of consistency , we use the criteria employed by @xcite and @xcite to study buried agns in optically non - seyfert ulirgs . we first search for ulirgs with luminous buried agn signatures in infrared 535 @xmath0 m spectra , and then estimate the extinction - corrected intrinsic luminosities of buried agns , based on the observed fluxes at @xmath210 @xmath0 m and dust extinction toward the 10 @xmath0 m continuum emitting regions @xcite . several other methods of evaluating the energetic importance of buried agns , based on _ spitzer _ irs infrared low - resolution spectra , have been proposed in the literature @xcite . in the methods of @xcite , agn and starburst zero - points were derived from _ unobscured _ agns and starburst galaxies , respectively . then , relative buried agn contributions to the _ observed _ fluxes of continuum and line emission at infrared 535 @xmath0 m were derived . no dust extinction correction was applied to estimate buried agn luminosities . even though dust extinction at 535 @xmath0 m is smaller than that at shorter wavelengths , using the _ observed _ fluxes could underestimate the intrinsic buried agn luminosities , because emission from buried agns is more highly flux - attenuated than the surrounding starbursts in individual ulirgs , as well as unobscured agns used to determine the agn zero - points . @xcite applied dust extinction correction to estimate the intrinsic buried agn luminosities , but their modeling utilized only 58 @xmath0 m spectra . our methods use a wider - wavelength range ( 535 @xmath0 m ) of _ spitzer _ irs spectra , and dust extinction is also taken into account to estimate the buried agn luminosities . aside from the strong 9.7 @xmath0 m silicate absorption peak , the flux attenuation of the continuum emission at @xmath21 @xmath6 5 @xmath0 m is small ( @xmath80.5 mag ) for dust extinction with a@xmath9 @xmath8 20 mag @xcite . thus , the observed pah emission luminosities at @xmath21 @xmath6 5 @xmath0 m can be used to roughly estimate the intrinsic luminosities of modestly obscured ( a@xmath9 @xmath8 20 mag ) pah - emitting normal starbursts ( with pdrs and modest metallicity ) , provided that the pah emission and infrared luminosities proportionally correlate in the starbursts @xcite . since the metallicity of ulirgs is estimated to be solar or even higher @xcite , the assumption of pah - emitting starbursts is valid in ulirgs @xcite . table 3 ( columns 8 and 9 ) lists the values of the 6.2 @xmath0 m pah to infrared luminosity ratio , l@xmath27/l@xmath5 , and the 11.3 @xmath0 m pah to infrared luminosity ratio , l@xmath28/l@xmath5 . in normal starburst galaxies with modest dust obscuration ( a@xmath9 @xmath8 20 mag ) , the ratios are estimated to be l@xmath27/l@xmath5 @xmath2 3.4 @xmath29 10@xmath30 @xcite and l@xmath28/l@xmath5 @xmath2 1.4 @xmath29 10@xmath30 @xcite . figure 4 compares the 6.2 @xmath0 m and 11.3 @xmath0 m pah luminosities with the observed infrared luminosities . the observed l@xmath27/l@xmath5 ratios are ( 0.32.4 ) @xmath29 10@xmath30 ( table 3 ) , or 971 % of 3.4 @xmath29 10@xmath30 for modestly obscured starburst galaxies . in the majority of the observed ulirgs , the ratios are @xmath81.7 @xmath29 10@xmath30 , or @xmath850 % of 3.4 @xmath29 10@xmath30 ( see also figure 4 ) . taken at face value , the modestly obscured starbursts detected in these ulirgs account for 971 % ( mostly @xmath850 % ) of their infrared luminosities . the same argument can be applied to the observed l@xmath28/l@xmath5 ratios for the ulirgs . the l@xmath28/l@xmath5 ratios are ( 0.31.8 ) @xmath29 10@xmath30 ( table 3 ) , or 21100 % of 1.4 @xmath29 10@xmath30 for modestly obscured starburst galaxies . on the basis of the l@xmath28/l@xmath5 ratios , there are some ulirgs whose infrared luminosities can be explained by the detected modestly obscured ( a@xmath9 @xmath8 20 mag ) starbursts . however , for ulirgs with @xmath80.7 @xmath29 10@xmath30 ( @xmath850 % of 1.4 @xmath29 10@xmath30 ) , additional energy sources would be required . the 11.3 @xmath0 m pah emission tends to provide higher contributions from modestly obscured starbursts to the infrared luminosities of ulirgs , than the 6.2 @xmath0 m pah emission ( figure 4 ) , possibly because the 11.3 @xmath0 m pah emission originates in larger - sized pahs which may be less susceptible to destruction by the energetic radiation from agns and intense starbursts in ulirgs @xcite . luminosities that are not accounted for by the detected modestly obscured starbursts might originate from ( 1 ) highly obscured ( a@xmath9 @xmath31 20 mag ) starbursts , in which the pah emission flux is substantially attenuated by dust extinction , and/or ( 2 ) buried agns that produce strong infrared radiation , but virtually no pah emission . we note that emission from _ extreme _ starbursts , which consist of hii - regions only , without pdrs and molecular gas , can be pah - free , because pah molecules can be destroyed inside the hii - regions themselves @xcite . if starbursts are exceptionally more centrally concentrated than the surrounding molecular gas and dust ( figure 1e of imanishi et al . 2007 ) , such hii - region - only starbusts could happen . however , in the case of ulirg s cores , such _ extreme _ starbursts require extremely high emission surface brightnesses , and so are unlikely @xcite , if not completely ruled out . we thus use the term `` buried agn signatures '' , rather than `` buried agn evidence '' throughout this manuscript . highly - obscured normal starbursts and buried agns are basically distinguishable , based on the equivalent width of the pah emission . a pah equivalent width ( ew@xmath22 ) must always be large in a normal starburst ( with pdrs ) with a mixed dust / source geometry , regardless of the amount of dust extinction , because both the pah and nearby continuum emission are similarly flux - attenuated . thus , a small ew@xmath22 value suggests a contribution from a pah - free continuum - emitting energy source , namely an agn @xcite . following @xcite and @xcite , we classify ulirgs with ew@xmath27 @xmath8 180 nm and ew@xmath28 @xmath8 200 nm as sources displaying clear signatures of luminous agns , because a considerable contribution from a pah - free continuum would be required for these sources . figure 5 is a plot of the distribution of ew@xmath27 and ew@xmath28 values . table 6 ( columns 23 ) records the detection or non - detection of buried agn signatures in terms of the pah equivalent width threshold . the ew@xmath27 method provides a much larger buried agn fraction ( 7/17 ; 41% ) than the ew@xmath28 method ( 2/17 ; 12% ) , as seen in @xcite and @xcite . an explanation for this is that the 11.3 @xmath0 m pah emission feature is inside the strong 9.7 @xmath0 m silicate dust absorption feature . thus , the buried agn continuum emission at @xmath21 @xmath2 11.3 @xmath0 m is severely attenuated , and has little effect on the equivalent widths of the 11.3 @xmath0 m pah emission from less obscured starburst regions outside the agns @xcite . based on the low ew@xmath22 method , a buried agn with very weak starbursts is easily detectable . if strong starburst activity is present , agn detection becomes difficult , but a _ weakly obscured _ agn is still detectable , because dilution of the pah emission by the agn s pah - free continuum can be significant . however , a _ deeply buried _ agn with strong starbursts is very difficult to detect . even if the intrinsic luminosity of a buried agn is large , the agn flux will be more highly attenuated by dust extinction than the surrounding starburst emission , keeping the observed ew@xmath22 value quite large . to determine the presence of a deeply buried agn with strong starbursts , we use the optical depths of the silicate dust absorption features . specifically , these values can be used to distinguish whether the energy sources are spatially well mixed with dust ( a normal starburst ) , or are more centrally concentrated than the dust ( a buried agn ) ( @xmath321 ) . @xcite obtained a maximum value of @xmath23 @xmath11 1.7 for a normal starburst with a mixed dust / source geometry . considering the possible uncertainties of the @xmath23 estimate , coming from continuum determination ambiguities and statistical errors of individual data points , we classify ulirgs with @xmath23 @xmath16 2 as potential locations of luminous , but deeply buried agns with centrally concentrated energy source geometries . five ulirgs have @xmath23 @xmath16 2 ( tables 4 and 6 ) . as noted by @xcite , although this large @xmath33 method is sensitive to deeply buried agns , it is obviously insensitive to weakly obscured agns , which are better probed with the low ew@xmath22 method . thus , the low ew@xmath22 and large @xmath33 methods are complementary to each other . at the same time , a normal starburst nucleus with a mixed dust / source geometry and a large amount of _ foreground screen dust in an edge - on host galaxy _ ( figure 1d of imanishi et al . 2007 ) , and an exceptionally centrally - concentrated starburst ( figure 1e of imanishi et al . 2007 ) , are capable of producing large @xmath23 values . however , it is unlikely that the majority of the ulirgs with @xmath23 @xmath16 2 fit these non - agn cases @xcite . as explained by @xcite , a buried agn with a centrally concentrated energy source geometry should have a strong dust temperature gradient , because the inner dust ( close to the central energy source ) has a higher temperature than the outer dust , whereas a normal starburst nucleus with a mixed dust / source geometry does not exhibit this behavior . the presence of a strong dust temperature gradient could be detected by comparing the optical depths of the 9.7 @xmath0 m and 18 @xmath0 m silicate dust absorption features , as long as the contaminations from weakly obscured starbursts to observed infrared fluxes are not severe . @xmath24/@xmath23 @xmath8 0.3 could be taken as the signature of a strong dust temperature gradient @xcite . this could provide an additional signature for a buried agn . although this was the case for 12 sources studied by @xcite and @xcite , no such case was found among the 17 newly observed ulirgs ( table 4 , column 4 ) . because of the contamination from weakly obscured starburst activity , as well as other possible ambiguities @xcite , not all buried agns actually satisfy @xmath24/@xmath23 @xmath8 0.3 . the criterion @xmath24/@xmath23 @xmath8 0.3 is a sufficient condition for an additional signature of a buried agn , but it is not a necessary condition . table 6 ( column 5 ) summarizes the strengths of the detected buried agn signatures in the _ spitzer _ irs 535 @xmath0 m spectra , based on the ( 1 ) low ew@xmath22 and ( 2 ) large @xmath23 methods . if buried agn signatures are found by both methods , or by the first method , the ulirgs are classified as _ strong _ buried agn candidates , marked with open circles . eight ulirgs show strong signatures of luminous buried agns ( table 6 ) . the detection rate of luminous buried agns in the newly observed optically non - seyfert ulirgs ( 8/17 ; 47% ) is significantly smaller than the rate obtained by imanishi ( 2009 ) ( 14/20 ; 70% ) . this may be partly due to the smaller fraction of ulirgs with l@xmath5 @xmath16 10@xmath17l@xmath4 ( 6/17 ; 35% ) in this sample , compared with @xcite ( 17/20 ; 85% ) . for ulirgs with very low pah equivalent widths , the observed fluxes are taken to be mostly the result of agn - heated , pah - free dust continuum emission . we can estimate the extinction - corrected intrinsic dust emission luminosity at @xmath210 @xmath0 m ( @xmath34f@xmath35 ) , heated by the agn , on the basis of the observed fluxes at @xmath21 @xmath2 10 @xmath0 m and dust extinction toward the 10 @xmath0 m continuum emitting regions inferred from @xmath23 @xcite . as argued by @xcite and @xcite , in a pure agn with a simple spherical dust distribution , dust emission luminosity is conserved at each temperature from hot inside regions to cool outside regions ( figure 2 of imanishi et al . the extinction - corrected 10 @xmath0 m luminosity ( @xmath34f@xmath35 ) should be comparable to the intrinsic luminosity of the agn s primary radiation . to obtain the estimate , we followed @xcite and @xcite . the flux attenuation of the 8 or 13 @xmath0 m continuum outside the 9.7 @xmath0 m silicate feature is 10@xmath36 @xcite , and ranges from a factor of 1.3 ( iras 14202 + 2615 ; @xmath23 @xmath2 0.7 ) to 3.3 ( iras 04074@xmath372801 ; @xmath23 @xmath2 3.0 ) . we found that the extinction - corrected intrinsic buried agn luminosities for selected ulirgs with very low ew@xmath22 values are a few @xmath29 10@xmath38 ergs s@xmath12 , or @xmath210@xmath3l@xmath4 in the maximum case ( table 7 ) . figure 6 compares the intrinsic buried agn luminosity and the infrared luminosity . the detected buried agns could explain a significant , but not dominant fraction ( 633% ) , of the luminosities of these ulirgs . in table 7 , the luminosities of the detected modestly obscured starbursts , estimated from the l@xmath27 and l@xmath28 values , are also listed for the sake of comparison . our method to estimate the intrinsic buried agn luminosities is very simple and straightforward , with a small amount of free parameters . the largest possible uncertainty is dust extinction curve in ulirgs , which might be different from that established in the galactic inter - stellar medium which we adopted . different dust grain size distribution can produce different dust extinction ratio between widely separate wavelengths , such as optical and 10 @xmath0 m . however , as mentioned by imanishi et al . ( 2007 ; their appendix ) , our estimate is dependent only on the dust extinction ratio between 8 or 13 @xmath0 m continuum and @xmath23 , which should be insensitive to dust size distribution , because of the proximity of the wavelengths . hence , our method is robust to this possible uncertainty . @xcite investigated the presence of luminous buried agns in 48 ulirgs at @xmath1 0.15 , classified optically as non - seyferts ( 28 liner and 20 hii - region ulirgs ) . @xcite did the same in 20 ulirgs at @xmath7 0.15 , classified optically as non - seyferts ( 10 liner , 6 hii - region , 4 unclassified ulirgs ) . in the present research , we have continued this work in 13 optically non - seyfert ulirgs at @xmath7 0.15 ( 5 liner , 6 hii - region , 2 unclassified ulirgs ) and four optically unclassified ulirgs at @xmath1 0.15 . taken together , these 85 observed sources comprise a _ complete _ sample of optically non - seyfert ulirgs in the _ iras _ 1 jy sample . following @xcite , we divide the observed ulirgs into those with 10@xmath3l@xmath4 @xmath18 l@xmath5 @xmath8 10@xmath17l@xmath4 and those with l@xmath5 @xmath16 10@xmath17l@xmath4 . after combining the results of @xcite , @xcite and this paper , we summarize the detectable buried agn fraction in table 8 . using this complete sample , we confirm the previously discovered trend @xcite that the detectable buried agn fraction is significantly higher in ulirgs with l@xmath5 @xmath16 10@xmath17l@xmath4 ( 22/31 ; 71% ) than in those with 10@xmath3l@xmath4 @xmath18 l@xmath5 @xmath8 10@xmath17l@xmath4 ( 15/54 ; 28% ) . figure 7 illustrates this trend by including galaxies with l@xmath5 @xmath8 10@xmath3l@xmath4 . figure 5 displays the distributions of ew@xmath27 , ew@xmath28 and @xmath23 as functions of galaxy infrared luminosity . it is clear that the fraction of galaxies that meet the requirements for a buried agn increases with increasing galaxy infrared luminosity in all plots . in table 8 , we also find that buried agns are more important in liner ulirgs ( 21/43 ; 49% ) than in hii - region ulirgs ( 12/32 ; 38% ) , as has been previously suggested @xcite . however , the difference is now small in this complete sample , and is possibly caused by a higher fraction of ulirgs with l@xmath5 @xmath16 10@xmath17l@xmath4 in the liner sample ( 18/43 ; 42% ) than in the hii - region sample ( 10/32 ; 31% ) ( table 8) . the buried agn fraction is also higher in ulirgs at @xmath7 0.15 than in those at @xmath39 0.15 , but this may be due to the presence of a larger fraction of more luminous ulirgs in a more distant sample . for an agn surrounded by dust with a torus - shaped distribution , the visibility of the central agn is expected to increase with increasing agn luminosity , because the innermost dust sublimation radius of the torus is proportional to the square root of the central agn luminosity , and thus ( assuming a constant torus scale height ) the opening angle of the agn is greater in luminous agns ( the so - called receding torus model : lawrence 1991 ; simpson 2005 ; arshakian 2005 ) . in this scenario , the fraction of optically elusive buried agns is not large in luminous agns . however , this model is applicable _ only if _ the total amount of dust surrounding the central agn does not vary significantly , depending on the galaxy and agn luminosity . ulirgs are driven by mergers of gas - rich galaxies , and gas / dust is quickly transported to the nuclear regions by the gravitational torques , producing a highly - concentrated nuclear gas / dust distribution @xcite . agns can be easily buried by the increasing amount of nuclear gas / dust surrounding the central agns , and thus we believe that a high buried agn fraction is a natural state of affairs in ulirgs . the higher fractions of buried agns ( found in this work ) and optical seyferts @xcite indicate that agns become intrinsically more important with increasing galaxy infrared luminosity . the agn - starburst connections depend on galaxy luminosity_. this is the primary consequence of our results . figure 6 suggests that agn s energetic contributions are generally 1050% , or @xmath230% in median value , in ulirgs with buried agn signatures , which roughly agrees with other independent estimates @xcite in overall ulirg sample . if 30% of the infrared luminosity comes from a buried agn in a ulirg with l@xmath5 = 10@xmath17l@xmath4 (= 2 @xmath29 10@xmath3l@xmath4 ) , then the remaining infrared luminosity , originating in starbursts ( including highly obscured a@xmath9 @xmath31 20 mag ones ) , is 1.4 @xmath29 10@xmath3l@xmath4 . in the meanwhile , buried agn contribution is generally insignificant in galaxies with l@xmath5 = 10@xmath40l@xmath4 (= 2 @xmath29 10@xmath41l@xmath4 ) ( figure 7 ) , and in this case , the starburst - originating luminosity is @xmath22 @xmath29 10@xmath41l@xmath4 . namely , in ulirgs , the _ relative _ energetic importance of buried agn is higher , but the _ absolute _ luminosities of starbursts , and thereby the star - formation rates , are also higher than galaxies with lower infrared luminosities . unless the starburst duration time is drastically different , ulirgs will produce more stars in the future , and evolve into more massive galaxies with larger stellar masses , than less infrared luminous galaxies . given the mass correlation between smbh and spheroidal stellar component @xcite , galaxies with lower infrared luminosities should also contain smbhs whose masses are proportional to those of spheroidal stars . the higher agn contributions in ulirgs suggest that smbhs in ulirgs ( progenitors of massive galaxies ) are actively mass - accreting , and consequently , can have stronger radiation feedbacks to the surrounding gas and dust than mildly mass accreting smbhs in less infrared luminous galaxies . a related phenomenon may be galaxy downsizing , where the more massive galaxies ( with currently larger stellar masses ) completed their major star formation during an earlier cosmic age @xcite . it has been suggested that in these more massive galaxies , agn feedback was stronger in the past , heating or expelling gas and suppressing further star formation over a shorter time scale @xcite . buried agns surrounded by a large amount of gas and dust may have particularly strong feedback , compared to already visible agns with a relatively thin dust covering , and thus are the important population for determining the interplay between agns and host galaxies . figure 8 illustrates our findings that _ buried agns are relatively more important energetically in galaxies that are currently more infrared luminous , and in the future , these galaxies will evolve into more massive galaxies with larger stellar masses . _ the energetic contributions from buried agns become discernible in ulirgs , which are thought to evolve into galaxies with stellar masses of several @xmath29 10@xmath42m@xmath4 , based on infrared velocity dispersion measurements of ulirg s host galaxies @xcite . a similar eventual spheroidal stellar mass is derived for ulirgs with detectable buried agn signatures , from the intrinsic buried agn luminosities @xcite , if we assume that agn luminosities are eddington limits and if the widely accepted mass correlation between smbhs and spheroidal steller components holds @xcite . this mass with several @xmath29 10@xmath42m@xmath4 roughly corresponds to the stellar mass that separates red , massive galaxies with low current star formation rates ( major star formation has already been completed ) and blue , less massive galaxies with ongoing active star formation in the local universe @xcite . thus , our results may offer support to the agn feedback scenario as the origin of the galaxy downsizing phenomenon . we have presented the results of _ spitzer _ irs infrared 535 @xmath0 m low - resolution ( r @xmath2 100 ) spectroscopy of 17 nearby ulirgs at @xmath1 0.2 , optically classified as non - seyferts ( liners , hii - regions , and unclassified , i.e. , no optical agn signatures ) . optically elusive , but intrinsically luminous buried agns were searched for in these optically non - seyfert ulirgs , on the basis of the strengths of pah emission and silicate dust absorption features . we then combined these results with those of our previous studies of nearby ulirgs , using _ irs , to investigate the energetic importance of buried agns in a _ complete sample _ of optically non - seyfert ulirgs in the local universe at @xmath1 0.3 ( 85 sources altogether ) . we arrived at the following primary conclusions . 1 . among the 17 newly observed optically non - seyfert ulirgs , the signatures of important energy contributions from buried agns were found in eight sources . in these sources , the extinction - corrected intrinsic buried agn luminosities were estimated at up to @xmath210@xmath3l@xmath4 , accounting for a significant fraction ( 633% ) of the observed infrared luminosities of these ulirgs . 2 . by combining our new results with those of our previous studies @xcite , we found that buried agns are energetically important in 37 sources of the complete ulirg sample of 85 ( 37/85 = 44% ) , confirming previous suggestion that optically elusive , luminous buried agns are common in the ulirgs of the local universe . we investigated the fraction of detectable luminous buried agns by separating ulirgs with 10@xmath3l@xmath4 @xmath18 l@xmath5 @xmath8 10@xmath17l@xmath4 ( 54 sources ) and l@xmath5 @xmath16 10@xmath17l@xmath4 ( 31 sources ) . we found that luminous buried agns were much more common in the latter ulirgs ( 22/31 = 71% ) than in the former ulirgs ( 15/54 = 28% ) , confirming the previous arguments that buried agns become more energetically important with increasing galaxy infrared luminosity . 4 . given the higher fraction of optical seyferts ( optically identified agns ) in ulirgs with higher infrared luminosities , luminous agns are more common in ulirgs with l@xmath5 @xmath16 10@xmath17l@xmath4 than in ulirgs with 10@xmath3l@xmath4 @xmath18 l@xmath5 @xmath8 10@xmath17l@xmath4 . because the detection rate of both optically identified seyfert agns and optically elusive buried agns is substantially lower in galaxies with lower infrared luminosities ( l@xmath5 @xmath8 10@xmath3l@xmath4 ) , we can conclude that the energetic importance of agns increases with increasing galaxy infrared luminosity , suggesting that _ the agn - starburst connections are luminosity dependent_. this may be related to the widely - proposed agn feedback scenario for the galaxy downsizing phenomenon . this work is based on observations made with the spitzer space telescope , operated by the jet propulsion laboratory at california institute of technology , under a contract with nasa . support for this work was provided by nasa , and also by an award issued by jpl / caltech . we thank the anonymous referee for his / her valuable comments which help significantly improve the clarity of the arguments in this manuscript . is supported by grants - in - aid for scientific research ( 19740109 ) . r.m . acknowledges partial support from inaf and asi , through contract asi - inaf i/016/07/0 . this research made use of the simbad database , operated at cds , strasbourg , france , and the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory at california institute of technology , under a contract with nasa . arshakian , t. g. 2005 , a&a , 436 , 817 armus , l. , et al . 2007 , apj , 656 , 148 bower , r. g. , benson , a. j. , malbon , r. , helly , j. c. , frenk , c. s. , baugh , c. m. , cole , s. , & lacey , c. g. 2006 , mnras , 370 , 645 brandl , b. r. , et al . 2006 , apj , 653 , 1129 bundy , k. , ellis , r. s. , & conselice , c. j. 2005 , apj , 625 , 621 caputi , k. i. , et al . 2007 , apj , 660 , 97 cowie , l. l. , songaila , a. , hu , e. m. , & cohen , j. g. 1996 , aj , 112 , 839 dasyra , k. m. , et al . 2006 , apj , 651 , 835 desai , v. , et al . 2007 , apj , 669 , 810 farrah , d. , et al . 2007 , apj , 667 , 149 ferrarese , l. , & merritt , d. 2000 , apj , 539 , l9 forster schreiber , n. m. , genzel , r. , lutz , d. , kunze , d. , & sternberg , a. 2001 , apj , 552 , 544 genzel , r. et al . 1998 , apj , 498 , 579 goto , t. 2005 , mnras , 360 , 322 granato , g. l. , de zotti , g. , silva , l. , bressan , a. , & danese , l. 2004 , apj , 600 , 580 hopkins , p. f. , hernquist , l. , cox , t. j. , di matteo , t. , martini , p. , robertson , b. , & springel , v. 2005 , apj , 630 , 705 hopkins , p. f. , hernquist , l. , cox , t. j. , di matteo , t. , robertson , b. , & springel , v. 2006 , apjs , 163 , 1 houck , j. r. , et al . 2004 , apjs , 154 , 18 imanishi , m. 2009 , apj , 694 , 751 imanishi , m. , & dudley , c. c. 2000 , apj , 545 , 701 imanishi , m. , & maloney , p. r. 2003 , apj , 588 , 165 imanishi , m. , dudley , c. c. , & maloney , p. r. 2006 , apj , 637 , 114 imanishi , m. , dudley , c. c. , maiolino , r. , maloney , p. r. , nakagawa , t. , & risaliti , g. 2007 , apjs , 171 , 72 imanishi , m. , nakagawa , t. , ohyama , y. , shirahata , m. , wada , t. , onaka , t. , & oi , n. 2008 , pasj , s489 kauffmann , g. , et al . 2003 , mnras , 341 , 54 kewley , l. j. , groves , b. , kauffmann , g. , & heckman , t. 2006 , mnras , 372 , 961 kim , d. -c . , & sanders , d. b. , 1998 , apjs , 119 , 41 kim , d. -c . , veilleux , s. , & sanders , d. b. , 2002 , apjs , 143 , 277 lawrence a. 1991 , mnras , 252 , 586 madden , s. c. , galliano , f. , jones , a. p. , & sauvage , m. , 2006 , a&a , 446 , 877 magorrian , j. , et al . 1998 , apj , 115 , 2285 maiolino , r. et al . 2003 , mnras , 344 , l59 mcleod , k. k. , rieke , g. h. , rieke , m. j. , & kelly , d. m. 1993 , apj , 412 , 111 moorwood , a. f. m. 1986 , a&a , 166 , 4 nardini , e. , risaliti , g. , salvati , m. , sani , e. , watabe , y. , marconi , a. , & maiolino , r. 2009 , mnras , 399 , 1373 nishiyama , s. , nagata , t. , tamura , m. , kandori , r. , hatano , h. , sato , s. , & sugitani , k. 2008 , apj , 680 , 1174 nishiyama , s. , tamura , m. , hatano , h. , kato , d. , tanabe , t. , sugitani , k. , & nagata , t. 2009 , apj , 696 , 1407 ohalloran , b. , satyapal , s. , & dudik , r. p. , 2006 , apj , 641 , 795 peeters , e. , spoon , h. w. w. , & tielens , a. g. g. m. 2004 , apj , 613 , 986 rieke , g. h. , & lebofsky , m. j. 1985 , apj , 288 , 618 puxley , p. j. 1991 , mnras , 249 , 11p roche , p. f. , aitken , d. k. , smith , c. h. , & ward , m. j. , 1991 , mnras , 248 , 606 rupke , d. s. n. , veilleux , s. , & baker , a. j. 2008 , apj , 674 , 172 sanders , d. b. , & mirabel , i. f. 1996 , ara&a , 34 , 749 sanders , d. b. , soifer , b. t. , elias , j. h. , madore , b. f. , matthews , k. , neugebauer , g. , & scoville , n. z. 1988a , apj , 325 , 74 sanders , d. b. , soifer , b. t. , elias , j. h. , neugebauer , g. , & matthews , k. 1988b , apj , 328 , l35 sellgren , k. 1981 , apj , 245 , 138 sijacki , d. , springel , v. , di matteo , t. , & hernquist , l. 2007 , mnras , 380 , 877 simpson , c. 2005 , mnras,360 , 565 smith , j. d. , et al . 2007 , apj , 656 , 770 soifer , b. t. et al . 2000 , aj , 119 , 509 soifer , b. t. et al . 2001 , aj , 122 , 1213 soifer , b. t. , neugebauer , g. , matthews , k. , egami , e. , & weinberger , a. j. 2002 , aj , 124 , 2980 tran , q. d. , et al . 2001 , apj , 552 , 527 veilleux , s. , kim , d. -c . , & sanders , d. b. 1999 , apj , 522 , 113 veilleux , s. , & osterbrock , d. e. 1987 , apjs , 63 , 295 veilleux , s. , et al . 2009 , apjs , 182 , 628 voit , g. m. 1992 , mnras , 258 , 841 werner , m. w. , et al . 2004 , apjs , 154 , 1 wu , y. , charmandaris , v. , hao , l. , brandl , b. r. , bernard - salas , j. , spoon , h. w. w. , & houck , j. r. 2006 , apj , 639 , 157 lcrrrrcrc iras 04074@xmath372801 & 0.153 & @xmath80.07 & 0.07 & 1.33 & 1.72 & 12.2 & 0.05 ( c ) & liner + iras 05020@xmath372941 & 0.154 & @xmath80.06 & 0.10 & 1.93 & 2.06 & 12.3 & 0.05 ( c ) & liner + iras 13106@xmath370922 & 0.174 & @xmath80.12 & @xmath80.06 & 1.24 & 1.89 & 12.3 & @xmath80.05 ( c ) & liner + iras 14121@xmath370126 & 0.151 & 0.06 & 0.11 & 1.39 & 2.07 & 12.3 & 0.08 ( c ) & liner + iras 21477 + 0502 & 0.171 & @xmath80.09 & 0.16 & 1.14 & 1.46 & 12.3 & 0.14 ( c ) & liner + iras 03209@xmath370806 & 0.166 & @xmath80.10 & @xmath80.13 & 1.00 & 1.69 & 12.2 & @xmath80.13 ( c ) & hii - region + iras 10594 + 3818 & 0.158 & @xmath80.09 & @xmath80.15 & 1.29 & 1.89 & 12.2 & @xmath80.12 ( c ) & hii - region + iras 12447 + 3721 & 0.158 & @xmath80.12 & 0.10 & 1.04 & 0.84 & 12.1 & 0.10 ( c ) & hii - region + iras 14202 + 2615 & 0.159 & 0.18 & 0.15 & 1.49 & 1.99 & 12.4 & 0.10 ( c ) & hii - region + iras 15043 + 5754 & 0.151 & @xmath80.12 & 0.07 & 1.02 & 1.50 & 12.1 & 0.07 ( c ) & hii - region + iras 22088@xmath371831 & 0.170 & @xmath80.09 & 0.07 & 1.73 & 1.73 & 12.4 & 0.04 ( c ) & hii - region + iras 02480@xmath373745 & 0.165 & @xmath80.05 & @xmath80.11 & 1.25 & 1.49 & 12.2 & @xmath80.09 ( c ) & unclassified + iras 08591 + 5248 & 0.158 & @xmath80.10 & @xmath80.16 & 1.01 & 1.53 & 12.2 & @xmath80.16 ( c ) & unclassified + iras 02021@xmath372103 & 0.116 & @xmath80.07 & 0.30 & 1.45 & 1.72 & 12.0 & 0.21 ( w ) & unclassified + iras 08474 + 1813 & 0.145 & @xmath80.10 & @xmath80.19 & 1.28 & 1.54 & 12.1 & @xmath80.15 ( c ) & unclassified + iras 14197 + 0813 & 0.131 & @xmath80.17 & @xmath80.19 & 1.10 & 1.66 & 12.0 & @xmath80.18 ( c ) & unclassified + iras 14485@xmath372434 & 0.148 & @xmath80.11 & @xmath80.15 & 1.02 & 1.05 & 12.1 & @xmath80.15 ( c ) & unclassified + lclcccc iras 04074@xmath372801 & 50008 & 2009 mar 8 & 480 & 480 & 240 & 240 + iras 05020@xmath372941 & 50008 & 2008 dec 5 & 240 & 240 & 240 & 240 + iras 13106@xmath370922 & 50008 & 2009 mar 6 & 480 & 480 & 240 & 240 + iras 14121@xmath370126 & 50008 & 2009 mar 5 & 240 & 240 & 240 & 240 + iras 21477 + 0502 & 50008 & 2008 dec 5 & 960 & 960 & 480 & 480 + iras 03209@xmath370806 & 50008 & 2009 mar 8 & 480 & 480 & 240 & 240 + iras 10594 + 3818 & 50008 & 2009 jan 11 , 15 & 960 & 960 & 480 & 480 + iras 12447 + 3721 & 50008 & 2009 jan 24 & 240 & 240 & 240 & 240 + iras 14202 + 2615 & 50008 & 2009 mar 3 , 2009 apr 2 & 480 & 480 & 480 & 480 + iras 15043 + 5754 & 50008 & 2009 feb 26 & 960 & 960 & 480 & 480 + iras 22088@xmath371831 & 50008 & 2008 dec 5 & 960 & 960 & 480 & 480 + iras 02480@xmath373745 & 50008 & 2009 jan 25 & 240 & 240 & 240 & 240 + iras 08591 + 5248 & 50008 & 2008 dec 9 & 240 & 240 & 240 & 240 + iras 02021@xmath372103 & 3187 + 50008 & 2005 jan 15 , 2009 jan 15 & 240 & 240 & 240 & 240 + iras 08474 + 1813 & 30407 & 2007 dec 5 & 168 & 168 & 180 & 180 + iras 14197 + 0813 & 3187 + 50008 & 2005 feb 13 , 2009 mar 3 & 240 & 240 & 240 & 240 + iras 14485@xmath372434 & 50008 & 2009 apr 6 & 240 & 240 & 240 & 240 + lccccccccc iras 04074@xmath372801 & 60 & 420 & 235 & 3.3 & 32.8 & 2.5 & 0.6 & 0.4 + iras 05020@xmath372941 & 130 & 605 & 330 & 7.0 & 47.1 & 4.2 & 0.9 & 0.5 + iras 13106@xmath370922 & 115 & 425 & 565 & 3.3 & 27.6 & 3.5 & 0.4 & 0.4 + iras 14121@xmath370126 & 270 & 635 & 345 & 11.9 & 38.2 & 7.8 & 1.7 & 1.2 + iras 21477 + 0502 & 235 & 575 & 185 & 7.4 & 24.2 & 5.4 & 1.0 & 0.7 + iras 03209@xmath370806 & 285 & 555 & 410 & 9.9 & 26.4 & 9.2 & 1.6 & 1.5 + iras 10594 + 3818 & 350 & 780 & 490 & 16.0 & 46.0 & 11.1 & 2.4 & 1.7 + iras 12447 + 3721 & 210 & 625 & 260 & 7.0 & 24.0 & 4.1 & 1.4 & 0.8 + iras 14202 + 2615 & 160 & 455 & 265 & 18.7 & 52.9 & 13.6 & 1.9 & 1.4 + iras 15043 + 5754 & 285 & 770 & 555 & 8.0 & 29.3 & 6.7 & 1.6 & 1.3 + iras 22088@xmath371831 & 90 & 455 & 195 & 3.0 & 23.6 & 2.3 & 0.3 & 0.3 + iras 02480@xmath373745 & 325 & 915 & 535 & 7.3 & 25.6 & 5.0 & 1.2 & 0.8 + iras 08591 + 5248 & 310 & 695 & 535 & 9.7 & 30.7 & 9.7 & 1.8 & 1.8 + iras 02021@xmath372103 & 285 & 490 & 275 & 5.5 & 13.0 & 4.2 & 1.4 & 1.0 + iras 08474 + 1813 & 170 & 985 & 285 & 2.2 & 13.1 & 1.4 & 0.4 & 0.3 + iras 14197 + 0813 & 305 & 565 & 325 & 3.8 & 11.4 & 3.7 & 0.9 & 0.9 + iras 14485@xmath372434 & 150 & 495 & 295 & 6.8 & 26.2 & 6.8 & 1.6 & 1.6 + lccc iras 04074@xmath372801 & 3.0 & 1.2 & 0.40 + iras 05020@xmath372941 & 2.4 & 0.9 & 0.38 + iras 13106@xmath370922 & 2.0 & 1.2 & 0.60 + iras 14121@xmath370126 & 1.3 & & + iras 21477 + 0502 & 0.8 & & + iras 03209@xmath370806 & 1.0 & & + iras 10594 + 3818 & 1.0 & & + iras 12447 + 3721 & 1.4 & 0.5 & 0.36 + iras 14202 + 2615 & 0.7 & & + iras 15043 + 5754 & 1.4 & 0.7 & 0.50 + iras 22088@xmath371831 & 2.6 & 1.0 & 0.38 + iras 02480@xmath373745 & 1.4 & 0.6 & 0.43 + iras 08591 + 5248 & 1.0 & & + iras 02021@xmath372103 & 1.4 & 0.4 & 0.29 + iras 08474 + 1813 & 2.0 & 1.0 & 0.50 + iras 14197 + 0813 & 0.8 & & + iras 14485@xmath372434 & 1.2 & & + lcc iras 04074@xmath372801 & 0.3 + iras 05020@xmath372941 & 0.4 + iras 13106@xmath370922 & 1.0 + iras 14121@xmath370126 & 0.8 + iras 21477 + 0502 & 0.6 + iras 22088@xmath371831 & 0.5 + iras 02480@xmath373745 & 0.9 + iras 08591 + 5248 & 0.5 + iras 14485@xmath372434 & 0.3 + lcccc iras 04074@xmath372801 & @xmath43 & x & @xmath43 & @xmath43 + iras 05020@xmath372941 & @xmath43 & x & @xmath43 & @xmath43 + iras 13106@xmath370922 & @xmath43 & x & @xmath43 & @xmath43 + iras 14121@xmath370126 & x & x & x & x + iras 21477 + 0502 & x & @xmath43 & x & @xmath43 + iras 03209@xmath370806 & x & x & x & x + iras 10594 + 3818 & x & x & x & x + iras 12447 + 3721 & x & x & x & x + iras 14202 + 2615 & @xmath43 & x & x & @xmath43 + iras 15043 + 5754 & x & x & x & x + iras 22088@xmath371831 & @xmath43 & @xmath43 & @xmath43 & @xmath43 + iras 02480@xmath373745 & x & x & x & x + iras 08591 + 5248 & x & x & x & x + iras 02021@xmath372103 & x & x & x & x + iras 08474 + 1813 & @xmath43 & x & @xmath43 & @xmath43 + iras 14197 + 0813 & x & x & x & x + iras 14485@xmath372434 & @xmath43 & x & x & @xmath43 + lcccc iras 04074@xmath372801 & 2 & 1 & 2 & 6 + iras 05020@xmath372941 & 1.5 & 2 & 3 & 8 + iras 13106@xmath370922 & 1 & 1 & 2.5 & 8 + iras 14202 + 2615 & 1 & 5.5 & 9.5 & 10 + iras 22088@xmath371831 & 1.5 & 1 & 1.5 & 10 + iras 08474 + 1813 & 0.3 & 0.7 & 1 & 5 + iras 14485@xmath372434 & 1 & 2 & 5 & 5 + llcc non - seyfert & total & 85 & 37 ( 44% ) + & l@xmath5 @xmath8 10@xmath17l@xmath4 & 54 & 15 ( 28% ) + & l@xmath5 @xmath16 10@xmath17l@xmath4 & 31 & 22 ( 71% ) + & @xmath39 0.15 & 52 & 18 ( 35% ) + & @xmath7 0.15 & 33 & 19 ( 58% ) + liner & total & 43 & 21 ( 49% ) + & l@xmath5 @xmath8 10@xmath17l@xmath4 & 25 & 9 ( 36% ) + & l@xmath5 @xmath16 10@xmath17l@xmath4 & 18 & 12 ( 67% ) + & @xmath39 0.15 & 28 & 10 ( 36% ) + & @xmath7 0.15 & 15 & 11 ( 73% ) + hii - region & total & 32 & 12 ( 38% ) + & l@xmath5 @xmath8 10@xmath17l@xmath4 & 22 & 4 ( 18% ) + & l@xmath5 @xmath16 10@xmath17l@xmath4 & 10 & 8 ( 80% ) + & @xmath39 0.15 & 20 & 6 ( 30% ) + & @xmath7 0.15 & 12 & 6 ( 50% ) + unclassified & total & 10 & 4 ( 40% ) + & l@xmath5 @xmath8 10@xmath17l@xmath4 & 7 & 2 ( 29% ) + & l@xmath5 @xmath16 10@xmath17l@xmath4 & 3 & 2 ( 67% ) + & @xmath39 0.15 & 4 & 2 ( 50% ) + & @xmath7 0.15 & 6 & 2 ( 33% ) +

there are various situations in which it is natural to ask whether a given collection of functions , , , defined on a set , are the first correlation functions of a point process on . here we describe some necessary and sufficient conditions on the s for this to be true . our primary examples are , , and an arbitrary finite set . in particular , we extend a result by ambartzumian and sukiasian showing realizability at sufficiently small densities . typically if any realizing process exists there will be many ( even an uncountable number ) ; in this case we prove , when is a finite set , the existence of a realizing gibbs measure with body potentials which maximizes the entropy among all realizing measures . we also investigate in detail a simple example in which a uniform density and translation invariant are specified on ; there is a gap between our best upper bound on possible values of and the largest for which realizability can be established . * realizability of point processes * t. kuna , j. l. lebowitz , and e. r. speer

math-ph0612075

a _ point process _ in a set @xmath3 is a random collection of points in @xmath3 , whose distribution is described by a probability measure @xmath12 on the set of all possible point collections . here we will take @xmath3 to be @xmath13 , a lattice such as @xmath14 , the torus @xmath15 , a lattice on the torus , or an open subset of any of these . we always assume that any bounded subset of @xmath3 contains only finitely many points of the collection ( this is of course automatically true if @xmath3 is a lattice ) ; the collection of points is then necessarily countable and we will write it as @xmath16 , with the understanding that the @xmath17 are all distinct . well known examples of point processes are gibbs measures for equilibrium systems of statistical mechanics . the points of the process are then interpreted as the positions of particles ; because the particle configuration is identified with a subset of physical space , the models satisfy an exclusion principle : no two particles can occupy the same position . point processes are also used to model phenomena other than those of statistical mechanics , such as trains of neural spikes @xcite , departure times from queues @xcite , and positions of stars in galaxies @xcite . in many of these cases the quantities of primary interest , partly because they are the ones most accessible to experiment , are the low order correlations , such as the one particle density @xmath18 and the pair density @xmath19 . these may be defined in terms of expectations ( averages ) , with respect to the measure @xmath12 , of products of the ( random ) _ empirical field _ @xmath20 describing the process . for continuum systems , i.e. , when @xmath3 is @xmath13 or an open subset of @xmath13 , @xmath20 is defined by @xmath21 where @xmath22 is the dirac delta function and the @xmath17 s are as above the ( random ) positions of the points of the process . then , with @xmath23 denoting expectation with respect to @xmath12 , @xmath24 and so on ; in general , @xmath25 equation ( [ defrhon ] ) defines @xmath26 as a measure , and we will always assume that this measure assigns finite mass to any bounded set in @xmath27 ( which means that the number of particles in a bounded set in @xmath13 is a random variable with finite moments up to order @xmath28 ) . in many cases this measure is absolutely continuous with respect to lebesgue measure and we identify @xmath29 with its density , i.e. , @xmath30 is the probability of finding a particle in the infinitesimal box @xmath31 at @xmath32 for @xmath33 . we will assume that , whenever possible , @xmath26 is extended by continuity to be defined at points where two of its arguments coincide . the @xmath26 , here and in the lattice case discussed in the next paragraph , are often referred to as the @xmath28-particle distribution functions or correlation functions . when @xmath3 is discrete , a finite set or a lattice , we will also use the notation ( [ eta ] ) and the definitions ( [ defrho1])([defrhon ] ) , interpreting @xmath34 as the kronecker delta @xmath35 , so that @xmath36 has value 0 or 1 , depending on whether the site @xmath37 is empty or occupied . note that if @xmath38 then @xmath29 , @xmath39 , vanishes whenever @xmath40 for some @xmath41 , and that for distinct sites @xmath42 , @xmath29 is the probability of having particles at these sites . in this paper we shall study the following _ realizability problem : _ given functions @xmath18 , @xmath43 , @xmath44 , @xmath45 , nonnegative and symmetric , does there exist a point process for which these are the correlations ? since only a finite number of correlations are prescribed , the problem may be regarded as an infinite dimensional version of the standard truncated moment problem @xcite . the full realizability problem , in which all the correlations @xmath4 , @xmath46 are given , was studied by a. lenard @xcite . realizability , and the related question of fully describing the realizing process , are long standing problems in the classical theory of fluids @xcite , @xcite , and @xcite , recently revived by torquato , stillinger , et al . this problem has been also investigated in the context of the quantum system see e.g. @xcite , @xcite , and @xcite . an important ingredient in that theory is the introduction of various approximation schemes for computing @xmath47 , such as the percus - yevick and hyper - netted chain approximations @xcite . it is then of primary interest to determine whether or not the resulting functions @xmath48 in fact correspond to any point process , that is , are in some sense internally consistent . if so they can provide rigorous bounds for the entropy of the system under consideration . a novel application of the realizability problem to the determination of the maximal density of sphere packing in high dimensions is discussed in @xcite . applications of the problem of describing a point process from its low order correlations occur in other contexts , for example , in the study of neural spikes @xcite . in this and other physical situations it is natural to consider a closely related problem in which the @xmath4 , @xmath49 , are specified only on part of the domain @xmath3 ; for example , on the lattice we might only specify the nearest neighbor correlations . see @xcite for a similar problem in error correcting codes . we will not consider this case further here , except for some comments at the end of section 6 . an important special case is that in which @xmath3 is @xmath13 , @xmath14 , or a periodic version of one of these ( a torus ) , and the point process is translation invariant . the specified correlation functions will then also be translation invariant and may be written in the form @xmath50 as we often work with @xmath51 , we write @xmath52 . we will often state our arguments and results in the translation invariant case , but these may frequently be extended to the more general situation ; when we do not impose translation invariance we will use a notation similar to ( [ g ] ) : @xmath53 we now make some general remarks in order to put the realizability problem in context . first , we observe that if the correlations ( [ rho])([g ] ) can be realized for some density @xmath8 , then they can also be realized , for the same functions @xmath54 , for any @xmath55 with @xmath56 @xcite . to see this , note that if @xmath57 is the empirical field with density @xmath8 , then @xmath58 , where the @xmath59 are independent , identically distributed bernoulli random variables with expectation @xmath60 , is a field with density @xmath55 having the same value for all @xmath61 s . in other words , the new measure is constructed by independently choosing to delete or retain each point in a configuration , keeping a point with probability @xmath60 . ( we will refer to this construction as _ thinning_. ) in this light it is thus natural to pose the realizability problem in the following form : given the @xmath61 , @xmath62 , what is the least upper bound @xmath63 of the densities for which they can be realized ? it is of course possible in the continuum case to have @xmath64 ; for example , if @xmath65 for @xmath66 then for any density @xmath67 a poisson process realizes the correlations . for the lattice systems considered here , on the other hand , we always have @xmath68 . lacking a full answer to this question , one may of course ask rather for upper and lower bounds on @xmath63 . a lower bound @xmath69 may be obtained by the construction of a process at density @xmath70 ; we discuss such constructions in sections [ lowrho][subseclee ] and show that @xmath71 under reasonable restrictions on the @xmath61 in ( [ g ] ) . upper bounds on @xmath63 may be obtained from necessary conditions for realizability , some of which are described in section [ secneccond ] . beyond the question of realizability one may ask about the number or more generally about the types of measures which give rise to a specified set of correlations @xmath4 , @xmath2 . a natural question in the theory of fluids , for example , is whether any of these measures are gibbsian for interactions in a particular class ; for example , given @xmath72 and @xmath9 , is there a gibbs measure realizing these correlations which involves only pair interactions ? this question is considered in section [ gibbs ] , where we discuss the nature of the realizing measure @xmath12 which maximizes the gibbs - shannon entropy . the outline of the rest of this paper is as follows . in section [ secneccond ] , we discuss some necessary conditions for realizability . ( in a separate paper @xcite we will present general necessary and sufficient conditions . ) sections [ lowrho ] , [ triplet ] , and [ subseclee ] cover proofs of realizability : in section [ lowrho ] we prove a theorem , a generalization of one proven by r. v. ambartzumian and h. s. sukiasian @xcite , showing that if @xmath73 is absolutely integrable and @xmath74 satisfies a certain stability condition then the pair @xmath75 is realizable for sufficiently small @xmath8 , with explicitly given higher correlations . in section [ triplet ] we show that the construction can be extended to the case in which the third correlation function @xmath76 is also specified , showing in particular that the realization determined by @xmath75 alone is not unique ; in fact since @xmath76 can take an uncountable number of values there are at low values of @xmath8 an uncountable number of measures realizing @xmath75 . we note also possible extensions to higher order @xmath61 . in section [ subseclee ] we give a variant construction for lattice systems , based on the lee - yang theorem @xcite . in section [ gibbs ] we show that a problem with specified @xmath75 , on a finite set , e.g. , a periodic lattice , may be realized by a gibbs measure with just one- and two - particle potentials whenever @xmath77 . we make some concluding remarks in section [ conclude ] , and in the appendices discuss in some detail an illustrative one - dimensional example and give some technical proofs . clearly , from ( [ defrhon ] ) , realizability requires that @xmath78 we also know that the covariance matrix of the field @xmath36 , @xmath79 must be positive semi - definite , i.e. , for all functions @xmath80 with bounded support , @xmath81 if we take @xmath82 for @xmath83 a bounded subset of @xmath3 then ( [ covform ] ) becomes @xmath84 d{\r_1}d\r_2 \geq 0 ; \label{ftl}\\\ ] ] conversely , if ( [ ftl ] ) holds for all @xmath85 and all @xmath83 , then ( [ covform ] ) holds for all @xmath80 . in the translation invariant case ( [ ftl ] ) is equivalent to the nonnegativity of the infinite volume structure function @xmath86 : @xmath87 d{\r } \geq 0 . \label{ft}\\\ ] ] here we assume @xmath88 ; otherwise ( [ ft ] ) holds in the sense of generalized functions , cf . there are corresponding conditions on the torus @xmath15 , the lattice @xmath14 , and the periodic lattice . if equality holds in ( [ ft ] ) for some @xmath89 then clearly @xmath8 is maximal : @xmath90 . we note also a necessary condition due to yamada @xcite : if @xmath91 denotes the number of of particles in a region @xmath92 , and if @xmath93 is the fractional part of the mean of @xmath91 , so that @xmath94 with @xmath95 and @xmath96 , then the variance @xmath97 of @xmath91 , @xmath98 d{\bf r}_1 d{\bf r}_2 , \label{var } \end{aligned}\ ] ] must satisfy @xmath99 because @xmath91 is an integer : @xmath100 the above necessary conditions all follow from the more general conditions that we prove in @xcite . in summary these say that , given any functions @xmath101 , @xmath102 and constant @xmath103 such that , for any @xmath28 points @xmath104 in @xmath3 , @xmath105 , we must have @xmath106 for all @xmath107 . we prove in fact that in the case @xmath51 , i.e. for the case that only @xmath72 and @xmath9 are given , ( [ all ] ) is also a sufficient condition for realizability under some mild assumptions on the point process . we remark that in the case @xmath51 all restrictions on @xmath8 and @xmath74 beyond those arising from nonnegativity of @xmath8 and of the covariance matrix @xmath108 of ( [ cov ] ) are due to the fact that we want @xmath36 to be a point process , since we can always find a gaussian process realizing any @xmath72 , @xmath9 with @xmath109 @xcite . we also note that for @xmath110 one has @xmath111 hence equality in ( [ ft ] ) implies that the variance @xmath97 is growing slower than the volume . processes with this property are called _ superhomogeneous _ and are of independent interest , see @xcite . as noted above , superhomogeneity can hold , for a given @xmath112 , only at the maximal density @xmath90 . in this section we show the realizability of a given translation invariant @xmath8 and @xmath112 , @xmath113 , for sufficiently small @xmath8 . our arguments extend immediately to the lattice @xmath14 and to the torus , as well as to non - translation invariant @xmath7 , @xmath47 . our results are an extension of those given by r. v. ambartzumian and h. s. sukiasian @xcite , and are based closely on the key idea of that paper : given @xmath8 and @xmath112 satisfying suitable conditions , one proves the existence of a translation invariant process for which the correlation functions @xmath26 , @xmath114 , are given by @xmath115 and which therefore solves the realization problem for @xmath8 and @xmath74 . the ansatz ( [ armansatz ] ) for the dependence on @xmath8 and @xmath74 of the higher order correlations ( which determines the point process uniquely ) is arguably the simplest one possible . it corresponds , for @xmath116 , to the well - known kirkwood superposition approximation in the theory of equilibrium fluids @xcite . here , however , we are not talking of an approximation to a particular given point process but rather of constructing a realizing process or measure @xmath12 whose correlations have the form ( [ armansatz ] ) . to find a point process corresponding to ( [ armansatz ] ) , ambartzumian and sukiasian used the inclusion - exclusion principle which , for any point process , relates the correlation functions @xmath26 to the probability densities @xmath117 for finding exactly @xmath28 particles , with positions @xmath42 , in a region @xmath118 : @xmath119 inserting the ansatz ( [ armansatz ] ) in ( [ inclus ] ) , one finds an expression for the proposed densities : @xmath120 it remains to verify that ( [ armmeasure ] ) in fact defines the probability densities of a point process . first , note that the quantities @xmath121 are well defined by ( [ armmeasure ] ) for any value of @xmath8 whenever there is , for every region @xmath83 , a constant @xmath122 such that ( [ armansatz ] ) satisfies @xmath123 for @xmath42 in @xmath83 . the condition ( [ clam ] ) is easily verified for many @xmath74 ( see , e.g. , theorem [ tharmeniancluster ] below ) . the remaining problem is to prove that the @xmath124 are all nonnegative . if this is done , then in each region @xmath83 the collection @xmath124 , @xmath125 , determines a measure @xmath126 defining a point process ; if @xmath127 then @xmath126 and @xmath128 are compatible and by general arguments ( kolmogorov s projective limit theorem ) there exists an infinite volume realizing measure @xmath12 . ambartzumian and sukiasian considered only the case @xmath110 , @xmath129 . for this case they constructed a cluster expansion of the penrose - ruelle type and obtained inequalities of the groeneveld - lieb - penrose type to show nonnegativity of each term in a reorganized expansion of the @xmath130 . in order to extend their result to @xmath74 s which can be bigger than one , when the cluster expansion is no longer positive term by term , we need to use a different approach . recall the definition of the standard grand canonical partition function , in the region @xmath83 , of a particle system with fugacity @xmath131 , one - particle potential @xmath132 , pair potential @xmath133 , and inverse temperature @xmath134 @xcite : @xmath135\biggr\}\ , d\y_1\cdots d\y_k . \end{aligned}\ ] ] then ( [ armmeasure ] ) takes the form @xmath136 \xi_\lambda(-\rho , v^{(1)},v),\ ] ] with @xmath137 and @xmath138 . note that in ( [ gcpf ] ) the one - particle potential @xmath139 , and hence also the partition function @xmath140 , is different for each @xmath141 , depending explicitly on @xmath28 and on the particle positions @xmath142 . the condition ( [ clam ] ) implies that @xmath140 is an entire function of @xmath131 . suppose now that @xmath143 is analytic in @xmath131 in some domain @xmath144 containing the origin ; then @xmath140 can not vanish in @xmath144 . in particular , if @xmath145 is the largest interval on the real axis which contains the origin and is contained in @xmath144 , then @xmath146 for @xmath147 , since @xmath148 . we will apply this observation by finding a domain @xmath144a disk centered at the origin , of radius @xmath149such that for all @xmath83 , all @xmath28 , and all @xmath42 , @xmath143 is analytic in @xmath144 ; then since @xmath150 , all @xmath151 will be nonnegative for @xmath152 . these considerations lead to our main result . we write @xmath153 [ tharmeniancluster ] let @xmath74 be a non - negative even function on @xmath13 , and suppose that ( i ) @xmath154 and ( ii ) there exists a constant @xmath155 , @xmath156 , such that for all @xmath157 , @xmath158 whenever @xmath159 satisfy @xmath160 . then ( [ clam ] ) is satisfied , and @xmath75 is realizable , for all @xmath8 satisfying @xmath161 for completeness we state the analogous result on the lattice . [ armenian - lat ] let @xmath74 be an even non - negative function on @xmath162 , and suppose that @xmath163 . let @xmath155 be a constant , with @xmath156 , such that @xmath164 whenever @xmath165 satisfy @xmath166 . then @xmath75 is realizable for all @xmath8 satisfying @xmath167 . [ armremark ] ( a ) the fact that ( [ clam ] ) holds under hypothesis ( ii ) of the theorem , with constant @xmath168 independent of @xmath83 , is immediate . in the language of statistical mechanics , this says that the interaction @xmath169 is _ stable_. \(b ) if @xmath170 then hypothesis ( ii ) holds with @xmath171 , and we recover the result of @xcite . \(c ) hypothesis ( ii ) also holds if there exists ( i ) a @xmath172 such that @xmath173 when @xmath174 , and ( ii ) a nonnegative decreasing function @xmath175 on @xmath176 , satisfying @xmath177 , such that @xmath178 @xcite . in the language of statistical mechanics , ( i ) says that @xmath133 has a _ hard core _ ; ( ii ) says that @xmath133 is _ lower regular _ @xcite . in this case one easily obtains an explicit possible value for the constant @xmath155 . \(d ) note that , despite the use of results from equilibrium systems , there is no reason to expect the realizing measure @xmath12 giving rise to the @xmath26 of ( [ armansatz ] ) to be a gibbs measure with pair potential ( unless @xmath179 , in which case @xmath180 and @xmath12 corresponds to a poisson process ) . denote by @xmath181 the @xmath182 correlation function for a grand canonical ensemble in @xmath83 with pair potential @xmath169 , one - particle potential @xmath139 , inverse temperature @xmath134 , and activity @xmath131 ; as for the partition function ( [ partfunct ] ) for this system , these correlation functions depend through @xmath139 on @xmath42 . by the above remarks it suffices to establish analyticity of @xmath183 in a disk @xmath184 with @xmath185 , because from @xmath186 it follows that @xmath143 is also analytic in this disk . to establish the analyticity of @xmath183 we in fact show analyticity of all @xmath181 ; we proceed as in the classical proof , following in particular section 4.2 of @xcite . in this proof the kirkwood - salsburg equations for the correlation functions are written in an appropriately chosen banach space in the form @xmath187 for some operator @xmath188 and fixed vector @xmath175 . one shows that @xmath189 when @xmath190 , so that @xmath191 is then invertible via a power series in @xmath131 , and a unique solution , analytic in @xmath131 , exists . the primary change in the proof required in our case is that one must introduce a dependence of the operator @xmath188 on the sites @xmath42 . we leave the details to appendix [ armproof ] , which is probably best read with @xcite in hand . we next state a generalization of theorem [ tharmeniancluster ] to non - translation invariant systems ; the proof is omitted . let @xmath192 be open , and recall the notation @xmath193 of ( [ ntig ] ) . let @xmath72 and @xmath194 be non - negative functions on @xmath3 and @xmath195 , respectively , with @xmath194 symmetric , and suppose that there exists a constant @xmath155 , with @xmath156 , such that for all @xmath157 , @xmath196 whenever @xmath197 satisfy @xmath198 . then the pair @xmath48 is realizable if @xmath199 we are interested in the decay of the truncated correlation functions @xmath200 for the realizing measure specified by ( [ armansatz ] ) , defined recursively by @xcite @xmath201 where @xmath202 denotes the set of all partitions of @xmath203 in @xmath0 disjoint sets . we consider only the case in which @xmath5 and @xmath72 and @xmath9 are translation invariant ; then @xmath204 and @xmath205 $ ] . for the correlation functions ( [ armansatz ] ) the corresponding truncated correlation functions have the form @xmath206 with @xmath207 the set of all connected subgraphs of the complete graph with vertex set @xmath208 . let @xmath209 denote the set of all undirected trees on @xmath210 . then from ( [ graphs ] ) and an estimate of penrose @xcite , @xmath211 where @xmath155 is defined as in theorem [ tharmeniancluster ] . using @xmath212 we then obtain the @xmath213 decay property @xmath214 compare theorem 4.4.8 of @xcite . one may also establish a pointwise decay bound : if @xmath215 decays polynomially or exponentially , then @xmath216 also decays polynomially or exponentially , respectively , with @xmath217 . for example , if @xmath218 for some @xmath219 and all @xmath220 , then @xmath221 where @xmath222 is the minimal length of a tree connecting all points @xmath42 and the length of a tree @xmath223 is @xmath224 . this decay implies that the realizing measure is mixing and therefore ergodic . we now consider briefly the application of the ideas of section [ lowrho ] to the realization problem under the specification of @xmath225 for @xmath226 . for simplicity we discuss only the case @xmath227 and restrict our considerations to translation invariant correlations in @xmath13 which have densities with respect to lebesgue measure . other cases can be treated analogously . we adopt the notation @xmath228 the first equation here is just ( [ g ] ) , and the second is justified by the fact that , from the definition ( [ defrhon ] ) , @xmath229 can not vanish on a set @xmath108 of positive ( lebesgue ) measure unless @xmath230 vanishes , for almost all @xmath231 , if ( @xmath232 . in analogy with ( [ armansatz ] ) we make now the ansatz @xmath233 for the higher correlation functions ( @xmath234 ) . as before , the probability densities @xmath130 for the point process defined by ( [ armansatztriplet ] ) can be written in terms of the correlations via the inclusion - exclusion principle ( [ inclus ] ) and thus in terms of a grand - canonical partition function @xmath235 for a particle system in @xmath83 with fugacity @xmath131 , one particle potential @xmath139 , non - translation - invariant pair potential @xmath236 , and translation invariant triplet potential @xmath237 : @xmath238 where @xmath239 in order to proceed as in section [ lowrho ] we have to show that there exists a domain for @xmath131 , independent of @xmath42 , in which @xmath240 is analytic . now , however , we must work with the cluster expansion for both the pair and the triplet interactions , as in @xcite . we give a result in which the hypotheses have been chosen to keep the proof simple and are thus far from optimal . we let @xmath241 denote the volume of the sphere in @xmath13 of diameter @xmath242 and write @xmath243 [ probtriplet ] let @xmath74 and @xmath244 be be given , and assume that ( i ) @xmath74 satisfies the conditions of remark [ armremark](c ) , and @xmath154 ; ( ii ) there exists a @xmath245 such that @xmath246 if @xmath247 for some @xmath248 , and @xmath249 . then @xmath250 is realizable whenever @xmath251^{-1},\ ] ] where @xmath155 is defined as in theorem [ tharmeniancluster ] and @xmath252 is a constant such that @xmath253 for all @xmath28 and all @xmath159 with @xmath254 for @xmath255 . . [ tripremark ] ( a ) without loss of generality we may assume that @xmath256 , since otherwise @xmath257 and the proposition reduces to theorem [ tharmeniancluster ] . \(b ) since there can be at most @xmath258 points within a distance @xmath259 of @xmath260 , all separated from @xmath260 and from each other by a distance at least @xmath261 , we may in particular take @xmath262 note that with this choice the upper bound of ( [ eqcondtrip ] ) converges as @xmath263 to the upper bound of ( [ tharmeniancluster req ] ) . the proof is a fairly straightforward extension of the proof of theorem [ tharmeniancluster ] . we sketch some details in appendix [ armproof ] . finally we would like to point out a consequence of proposition [ probtriplet ] which illustrates the fact that the same ( finite ) family of correlation functions may be realized by distinct measures , and in fact by mutually singular measures , where two measures are called _ mutually singular _ if configurations of points typical for one of the realizing point process are atypical for the other one , i.e. , if there exists a set of point configurations @xmath264 such that @xmath264 has probability @xmath242 for one measure and probability @xmath265 for the other . [ nonuni ] let @xmath74 be a function fulfilling the conditions of remark [ armremark](c ) . then for any @xmath8 satisfying the bound ( [ tharmeniancluster req ] ) of theorem [ tharmeniancluster ] with strict inequality there exist uncountably many distinct and in fact mutually singular realizations of @xmath75 by point processes . under the hypotheses one may choose @xmath266 quite arbitrarily , subject only to a condition that @xmath267 be sufficiently small ( how small depends on @xmath259 ) , and still have @xmath8 satisfy the bound ( [ eqcondtrip ] ) . thus there are certainly uncountably many realizations with distinct three point functions . to show that these are mutually singular one first establishes , following the procedure of subsection [ decay ] , the decay of the truncated correlation functions for the measure constructed in proposition [ probtriplet ] . a direct consequence of this decay is that the corresponding realizing point process is mixing and therefore ergodic . since any two translation invariant ergodic measures are either identical or mutually singular , the result follows . while for given @xmath4 we can not in general improve the region of realizability beyond that described in section [ lowrho ] and [ triplet ] , there are special situations in which more can be said . one class of examples is treated in appendix [ examples ] . in this section we consider a lattice gas on a countable set @xmath3 , e.g. , @xmath6 , with @xmath268 for @xmath269 , @xmath270 . this enables us to to establish realizability using techniques developed for proving the lee - yang theorem @xcite . [ theleeyang ] let @xmath3 be a countable set and suppose that @xmath271 for all @xmath272 with @xmath273 and that @xmath274 then @xmath275 is realizable for all @xmath276 the condition ( [ ineqleeyang ] ) improves the result of theorem [ armenian - lat ] , increasing the upper bound on @xmath8 by a factor of @xmath277 , with @xmath278 as in that theorem . note that @xmath279 since @xmath280 . we again use the ansatz ( [ armansatz ] ) for the higher correlation functions . let @xmath83 be a finite subset of @xmath3 ; then ( [ defb ] ) implies that @xmath26 fulfills the bound ( [ clam ] ) with @xmath281 . hence we can write the probability densities @xmath130 in terms of the correlation functions via ( [ inclus ] ) . since the correlation function @xmath282 vanishes when any of its arguments coincide , one can work with the variables @xmath283 and @xmath284 , where @xmath285 and @xmath286 vary over all finite subsets of @xmath83 with @xmath287 . then ( [ inclus ] ) may be written in terms of @xmath286 and @xmath285 as @xmath288 , where @xmath289 ( compare ( [ partfunct])([gcpf ] ) ) . as before the main problem is to verify that @xmath290 . to apply techniques used for the lee - yang theorem we write @xmath291 in terms of the set @xmath292 of empty sites rather than in terms of @xmath286 . writing @xmath293 we obtain @xmath294 clearly the prefactor here is non - negative in general and is positive for @xmath67 . to prove that the sum is non - negative we rewrite it in the form @xmath295 where @xmath296 note that from @xmath268 for @xmath273 it follows that @xmath297 . then proposition 5.1.1 . of @xcite implies that ( [ leeyangform ] ) is not zero if @xmath298 for all @xmath299 . @xmath300 can be bounded below by @xmath301 . we have thus shown that @xmath302 has no zeros for @xmath303 . but @xmath304 for @xmath305 , so that @xmath306 and hence @xmath307 is non - negative for all @xmath308 . in this section we ask whether a specified set of correlation functions @xmath4 , @xmath2 , which can be realized by at least one point process , can also be realized by a gibbs measure involving at most @xmath0-particle potentials . here we will first consider this problem for the case in which our system lives on a finite set @xmath83 , e.g. , a subset of the lattice . on a finite set every measure is gibbsian in a general sense , so the important restriction is to be gibbsian for a set of potentials involving at most @xmath0 particles : we will say that a measure @xmath309 on @xmath310 is _ @xmath0-gibbsian _ if it has the form @xmath311 where @xmath312 , @xmath313 is a normalization constant , and @xmath314 . as in ( [ ntig ] ) we write , for @xmath62 , @xmath315 and again think in terms of specifying the @xmath316 , @xmath317 , and asking for what densities @xmath318 the correlations ( [ gnti ] ) may be realized by a @xmath0-gibbs measure . we will prove that this is possible whenever @xmath318 satisfies @xmath319 for all @xmath320 , with equality allowed if @xmath321 , for some @xmath322 with the property that @xmath322 and the @xmath316 , @xmath62 , are realizable . the proof is presented only for @xmath51 , but the result could easily be extended to general @xmath0 . the key ingredient in the argument is the fact that gibbs measures are those which maximize the gibbs - shannon entropy of the measure @xmath12 , @xmath323 subject to some specified constraints @xcite . in particular , if one can use the method of lagrange multipliers to find a measure which maximizes the entropy , subject to the constraint of a given @xmath72 and @xmath9 , then the maximizing measure will be 2-gibbsian and the lagrange multipliers obtained in this way will be the desired one body and pair potentials @xcite . here we verify that if @xmath319 ( with equality allowed if @xmath321 , as described above ) then the method of lagrange multipliers will indeed apply . [ gibbsmeasure ] suppose that the pair @xmath324 is realizable on @xmath83 . if @xmath72 satisfies @xmath325 for all @xmath320 , with @xmath319 unless @xmath321 , then @xmath326 is realizable by a 2-gibbsian measure @xmath309 for some uniquely determined potentials @xmath327 , @xmath320 and @xmath328 , @xmath329 . moreover , @xmath309 maximizes the gibbs - shannon entropy ( [ gs ] ) over all measures @xmath12 realizing @xmath326 . we begin with a preliminary remark . suppose that we have verified the theorem in the case in which all @xmath318 ( and hence also all @xmath330 ) are strictly positive . then the case in which , say , @xmath331 for @xmath332 , is a direct corollary : we obtain immediately a 2-gibbsian measure on @xmath333 realizing the correlations there , and then take @xmath334 for @xmath335 . similarly , if @xmath336 for some pair of sites @xmath337 , @xmath338 , with @xmath318 and @xmath339 nonzero , then we set @xmath340 , which guarantees that if @xmath309 is given by ( [ kgibbs ] ) then @xmath341 whenever @xmath342 . thus in the remainder of the proof we will assume that @xmath343 for all @xmath320 and prove the existence of a realizing measure @xmath309 , of the form ( [ kgibbs ] ) with _ finite _ potentials and with @xmath51 , on the set of configurations @xmath344 we now turn to the main body of the proof . for any @xmath20 we let @xmath345 be the number of particles in the configuration @xmath20 . we first show that there exists a measure @xmath346 realizing @xmath326 for which @xmath347 whenever @xmath348 and @xmath349 . by hypothesis there exists a measure @xmath350 realizing @xmath324 . we may thin this measure as in section [ intro ] , deleting a particle at the site @xmath351 with probability @xmath352 , independently for each site , to obtain a measure @xmath346 realizing @xmath326 . now we observe that if @xmath353 and @xmath349 then @xmath347 . for example , if @xmath354 with @xmath342 for @xmath270 , then @xmath355 by ( [ etacond ] ) and hence , since @xmath350 realizes @xmath324 , @xmath356 at least for one @xmath357 with @xmath358 , and there is a positive probability that @xmath20 will result from @xmath357 applying the thinning process . next we construct a measure @xmath359 realizing @xmath326 for which @xmath360 for all @xmath353 . we first fix @xmath361 and for @xmath362 define @xmath363 . now the condition that @xmath359 realize @xmath326 is that @xmath364 equations ( [ constraint0])([constraint2 ] ) may be regarded as a system of linear equations for the ( as yet ) undefined @xmath365 , @xmath349 ; note that the number of these unknowns is the same as the number of equations . the coefficient matrix in this system is ( after an appropriate ordering of the @xmath20 , @xmath366 ) upper triangular , with unit diagonal ; thus these equations can be solved uniquely for the @xmath365 , @xmath367 , in terms of these given @xmath365 , @xmath362 . the resulting @xmath359 will differ from @xmath346 by a perturbation of order @xmath368 ; in particular , since @xmath347 for @xmath367 , we can by choice of @xmath368 guarantee that also @xmath360 for @xmath367 . but @xmath360 for @xmath362 by construction , so that @xmath359 has the desired properties . finally we show that the gibbsian measure we seek is the measure which maximizes @xmath369 ( see ( [ gs ] ) ) among all measures realizing @xmath326 . let @xmath309 be such a maximizer ; @xmath309 is unique by the strict concavity of @xmath108 . we first observe that @xmath309 must lie in the interior of @xmath370 , i.e. , that @xmath371 for all @xmath372 ; otherwise define @xmath373 and note that then @xmath374 so that @xmath375 for some @xmath376 and @xmath309 can not maximize @xmath108 . hence @xmath309 may be obtained by the method of lagrange multipliers , with ( [ constraint0])([constraint2 ] ) ( written for @xmath309 rather than @xmath359 ) as constraints . a simple computation shows that the @xmath377 then have the form ( [ kgibbs ] ) with @xmath51 , where the @xmath327 and @xmath328 are the lagrange multipliers associated with ( [ constraint1 ] ) and ( [ constraint2 ] ) , respectively ( the multiplier for ( [ constraint0 ] ) is related to the factor @xmath313 ) . to verify uniqueness of the potentials , note that any 2-gibbsian measure realizing @xmath326 with , say , potentials @xmath378 and @xmath379 , must satisfy the lagrange multiplier equations with these potentials as multipliers and is hence an extremum of the entropy . from the uniqueness of the extremum and the non - degeneracy of the constraint equations ( [ constraint0])([constraint2 ] ) it follows that these multipliers are uniquely defined , i.e. , that @xmath380 and @xmath381 . it is natural to ask if there exists an analogue of theorem [ gibbsmeasure ] for an infinite lattice , such as @xmath14 , for example when the given correlation functions are defined from the beginning on @xmath14 as translation invariant quantities , i.e. , @xmath382 , @xmath383 as in ( [ rho ] ) , ( [ g ] ) ; on @xmath14 we will call a measure 2-gibbsian if it satisfies the dlr equations for an interaction with only one and two body potentials . a result in this direction is due to l. koralov @xcite ; using cluster expansion techniques , he has established the existence of an infinite - volume 2-gibbsian measure in the lattice case for @xmath51 , @xmath8 small , and @xmath74 sufficiently close to @xmath242specifically , for @xmath384 . an attractive alternative approach would be to apply theorem [ gibbsmeasure ] in large boxes @xmath385 to obtain potentials @xmath386 and @xmath387 , @xmath337 , realizing @xmath75 in @xmath83 , and then to show that under suitable restrictions on @xmath8 and @xmath74 the @xmath388 limits of these potentials exist and are summable and translation invariant . we do not know how to carry out such a program , as we have no control on the behavior of the @xmath389 as @xmath83 changes . in fact we do not know if the 2-gibbsian measures @xmath390 realizing @xmath75 in @xmath83 converge as @xmath388 to any measure @xmath309 on @xmath14 . any sequence of such measures must have a ( weakly ) convergent subsequence , however , by compactness , and the limiting measure @xmath309 will realize the translation invariant @xmath75 on @xmath14 . in @xcite we showed that any such realizing measure which is translation invariant and gibbsian , with summable potentials , is necessarily a 2-gibbsian measure with uniquely determined potentials which maximizes the entropy density among all realizing translation invariant measures . it follows from the above that if there is in fact a translation invariant entropy maximizing 2-gibbsian measure @xmath309 on @xmath14 , realizing @xmath75 , with summable translation invariant potentials @xmath378 and @xmath391 ( @xmath378 is just the chemical potential ) , then the conditional measure @xmath392 on @xmath83 , for a specified configuration @xmath393 on @xmath394 , will be the 2-gibbsian measure which maximizes the entropy for @xmath395 , the one and two particle distributions obtained from @xmath396 . if furthermore this measure is unique for these potentials , then clearly @xmath397 for every @xmath393 as @xmath388 . to obtain translation invariant measures we may take for the domain @xmath83 in theorem [ gibbsmeasure ] a periodic lattice @xmath398 , where @xmath399 with periodic boundary conditions , and find a measure @xmath400 realizing @xmath401 for some periodic @xmath402 defined for those @xmath220 satisfying @xmath403 , @xmath404 , and such that @xmath405 as @xmath406 in this case any subsequence limit @xmath309 of @xmath400 will be translation invariant in @xmath14 and realize @xmath407 . conversely , from a translation invariant @xmath12 we can construct @xmath408 by first projecting @xmath12 into a cubical box @xmath83 of side @xmath409 to obtain @xmath410 , then defining , for @xmath20 a configuration in @xmath83 or equivalently @xmath398 , @xmath411 , where @xmath412 is the shift operator on @xmath398 . this yields a periodic measure @xmath413 with density @xmath8 and with @xmath414 for fixed @xmath220 . the real question then is whether any subsequence limit of the @xmath415 will be a 2-gibbs measure with summable pair potentials . to answer this requires some control of the potentials @xmath416 , which we lack at present ( see question 4 at section [ conclude ] ) . there are clearly many natural questions left unanswered by our results . we list some of them for the case of a specified @xmath8 and @xmath112 . \1 ) is there any practical way to bridge the gap between the obvious necessary conditions described in section [ secneccond ] and the sufficiency conditions given in sections [ lowrho ] and [ subseclee ] ? \2 ) when is the measure defined by ( [ armansatz ] ) gibbsian or quasi - gibbsian ? \3 ) can one extend theorem [ gibbsmeasure ] to continuum systems in a finite domain @xmath118 ? we expect this to be true under some reasonable assumptions on @xmath112 , e.g. , the hard core condition that @xmath417 for @xmath418 , @xmath419 , under which there can only be a finite number of particles in @xmath83 . \4 ) what can one say about the existence and nature of an entropy maximizing measure on @xmath14 for @xmath75 ? in particular are there situations for @xmath77 when such a measure is not a translation invariant 2-gibbsian measure ? as pointed out at the end of section [ gibbs ] , if the answer to this question is no then the @xmath309 obtained from the periodic @xmath415 of formula ( [ kgibbs ] ) will be the 2-gibbs entropy maximizing measure on @xmath14 . \5 ) what happens to the realizability problem if one does not specify @xmath112 for all @xmath420 ( or @xmath13 ) , but only for @xmath220 in some finite domain , say @xmath421 ? ( this question was mentioned briefly in section[intro ] . ) as the notation indicates , we are still considering translation invariant correlations ; we may in addition require that the two - point correlations @xmath229 of the realizing measure all be translation invariant and approach @xmath422 as @xmath423 . this is the problem discussed in @xcite for @xmath424 , or just on a ring . it turns out that , at least for the case @xmath424 and specified @xmath425 , one can compute , via a finite number of operations , whether the correlations are realizable . on the @xmath426 dimensional periodic lattice @xmath398 , for general @xmath426 , the entropy maximizing measure @xmath415 in ( [ kgibbs ] ) will now contain only a finite number of terms , so that the transition to an entropy maximizing 2-gibbs measure on @xmath14 with only finite range potentials appears feasible . this will be described in a separate publication . in this appendix we collect some realizability results for certain concretely given @xmath74 . let us first mention a realizability problem in @xmath13 which has been extensively studied by torquato and stillinger @xcite : to determine for which densities @xmath8 there exists a translation invariant point process on @xmath13 with @xmath427 condition ( [ ft ] ) implies that @xmath75 can be only realized if @xmath428 , where @xmath241 is the volume of the ball with diameter @xmath242 in @xmath13 ( @xmath429 , @xmath430 , etc . ) . in the other direction , theorem [ tharmeniancluster ] implies that for general @xmath426 these correlations are indeed realizable if @xmath431 . thus the maximum density @xmath432 for which @xmath74 is realizable satisfies @xmath433 ideas introduced in recent work on the radius of convergence for pure hard - core systems @xcite can be used to give a stronger version of theorem [ tharmeniancluster ] , improving the lower bound in ( [ lowup ] ) beyond @xmath434 , e.g. , for @xmath435 to @xmath436 and for @xmath437 to @xmath438 ( numerically ) . in one dimension we can say more : a simple construction of @xcite shows realizability by a renewal process if @xmath439 ( it is also shown in @xcite that @xmath440 is the maximal density for which a renewal process can realize @xmath74 ) . a more complicated construction @xcite , using hidden markov processes , gives realizability for all @xmath441 , so that @xmath442 . the gap between these upper and lower bounds remains as a challenge to further rigorous analysis . certain simulation results @xcite have suggested that in low dimensions the process may in fact have @xmath443 . however , we can show that this is false in one dimension , i.e. , that @xmath444 ; we discuss this further below . this continuum problem is , in dimension @xmath435 , related to the following lattice problem : for what densities @xmath8 can the second correlation function @xmath445 , where @xmath446 be realized by a point process on @xmath10 ? from the remarks in section [ intro ] we know that for fixed @xmath447 the set of realizable densities @xmath8 is an interval @xmath448 $ ] with @xmath449 . there is a superficial similarity between the continuum problem ( [ gc ] ) and the lattice problem ( [ hnew ] ) for @xmath450 , but there is also a deeper relation . for suppose that @xmath451 is a point process in @xmath452 which realizes ( [ gc ] ) at density @xmath8 . then , for any @xmath453 , define @xmath454}=\int_{k-1}^k\eta_c(x)\,dx$ ] ( that is , @xmath455 is the number of points of the process @xmath451 lying in the interval @xmath456 $ ] ) . then @xmath455 has value 0 or 1 , @xmath457 , and for @xmath458 , @xmath459 thus @xmath20 solves the lattice problem with the same density @xmath8 and with @xmath460 , from which @xmath461 . we will see below that @xmath462 , so that the relation @xmath461 is consistent with the possibility @xmath463 discussed above . we now discuss the lattice problem ( [ hnew ] ) in some detail , as an illustration of the difficulties to face in the general situation . clearly @xmath464 , since for each @xmath465 $ ] the bernoulli or product measure @xmath466 realizes ( [ hnew ] ) ; for @xmath467 there are in fact uncountably many mutually inequivalent realizing measures @xcite , while for @xmath468 the realization is unique . for other values of @xmath447 , upper bounds on @xmath469 are provided by ( [ ft ] ) and ( [ yamada ] ) . in particular , ( [ ft ] ) yields the upper bound @xmath470 , where @xmath471 the yamada condition ( [ yamada ] ) gives an upper bound @xmath472 ; a straightforward but somewhat lengthy computation shows that @xmath473 for @xmath460 , for @xmath474 , @xmath475 , and for @xmath476 . for other values of @xmath447 , @xmath477 , so that certainly the bound @xmath478 is not always sharp . these bounds , together with several lower bounds for @xmath479 obtained below , are plotted in figure 1 ; for values of @xmath447 at which @xmath477 , @xmath480 is determined numerically . [ ! t ] lower bounds on @xmath479 come , essentially , from procedures for explicitly realizing the desired process at some value of @xmath8 . for example , from theorem [ tharmeniancluster ] , for @xmath481 , and theorem [ theleeyang ] , for @xmath476 , we obtain @xmath482 , where @xmath483 comparison with ( [ srhoalpha ] ) shows that , as might be expected , the lower bounds from these general construction methods do not approach the upper bound very closely . to get better bounds or exact values for @xmath469 one must turn to more _ ad hoc _ methods . in this spirit we next describe two families of processes which realize ( [ hnew ] ) and which provide improvements in the lower bounds ( [ genbds ] ) . we studied other constructions , partially improving some of the results below , but these are omitted for conciseness . the first construction , valid for @xmath484 , achieves a density @xmath485 , where @xmath486 thus @xmath487 . comparison of ( [ super ] ) with ( [ srhoalpha ] ) shows that @xmath488 for @xmath460 and @xmath476 , so that @xmath489 for these values . the measures for these processes are superpositions of two measures of period two . to construct them , we first choose with equal probability one of two partitions of @xmath10 , either @xmath490 or @xmath491 , and then assign a configuration to each pair @xmath492 of sites in the partition independently , taking @xmath493 to have value @xmath494 , @xmath495 each with probability @xmath496 , @xmath497 with probability @xmath498 , and @xmath499 with probability @xmath500 . the optimal choices of parameters which lead to ( [ super ] ) are @xmath501 , @xmath502 for @xmath503 and @xmath504 , @xmath505 for @xmath506 . the measures constructed above , as superpositions of period - two measures , do not have good mixing properties ( this defect will be inherited by measures with lower values of @xmath8 obtained via the thinning process described in section [ intro ] ) . for the case @xmath507 this decomposability is inevitable ; the system is then superhomogeneous ( see section [ secneccond ] ; in fact the variance of the number of points on any set of consecutive lattice sites is uniformly bounded by 3/4 ) and the decomposability of any realizing measure then follows from a result of aizenmann , goldstein , and lebowitz @xcite . we now sketch briefly the argument , mentioned in the first paragraph of this appendix , that yields @xmath444 . given a point process @xmath451 in @xmath452 realizing ( [ gc ] ) at density @xmath8 , we define a lattice process @xmath508 , @xmath157 , by @xmath509 . then @xmath508 has density @xmath510 and two point function @xmath511 for @xmath512 this is just the construction given earlier ( see ( [ neq1 ] ) ) , and as there we see that if @xmath513 is the maximum density for which ( [ hn ] ) can be realized then necessarily @xmath514 ; thus @xmath444 follows from @xmath515 . to verify the latter , one first shows that any realizing measure @xmath516 for ( [ hn ] ) with @xmath512 , @xmath517 is supported on configurations such that for some @xmath518 with @xmath519 , each pair of sites @xmath520 , @xmath521 , contains exactly one particle ; this follows from the fact that the random variable @xmath522 ( @xmath521 , @xmath523 ) has minimal variance @xmath524 and hence takes only the values @xmath0 and @xmath525 . ( note that these configurations form the support of the @xmath526 measure constructed above . ) such a @xmath516 may be obtained by projecting a realizing measure @xmath527 for ( [ hn ] ) with @xmath528 , @xmath517 via either @xmath529 or @xmath530 , and it then follows that @xmath527 is supported on configurations with the property that for some @xmath518 with @xmath531 , all sites @xmath532 , @xmath533 , are empty and each triple of sites @xmath534 contains exactly one particle . this leads to a decomposition @xmath535 and from this we derive a contradiction , since the variance with respect to @xmath536 of one of the random variables @xmath537 , @xmath538 , would be negative for large enough @xmath539 . for details see @xcite . our second construction is valid for @xmath540 . for this process we first distribute particles on @xmath541 with a bernoulli measure such that each site is occupied with probability @xmath542 ; then if in this initial configuration a site is occupied we delete the particle occupying its left neighbor , if it exists , with probability @xmath543 . with the optimal choices @xmath544 and @xmath545 we obtain a realization of ( [ hnew ] ) with density @xmath546 so that @xmath547 . for @xmath484 this is of no interest , since @xmath548 , but for @xmath549 it improves on the bounds ( [ genbds ] ) . the case @xmath450 merits special discussion . from ( [ bern ] ) , @xmath550 ; in this case the point process used to obtain @xmath551 is a renewal process and the construction is a lattice version of that given in @xcite to establish that @xmath552 for the continuum problem ( [ gc ] ) . a construction based on a hidden markov process @xcite improves this lower bound to @xmath553 . the upper bound @xmath554 can be improved @xcite to @xmath555 . as in the continuum problem ( [ gc ] ) , it remains a challenge to diminish the rather large gap between these upper and lower bounds . we must show that the functions @xmath556 are analytic for @xmath190 . as indicated in section [ lowrho ] , the proof follows closely the proof of theorem 4.2.3 of @xcite , and we content ourselves with pointing out a few key steps and the necessary changes . let @xmath557 be the banach space of all sequences @xmath558 , where @xmath559 , for which @xmath560 and let @xmath561 be the characteristic function of the set @xmath562 define the operator @xmath188 on @xmath563 by @xmath564 the factor @xmath565 in ( [ kdef ] ) is needed in the estimate ( [ kest ] ) below because the one - body potential @xmath139 in ( [ kdef ] ) depends on @xmath42 , and it is primarily these aspects which distinguish the proof here from that of @xcite . note that , with this factor , @xmath188 depends on @xmath83 , on @xmath74 , and on @xmath42 . then @xmath566 so that @xmath567 now as in @xcite the sequence of correlation functions @xmath181 satisfies the kirkwood - salsburg equation @xmath568 where @xmath569 ; by an optimal choice of @xmath285 we may insure that @xmath570 whenever @xmath190 . for such @xmath131 , @xmath191 is invertible and the equation ( [ ks ] ) has a unique solution in @xmath557 ; one proceeds as in @xcite . as in the proof of theorem [ tharmeniancluster ] we denote by @xmath181 the @xmath182 correlation function for a grand canonical ensemble in @xmath83 with activity @xmath131 , inverse temperature @xmath571 , and potentials ( [ trippots3])([trippots1 ] ) ; we do not write explicitly the dependence of these functions on @xmath42 . the @xmath572 again satisfy the kirkwood - salsburg equation ( [ ks ] ) in the banach space @xmath557 , but with a modified operator @xmath188 . to define @xmath188 we write @xmath573 for @xmath574 , @xmath575 for @xmath576 , and @xmath577 for @xmath578 . then @xmath579 with @xmath580 here @xmath581 extends over the set of graphs which have vertex set @xmath582 , where @xmath583 and @xmath584 ( a graph being identified as a set of edges , i.e. , of unordered pairs of vertices ) , and in which every edge has at least one of its vertices lying in @xmath585 , and every vertex in @xmath585 has at least one edge incident on it . the derivation of ( [ kirkwood][k3def ] ) is similar to that of the usual kirkwood - salsburg equation @xcite . the operator @xmath188 can be bounded as follows . first , the hypotheses of the theorem imply that @xmath586 . next , since the factor @xmath587 in ( [ k3def ] ) vanishes unless @xmath588 and @xmath589 lie inside the ball of radius @xmath259 around @xmath351 , the sum over @xmath20 is nonzero only for those @xmath20 s such that all points @xmath590 are inside this ball . on the other hand , the factor @xmath591 implies that @xmath592 vanishes unless all the @xmath588 are separated by a distance at least @xmath261 , so that we may suppose that there are at most @xmath593 ( see ( [ ndef ] ) ) of these points inside the ball . we thus have the bound @xmath594 here @xmath595 , @xmath596 , @xmath597 is the least integer not smaller than @xmath598 , and we have used the inequality @xmath599 , valid for @xmath600 , which follows from ( [ c3def ] ) by considering separately the cases @xmath601 and @xmath602 . from ( [ ndef ] ) we have that @xmath603 , and from ( [ kirkwood ] ) and ( [ best ] ) it then follows that the norm of @xmath188 in the banach space @xmath557 satisfies @xmath604 an optimal choice of @xmath285 again shows that @xmath570 when @xmath605 , where @xmath606 denotes the right hand side of ( [ eqcondtrip ] ) , completing the proof . * acknowledgments * : we thank e. caglioti , r. fernandez , g. gallavotti , i. kanter , yu . g. kondratiev , p. p. mitra , j. k. percus , f. stillinger , s. torquato , a. van enter , and r. varadhan for valuable comments . we also thank the ihes , and j.l.l . and thank the ias for hospitality during the course of this work . the work of t.k . was supported by the a. v. humboldt foundation . the work of j.l.l . and t.k . was supported by nsf grant dmr-0442066 and afosr grant af - fa9550 - 04 . we also thank dimacs and its supporting agencies . any opinions , findings , conclusions , or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the national science foundation . o. penrose , convergence of fugacity expansions for classical systems , in _ statistical mechanics : foundations and applications ; proceedings of the i.u.p.a.p . meeting , copenhagen , 1966 _ , ed . t. bak . new york , benjamin , 1967 .

we study the full entanglement dynamics of two uniformly accelerated unruh - dewitt detectors with no direct interaction in between but each coupled to a common quantum field and moving back - to - back in the field vacuum . for two detectors initially prepared in a separable state our exact results show that quantum entanglement between the detectors can be created by the quantum field under some specific circumstances , though each detector never enters the other s light cone in this setup . in the weak coupling limit , this entanglement creation can occur only if the initial moment is placed early enough and the proper acceleration of the detectors is not too large or too small compared to the natural frequency of the detectors . once entanglement is created it lasts only a finite duration , and always disappears at late times . prior result by reznik derived using the time - dependent perturbation theory with extended integration domain is shown to be a limiting case of our exact solutions at some specific moment . in the strong coupling and high acceleration regime , vacuum fluctuations experienced by each detector locally always dominate over the cross correlations between the detectors , so entanglement between the detectors will never be generated .

0910.5858

quantum entanglement , the uniquely quantum mechanical feature of a quantum system as schrdinger emphasized @xcite , is the distinguishing resource of quantum information processing ( qip ) . its evolutionary behavior in time is thus of primary importance to the viability and functioning of quantum computing . many qip candidate systems involve qubits interacting with a quantum field , such as two level atoms ( 2la ) in a cavity . because of the intrinsically relativistic nature of a quantum field , in that physical information propagates at a finite speed , issues of causality are unavoidably imbued with qip , even though the qubits in popular schemes are likely to move at nonrelativistic speed or remain stationary . it is of interest to ask whether quantum entanglement can transcend such limitations . this underlies the frequently mentioned yet often - misconjured notion of quantum nonlocality " such as usually associated with the famous epr paper @xcite . following our two recent work discussing the behavior of quantum entanglement in a relativistic setting , one with a detector ( an object with internal degrees of freedom such as a harmonic oscillator or an atom ) in relativistic motion @xcite and another focusing on the relativistic features of a quantum field mediating two inertial detectors @xcite in this paper we study the conditions whereby quantum entanglement between two causally disconnected detectors ( spacelike separated , outside of each other s light cone ) can be created and if so how it evolves in time . by pushing to extreme conditions we can appreciate better this unique feature of quantum mechanics assessed in the more complete setting of relativistic quantum fields . the question in focus here is , whether quantum entanglement between two localized causally disconnected atoms can be created by the vacuum of a common mediating quantum field they both interact with , but not directly with each other . @xcite affirm such a possibility whereas refs . @xcite see no such evidence . these are not contradictory claims because the setups of the problem are not exactly the same . using the time - dependent perturbation theory ( tdpt ) with extended integration domain , reznik discovered that a pair of two - level atoms initially in their ground states will become entangled when they are uniformly accelerated back - to - back in the minkowki vacuum of the field , even though in this setup the two atoms are causally disconnected @xcite . later massar and spindel ( ms ) @xcite considered an exactly solvable model with two unruh - raine - sciama - grove ( u - rsg ) harmonic oscillators @xcite moving in a similar fashion in ( 1 + 1 ) dimensions but both in equilibrium with the minkowski vacuum . they discovered that entanglement can indeed be created , but only in a finite duration of minkowski time after the moment that the distance between the detectors is the minimum . however , contrary to reznik s perturbative results , ms found that such entanglement creation process does not occur in perturbative regime . in our assessment , since ms were looking at the late - time steady - state behavior ( where the detectors are in equilibrium with the unruh heat bath " @xcite ) , their result can not be compared with reznik s tdpt result , which corresponds to early - time behavior of the detectors or atoms . to resolve the difference and understand the discrepancy our present investigation adopts the physical system used in @xcite to a situation similar to that considered in @xcite or @xcite . taking advantage of the existence of exact solutions to the model under study we can perform a thorough analysis and follow the system s evolution through their whole history which enables us to identify the conditions ( in the motion of the two atoms ) , the approximations invoked ( e.g. , time dependent perturbation theory used in @xcite ) and the parameter ranges where quantum entanglement may be generated . this paper is organized as follows . in sec . ii we introduce the model and describe the setup in the problem . in sec . iii we derive explicitly the expression for the cross correlators , and discuss their evolutionary behavior . using these correlators we examine the exact dynamics of quantum entanglement in different conditions in sec . we then give a comparison between the exact entanglement dynamics and the one from the reduced density matrix ( rdm ) of the truncated system . we conclude in sec . v with some discussions . in appendix a we give the expression for some elements of rdm from a first order time - dependent perturbation theory , and discuss some subtleties in the regularization and integration domain . in appendix b we derive explicitly the rdm of the two truncated detectors in an eigen - energy representation . consider two identical , localized but spacelike separated unruh - dewitt detectors , whose internal degrees of freedom @xmath0 and @xmath1 are coupled to a relativistic massless quantum scalar field as described in @xcite , undergoing uniform acceleration in opposite directions as described in @xcite . the action is given by eq.(1 ) in @xcite , but now the trajectories of the detectors are chosen as @xmath2 , and @xmath3 , parametrized by their proper times @xmath4 and proper acceleration @xmath5 . in this setup the detectors never enter into the light cone of each other , so they can not exchange classical information and energy . suppose the initial state of the combined system is a direct product of the minkowski vacuum of the field and a separable state of the detectors in the form of a product of the gaussian states with minimum uncertainty for each free detector , represented by the wigner function @xmath6 where @xmath7 and @xmath8 are the conjugate momenta of @xmath0 and @xmath1 , respectively , and @xmath9 and @xmath10 are the squeeze parameters . then the quantum state of the combined system is always gaussian during the evolution since the action is quadratic . in this case the separability of the two detectors can be well defined by the covariance matrix of the detectors throughout their history . we study the entanglement dynamics by calculating the exact evolution of the two - point functions or correlators , as in @xcite . to compare with the results in @xcite in the time - dependent perturbation theory regime , we will show the reduced density matrix ( rdm ) of the two detectors in an eigen - energy representation , then truncate the energy spectra of the detectors to include only their ground states and first excited states so that they look like two two - level systems . using the explicit expressions for the elements of the rdm of the truncated system we can compare the criteria of separability in the dynamics of the exact and truncated systems . when the detectors are in a gaussian state the rdm obtained from integrating over the quantum field is fully determined by the two - point functions or correlators of the detectors . if we could obtain the time evolution of each correlator of the detectors , we have the full dynamics of this model . by virtue of the factorized initial state for the combined system and the linear interaction , each correlator splits into two parts as @xmath11 . the a - part describes the evolution of the initial zero - point fluctuation in the detector , while the v - part accounts for the response to the vacuum fluctuations of the field . since the quantum field effects such as retardation from one detector will never reach the other in the setup of this paper , no higher order correction from mutual influences is needed . so in this setup the expressions for the self correlators of a single detector in @xcite and @xcite are actually exact , and the v - part of the self correlators there can be directly applied here with @xmath12 , @xmath13 , and @xmath14 . the a - part of the self correlators for @xmath0 and @xmath1 could be different if we take different values of @xmath15 and @xmath16 in the initial state @xmath17 , but the calculation is still straightforward . the remaining task is to calculate the cross correlators between the two detectors . the a - part of the cross correlators always vanishes since the initial state @xmath17 is factorizable for detectors @xmath0 and @xmath1 at the initial moment and no retarded mutual influence between them ever arises after the coupling is switched on . so the cross correlator @xmath18 contains contributions solely from the v - part , or vacuum fluctuations of the field . explicitly , it is obtained by performing the following two - dimensional integration , @xmath19 where @xmath20 is the coupling constant between each detector and the field , @xmath21 , @xmath22 is the natural frequency of each detector , @xmath23 , @xmath24 are the durations of interaction from the initial moments @xmath25 and @xmath26 when the coupling with the field is switched on to the proper times @xmath4 and @xmath27 of detectors @xmath0 and @xmath1 , respectively , and @xmath28 is the positive frequency wightman function of the massless scalar field , eq . @xmath29 . in appendix [ tdptapx ] , we learned that one should take the value of the mathematical cutoff @xmath30 in the expression @xmath29 to be zero in calculating the cross correlators . then the width of the @xmath31 ridge on @xmath32 plane ( see appendix [ dpridge ] ) is infinity in the @xmath33 direction , and @xmath34 can be written in the closed form , @xmath35 f_{k_-}\left(-e^{a(\tau+\tau')}\right ) \right . \nonumber\\ & & + e^{-\gamma\eta'}\left[-{i\gamma\over \omega}\cos\omega\eta ' + i\sin\omega\eta'+ e^{i\omega\eta'}(\omega+i\gamma ) \partial_\omega\right ] f_{k_-}\left(-e^{a(\tau+\tau'_0)}\right ) \nonumber\\ & & + e^{-\gamma\eta}\left[-{i\gamma\over \omega}\cos\omega\eta + i\sin\omega\eta + e^{i\omega\eta}(\omega+i\gamma ) \partial_\omega\right ] f_{k_-}\left(-e^{a(\tau'+\tau_0)}\right ) \nonumber\\ & & + \left . e^{-\gamma(\eta+\eta')}\left [ -e^{i\omega(\eta+\eta ' ) } \left(1+(\omega+i\gamma)\partial_\omega\right)+ \left({i\gamma\over\omega}+1\right)\cos\omega(\eta-\eta')\right ] f_{k_-}\left(-e^{a(\tau_0+\tau'_0)}\right ) \right\ } , \label{exactqx}\end{aligned}\ ] ] where @xmath36 , and @xmath37 . if @xmath25 , @xmath38 and @xmath39 , @xmath40 while @xmath41 , @xmath42 are finite , only the first line of @xmath43 which is a function of @xmath44 will survive . this is the counterpart of the equilibrium result obtained by massar and spindel in the u - rsg model @xcite . other cross correlators are obtained straightforwardly from @xmath45 by proper - time derivatives , for example , @xmath46 , and so on . below we consider the case with @xmath47 and @xmath48 . in this case the time - slicing scheme is equivalent to minkowski times . given @xmath49 as @xmath50 , one can see that @xmath51 always vanishes as @xmath41 , @xmath52 , and @xmath53 . thus the detectors must be separable at late times . but the transient behavior of the cross correlators could be more complicated . in particular , in the regime with @xmath54 and @xmath55 , but @xmath56 , one can see that the cross correlators manifest the following multi - stage behaviors . in the cases with large @xmath57 but still @xmath58 , @xmath59 behaves differently in three stages respectively , as illustrated in fig . [ qaqbdemo](left ) : \(i ) when @xmath60 , the value of @xmath59 is extremely small though exponentially growing ( @xmath61 ) if @xmath62 is not very small . \(ii ) when @xmath63 , the first line of @xmath43 dominates , so @xmath59 oscillates like @xmath64 , \label{t0ggt}\ ] ] where the @xmath65 term dominates at large @xmath4 , when the amplitude of the oscillation grows almost linearly if @xmath66 . the amplitude of the oscillating @xmath59 will reach the maximum at @xmath67 if @xmath68 with the maximum amplitude about @xmath69 otherwise at @xmath70 with the maximum amplitude about @xmath71 \(iii ) after @xmath72 , the second and the third lines of @xmath43 become important and cancel the @xmath65 behavior , so one has @xmath73 , \label{tggt0}\ ] ] which oscillates with slowly decaying amplitude ( @xmath74 if @xmath75 ) . the underlying reason for the above three - stage behavior is similar to the one for explaining the behavior of @xmath76 , which is discussed below @xmath29 . later we will see that this three - stage profile of @xmath77 will be present in the entanglement dynamics in the same regime . two special cases are worthy of mention here . first , if @xmath78 , @xmath79 will never enter stage ( iii ) . it will always behave like @xmath80 at large positive @xmath4 . second , if @xmath81 , @xmath59 has no stages ( i ) and ( ii ) . rather , it starts with stage ( iii ) and @xmath82\cos 2\omega\tau + { 1\over 2}{\rm re}\ , \left [ \psi\left ( -{i\omega\over 2a}\right)- \psi\left({1\over 2}-{i\omega\over 2a}\right)\right ] ( 1-\cos 2\omega\tau)\right\ } , \label{zerot0}\end{aligned}\ ] ] which oscillates in proper time @xmath4 with a very small and slowly decaying amplitude , as shown in fig . [ qaqbdemo](right ) . for the detectors initially in their ground state , with @xmath83 , @xmath84 , @xmath85 , @xmath86 , and starting at @xmath87 ( left ) and @xmath88 ( right ) . in the left plot one can see that when @xmath89 , @xmath59 behaves according to @xmath80 , while after @xmath90 it behaves according to @xmath91 . in the right plot one sees that for @xmath81 , @xmath92 behaves like @xmath93 and its amplitude never grows.,title="fig:",width=302 ] for the detectors initially in their ground state , with @xmath83 , @xmath84 , @xmath85 , @xmath86 , and starting at @xmath87 ( left ) and @xmath88 ( right ) . in the left plot one can see that when @xmath89 , @xmath59 behaves according to @xmath80 , while after @xmath90 it behaves according to @xmath91 . in the right plot one sees that for @xmath81 , @xmath92 behaves like @xmath93 and its amplitude never grows.,title="fig:",width=302 ] note that the behavior of cross correlators is non - markovian because it depends on the fiducial time , namely , the initial moment @xmath25 . so is the entanglement dynamics . the degree of quantum entanglement between the two detectors in gaussian states can be measured by the values of the logarithmic negativity @xmath94 or the quantity @xmath95 defined in @xcite . to demonstrate how the evolution of the cross correlators affect the entanglement dynamics , however , we calculate below @xmath96 defined by @xmath97^{1/2 } \label{sympspec}\ ] ] with the covariance matrix @xmath98 in @xmath99 and @xmath100 @xmath96 ( @xmath101 ) is the smaller ( larger ) eigenvalue in the symplectic spectrum of the partially transposed covariance matrix plus a symplectic matrix @xcite . if @xmath102 , the two detectors are entangled , otherwise separable . from @xmath103 one can easily determine the values for the quantity @xmath104[c_-^2-(\hbar^2/4)]$ ] we used before , and the logarithmic negativity by @xmath105 , where the information in the range @xmath106 is beyond reach . let us first consider the case with both detectors initially in their ground states , namely , @xmath107 , where @xmath108 is the renormalized natural frequency of the detectors . in fig . [ exactdyn001 ] ( left ) we can see that the three - stage profile of @xmath109 in fig . [ qaqbdemo](left ) emerges in the evolution of @xmath96 . a transient entanglement is created as the amplitude of the cross correlators grows , then decreases as the amplitude of the cross correlators decays , and totally disappears at a finite time . the created entanglement could remain in a duration much longer than the natural period of each detector . in the ultraweak coupling limit ( @xmath110 ) the feature of the cross correlators is even clearer in the entanglement dynamics . indeed , in this limit we have @xcite @xmath111 @xmath112 with the constants @xmath113 and @xmath114 corresponding to the time scale of switching on the interaction and the time resolution of the detector , respectively , and @xmath115,\ ] ] while @xmath116 and @xmath117 are @xmath118 and negligible . when @xmath4 is large , one can write the cross correlators as @xmath119 , @xmath120 , and @xmath121 , where @xmath122 \label{chiweak}\ ] ] is the envelop of the oscillating cross correlators . then one has @xmath123 which is less than @xmath124 if the detectors are entangled . now one can easily see how the profile of the cross correlators ( @xmath125 , the analysis is the simplest : when @xmath126 , @xmath127 has been in its late - time constant value and @xmath113 term decays away . then one has @xmath128 which , together with ( @xmath129 ) and @xmath130 , imply that there will be transient entanglement creation outside the light cone after @xmath126 if @xmath131 for the cases with @xmath132|$ ] in the ultraweak coupling limit . that is , to generate entanglement , the proper acceleration of the detectors or the unruh temperature experienced by the detectors can not be too small or too large , otherwise the self correlators will always dominate and the entanglement will never be created in this case . for the cases with the value of @xmath133 satisfying @xmath134 , by substituting @xmath135 into @xmath136 , one can further estimate the moment of entanglement creation @xmath137 and the disentanglement time @xmath138 , where @xmath139 @xmath140 is the @xmath141-th product log or lambert @xmath142 function , which is the inverse function of @xmath143 . for example , for @xmath144 with other parameters the same as those in fig . [ exactdynweak ] , one has @xmath145 and @xmath146 . note that in this setup the moment of entanglement creation @xmath147 is always positive , when the two detectors are moving apart . while the @xmath148 term in @xmath136 is small compared to the @xmath149 term , it can strongly affect the values of @xmath147 and @xmath150 if @xmath151 is very small , _ i.e. _ , @xmath152 is very close to @xmath124 . for the cases with a smaller @xmath153 the situation is similar . different values of @xmath25 would give about the same @xmath147 , if entanglement creation still happens , while the disentanglement time @xmath150 and the minimum value of @xmath96 ( thus the upper limit of @xmath154 for entanglement creation ) can be quite different but of the same order as those with @xmath155 . an example of entanglement creation in ultraweak coupling limit is given in fig . [ exactdynweak ] . when @xmath5 gets smaller than @xmath156 , the above approximation in ultraweak coupling limit fails @xcite . the entanglement dynamics for the cases with both detectors at rest ( @xmath157 ) has been discussed in ref . @xcite , while the separation @xmath158 there should be taken as @xmath159 in this setup . in the strong coupling limit , the self correlators always dominate over the cross correlators , due to the manifestation of the long - range autocorrelation in the field that each detector experienced locally in space . so quantum entanglement is never created if the coupling is sufficiently strong ( see fig . [ dynstrong ] ) . of the detectors initially in their ground states ( @xmath160 ) . ( left ) the parameters in the left plot are the same as those in fig . [ qaqbdemo](left ) ( i.e. , @xmath161 , etc . ) to facilitate direct comparison . one can see that the three - stage profile of @xmath109 emerges in the evolution of @xmath96 , and creates transient entanglement during @xmath162 , which is much longer than the natural period of the detectors . ( right ) the right plot assumes the same conditions as the left plot except that the initial moment is @xmath163 . now @xmath96 is always greater than @xmath164 because the cross correlators have not had sufficient time to become strong enough to create quantum entanglement . this shows clearly that the entanglement creation process is intrinsically non - markovian in nature.,title="fig:",width=302 ] of the detectors initially in their ground states ( @xmath160 ) . ( left ) the parameters in the left plot are the same as those in fig . [ qaqbdemo](left ) ( i.e. , @xmath161 , etc . ) to facilitate direct comparison . one can see that the three - stage profile of @xmath109 emerges in the evolution of @xmath96 , and creates transient entanglement during @xmath162 , which is much longer than the natural period of the detectors . ( right ) the right plot assumes the same conditions as the left plot except that the initial moment is @xmath163 . now @xmath96 is always greater than @xmath164 because the cross correlators have not had sufficient time to become strong enough to create quantum entanglement . this shows clearly that the entanglement creation process is intrinsically non - markovian in nature.,title="fig:",width=302 ] of the detectors initially in their ground state ( @xmath160 ) in the ultraweak coupling limit , with @xmath165 , @xmath166 , @xmath167 , @xmath168 , @xmath169 ( left ) and @xmath170 ( right ) . quantum entanglement is created ( @xmath171 ) after @xmath172 , but disappears @xmath173 at late times.,title="fig:",width=302 ] of the detectors initially in their ground state ( @xmath160 ) in the ultraweak coupling limit , with @xmath165 , @xmath166 , @xmath167 , @xmath168 , @xmath169 ( left ) and @xmath170 ( right ) . quantum entanglement is created ( @xmath171 ) after @xmath172 , but disappears @xmath173 at late times.,title="fig:",width=302 ] of the detectors initially in their ground states in the strong coupling limit , with @xmath174 , @xmath84 , @xmath167 , @xmath168 , and @xmath175 . owing to the strong coupling the self correlators always dominate over the cross correlators , so quantum entanglement is never created.,width=302 ] for the cases with the initial state as a direct product of two squeezed states of free detectors with minimum uncertainty , namely , @xmath176 , the only difference is the a - part of the self correlators . detectors in such cases will still be separable at late times , because all of the a - part of the correlators eventually decays away , and the late - time steady state is independent of the initial state of the detectors . in transient , the a - part of the self correlators with the initial squeezed state oscillates in time about the one with the initial ground state : at some moment the former is greater than the latter , and at another moment , lesser . but the oscillations of @xmath177 and @xmath178 are out of phase , so the overall effect of the a - part of the correlators is to increase the domination of the self correlators and decrease the degree of entanglement ( associated with the value of @xmath95 or @xmath94 ) if the values of @xmath179 are sufficiently far from @xmath180 ( or the squeeze parameters are sufficiently large ; see fig . [ dyn001albe ] ) . from our numerical results we observe , however , that it is still possible to enhance the tendency to entangle if one takes the values of @xmath181 very close to the ones for the initial ground state , while the enhancement is very tiny and the correction to the values of @xmath95 is usually of the next order in @xmath156 . with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] + with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] with each detector initially in a squeezed state with minimum uncertainty . the parameters are the same as those in fig . [ qaqbdemo ] ( left ) except @xmath15 and @xmath16 , which are indicated in the plots . in the dark spots of the above plots , @xmath182 and the detectors are entangled . one can see that the parameters in @xmath181 space far away from the ones for the ground state ( at @xmath183 , in the third plot from the left in the upper row ) tend to increase the value of @xmath95 , namely , decrease the degree of entanglement.,title="fig:",width=151 ] at early times in the perturbative regime the leakage of amplitudes to higher excited states are negligibly small , so we expect the two detectors can be approximately seen as a two 2-level systems ( 2ls ) . for this kind of @xmath184 system the two 2ls are entangled if the partial transpose of the rdm @xmath185 in @xmath186 has a negative eigenvalue , namely if one of the inequalities @xmath187 and @xmath188 is violated . reznik discovered @xcite that the inequality @xmath189 for the rdm of the two 2ls will always be violated in tdpt with integration domain extended to @xmath190 ( denoted as tdpt@xmath191 ; cf . @xmath192-@xmath193 ) , according to which he claimed that quantum entanglement can be generated outside the light cone . in appendix [ rdm2dec ] we derive the truncated rdm for two ud detectors explicitly . comparing this result with the full dynamics , we can make the following observations : \1 ) reznik s conclusion based on tdpt@xmath191 is not generic for all values of the initial moment @xmath25 , the proper acceleration @xmath5 , the natural frequency @xmath22 , or the coupling strength @xmath156 . actually , the key result eq.(18 ) in @xcite is correct only at the moment @xmath194 in the tdpt regime that entanglement creation can occur . \2 ) beyond the tdpt@xmath191 regime , our result shows that once quantum entanglement is created , it only survives in a finite duration . the two detectors are always separable at @xmath195 and at late times . \3 ) the inequality @xmath189 implies that , if the quantum state of the detectors is separable , then from @xmath186 one has @xmath196 ^ 2},\ ] ] which is a product of the quantity @xmath95 and a positive definite function of time . ( here @xmath197 is the covariance matrix for two free detectors in their ground states . ) thus the criterion of separability @xmath189 , though derived from the rdm of the truncated oscillator , is actually equivalent to @xmath198 , which is the sufficient and necessary criterion for gaussian states in the full dynamics @xcite . in other words , the rdm of the two oscillators each with a truncated spectrum , including only up to the first excited state , possesses the complete information pertaining to the separability of the two detectors in gaussian states . we illustrate the time evolution of the exact @xmath199 and @xmath200 from @xmath186 in fig . [ truncdyn001 ] , where the value of @xmath199 is highly dependent on the behavior of the cross correlators ( see eqs . @xmath201-@xmath202 ) : it grows as the amplitude of the cross correlators grows . at the same time the growing cross correlators tend to decrease the value of @xmath200 . combining these two factors we find , in exactly the same duration that the detectors are entangled in the full dynamics , @xmath199 exceeds @xmath200 and violates the inequality @xmath189 . ( dotted curve ) and @xmath200 ( solid curve ) of the truncated rdm under the same condition as in fig . [ exactdyn001 ] ( left ) . the duration the detectors are entangled determined by the condition @xmath203 matches exactly with the result determined by @xmath102.,width=302 ] we have studied the entanglement creation process of two uniformly accelerated ud detectors moving in opposite directions in the minkowski vacuum of a massless scalar field . the two detectors are causally disconnected during the whole history and are far apart at late times . for two detectors initially in their ground states , entanglement creation does occur under some specific conditions : if the initial time is negative and not very close to zero , the coupling strength @xmath156 between each detector and the field is not too large , and the ratio of the proper acceleration @xmath5 to the natural frequency of the detectors @xmath22 is at some moderate value . the moment of entanglement creation , if any , is always at positive @xmath4 . once quantum entanglement is created it can last for a lifetime much longer than the natural period of the detectors in some parameter range , while entanglement always disappears in a finite time . for two detectors each initially in a squeezed state , similar entanglement creation can also occur if the squeeze parameters in the initial state of both detectors are sufficiently small so the initial state is close enough to the direct product of the ground states . moreover , we find that the rdm of two oscillators each with a truncated spectrum up to the first excited state contains the complete information about the separability of the oscillators in gaussian states . in appendices a and b we see that in tdpt@xmath191 regime , where the integration domain has been extended to @xmath190 , the growing rates of the matrix elements of the truncated rdm can be estimated well , but one has to be careful in obtaining the ratio of different matrix elements . while tdpt@xmath191 results are markovian , we find tdpt still keeps some of the non - markovian features . the dependence of entanglement dynamics on the initial state , the fiducial time , and the non - markovian features shown above are beyond the scope of massar and spindel s steady - state calculation @xcite . in addition , there are some other interesting differences between our findings in ( 3 + 1)d ud detector theory and ms s in ( 1 + 1)d u - rsg model . first , in our analysis @xmath113 and @xmath114 corresponding to the ultraviolet cutoffs in ( 3 + 1)d ud detector theory explicitly enter the entanglement dynamics , while the infrared cutoff in ( 1 + 1)d u - rsg model does not affect the degree of entanglement in ms s result because they only considered the cases where their constant @xmath204 , which is a product of the coupling strength and the very long duration of the interaction corresponding to the infrared divergence , is always much greater than all other parameters , such that the higher - order terms in the @xmath204-expansion of the logarithmic negativity can be neglected . second , while in both cases quantum entanglement between the detectors can be created after the moment in minkowski time that the distance between the detectors is a minimum , the lifetimes of the created entanglement in our case are usually much longer than the natural period of the detectors , in contrast to the short lifetimes of entanglement in ms s ( 1 + 1)d results . finally , in our ( 3 + 1)d case entanglement creation is totally suppressed in the strong coupling regime and can manifest in the weak coupling limit , while in ms s result entanglement generation is a non - pertubative phenomena so there is no entanglement creation in their weak coupling regime . entanglement creation between two causally disconnected objects as we have captured in this work may be viewed as _ a manifestation of quantum nonlocality " _ in quantum physics . quantum entanglement between two localized objects can be generated by allowing them to interact with a common quantum field even though they do not exchange any classical information . foremost our results testify to the important fact that _ quantum nonlocality does not violate causality _ notice that quantum entanglement can be recognized and put to use ( as in qip ) only by those spectators " who can access the information from both detectors . from the viewpoint of the separate detectors each can never find out the existence of the other from its own rdm , nor the quantum entanglement between them . there is no transmittal of physical information between them and causality ( of information ) is always respected . this fact also means that the existence of aspectator " is essential when we refer to the dynamics of entanglement between the two localized parties outside of causal contact . the spectator is causally connected to ( and thus can see " ) both parties and can recognize the quantum entanglement between them through its own observation of both . recall that in ref . @xcite we found that entanglement dynamics for two quantum objects in relativistic motion depend on the choice of reference frames or coordinates ( minkowski or rindler , for example ) . one may debate on which coordinate is better " or more objective " for the depiction of entanglement dynamics . our results show that the only physically meaningful one in a given physical setup for describing the entanglement dynamics of the system is that of the spectator . this is of course not unique , but there are well - established ways to relate what is measured by one spectator to another with considerations of time - slicing ( see , e.g. , @xcite ) and transformation laws of reference frames in relativity theory . the setup of our problem can also mimic the situation of two ions of the same charge in a head - on collision . quantum entanglement is generated in the later stage of the collision when the two ions are moving apart , once the trajectories of the ions possess the right kind of symmetry to enhance the field correlations as those in the present problem ( see appendix [ dpridge ] ) . there is no energy exchange between the two ions during the entanglement creation process of this kind . it has been shown that two atoms coupled with a common field vacuum in a cavity can get entangled in a collision " @xcite . these entangling processes could be interpreted as a consequence of virtual quanta exchange in a van der waal potential , analogous to those with the coulomb interaction in electron - electron scattering . in @xcite the interaction time is much longer than the propagation time of photons across the spatial separation between the atoms , and energy exchange between two atoms is clearly present . these cases where entanglement generation occurring well inside the light cone is different from the present setup but closer to the situation we considered in @xcite , where quantum entanglement is established mainly by retarded mutual influences . note that particle ( or atom , molecule ) collisions in non - relativistic classical and quantum mechanics are usually described by an effective hamiltonian with a direct , nonlocal , inter - particle interaction ( e.g. a potential @xmath205 , where @xmath206 are the positions of the particles ) . direct interaction always generates correlation between the particles , classical or quantum . the nonlocality of the interaction , on the other hand , is a consequence of coarse graining from a more fundamental local theory , resulting in an effective theory description . the particles in this non - relativistic regime can not resolve the time scale for information propagating back and forth between them . thus the generation of correlation by this kind of nonlocal direct interaction happens in a time scale during which two particles have causal contact for a long time . one can not tell whether correlations can be created between two causally disconnected particles in this kind of effective theories . moreover , with this kind of nonlocal direct interaction , if one does not further introduce a spatial range of interaction associated with a coarse - graining in time , even classical correlation will be nonlocal and can not be described by any local hidden - variable theory . + * acknowledgment * syl wishes to thank daniel braun for illuminating discussions during his visit to the joint quantum institute , university of maryland and national institute of standards and technology , where this work was first motivated and commenced . blh wishes to thank the hospitality of national center for theoretical sciences and the qis group at national cheng kung university of taiwan . this work is supported partially by grants from lps , nsf grants phy-0426696 , phy-0801368 and darpa - hr0011 - 09 - 1 - 0008 . in tdpt , one expands the wave function of the two detectors in terms of the energy eigenstates @xmath207 and @xmath208 of the free detectors as @xmath209 then calculates the first few factors in the perturbative series of @xmath210 assuming a small coupling constant @xmath20 , namely , @xmath211 . here @xmath212 $ ] . for the initial state @xmath17 , to lowest order in @xmath20 , the elements of the rdm of the detectors are @xmath213 , \label{r1111tdpt}\end{aligned}\ ] ] etc . , where @xmath214 is the positive frequency wightman function of the massless scalar field in minkowski vacuum , given by @xmath215 with @xmath216 and the frequency @xmath217 for the massless scalar field . @xmath30 is a small positive real number serving as a regulator at high frequency in @xmath218-integration . in obtaining @xmath219 in @xmath192 , one could insert the trajectory of the detector a into @xmath220 , which gives the wightman function as @xmath221 with @xmath222 and @xmath223 . conventional wisdom says that the value of @xmath30 is extremely small , so the expression of the wightman function could be replaced by @xcite @xmath224 which is independent of @xmath225 . indeed , when @xmath226 is small , the original @xmath214 and the modified @xmath227 have similar positions of poles and similar behavior around the poles in the complex @xmath33 plane . however , as shown in fig . [ comparedp ] , only the modified @xmath227 will give a linear - growing phase in transient like @xmath228 where the growing rate @xmath229 can be obtained by extending the integration domain of @xmath32 to @xmath190 , namely , from tdpt@xmath191 . in contrast , the original @xmath214 leads to a totally different behavior of @xmath230 : it is linearly decreasing in the beginning and saturates after @xmath231 . learning from what we did in taking the coincidence limit of the self correlators @xcite , we find that , if we put the small shifts @xmath232 and @xmath233 at the upper and lower bounds of the @xmath234 and @xmath235 integration domain , such as @xmath236 and take the @xmath30 in @xmath214 to zero limit , then the exotic behavior of @xmath214 will be altered and we recover the results obtained by using the modified @xmath227 with @xmath237 corresponding to @xmath232 when @xmath238 . so we interpret @xmath30 as the mathematical cutoff , which should be taken the zero limit at some point of calculation , while the @xmath232 , @xmath233 in @xmath239 and the @xmath237 in @xmath240 are physical cutoffs corresponding to the time resolution of the detectors and related to the constants @xmath113 and @xmath114 in the expressions for the self correlators @xcite . the replacement from @xmath241 to @xmath240 actually changes the interpretation of the regulator . in tdpt with the original positive frequency wightman function , @xmath242 in eq . @xmath241 ( dashed curve ) with @xmath243 ( @xmath244 is the euler s constant ) , the @xmath230 with the modified one , @xmath245 in eq . @xmath240 ( solid curve ) with @xmath246 , and the @xmath230 in @xmath239 with the original @xmath242 and @xmath247 , together with the physical cutoffs @xmath248 ( dots ) . the left and the right plots are the same results but shown in different scales to stress that the early - time behavior of the dotted curve really matches with the climbing solid curve . here @xmath165 and @xmath166 . the slope of the dotted curve in its linear growing phase agrees quite well with @xmath249 in @xmath250.,title="fig:",width=302 ] in tdpt with the original positive frequency wightman function , @xmath242 in eq . @xmath241 ( dashed curve ) with @xmath243 ( @xmath244 is the euler s constant ) , the @xmath230 with the modified one , @xmath245 in eq . @xmath240 ( solid curve ) with @xmath246 , and the @xmath230 in @xmath239 with the original @xmath242 and @xmath247 , together with the physical cutoffs @xmath248 ( dots ) . the left and the right plots are the same results but shown in different scales to stress that the early - time behavior of the dotted curve really matches with the climbing solid curve . here @xmath165 and @xmath166 . the slope of the dotted curve in its linear growing phase agrees quite well with @xmath249 in @xmath250.,title="fig:",width=302 ] usually in obtaining @xmath251 , tdpt is good from @xmath252 up to @xmath253 @xcite with @xmath254 ( here the masses of both harmonic oscillators in the detectors have been set to @xmath255 and the proper accelerations of the detectors are not extremely large ) . this is a large duration if @xmath156 is small , which is consistent with the assumption of tdpt , so it is commonly argued that extending the domain of @xmath234 integration from @xmath256 to @xmath257 in this duration would not change the results too much , that is , the tdpt@xmath191 results should be very close to the tdpt results . in ref.@xcite , reznik compared the absolute values of @xmath199 and @xmath200 in tdpt@xmath191 and discovered that the former is always greater than the latter , then drew the conclusion that the two detectors are always entangled . nevertheless , with the extended integration domain , both @xmath199 and @xmath200 are infinities , which are meaningless as elements of any density matrix . usually one divides @xmath200 by the infinite duration of interaction and interprets the result as the transition probability per unit time in the linear - growing phase in tdpt regime @xcite . but the choice of proper duration for @xmath199 and the interpretation for the result could be tricky ( see appendix [ rdm2dec ] for more details ) . further , our numerical results show that if we calculate @xmath258 with a finite integration domain and a nonzero regulator @xmath30 , the value of @xmath259 after integration can be quite different from those obtained in ref.@xcite . in eq . @xmath29 . here @xmath260 and @xmath261 . one can see that the correlation of the massless scalar field at @xmath262 and @xmath263 is enhanced around @xmath264 . such an enhancement will be suppressed if the trajectories of the detectors are off the @xmath265 and @xmath263 in this paper or the scalar field is massive . ( right ) contour plot of the same @xmath266 . the long - dashed lines , solid lines , short - dashed lines , and dotted lines are borders of the integration domain at @xmath267 , @xmath268 , @xmath269 , and @xmath270 , respectively.,title="fig:",width=302 ] in eq . @xmath29 . here @xmath260 and @xmath261 . one can see that the correlation of the massless scalar field at @xmath262 and @xmath263 is enhanced around @xmath264 . such an enhancement will be suppressed if the trajectories of the detectors are off the @xmath265 and @xmath263 in this paper or the scalar field is massive . ( right ) contour plot of the same @xmath266 . the long - dashed lines , solid lines , short - dashed lines , and dotted lines are borders of the integration domain at @xmath267 , @xmath268 , @xmath269 , and @xmath270 , respectively.,title="fig:",width=264 ] to calculate @xmath271 , one substitutes the trajectories @xmath272 and @xmath273 into the positive frequency wightman function and obtain @xmath274 now @xmath275 is regular everywhere in @xmath276 plane . for small finite @xmath30 , since @xmath277 , the @xmath278 term in the denominator can always be neglected , but the @xmath30 term can not . if @xmath279 , then @xmath280 , which is a function of @xmath225 only and peaks around @xmath281 . for @xmath33 sufficiently large such that @xmath282 , however , @xmath31 will be suppressed . so @xmath266 looks like a ridge at @xmath283 and spread with the width @xmath284 in @xmath33 direction , where @xmath285 ( see fig . [ wightman ] ) . suppose the coupling is switched on at the initial moment @xmath286 . from the landscape of @xmath275 shown in fig . [ wightman ] , the evolution of @xmath199 will have three stages : \(i ) when @xmath287 , @xmath275 is small all over the @xmath288-integration domain ( within the square with two sides in long - dashed lines in fig . [ wightman ] ( right ) ) , so the change in @xmath199 is tiny . \(ii ) at about @xmath289 , the edge of the @xmath288-integration domain [ within the square with all sides in solid lines in fig . [ wightman ] ( right ) ] touches the ridge of @xmath290 around @xmath281 , which gives obvious contribution to the result . as @xmath4 grows , more and more portion of the ridge are included in the integration domain . since the height of the ridge is almost constant in the @xmath33 direction , @xmath199 grows linearly in @xmath4 . \(iii ) the linear growth terminates at about @xmath291 . then the integration domain has covered almost all the ridge so that @xmath292 does not add obvious contribution after @xmath293 and @xmath199 saturates at some constant . from the argument in sec . [ regu ] , one should take @xmath294 so @xmath295 and the width of the ridge in the @xmath33 direction is virtually infinite . then stage ( ii ) will be terminated around @xmath296 because only a portion of the ridge with @xmath297 is covered in the integration domain . this could be enough to make the value of @xmath199 exceeds @xmath200 and the two detectors get entangled , if @xmath298 is sufficiently large . on the other hand , for @xmath299 , the linear growth of stage ( ii ) never occurs , and @xmath199 is always less than @xmath200 . thus one can see that the value of @xmath199 as well as the ratio @xmath300 have complicated evolution in time even in tdpt . their behaviors are quite different from the constant ones obtained in ref.@xcite . generalize the method applied in @xcite for one single detector , the elements of the rdm in eigen - energy representation can be read off by comparing the coefficients of the @xmath301 terms in both sides of the following equation , @xmath302 = \sqrt { \prod_j k_j\over { \cal g}\det \tilde{g } } \ , \exp \sum_{i , j } s_i \left [ k_i ( \tilde{g}^{-1})^{ij}k_j -\delta^{ij}\right]s_j , \ ] ] where @xmath303 , @xmath304 , @xmath305 is defined by @xmath306 , \label{calg}\ ] ] and @xmath307 is parametrized in @xmath308 with the factors @xmath309,\\ a_{b\pm } & = & { 1\over 2{\cal g}}\left [ v^{33 } \pm { 4\over \hbar^2}\left(v^{-1}\right)^{44}\det v\right],\\ a_{x\pm } & = & -{1\over 2{\cal g}}\left [ v^{13 } \pm { 4\over \hbar^2}\left(v^{-1}\right)^{24}\det v\right],\\ b_a & = & { 2\over\hbar{\cal g}}\left ( \left<\right.p_a , q_b\left.\right>\left<\right.q_a , q_b\left.\right > -\left<\right.p_a , q_a\left.\right>\left<\right.q_b^2\left.\right>\right),\\ b_b & = & { 2\over\hbar { \cal g}}\left ( \left<\right.p_b , q_a\left.\right>\left<\right.q_a , q_b\left.\right > -\left<\right.p_b , q_b\left.\right>\left<\right.q_a^2\left.\right>\right),\\ b_{x\pm } & = & { 1\over\hbar { \cal g}}\left [ \left(\left<\right.p_a , q_a\left.\right>\left<\right.q_a , q_b\left.\right > -\left<\right.p_a , q_b\left.\right>\left<\right.q_a^2\left.\right>\right ) \pm \right . \nonumber\\ & & \,\,\ , \left . \left(\left<\right.p_b , q_b\left.\right>\left<\right.q_a , q_b\left.\right > -\left<\right.p_b , q_a\left.\right>\left<\right.q_b^2\left.\right>\right ) \right].\end{aligned}\ ] ] these @xmath5 and @xmath310 factors are obtained by solving the coefficients @xmath311 in the gaussian rdm of the detectors @xmath312 in terms of the two - point correlators . here @xmath313 are the elements of the covariance matrix @xmath314 in which the elements of the @xmath184 matrices @xmath315 are symmetrized two - point correlators @xmath316 with @xmath317 , @xmath318 and @xmath319 . up to the first excited states of both detectors , the truncated rdm of the two detectors in eigen - energy representation reads @xmath320 \end{array}\right ) \label{truncrho}\ ] ] with @xmath321 , @xmath322 , and @xmath323 here we use the basis @xmath324 for @xmath325 . what tdpt@xmath191 could describe well is the transient behavior during @xmath326 while @xmath327 with @xmath328 , when the amplitudes of the oscillating cross correlators are linearly increasing , while the magnitudes of the self correlators are still in transient and growing linearly , too . in this regime , @xmath59 can be approximated by @xmath80 , which also yields @xmath329 , \\ & & \left<\right . p_a , p_b\left.\right > \approx { \hbar\gamma\over \sinh{\pi\omega\over a}}\left [ 2\omega\tau \cos 2\omega\tau - \left({\pi\omega\over a}\coth{\pi\omega\over a } -1\right)\sin 2\omega\tau\right],\end{aligned}\ ] ] up to @xmath118 . together with the @xmath118 results of the self correlators in tdpt regime from @xcite with @xmath255 , @xmath330,\\ \left<\right . q_a , p_a\left.\right > = \left<\right . q_b , p_b\left.\right > & \approx & { \hbar\gamma\over 2\omega}\left [ \left(\coth { \pi\omega\over a}-1\right ) + { 2\over\pi}(\lambda_0-\ln a ) \sin 2\omega\eta + \cos 2\omega\eta\right],\\ \left<\right . p_a^2\left.\right > = \left<\right . p_b^2\left.\right > & \approx & { \hbar\omega\over 2 } + \nonumber\\ & & { \hbar\gamma\omega}\left [ \left(\coth { \pi\omega\over a}-1\right)\eta+ { 2\over\pi\omega}\left [ \lambda_1-\ln a + ( \lambda_0-\ln a)\cos^2\omega\eta\right ] - { \sin 2\omega\eta\over 2\omega}\right],\end{aligned}\ ] ] one finds that , for the detectors initially in their ground states , @xmath331 - { 1\over 2}\sin 2\omega\eta \right ] , \\ % a5 a_{a-}=a_{b- } & \approx & { \gamma\over\hbar}\left [ { 2\omega\eta\over e^{2\pi\omega / a}-1 } + { 1\over\pi}(\lambda_0+\lambda_1 - 2\ln a)\right],\\ % a3 a_{x+ } & \approx & { \gamma\over\hbar\sinh{\pi\omega\over a}}\left [ 2\omega\tau \cos 2\omega\tau + \left ( 1-{\pi\omega\over a}\coth{\pi\omega\over a}\right ) \sin 2\omega \tau\right],\\ % b3 b_{x+ } & \approx & { \gamma\over\hbar\sinh{\pi\omega\over a}}\left [ -2\omega\tau \sin 2\omega\tau + \left ( 1-{\pi\omega\over a}\coth{\pi\omega\over a}\right ) \cos 2\omega \tau\right],\end{aligned}\ ] ] up to @xmath118 , so that @xmath332 now the growing rates of @xmath199 and @xmath333 at large @xmath4 agree well with those estimated by tdpt@xmath191 , where both @xmath334 and @xmath4 are replaced by infinities . nevertheless , one has to be careful in calculating the ratio of different matrix elements in tdpt@xmath335 . indeed , at @xmath336 , @xmath114 , @xmath333 is approximately proportional to @xmath337 while @xmath199 is approximately proportional to @xmath4 , so that @xmath338 , where @xmath4 and @xmath334 do not simply cancel each other . @xmath300 is actually time - varying , and eq.(18 ) in @xcite is correct only at the moment @xmath339 , or @xmath194 , which is almost at the end of the tdpt@xmath191 regime . note that , in weak coupling limit , @xmath340,\\ a_{b- } & \approx & { 1\over 8\left<\right.q_b^2\left.\right > } - { 1\over 2\hbar^2}\left<\right.p_b^2\left.\right > , \label{abmwcl}\end{aligned}\ ] ] so the behavior of @xmath271 in @xmath201 is highly dependent on the behavior of the cross correlators , while @xmath341 in @xmath342 is not . note also that @xmath186 in ultraweak coupling limit is numerically consistent with the tdpt results in appendix [ tdptapx ] in the time interval @xmath343 for @xmath344 . here @xmath4 can be negative or greater than @xmath298 . from the three - stage behavior of the @xmath199 in appendix [ tdptapx ] , which depends on the fiducial time @xmath25 , one can see that tdpt still keeps some of the non - markovian features , though the damping behavior ( @xmath345 ) has been lost in the coupling constant expansion ( cf . the integrands of @xmath192-@xmath259 with @xmath346 ) . on the other hand , the validity range of tdpt@xmath191 is more restricted . tdpt@xmath191 is valid only when both @xmath199 and @xmath199 are linearly growing , namely , only in the section @xmath347 of the time interval @xmath343 for tdpt . tdpt@xmath335 results are markovian because even the memory of the initial moment is lost after the integration domain is extended .

recent developments in simulating fundamental quantum field theoretical effects in the kinematical context of analogue gravity are reviewed . specifically , it is argued that a curved space - time generalization of the unruh - davies effect the gibbons - hawking effect in the de sitter space - time of inflationary cosmological models can be implemented and verified in an ultracold gas of bosonic atoms .

cond-mat0406086

curved space - times are familiar from einstein s theory of gravitation,@xcite where the metric tensor @xmath0 , describing distances in a curved space - time with local lorentz invariance , is determined by the solution of the einstein equations . a major problem for an experimental investigation of the ( kinematical as well as dynamical ) properties of curved space - times is that generating a significant curvature , equivalent to a ( relatively ) small curvature radius , is a close to impossible undertaking in manmade laboratories . for example , the effect of the gravitation of the whole earth is to cause a deviation from flat space - time on this planet s surface of only the order of @xmath1 ( the ratio of schwarzschild and earth radii ) . the fact that proper gravitational effects are intrinsically small is basically due to the smallness of newton s gravitational constant @xmath2m@xmath3kg@xmath4sec@xmath5 . various fundamental classical and quantum effects in strong gravitational fields , a few of which we discuss below , are thus inaccessible for earth - based experiments . the realm of strong gravitational fields ( or , equivalently , rapidly accelerating a reference frame to simulate gravity according to the equivalence principle ) , is therefore difficult to reach . however , earth - based experiments are desirable , because they have the obvious advantage that they can be prepared and , in particular , repeated under possibly different conditions at will . the formalism to be described in what follows is aimed at the realization of _ effective _ curved space - time geometries in perfect fluids , which can indeed be prepared on earth , and which mimic the effects of gravity inasmuch the kinematical properties of curved space - times are concerned . among such perfect fluids are bose - einstein condensates , i.e. , the dilute matter - wave - coherent gases formed if cooled to ultralow temperatures , where the critical temperatures are of order @xmath6k ; for a short review of the ( relatively ) recent status of this rapidly developing field see ref.@xcite . in what follows , it will be of some importance that bose - einstein condensates belong to a special class of quantum perfect fluids , so - called superfluids.@xcite the curved space - times we have in mind are experienced by sound waves propagating on the background of a spatially and temporally inhomogeneous perfect fluid . of primary importance is , first of all , to realize that the identification of sound waves propagating on an inhomogeneous background , which is itself determined by a solution of euler and continuity equations , and photons propagating in a curved space - time , which is determined by a solution of the einstein equations , is of a _ kinematical _ nature . that is , the space - time metric is fixed externally by a background field obeying the laws of hydrodynamics ( which is prepared by the experimentalist ) , and not self - consistently by a solution of the einstein equations @xmath7 ( where @xmath8 and the speed of light are set to unity ) . the latter equations relate space - time curvature represented by the einstein tensor @xmath9 , where @xmath10 is the ricci tensor and @xmath11 the ricci scalar with the energy - momentum content of all other fundamental quantum fields . this energy - momentum content is represented by the classical quantity @xmath12 , which is the regularized expectation value of a quantum energy - momentum tensor.@xcite as a first introductory step to understand the nature of the proposed analogy , consider the wave equation for the velocity potential of the sound field @xmath13 , which in a homogeneous medium at rest reads @xmath14\phi = 0 , \label{waveeq}\ ] ] where @xmath15 is the sound speed . it is a constant in space and time for such a medium . this equation has lorentz invariance , that is , if we replace the speed of light by the speed of sound , it retains the form shown above in the new space - time co - ordinates obtained after lorentz - transforming to a frame moving at a constant speed less than the sound speed . just as the light field _ in vacuo _ is a proper relativistic field , sound is a `` relativistic '' field . the lorentz invariance can be made more manifest by writing equation ( [ waveeq ] ) in the form @xmath16 , where @xmath17diag(@xmath18 is the ( contravariant ) flat space - time metric ( we choose throughout the signature of the metric as specified here ) , determining the fundamental light - cone structure of minkowski space;@xcite we employ the summation over equal greek indices @xmath19 assuming , then , the sound speed @xmath20 to be local in space and time , and employing the curved space - time version of the 3 + 1d laplacian @xmath21,@xcite one can write down the sound wave equation in an _ inhomogeneous medium _ in the generally covariant form@xcite @xmath22 here , @xmath23 $ ] is the determinant of the ( covariant ) metric tensor . it is to be emphasized at this point that because the space and time derivatives @xmath24 are covariantly transforming objects in ( [ curvedspacewaveeq ] ) , the primary object in the condensed - matter identification of space - time metrics via the wave equation ( [ curvedspacewaveeq ] ) is the contravariant metric tensor @xmath25.@xcite in the condensed - matter understanding of analogue gravity , the quantities @xmath25 are _ material - dependent _ coefficients . they occur in a dispersion relation of the form @xmath26 , where @xmath27 is the covariant wave vector , with @xmath28 the ordinary spatial momentum ( or quasi - momentum in a crystal ) . the contravariant tensor components @xmath25 , for a perfect , irrotational liquid turn out to be@xcite @xmath29 where @xmath30 is the unit matrix and @xmath31 a space and time dependent function , to be determined from the proper equations of motion for the sound field ( see below ) . inverting this expression according to @xmath32 , to obtain the covariant metric @xmath0 , the fundamental tensor of distance reads @xmath33 where the line element is @xmath34 . this form of the metric has been derived by unruh for an irrotational perfect fluid described by euler and continuity equations;@xcite its properties were later on explored in more detail in particular by visser.@xcite i also mention that an earlier derivation of unruh s form of the metric exists , from a somewhat different perspective ; it was performed by trautman.@xcite the conformal factor @xmath31 in ( [ gdown ] ) depends on the spatial dimension of the fluid . it may be unambiguously determined by considering the action of the velocity potential fluctuations above an inhomogeneous background , identifying this action with the action of a minimally coupled scalar field in @xmath35-dimensional space - time:@xcite @xmath36 \nonumber\\ & \equiv & \frac12 \int d^{d+1}x % dt dz \sqrt{-{\sf g } } { \sf g}^{\mu\nu } \partial_\mu \phi \partial_\nu \phi \ , , \label{action}\end{aligned}\ ] ] where it is assumed that the compressibility @xmath37 of the ( barotropic ) fluid , @xmath38 , where @xmath39 is the pressure and @xmath40 the density of the fluid , is a constant . using the above canonical identification , it may easily be shown that the conformal factor is given by @xmath41 . it is mentioned here that the case of one spatial dimension ( @xmath42 ) is special , in that the so - called conformal invariance in two space - time dimensions implies that the classical equations of motion are invariant ( take the same form ) for any space and time dependent choice of the conformal factor @xmath31 . the line element @xmath43 gives us the distances travelled by the phonons in an effective space - time world in which the scalar field @xmath13 `` lives '' . in particular , quasiclassical ( large momentum ) phonons will follow _ light_-like , that is , here , _ sound_-like geodesics in that space - time , according to @xmath44 . particularly noteworthy is the simple fact that the constant time slices obtained by setting @xmath45 in the line element are conformally flat , i.e. the quasiparticle world looks on constant time slices like the ordinary ( newtonian ) lab space , with a simple euclidean metric in the case of cartesian spatial co - ordinates we display . all the intrinsic curvature of the effective space - time is therefore encoded in the metric tensor elements @xmath46 and @xmath47 . together with matt visser , i described this curvature and its properties for an isotachic fluid ( i.e. , having a speed of sound independent of space and time).@xcite using that phonons move on geodesics , we discovered in@xcite the phenomenon that a vortex acts on quasiclassical phonons as an effective gravitational lens . in ref.@xcite , using the fact that there is curvature in any spatially inhomogeneous flow ( that is , a flow which is not a simple superposition of translational motion and rigid body rotation ) , we have shown that there exists a sonic analogue of the `` warp - drive '' in general relativity permitting superluminal , i.e. , here `` superphononic '' motion.@xcite the point made by us is that in the acoustic curved space - times we consider , there is no violation of any condition on the positivity of energy necessary , which is in marked contrast to the original warp drives , where local energy densities by necessity must be negative to permit superluminal travel.@xcite this is due to the fact that the einstein equations , relating curvature and energy - momentum content of all fields other than gravitational fields , as already mentioned in the above do not need to be imposed in the analogy . it is important to recognize that the form ( [ curvedspacewaveeq ] ) of the wave equation is valid generally ( with a possible additional scalar potential term ) . that is , a generally covariant curved space - time wave equation can be formulated not just for the velocity perturbation ( sound ) potential in an irrotational euler fluid , for which we have introduced the effective metric concept . if the spectrum of excitation ( in the local rest frame ) is linear , @xmath48 , where @xmath49 is the propagation speed of _ some _ collective excitation , the statement that an effective space - time metric exists is true , provided we only consider wave perturbations of a single scalar field which constitutes the fixed classical background . the argument to reach this conclusion is as follows . given that the action density @xmath50 is a functional of @xmath51 and its space - time derivatives @xmath52 , i.e. @xmath53 $ ] , we expand the action to quadratic order in the fluctuations around some stationary classical background solution @xmath54 of the euler - lagrange equations . for any lagrangian of the specified form , the wave equation for perturbations @xmath55 above the background @xmath54 is@xcite @xmath56 the above equation in covariant notation reads @xmath57 where @xmath58 is a background - field dependent effective potential ( equal to the second term in round brackets in equation ( [ general ] ) divided by @xmath59 ) . the potential @xmath60 may , for example , in the simplest case contain an effective mass of the scalar field , such that the wave equation becomes klein - gordon like , @xmath61 . the effective metric coefficients are , up to an ( again dimension dependent ) conformal factor given by : @xmath62 the concept of an effective space - time metric therefore applies to every system having a single scalar wave equation of second order in both space and time derivatives , corresponding to perturbations propagating on a fixed classical background , where this background itself determines the metric coefficients . we assumed in eq.([action ] ) that the compressibility @xmath37 is a constant . this entails that the ( barotropic ) equation of state reads @xmath63 . we then have , in terms of the interaction between the particles ( atoms ) constituting the fluid , a contact interaction ( pseudo-)potential , @xmath64 . this is indeed the case for the _ dilute _ atomic gases forming a bose - einstein condensate . well below the transition temperature , they are described to good accuracy by the gross - pitaevski mean field equation for the order parameter @xmath65 , breaks particle number conservation ( the global u(1 ) invariance ) ; for a review of the consequences see ref.@xcite . ] representing the expectation value of the quantum field operator @xmath66:@xmath67 \psi ( { \bm x},t ) . \label{gpeq}\ ] ] the madelung transformation reads @xmath68 $ ] , where @xmath40 yields the condensate density and @xmath51 is the velocity potential ; it allows for an interpretation of quantum theory in terms of hydrodynamics.@xcite namely , identifying real and imaginary parts on left- and right - hand sides of ( [ gpeq ] ) , respectively , gives us the two equations @xmath69 the first of these equations is the josephson equation for the superfluid phase , which corresponds to the bernoulli equation of classical hydrodynamics , where the usual velocity potential of irrotational hydrodynamics equals the superfluid phase @xmath51 times @xmath70 , such that @xmath71 . the latter equation implies that the flow is irrotational save on singular lines , around which the _ wave function phase _ @xmath51 is defined only modulo @xmath72 . therefore , circulation is quantized,@xcite and these singular lines are the center lines of quantized vortices . the isothermal chemical potential @xmath73 ( which we have chosen to incorporate the kinetic energy term @xmath74 ) , is augmented by the `` quantum pressure term '' @xmath75 , which is the one genuine quantum term in ( [ jaeq ] ) , because @xmath76 ( observe that the first order in @xmath77 may be incorporated into the velocity potential ) . the second equation ( [ conteq ] ) is the continuity equation for conservation of particle number , i.e. , atom number in the superfluid gas . the dynamics of the weakly interacting , dilute ensemble of atoms is thus that of a perfect euler fluid with quantized circulation of singular vortex lines , which are the only vortical excitations in a superfluid . the fact that it is an euler fluid is true save for regions in which the density rapidly varies and the quantum pressure term @xmath78 becomes relevant , which happens on scales of order the coherence length @xmath79 , where @xmath80 is a constant asymptotic density far away from the density - depleted ( or possibly density - enhanced ) region . the quantum pressure becomes relevant in the depleted - density cores of quantized vortices , or at the low - density boundaries of the system , and is negligible outside these domains of rapidly varying or low density . the whole armoury of space - time metric description of excitations , explained in the last section , which is based on the euler and continuity equations , is then valid for phonon - like excitations of a bose - einstein condensate , with the space - time metric ( [ gdown ] ) , as long as we are outside the core of quantized vortices , ) arising therefrom , may be found in ref.@xcite . ] and far enough from the boundaries of the condensate . quasiparticles are the fundamental entities used to describe an interacting condensed matter system in a particle picture , that is , in a suitable fock space . on the microscopic level , if the elementary constituents interact by two - body forces , we are given a second quantized hamiltonian operator of the form @xmath81 , where @xmath82 are the matrix elements for two - particle interaction in a plane wave basis , and @xmath83 are the bare single particle energies of the `` elementary '' bosons or fermions , which are created by the operators @xmath84 from the proper particle vacuum . one then employs a unitary , i.e. operator - algebra - conserving bogoliubov transformation@xcite to another set of quasiparticle operators @xmath85 , @xmath86 , see eq.([bogotrafo1 ] ) below . this gives the hamiltonian the reduced diagonal form @xmath87 , where @xmath88 represents additional terms of higher order than quadratic , which are supposed to be small compared to the leading diagonalized part of the hamiltonian , for the picture of noninteracting quasiparticles to make sense . these quasiparticles possess a ( possibly spatially anisotropic ) dispersion relation @xmath89 which is linear for various important classes of quasiparticles , @xmath90 . among these classes of collective excitations are , e.g. , phonons , antiferromagnetic magnons,@xcite or the excitations around the gap nodes of the @xmath39-wave superfluid @xmath91he - a.@xcite phonons are the small momentum quanta of the sound field in solids or fluids , with @xmath92 for a medium at rest . their classical equation of motion in a perfect fluid is the generally covariant wave equation ( [ curvedspacewaveeq ] ) , with the effective space - time metric ( [ gdown ] ) . in the simplest case , for a homogeneous medium with space and time independent density , and a constant speed of sound @xmath15 , the quantized velocity potential of the phonons reads@xcite @xmath93 \left[{\hat{b}}_{\bm k}e^{-ic_s k t + i{\bm k}\cdot { \bm x } } + { \hat{b}^\dagger}_{\bm k } e^{ic_s k t - i{\bm k}\cdot { \bm x } } \right ] , \label{homo}\ ] ] where @xmath94 is the quantization volume of the system , and @xmath80 the constant background density . that is , the quantum excitation field in a homogeneous medium may generally be decomposed into plane waves , with the appropriate frequencies as a function of momentum stemming from the dispersion relation , here @xmath95 . note , in particular , that for this spatially and temporally homogeneous case , the statement that we observe positive frequency ( energy ) with respect to the laboratory time interval @xmath96 is unique , that is , it can be made independent of time and space . in an inhomogeneous fluid , this is ( generally ) no longer the case , and the notion of an excitation having positive energy may depend on where the detector is located in the fluid , if it is at rest relative to the fluid or moves , and what its natural time interval is . the latter may be different from that of the laboratory , due to the particular way the detector couples to the fluid , see section [ gibbonshawking ] . the number of particles assigned to the quantum field @xmath97 is unique with respect to a certain given state @xmath98 of the quantum field , @xmath99 , provided we decompose @xmath97 into modes associated to the operators @xmath100 and their hermitian conjugates . the number of particles is , in particular , zero for the _ vacuum state with respect to the field operator @xmath101 _ , defined by @xmath102 . however , it need not be zero with respect to another set of quasiparticle operators @xmath85 , which has a _ different _ vacuum @xmath103 . to demonstrate the basis dependence of quasiparticle number , we use a general bogoliubov transformation for bosons , the class of ( quasi-)particles we consider , of the general form @xmath104 where @xmath105 represents a set of quantum numbers , not necessarily the plane waves used in eq.([homo ] ) . the coefficients in the bogoliubov transformation must fulfill certain conditions for the transformation to be unitary , i.e. , to preserve the bosonic commutation relations for the new operators , @xmath106= \delta_{{\bm i}{\bm k}},\ , [ \hat b_{\bm i } , \hat b_{\bm k}]=0,\ , [ \hat b^\dagger_{\bm i } , \hat b^\dagger_{\bm k}]=0 $ ] . as a consequence of this defining unitary character , the following conditions on the transformation coefficients must hold : @xmath107 by using the transformation ( [ bogotrafo1 ] ) , it is straightforwardly shown that the number of @xmath101 particles in @xmath103 is given by @xmath108 : the old operator @xmath100 does not annihilate the new vacuum @xmath109 ( and vice versa ) , and what looks empty in one quasiparticle vacuum may be full of quasiparticles in another . related to this ( formal ) operator - basis dependence is the fact that the actually _ detected _ number of particles is strongly _ observer _ dependent , as opposed to the formally defined quantity @xmath110 , which refers to one particular quasiparticle state @xmath111 . the detected number of particles depends , in particular , on how the detector actually couples to the field @xmath97 whose quanta it measures . various couplings of the detector , for example to different powers of the fluid density , will influence the quasiparticle basis in which the detector measures , and thus the quasiparticle number detected . now , the salient point is that because the phonon field @xmath97 is a _ relativistic _ quantum field , we will be able to map the observer dependence just described , which is general and holds for any sort of quasiparticles , to the observer dependence experienced by proper relativistic quantum fields in curved or flat space - time.@xcite the observer dependence is of kinematical origin , i.e. , it originates in the fact that the relativistic quantum field propagates in a space - time of lorentzian signature , cf . the discussion of the hawking radiation analogue phenomenon in ref.@xcite , and is therefore fully within the capabilities of our proposed analogy . to model the quasiparticle analogue of expanding universes , we will make use of expanding bose - einstein condensates , which are produced by reducing the trapping potential strength , i.e. the harmonic trapping frequency with time , or by increasing the interaction coupling constant . firstly , it is thus appropriate to describe the evolution of density and velocity distribution in the expanding gas by discussing the so - called scaling procedure established in refs.@xcite . the scaling procedure introduces a set of generally three scaling variables , @xmath112 , which are a function of time only . using these scaling variables , one writes for the ( cartesian ) co - ordinate vector components @xmath113 ; for the scaled co - ordinate vector we use the shorthand notation @xmath114 . it may then be shown that the evolution of the bose - einstein - condensed gas cloud is described , starting from the initial density and velocity potential distributions @xmath115 , @xmath116 , by the following density and velocity distributions ( from here on we take @xmath117):@xcite @xmath118 this is true provided the scaling parameters @xmath119 fulfill the equations@xcite @xmath120 where @xmath121 are the initial trapping frequencies and the dimensionless `` scaling volume '' reads @xmath122 . the dots are time derivatives with respect to laboratory time @xmath123 here and in what follows . , but only over greek indices @xmath124 . ] we have taken into account in the above equation that the particle interaction can be varied in lab time , @xmath125 , by means of a suitable feshbach resonance;@xcite @xmath126 is the initial coupling constant . we designate `` scaling basis '' quantities with a tilde . for example , the ( stationary ) initial density distribution @xmath127 gives us @xmath128 if we replace @xmath129 . the scaling evolution is exact in a thomas - fermi approximation which neglects the quantum pressure term @xmath78 and the kinetic energy @xmath74 . since bose - einstein condensates were experimentally created , the scaling solution has routinely been employed to interpret the time - of - flight pictures with which they are visualized as well as analyzed.@xcite we define a `` scaling time '' variable by @xmath130 and the @xmath131 dependent _ scaling functions _ @xmath132 by @xmath133 in terms of these quantities , the effective second order action for the scaling basis fluctuations of the phase of the superfluid order parameter @xmath134 takes on the particularly simple diagonal form@xcite @xmath135 , \label{kappaaction}\ ] ] where the scaling speed of sound @xmath136 . because of the fact that this action does not mix spatial and temporal derivatives , the resulting line element _ in the scaling variables _ , according to the identification displayed in ( [ action ] ) is diagonal ( does not possess @xmath47 terms ) , and reads @xmath137.\label{blineelement}\ ] ] we now consider for simplicity the isotropic case , @xmath138 , which implies @xmath139 the generalization of the mode expansion in eq.([homo ] ) to inhomogeneous expanding bose - einstein condensates then takes , in this isotropic case , the form@xcite @xmath140.\label{3dsol}\end{aligned}\ ] ] the functions @xmath141 are the stationary solutions of the gross - pitaevski equations for excitations above the initial ground state , designated by the ( set of ) quantum numbers @xmath142 with initial energies @xmath143 . the initial thomas - fermi quantization volume is @xmath144 ; in the hard - walled cubic box limit the modes are plane waves , @xmath145 $ ] . the temporal mode functions @xmath146 satisfy the second order ordinary differential equation@xcite @xmath147 the case when @xmath148 is a constant ( unity ) is particular . in this case , the quantum state of the quasiparticle excitations remains unchanged for increasing @xmath131 , and a given initial quasiparticle vacuum , _ in the scaling basis _ with the associated quasiparticle operators defined by ( [ 3dsol ] ) , remains empty forever . that is , no quasiparticles are created in that basis , although the superfluid may be in a highly nonstationary motional state , obtained by changing the trapping @xmath149 rapidly with time . however , a detector which measures in a quasiparticle basis different from the scaling basis , for example due to its particular coupling to the superfluid , may still detect that quasiparticles are `` created '' . we will come back to this possibility in section [ gibbonshawking ] below , when we discuss the purely choice - of - observer related phenomenon of a thermal state in a quasiparticle basis belonging to one particular space - time , the de sitter space - time of cosmology . consider now the general case that the scaling function @xmath150 is a function of scaling time @xmath131 . the fact that @xmath148 depends on time implies that the statement `` the excitation is of positive frequency '' ( a particle ) or of `` negative frequency '' ( antiparticle ) for a given propagating wave can not be held up for all times @xmath131 . this _ frequency mixing _ implies that quasiparticles are created from the quasiparticle vacuum , because an initially empty scaling basis vacuum state does not remain empty during the evolution of the system , i.e. initially @xmath151 , but at a later stage @xmath152 . the fact that annihilation and creation operator parts of the initial vacuum are mixed , as a consequence of ( [ bogotrafo1 ] ) , is physically due to the fact that quasiparticles are scattered within the course of time at ( time dependent ) effective potentials . physically , the @xmath131 dependence of @xmath148 furnishes such an effective potential , at which excitations are scattered from negative to positive frequency and vice versa : the equation ( [ chin ] ) is formally equivalent to scattering of a non - relativistic particle with energy @xmath153 by a potential in @xmath131 space , @xmath154 at large @xmath131 , the wkb scattering solution of ( [ chin ] ) therefore reads:@xcite @xmath155 + \beta_{n}\exp\left[i\epsilon_{n}\int_{-\infty}^{\tau_s } { d\tau_s'}\sqrt{f(\tau_s')}\right]\right ) , \label{chineqs}\ ] ] where the scattering amplitudes are related via the particle flux conservation condition @xmath156 . the quantity @xmath157 can be interpreted as the number of _ scaling basis _ quasiparticles created from the initially empty scaling vacuum , due to the time dependent scattering of excitations moving in the nonstationary condensate . in the wkb approximation , the amplitudes @xmath158 and @xmath159 are connected in a simple way : @xmath160\alpha_{n } , \label{alphat0beta}\ ] ] where the inverse temperature @xmath161 is given by the integral @xmath162 , \label{deft0}\ ] ] and @xmath163 is the contour in the complex @xmath131-plane enclosing the closest to the real axis singular point of the function @xmath164.@xcite this gives us the number of particles created in the mode @xmath142 , using the flux conservation condition @xmath156 , @xmath165 - 1}.\ ] ] the distribution of the created quasiparticles follows a thermal bosonic distribution ( planck spectrum ) , at a temperature @xmath166 . the adiabatic evolution of trapped gases hence leads to `` cosmological '' quasiparticle creation with thermal occupation numbers in the scaling basis . the temperature @xmath167 occurring in the planck distribution above depends on the details of the scaling evolution , i.e. , on the specific superfluid dynamics imposed by the solution of eqs.([genbeq ] ) , that is , @xmath166 is a functional of the temporal evolution @xmath168.@xcite we use the term `` cosmological '' in context with the plain condensed - matter fact that quasiparticles are created in the scaling basis . we now justify this by comparing our effective bose - einstein condensate metric ( [ blineelement ] ) to line elements which constitute cosmological solutions of the einstein equations . for example , let @xmath169 by properly adjusting @xmath125 , thus choosing the coupling constant s time evolution to be given by @xmath170 , cf . eq.([deftau ] ) . we then obtain that ( [ blineelement ] ) equals ( up to the conformal factor @xmath171 ) ) an anisotropic version of the spatially flat friedmann - robertson - walker ( frw ) universe:@xcite @xmath172 . \label{blineelementiso}\ ] ] in the standard spatially isotropic form of the frw metric , all @xmath173 are equal . to obtain the exact equivalence to a spatially flat frw metric , we have to assume in addition that @xmath174 is spatially independent , which is fulfilled close to the center of the gas cloud , where the parabolic density profile is approximately flat , and @xmath174 is essentially a constant . in the spatially isotropic case , @xmath175 is the hubble `` constant '' , which obviously is a constant only if the gas expands exponentially in laboratory time , @xmath176 $ ] , just as we need exponentially rapid expansion for a constant hubble parameter in inflationary cosmological models.@xcite we therefore come to the remarkable conclusion that the co - ordinate scaling factor @xmath177 of the bose - einstein condensate quasiparticle universe , occurring in the equations of motion of a nonrelativistic condensed matter system , has exactly the same meaning , in the quasiparticle world , as the cosmological scale factor of the universe proper . we now treat the case that the scaling factors @xmath132 in ( [ deff ] ) are constants , i.e. , do not depend on scaling time . therefore , following eqs . ( [ 3dsol ] ) and ( [ chin ] ) , no scaling basis quasiparticles are created through negative and positive frequency mixing . the superfluid can still be in highly nonstationary motion , though : the time evolution is according to ( [ deff ] ) prescribed by @xmath178 cf . ( [ deff ] ) , where the @xmath179 are constants . however , although the superfluid is in motion , no dissipation through intrinsic quasiparticle creation takes place , because there exists the fock space `` scaling '' basis , in which no quasiparticles are created from the scaling vacuum , and the energy of that particular superfluid vacuum is conserved . a particular instance is the isotropic 2d case , @xmath180 , where for constant @xmath181 the condition on @xmath148 being constant is fulfilled . in de sitter space - time space is empty and flat ( that is , the constant time slices are euclidean space ) , and all of the curvature of space - time is encoded in a nonvanishing cosmological constant @xmath182 . for obvious reasons , the de sitter space - time is very popular in the quantum field theoretical treatment of cosmological theories,@xcite because it highlights the crucial cosmological role the energy density of all conceivable quantum fields taken together might play : the vacuum energy density may constitute the dominant effective source term for space - time curvature in the einstein equations . the gibbons - hawking effect for geodesic observers@xcite in such a de sitter space - time is the curved space - time analogue of the unruh - davies effect.@xcite the latter consists in the fact that a constantly accelerated detector moving in the flat ( purportedly `` empty '' ) space - time vacuum , responds as if it were placed in a thermal bath of ( quasi-)particles with temperature proportional to its acceleration . observer - related phenomena are at the heart of quantum field theory on nontrivial , and generally curved , space - time backgrounds . they tell us , in particular , that the particle content of a given quantum field depends on the ( motional ) state of the detection apparatus , which is verifying that there are particles by its `` clicks '' . more technically speaking , the dependence of the particle content of quantum fields in curved space - time is rooted in the non - uniqueness of canonical field quantization in riemannian spaces.@xcite it is of fundamental importance to make these observer - dependent effects measurable , because such a measurement constitutes , _ inter alia _ , a consistency check for an all - important concept of standard quantum field theory , namely that the quantization of a given field is carried out on a _ fixed _ space - time background . that the experimental verification of the observer dependence is exceedingly difficult with light becomes readily apparent if we calculate the unruh - davies temperature : the result is that it equals @xmath183 a = 4\ , { \rm k } \times a[10^{20 } g_{\oplus } ] \,,$ ] where @xmath184 is the acceleration of the detector in minkowski space ( @xmath185 is the gravity acceleration on the surface of the earth ) , and @xmath186 the speed of light . the huge accelerations needed to obtain measurable values of @xmath187 make it obvious that an observation of the effect with light ( photons ) is decidedly a less than trivial undertaking . although proposals for a measurement with ultraintense short pulses of electromagnetic radiation have been put forward in , e.g. , refs.@xcite , it is less than obvious how the thermal spectra associated to the effect , which still furnish tiny contributions to the total energy , should be discernible from the background dominated by the ultraintense lasers used to create large accelerations of the elementary particles contained in the plasma . we first discuss the simplest case of an isotropically expanding gas in a harmonic trap . we will see that this case also serves a pedagogical purpose , because it forces us to distinguish between purely observer - related phenomena with thermal spectrum , which by definition all have @xmath188 , and `` cosmological '' particle creation with a thermal spectrum , as defined in section [ cosmological ] , for which the scaling functions @xmath189 in ( [ deff ] ) depend explicitly on scaling time . one can create de sitter universes in an expanding gas by letting it expand exponentially , @xmath190 $ ] where @xmath191 is the hubble constant of cosmological expansion , equalling @xmath192 for the de sitter universe discussed , where @xmath193 is the spatial dimension . the de sitter metric in its standard form , using as its time co - ordinate the `` cosmological '' time @xmath194 , reads@xcite @xmath195,\label{desittercomoving}\ ] ] where @xmath196 is the central scaling ( i.e. , initial ) speed of sound . in the present case of exponential expansion , the `` cosmological '' time interval occurring in the metric above is equal to both the laboratory and the scaling time interval , @xmath197 ; cf . eq.([blineelementiso ] ) with @xmath198 $ ] . however , using the isotropic harmonic expansion setup , one has to face the difficulty that in order to give the frw metric ( [ blineelementiso ] ) near the center of the trap ( i.e. , close to @xmath199 ) the de sitter form , one has to increase exponentially the interaction with laboratory time : because of ( [ defh ] ) , @xmath200 = \exp[c_0 \lambda t]$ ] . though the central density also decreases exponentially ( like @xmath201 ) , the exponential increase of the coupling `` constant '' incurs strong three - body recombination losses,@xcite whose total rate ( in the dilute gas case and in three spatial dimensions ) is proportional to @xmath202 $ ] . therefore , within a short time of order @xmath203 , the bose - einstein - condensed gas of interacting single atoms will simply no longer be in existence , because the atoms rapidly form bound states . such an experiment will leave no time to measure phenomena which depend on the fact that an equilibrium is established ; in particular , the thermal equilibrium for the occupation numbers in the de sitter quasiparticle basis will not be established on such a short time scale . even more importantly , though one obtains indeed a thermal spectrum in the de sitter quasiparticle basis corresponding to the metric ( [ desittercomoving ] ) , @xmath204 $ ] depends on time , i.e. , it is not a constant in the quasiparticle basis corresponding to the mode functions @xmath205 $ ] . the thermal spectrum obtained therefore is , by our physical definition , the thermal `` cosmological '' quasiparticle creation discussed in the previous section . it is _ not _ what we want to observe in our effective de sitter space - time , namely the purely choice - of - observer related phenomenon gibbons - hawking effect , for which no actual quasiparticle `` creation '' in the scaling basis should take place . to circumvent the problem that the interaction coupling needs to be increased exponentially with time to obtain a de sitter universe in 2 + 1 or 3 + 1 isotropic space - time dimensions in an isotropic harmonic trap , i and petr fedichev have developed the model of a 1 + 1d de sitter universe . this 1 + 1d toy model can be realized in a strongly anisotropic , cigar - shaped bose - einstein condensate,@xcite cf . fig.[ellipse ] . in particular , in the proposed experimental setup , no time variation of the coupling constant at all is necessary , which is thus a true `` constant '' also in time . the analysis of the excitation modes in a strongly anisotropic , elongated bose - einstein condensate is based on the adiabatic separation ansatz@xcite @xmath206 where @xmath207 is the radial wavefunction characterized by the quantum number @xmath142 ( only zero angular momentum modes are considered here ) . the above ansatz incorporates the fact that for strongly elongated traps , i.e. , traps for which @xmath208 , the dynamics of the condensate motion separates into a fast radial motion and a slow axial motion , which are essentially independent . the @xmath209 are the mode functions for travelling wave solutions in the @xmath210 direction ( plane waves for a condensate at rest read @xmath211 $ ] ) . the radial motion is assumed to be `` stiff '' such that the radial part is effectively time independent , because the radial time scale for adjustment of the density distribution after a perturbation is much less than the axial oscillation time scales of interest . the ansatz ( [ sepansatz ] ) works independent from the ratio of healing length and radial size of the superfluid cigar . in the limit that the healing length is much less than the radial size , thomas - fermi wave functions are used , in the opposite limit , a gaussian ansatz for the radial part of the wave function @xmath207 is appropriate . the squared oscillation spectrum of the cigar - shaped condensate cloud reads @xmath212 , where @xmath213.@xcite a de sitter space - time is then obtained from the effective action for the phase fluctuations of the phonon ( @xmath214 ) modes near the center of the trap . the corresponding 1 + 1d action is obtained after integrating out the transverse , strongly confined directions , and reads:@xcite @xmath215 \nonumber\\ & \equiv & \frac12 \int d^{d+1}x % dt dz \sqrt{-{\sf g } } { \sf g}^{\mu\nu } \partial_\mu \chi_0 \partial_\nu \chi_0 \ , . \label{1daction}\end{aligned}\ ] ] the scaling parameters @xmath177 in the axial ( @xmath216 ) and @xmath217 in the perpendicular ( @xmath218 ) directions are functions of time , such that the action fulfills the identification with the action of a scalar field minimally coupled to gravity , as indicated in the second line above , where the metric coefficients are the @xmath0 in eq.([gdown ] ) . the identification of the above phase - fluctuations action with the action of a scalar minimally coupled to gravity works , that is , the two actions in the first and second lines of ( [ 1daction ] ) are consistent , if we impose the consistency condition that @xmath219 where @xmath220 is a renormalization factor according to @xmath221 , with @xmath222 the renormalized wave function , and @xmath223 the initial central density . the factor @xmath220 does not influence the ( classical ) equation of motion @xmath224 ( it simply drops out ) , but does influence the response of a detector . we will come back to this point in section [ desitterdetect ] below , when we discuss the explicit dependence on @xmath220 of the equilibration time scale for the stationary detector state considered , see eq.([tauequil ] ) . in refs.@xcite , we were using an alternative form of the 1 + 1d de sitter metric ( [ desittercomoving ] ) . this alternative form is the one used in the original gibbons - hawking paper.@xcite it reads ( we leave out the conformal factor ) @xmath225 the time interval @xmath226 in the above metric is not the `` cosmological '' time interval @xmath227 in the version of the metric displayed in ( [ desittercomoving ] ) . the two metrics may be transformed into each other using the following co - ordinate transformations : @xmath228 & = & \exp[-2\sqrt \lambda c_0 \tau_c ] -\lambda z_b^2 , % \quad \leftrightarrow \quad \tau = \tau ( \tau_c , z_b ) \nonumber\\ z & = & z_b \exp[\lambda c_0 \tau_c ] , \label{coordtrafo } % \quad \leftrightarrow \quad z = z ( t , z_b),\end{aligned}\ ] ] giving us @xmath229 , @xmath230 , which transforms ( [ desitterlineelement ] ) into ( [ desittercomoving ] ) . the advantage of the form ( [ desitterlineelement ] ) is that it is plain in this form that the de sitter space - time has an event horizon , located in our 1 + 1d case at the constant values @xmath231 . the quantity @xmath232 in the de sitter metric ( [ desitterlineelement ] ) must be independent of time . this leads us to the requirement @xmath233 relating the dynamical parameters of our problem to each other ( @xmath234 is the instantaneous sound velocity at the center of the cloud ) . these relations imply that @xmath235 const . , and thus that @xmath236 . then , the cosmological constant @xmath182 becomes independent of time , as is necessary for an analogue de sitter space - time to be established . the experimental procedure now is determined to be as follows : prepare a thomas - fermi ( i.e. , sufficiently large ) , strongly anisotropically trapped , cigar - shaped condensate . let it expand in the axial direction linear in lab time by changing the trapping according to the scaling equations ( [ genbeq ] ) , such that @xmath237 , _ and _ simultaneously expand in the perpendicular direction with the square root of lab time , @xmath238 , such that @xmath239 in ( [ cond1 ] ) is a constant . then , a detector `` tuned '' to the de sitter space - time ( [ desitterlineelement ] ) , i.e. , which works in the de sitter quasiparticle basis , will measure a thermal quasiparticle spectrum , with the de sitter temperature:@xcite @xmath240 the fact that a thermal spectrum is obtained can be directly derived from the equations of superfluid hydrodynamics , as expounded in refs.@xcite . that is , it is not simply postulated due to the ( kinematical ) analogy with quantum field theory in de sitter space - time , but is a result of the quantized hydrodynamic equations determining the evolution of the quasiparticle content of the sound field in the de sitter basis . the relation between de sitter time @xmath241 and the laboratory time is fixed by @xmath242 . [ note that @xmath226 and the scaling time interval @xmath243 defined in eq.([deftau ] ) differ ; @xmath244 . ] the transformation law between @xmath123 and the de sitter time @xmath241 ( on a constant @xmath210 detector , such that @xmath245 ) , is given by @xmath246 , \label{trafo}\ ] ] where the unit of lab time @xmath247 is set by the initial conditions for the scaling variables @xmath177 and @xmath217 . it is important to recognize that an effective exponential `` acceleration '' of the oscillation frequencies , either because of the exponential dependence of laboratory time on scaling time , represented by ( [ trafo ] ) , or coming from the wkb approximation for the scattering amplitudes , and the exponentially small mixing of positive and negative frequency parts resulting therefrom , eq.([alphat0beta ] ) , is sufficient for the thermal spectrum to be obtained . in particular , though we have discussed a space - time which possesses a horizon , no pre - existing horizons in the parent space - time are necessary _ per se _ for thermal occupation number distributions to be established , as has been pointed out in ref.@xcite . an explicit example is the unruh - davies effect , where this parent , global space - time is simply minkowski space . to detect the gibbons - hawking effect in de sitter space - time , one has to set up a detector which measures frequencies in units of the inverse de sitter time @xmath241 , rather than in units of the inverse laboratory time @xmath123 ; this corresponds to measuring in the proper de sitter quasiparticle vacuum , where , in particular , positive and negative frequency are defined with respect to @xmath241 . only then does one detect quasiparticles which are defined with respect to the de sitter quasiparticle basis , and refer to a vacuum corresponding to exactly that space - time . in refs.@xcite , i and petr fedichev have provided such a detector . we have shown that a `` de sitter basis '' detector is realized by an `` atomic quantum dot '' ( aqd ) ( for a detailed exposition of aqd properties cf . ref.@xcite ) . the aqd can be implemented in a bose gas of atoms possessing two hyperfine ground states @xmath248 and @xmath249 ; the level scheme is represented in fig.[aqd ] . the atoms in the state @xmath248 represent the expanding bose - einstein condensate , and are used to model the expanding de sitter universe . the aqd itself is formed by trapping atoms in the state @xmath249 in a tightly confining optical potential @xmath250 created by a laser at the center of the cloud . the interaction of atoms in the two internal levels is described by a set of coupling parameters @xmath251 ( @xmath252 ) , where @xmath253 are the @xmath254-wave scattering lengths characterizing short - range intra- and inter - species collisions ; @xmath255 , @xmath256 , and @xmath257 . the on - site repulsion between the atoms @xmath249 in the dot is given by the energy level spacing @xmath258 between states with a occupation difference of one @xmath249 atom , where @xmath259 is the characteristic size of the ground state wavefunction of atoms @xmath249 localized in @xmath250 . we consider the so - called collisional blockade limit of large @xmath260 , where only one atom of type @xmath249 can be trapped in the dot . this limit assumes that @xmath261 is much larger than all other relevant frequency scales in the dynamics of both the aqd and the expanding superfluid , and corresponds to a large `` coulomb blockade gap '' in electronic quantum dots.@xcite as a result of these assumptions , the collective co - ordinate of the aqd is modeled by a pseudo - spin-@xmath262 degree of freedom @xmath263 , with spin - up / spin - down state corresponding to occupation of the aqd by a single atom or no atom in the hyperfine state @xmath249 . a rabi laser of frequency @xmath264 , with a detuning @xmath265 from resonance between the two hyperfine levels @xmath248 and @xmath249 , couples atoms of the hyperfine species @xmath248 , constituting the expanding cigar - shaped superfluid , into the aqd . the detector lagrangian reads ( ref.@xcite , cf . the hamiltonian formulation in ref.@xcite ) : @xmath266 \bar \eta^ * \bar \eta\ , . \label{lbareta}\end{aligned}\ ] ] the fact that the detector has the de sitter basis as its `` natural '' quasiparticle basis , and therefore measures in de sitter time , is due to the fact that the term linear in the detector co - ordinate @xmath267 in the lagrangian couples in a certain manner to the superfluid cigar in which it is embedded . namely , the laser with frequency @xmath264 , causing transitions between the two hyperfine levels @xmath248 and @xmath249 , couples to the _ square root _ of the central mean - field particle density . this particular coupling ( represented by the second term in the first line of the above lagrangian ) is what we need , because @xmath268 . the fact that the coupling coefficient is proportional to @xmath269 is required to establish that the detector can work as a de sitter detector , because it transforms the detector equations , which are obviously _ a priori _ in laboratory time , into equations in de sitter time ; see the equations ( 43 ) for the temporal evolution of the detector level occupations below . adjusting the detuning @xmath265 properly , such that @xmath270 , the first and the second term in the square brackets in the second line of ( [ lbareta ] ) cancel . one then obtains a simple set of coupled equations for the occupation amplitudes of the state @xmath271 of the aqd : @xmath272 where @xmath226 is the _ de sitter time interval . _ were it not for the density oscillations in the cigar , represented by the potential @xmath273 , the above equations ( [ desitterevol ] ) would represent a simple two - level system , with frequency splitting @xmath274 . the density oscillations contained in the perturbation operator @xmath275 cause transitions between the two undisturbed _ detector eigenstates _ @xmath276 of the two - level system , which are separated by the energy @xmath277 . the density perturbations in the expanding host superfluid lead to a damping of the rabi oscillations with frequency @xmath277 between these two states . this constitutes the effect of the de sitter thermal bath to be observed , where the damping happens on the time scale displayed in ( [ tauequil ] ) below . the response of the detector , that is , the transition rates between the detector states , can be calculated by evaluating a response function@xcite which makes use of the expectation value of the product of two @xmath278 operators . the probability per unit time for excitation ( @xmath279 , transition from @xmath280 to @xmath281 ) and de - excitation ( @xmath282 , transition from @xmath283 to @xmath284 ) of the detector takes the form:@xcite @xmath285 the second - quantized solution of the hydrodynamic equations for the density fluctuations above the superfluid ground state in the expanding cigar - shaped bose - einstein condensate reads@xcite @xmath286 \right ) \ ! + { \rm h.c . } \label{deltarho } % { \hat a}^\dagger_{k } \exp\left[i\int \frac{dt \epsilon_{0k}}{bb^2}- ikz % \right\ } \nonumber\\\end{aligned}\ ] ] using this solution , and inserting into ( [ detectorresponse ] ) , we have shown@xcite that at late times @xmath241 the transition probabilities per unit de sitter , i.e. , per unit detector time satisfy detailed balance conditions . they correspond to _ thermodynamic equilibrium _ at the temperature @xmath287 displayed in ( [ tds ] ) : @xmath288 where the bose - planck distribution function takes the form familiar from thermostatistics : @xmath289 - 1]}.\ ] ] the frequency @xmath290 can be varied by changing the undressed rabi frequency @xmath264 , varying the intensity of the rabi laser . the detector thus has a changeable and therefore _ tunable _ frequency standard , which can be adjusted to scan the above distribution function for a given de sitter temperature @xmath287 . from the relation ( [ occup ] ) , we come to the remarkable conclusion that a properly designed detector can `` see '' a thermal equilibrium distribution in its quasiparticle basis , though it is embedded in a highly nonstationary system with respect to the laboratory frame . the rapidly expanding bose - einstein condensate represents this highly nonstationary system , which hosts the de sitter quasiparticle detector aqd . i stress here again that ( [ occup ] ) is an exact _ result _ obtained by quantizing hydrodynamic fluctuations in a nonstationary superfluid , and not just concluded from a mere comparison of the phonon dynamics in our expanding superfluid with the quantum field theory of photons in de sitter space - time . the equilibration time scale of the detector , and thus the time scale on which the rabi oscillations between the detector states are damped out , is set by the detector frequency standard ( the level spacing ) @xmath277 , and by the renormalization factor @xmath220:@xcite @xmath291 the renormalization factor @xmath220 contained in ( [ cond1 ] ) determines the equilibration rapidity because it physically expresses the strength of detector - field coupling . it is related to the initial diluteness parameter @xmath292 of the bose - einstein condensate and to the ratio @xmath293 , which determines inasmuch the system is effectively one - dimensional , @xmath294 . to obtain sufficiently fast equilibration , the condensate thus has to be initially not too dilute as well as close to the quasi-1d rgime , for which the transverse harmonic oscillator energy scale is of order the energy per particle , @xmath295 . these two conditions have another important implication . the ratio of the instantaneous coherence length @xmath296 and the location of the horizons @xmath231 , which are _ stationary _ in the present setup , has to remain less than unity within the equilibration time scale . if this is not the case , the coherence length , which plays the role of the analogue planck scale ( which is time dependent here ) , exceeds the length scale of the horizon at equilibration , and the concept of `` relativistic '' phonons propagating on a fixed curved space - time background with local lorentz symmetry becomes invalid . the ratio @xmath297 at the lab equilibration time scale @xmath298 $ ] following from the de sitter equilibration time in eq.([tauequil ] ) , expressed in parameters relevant to the experiment , is given by @xmath299.\ ] ] we see that this ratio changes exponentially with both the initial diluteness parameter @xmath300 and the quasi-1d parameter @xmath293 . in most currently realized bose - einstein condensates,@xcite the diluteness parameter @xmath301 . here , we initially need @xmath302 to have the condition @xmath303 fulfilled , assuming a reasonably large value of the de sitter temperature @xmath287 . though the condensate has to be _ initially _ quite dense , it is to be stressed that the central density decays like @xmath304 during expansion . therefore , the rate of three - body recombination losses quickly decreases during the expansion of the gas , and the initially relatively dense bose - einstein condensate , which would rapidly decay if left with a @xmath305 close to unity , can live sufficiently long , the total rate of three - body losses decreasing like @xmath306 . the primary statement to be drawn from the present article is that phonons , i.e. , low - energy linear - dispersion quasiparticles , moving in a spatially and temporally inhomogeneous bose - einstein - condensed superfluid gas , are equivalent to photons , the quanta of the electromagnetic field , moving on geodesics in a given curved space - time . we have explored the classical as well as the quantum aspects of this statement . on the classical side , the analogy helps to provide us with a simple general means to study quasiparticle propagation in an inhomogeneous medium in motion . an example of such an application is the gravitational lensing effect exerted by a superfluid vortex.@xcite on the quantum field theoretical side , we can access within the analogy phenomena which are extremely difficult if not impossible to access with light . one of the fundamentals of quantum field theory , the fact that the particle content of a quantum field depends on the observer , can thus be experimentally verified for the first time . the basic reason that the phenomena in question are ( comparatively ) easy to simulate in a condensed matter system is that the energy and temperature scales , under which they occur , relative to the typical energy scales of the system , can be changed at will by the experimentalist in a very controlled manner . more particularly , the temperature of the thermal spectrum of phonons to be measured in the gibbons - hawking effect can be made relatively large compared to the axial phonon frequencies and the actual temperature of the gas itself , by expanding the condensate cloud rapidly enough . in the condensed matter analogue , the typical energy of the quanta produced can in principle be even made to approach the relevant `` planck '' scale , i.e. , the point in the quasiparticle energy spectrum where it begins to deviate from being linear . finally , on a more adventurous side , one could conceive of carrying out experiments in `` experimental '' cosmology , as opposed to the currently existing purely `` observational '' cosmology . in such an experimental approach to matters cosmological , one would try to reproduce under certain specified and , in particular , well - defined initial conditions large - scale features of the cosmos , in the laboratory setting of nonstationary , inhomogeneous superfluid gases . misner , k.s . thorne , and j.a . wheeler , _ gravitation _ ( freeman , san francisco , 1973 ) . j.r . anglin and w. ketterle , nature * 416 * , 211 ( 2002 ) . khalatnikov , _ an introduction to the theory of superfluidity _ ( addison wesley , reading , ma , 1965 ) . n.d . birrell and p.c.w . davies , _ quantum fields in curved space _ ( cambridge university press , cambridge , england , 1984 ) . h. minkowski , physik . zeitschr . * 10 * , 104 ( 1909 ) . unruh , phys . * 46 * , 1351 ( 1981 ) . m. visser , class . quantum grav . * 15 * , 1767 ( 1998 ) . a. trautman , _ comparison of newtonian and relativistic theories of spacetime _ , in : `` perspectives in geometry and relativity '' ( indiana university press , bloomington , 1966 ) . fedichev and u.r . fischer , phys . rev . lett . * 91 * , 240407 ( 2003 ) . fedichev and u.r . fischer , phys . d * 69 * , 064021 ( 2004 ) . m. visser , gr - qc/0309072 . u.r . fischer and m. visser , ann ( n.y . ) * 304 * , 22 ( 2003 ) . fischer and m. visser , phys . * 88 * , 110201 ( 2002 ) . fischer and m. visser , europhys . 62 * , 1 ( 2003 ) . m. alcubierre , class . quantum grav . * 11 * , l73 ( 1994 ) . olum , phys . 81 * , 3567 ( 1998 ) . c. barcel , s. liberati , and m. visser , class . quantum grav . * 18 * , 3595 ( 2001 ) . leggett , rev . * 73 * , 307 ( 2001 ) ; _ ibid . _ * 75 * , 1083(e ) ( 2003 ) . e. madelung , z. phys . * 40 * , 322 ( 1927 ) . l. onsager , nuovo cimento suppl . * 6 * , 279 ( 1949 ) . perez bergliaffa , k. hibberd , m. stone , and m. visser , physica d * 191 * , 121 ( 2004 ) . bogoliubov , _ selected works , part ii : quantum and statistical mechanics _ ( gordon and breach , new york , 1991 ) . c. kittel , _ quantum theory of solids _ ( wiley and sons , new york , 2nd revised printing 1987 ) . volovik , _ the universe in a helium droplet _ ( oxford university press , oxford , 2003 ) . e.m . lifshitz and l.p . pitaevski , _ statistical physics , part 2 _ ( pergamon press , new york , 1980 ) . unruh , phys . d * 14 * , 870 ( 1976 ) . s.w . hawking and g.f.r . ellis , _ the large scale structure of space - time _ ( cambridge university press , cambridge , england , 1973 ) . gibbons and s.w . hawking , phys . d * 15 * , 2738 ( 1977 ) . m. visser , phys . * 80 * , 3436 ( 1998 ) . fulling , phys . d * 7 * , 2850 ( 1973 ) . e. yablonovitch , phys . rev . lett . * 62 * , 1742 ( 1989 ) . p. chen and t. tajima , phys . lett . * 83 * , 256 ( 1999 ) . kagan , e.l . surkov , and g.v . shlyapnikov , phys . a * 54 * , r1753 ( 1996 ) . y. castin and r. dum , phys . * 77 * , 5315 ( 1996 ) . kagan , e.l . surkov , and g.v . shlyapnikov , phys . lett . * 79 * , 2604 ( 1997 ) . s. inouye _ et al . _ , nature * 392 * , 151 ( 1998 ) . lett . * 89 * , 283202 ( 2002 ) . c. barcel , s. liberati , and m. visser , phys . a * 68 * , 053613 ( 2003 ) . fedichev and u.r . fischer , phys . a * 69 * , 033602 ( 2004 ) . landau , e.m . lifshitz , and l.p . pitaevski , _ quantum mechanics : non - relativistic theory _ ( 3rd edition , butterworth - heinemann , newton , ma , 1981 ) . s. weinberg , _ gravitation and cosmology _ ( wiley and sons , new york , 1972 ) . guth , phys . d * 23 * , 347 ( 1981 ) . lidsey , a.r . liddle , e.w . kolb , e.j . copeland , t. barreiro , and m. abney , rev . phys . * 69 * , 373 ( 1997 ) . l.p . pitaevski and a. rosch , phys . a * 55 * , r853 ( 1997 ) . f. chevy , v. bretin , p. rosenbusch , k.w . madison , and j. dalibard , phys . * 88 * , 250402 ( 2002 ) . fedichev , m.w . reynolds , and g.v . shlyapnikov , phys . lett . * 77 * , 2921 ( 1996 ) . e. zaremba , phys . rev . a * 57 * , 518 ( 1998 ) . b.l . hu and a. raval , mod . phys . lett . a * 11 * , 2625 ( 1996 ) . a. recati , p.o . fedichev , w. zwerger , j. von delft , and p. zoller , cond - mat/0404533 . s.m . reimann and m. manninen , rev . phys . * 74 * , 1283 ( 2002 ) .

the-particle partition function of a one - dimensional-function bose gas is calculated explicitly using only the periodic boundary condition ( the bethe ansatz equation ) . the-particles cluster integrals are shown to be the same as those by the thermal bethe ansatz method . + + keywords : thermal bethe ansatz method ,-function bose gas , integrable system , partition function , cluster expansion

cond-mat0212321

extending the work by lieb and liniger @xcite , yang and yang @xcite presented an ingenious method to study the equilibrium thermodynamics of one - dimensional system of bosons with repulsive @xmath1-function interaction . the hamiltonian of the system is @xmath2 where @xmath3 is the coupling constant and is assumed to be positive . throughout the paper , the planck constant and the mass of a particle are chosen to be @xmath4 and @xmath5 . this system is a first quantized version of the nonlinear schrdinger model described by the hamiltonian operator , @xmath6,\end{aligned}\ ] ] where @xmath7 and @xmath8 are boson field operators . the method invented by yang and yang is called the thermal bethe ansatz ( tba ) method since it starts from the bethe ansatz wave function @xcite and enables us to calculate the thermodynamic quantities at finite temperature . while there have been many successful applications to quantum particle and spin systems , not much study on the method itself has been done . a crucial assumption is the form of the entropy . thacker@xcite studied the @xmath1-function bose gas in infinite volume by the field - theoretic perturbation method and reproduced the results of the tba method . one of the authors ( m.w . ) @xcite presented a bosonic formulation of the tba method and calculated the grand partition function of the system at infinite volume . in this paper , we calculate explicitly the @xmath0-particle partition function of the hamiltonian ( [ eq : hamiltonian ] ) under the periodic boundary condition . that is , the assumptions of the tba method are not used and all the calculations are carried out exactly at finite volume . a preliminary result for @xmath9 has been reported in @xcite . the outline of the paper is the following . in 2 , we present a new method to evaluate the partition function for the @xmath0-particle system . a key idea is that we use only the periodic boundary condition of the wave function . in 3 , we calculate the @xmath0-particle cluster integrals using the partition function . an explicit calculation for the @xmath0-particle system has been done for the first time . the results are compared with those given by the tba . we prove that both results completely agree . the last section is devoted to concluding remarks . technical details of calculations are summarized in [ sec : proof_on_summation]@xmath10d . we study a statistical mechanics of the quantum @xmath0-particle system ( [ eq : hamiltonian ] ) . let @xmath11 , @xmath12 and @xmath13 denote the total energy , the system size and the wave numbers . it is known that the total energy and the wave numbers are determined by the following relations , @xmath14 where @xmath15 are integers satisfying the condition , @xmath16 and @xmath17 is the phase - shift for two - body scattering , @xmath18 the relation ( [ eq : bethe_2 ] ) is obtained from the periodic boundary condition imposed on the @xmath0-particle eigenfunction and is called the bethe ansatz equation . the range of the function ( [ eq : define_delta ] ) is assumed to be @xmath19 , and this phase - shift is not a `` true '' one in the sense of @xcite . that is , the phase - shift function ( [ eq : define_delta ] ) has analyticity on the real axis . the appearance of only two - body s - matrix in ( [ eq : boundary_condition ] ) implies the factorization of @xmath20-matrices , which is one of the remarkable properties of integrable systems . the partition function for the @xmath0-particle system is defined by @xmath21 where , with @xmath22 being the absolute temperature , @xmath23 and the summation is over all possible configurations of @xmath24 under the condition ( [ eq : condition_of_n ] ) . without recourse to the tba method , we calculate the partition function @xmath25 only by using the relations ( [ eq : bethe_1])@xmath10([eq : bethe_3 ] ) . we explain a method in three steps . in terms of new variable and function , @xmath26 the periodic boundary condition ( [ eq : bethe_2 ] ) is changed into @xmath27 we interpret ( [ eq : bethe_2_m1 ] ) as analytic relations between real numbers @xmath28 and @xmath13 . from the symmetry of ( [ eq : bethe_2_m1 ] ) , we see that , when @xmath29 corresponds to @xmath30 + @xmath31 , @xmath32 should correspond to @xmath33 . it is convenient to introduce the following set function , @xmath34 a domain @xmath35 of the function is a family of arbitrary finite sets . the first condition , @xmath36 , @xmath37 , @xmath38 , means that the sum of the elements in @xmath39 is @xmath35 . the second condition , @xmath40 , @xmath41 @xmath42 @xmath43 , @xmath44 , @xmath45 , means that the sum is a direct sum . that is , the image @xmath46 is all the families of sets whose direct sum is the set @xmath35 . if the argument is integer , we define @xmath47 for example , @xmath48 figure [ fig : theta ] illustrates @xmath49 in ( [ eq : example_theta ] ) . and we define a symbol , for arbitrary function @xmath50 , @xmath51 this symbol , @xmath52 , indicates that a function @xmath53 is summed up over all @xmath54 on the condition that @xmath55 if an element which contains both @xmath56 and @xmath57 exists . the following formula can be proved ( [ sec : proof_on_summation ] ) , @xmath58 where @xmath59 @xmath60 here and hereafter , the number of elements in a set @xmath35 is denoted by @xmath61 . by use of the formula([eq : sumrule_h ] ) , the partition function is rewritten as @xmath62 to evaluate ( [ eq : partition_function_m1 ] ) , we have a useful formula , @xmath63 this can be proved by the fourier transform and the jacobi s imaginary transformation for elliptic theta functions . applying the formula ( [ eq : summation_formula ] ) , we replace summations in ( [ eq : partition_function_m1 ] ) by integrals , @xmath64 to proceed further , we define four symbols . first , @xmath65 the set function @xmath66 consists of elements @xmath67 . @xmath67 contains elements @xmath68 , each of which has two elements @xmath69 . we call @xmath68 connection between the two elements @xmath69 . the connectivity is referred to as pattern of connections . @xmath66 represents a set of all the patterns of connections that have no ring of the connections . in fig.[fig : ring ] , the connection and ring of connections are illustrated . for example of @xmath70 , @xmath71 the pattern of connection @xmath72 is shown in fig . [ fig : pattern ] . as a special case , @xmath66 contains a pattern of no connection . i.e. @xmath73 . second , @xmath74 in words , @xmath75 is a number of elements which are linked with @xmath35 . third , @xmath76 that is , @xmath77 means a set of connection patterns each of which is a cluster of the connection pattern @xmath78 . here , the cluster means that if two connections @xmath68 and @xmath79 in @xmath78 are linked with a common element @xmath35 , then @xmath68 and @xmath79 are in the same cluster . for example , @xmath80 a pattern of no connection is included in @xmath77 . fourth , @xmath81 \right ) & \equiv & \left\ { \sigma \left| \begin{array}{c } \sigma\in\theta,\quad % { } ^{\forall } \lambda^{\prime}\in g\left(\lambda\right),\quad % { } ^{\forall } \xi\in \lambda,\quad \sigma\not\in \xi \\ { \rm or}\\ % { } ^\exists \lambda^{\prime}\in g\left(\lambda\right),\quad \lambda^{\prime}\neq\emptyset,\quad \sigma = \bigcup_{\left\{\sigma^{\prime},\sigma^{\prime\prime}\right\ } \in \lambda^{\prime } } \left ( \sigma^{\prime}\cup\sigma^{\prime\prime } \right ) \end{array } \right . \right\ } , \label{eq : define_g } \\ & & \theta = \left\{\sigma_1,\sigma_2,\cdots\right\ } , \lambda \in \lambda\left ( \theta \right).\nonumber\end{aligned}\ ] ] @xmath82 $ ] means that @xmath67 is consistent with the connection pattern of @xmath39 ; an element which is linked with @xmath67 is in @xmath39 . @xmath83 \right)$ ] indicates a set of direct sum `` elements of @xmath84 '' which are connected by one cluster in the connection pattern @xmath67 . the reason why an element is quoted here is that the element , at the same time , is a set . if an element of @xmath39 has no connection , the `` element '' belongs to @xmath83 \right)$ ] . for example , @xmath85\right ) & = & \left\ { \left\{1,2,3,4\right\ } , \left\{5,6,7,8\right\ } , \left\{9\right\ } \right\}.\end{aligned}\ ] ] the following formula is proved ( see a proof in [ sec : proof_on_jacobian ] ) : @xmath86\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right ) , \label{eq : jacobian_h}\end{aligned}\ ] ] with @xmath87\right ) & \equiv & l^{n-\sharp \lambda } \left ( \prod_{\sigma\in \theta } \sharp \sigma^{n\left(\sigma,\lambda\right)-1 } \right ) \left ( \prod_{\sigma^{\prime } \in g\left(\left[\lambda,\theta\right]\right ) } \sharp \sigma^{\prime } \right ) , \label{eq : define_i}\\ a_{\sigma,\sigma^{\prime}}\left(\theta\right ) & = & \left\ { \begin{array}{ll } l+\sum_{\sigma^{\prime\prime } \in \theta } \sharp \sigma^{\prime\prime } \ ; x_{\sigma , \sigma^{\prime\prime}}&{\rm if}\quad \sigma = \sigma^{\prime}\\ -\sharp \sigma^{\prime } \;x_{\sigma,\sigma^{\prime } } & { \rm if}\quad \sigma \neq\sigma^{\prime } , \end{array } \right . \label{eq : define_a}\end{aligned}\ ] ] @xmath88 using the formula([eq : jacobian_h ] ) , we change variables in ( [ eq : partition_function_m2 ] ) from @xmath28 to @xmath13 . in terms of the above introduced symbols , we obtain @xmath89\right ) \left ( \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } \frac{d\delta}{dk}\left(k_{\sigma}-k_{\sigma^{\prime}}\right ) \right ) \nonumber\\&&\times { } \exp \left [ \sum_{\sigma\in\theta } \left\ { -\beta \#\sigma k_{\sigma}^2 -i n_{\sigma } \left ( lk_{\sigma } -\sum_{\sigma^{\prime}\in\theta,\neq \sigma } \#\sigma^{\prime}\tilde{\delta}\left(k_{\sigma^{\prime}}-k_{\sigma}\right ) \right ) \right\ } \right ] . \label{eq : partition_function_m3}\end{aligned}\ ] ] @xmath90 is defined in ( [ eq : define_f ] ) . note that because of a relation , if @xmath91 , then @xmath92 , the jacobian of the transformation can be written explicitly . we change the integral paths in ( [ eq : partition_function_m3 ] ) from @xmath93 to @xmath94 . in what follows , ` an integral path steps over a residue ' means that when an integral path is moved , there is a residue in the region surrounded by the initial and final integral paths . it is important to note that each integral path in ( [ eq : partition_function_m3 ] ) does not step over any residue by this change of the integral path . performing the change of integral paths as introduced above , we arrive at an expression of the partition function , @xmath95\right ) s\left(\left[\lambda,\theta\right]\right ) \label{eq : partition_function}\end{aligned}\ ] ] with @xmath96\right ) & \equiv & \frac{l^{n^{\prime}-\sharp \lambda } } { n^{\prime } ! } \left ( \prod_{\sigma^{\prime } \in g\left(\lambda,\theta\right ) } \sharp \sigma^{\prime } \right ) \left ( \prod_{\sigma\in \theta } \sharp \sigma^{n\left(\sigma,\lambda\right)-1 } \left(-1\right)^{\left(\sharp \sigma -1\right ) } \left(\sharp \sigma -1\right ) ! \right ) , \label{eq : define_j } \\ % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % \!\!\!\!\!\!\!\!\!\!\!\ ! s\left(\left[\lambda,\theta\right]\right ) & \equiv & \sum_{\left\{n_{\sigma}\right\},\sigma\in\theta } ( -1)^ { \left(n^{\prime}-1\right ) \sum_{\sigma\in\theta}n_{\sigma } } \exp \left ( -\frac{l^2}{4\beta } \sum_{\sigma\in\theta } \frac{n_{\delta}^2}{\#\delta } \right ) \nonumber\\&&\times { } \int\prod_{\sigma\in\theta}\frac{dk_{\sigma}}{2\pi } \left ( \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } \frac{d\delta}{dk } \left ( k_{\sigma } -k_{\sigma^{\prime } } -\frac{ln_{\sigma}}{2\beta\#\sigma}i + \frac{ln_{\sigma^{\prime}}}{2\beta\#\sigma^{\prime}}i \right ) \right ) \nonumber\\&&\times { } \exp \left ( \sum_{\sigma\in\theta } \left ( \begin{array}{c } -\beta \#\sigma k_{\sigma}^2\\ + in_{\sigma}\sum_{\sigma^{\prime}\in\theta,\neq \sigma } \#\sigma \tilde{\delta } \left ( k_{\sigma^{\prime } } -k_{\sigma } -\frac{ln_{\sigma}}{2\beta\#\sigma}i + \frac{ln_{\sigma^{\prime}}}{2\beta\#\sigma^{\prime}}i \right ) \end{array } \right ) \right ) , \label{eq : define_s}\\ n^{\prime}&\equiv & \sum_{\sigma\in\theta}\#\sigma . \label{eq : define_n } \end{aligned}\ ] ] this formula enables us to calculate the partition function in any order of the system size @xmath12 . in other way of writing , we can calculate the partition function for the system with finite size as @xmath97 it should be remarked that @xmath98 for arbitrary @xmath99 is explicitly obtained from ( [ eq : partition_function])-([eq : define_n ] ) . the cluster expansion for the equation of state is defined by @xmath100 where @xmath101 is the fugacity ( rigorously , the absolute activity ) , @xmath102 the partition function @xmath103 and the cluster integral @xmath104 are related as @xmath105 expanding the right hand side of eq.([eq : grand_partition_function ] ) in powers of @xmath101 and combinatorially summing up the coefficients , we can prove the relation ( [ eq : cluster_integral ] ) . we define that @xmath106 @xmath107 is a set of connection pattern which consists of one cluster ( this cluster is nothing to do with the cluster integral ) . substituting the expression ( [ eq : partition_function ] ) of @xmath25 into ( [ eq : cluster_integral ] ) , we obtain @xmath108\right ) s_{\infty}\left(\left[\lambda,\theta\right]\right ) , \label{eq : cluster_integral_21}\end{aligned}\ ] ] where @xmath109 is defined in ( [ eq : define_j ] ) , and @xmath110\right ) & \equiv & \int\prod_{\sigma\in\theta}\frac{dk_{\sigma}}{2\pi } \left ( \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } \frac{d\delta}{dk } \left ( k_{\sigma } -k_{\sigma^{\prime } } \right ) \right ) \exp \left ( \sum_{\sigma\in\theta } \left ( -\beta \#\sigma k_{\sigma}^2 \right ) \right ) . \label{eq : define_s_inf}\end{aligned}\ ] ] this is proved in [ sec : calculation_of_cluster_integral ] . to compare the results in this paper with those in @xcite , we define a function @xmath111 by @xmath112 note that for the non - interacting case , @xmath1130 . in term of @xmath111 , the cluster integral @xmath114 can be expressed as follows , @xmath115\right ) s^{\prime}_{\infty}\left(\left[\lambda,\theta\right]\right ) , \label{eq : cluster_integral_22}\end{aligned}\ ] ] with @xmath116\right ) & \equiv & \frac{l } { \left(n-1\right ) ! } \left ( \prod_{\sigma\in \theta } \sharp \sigma^{n\left(\sigma,\lambda\right)-1 } \left(\sharp \sigma -1\right ) ! \right ) , \label{eq : define_j_prime}\\ s^{\prime}_{\infty}\left(\left[\lambda,\theta\right]\right ) & \equiv & \int\prod_{\sigma\in\theta}\frac{dk_{\sigma}}{2\pi } \left ( \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } k \left ( k_{\sigma } -k_{\sigma^{\prime } } \right ) \right ) \exp \left ( \sum_{\sigma\in\theta } \left ( -\beta \#\sigma k_{\sigma}^2 \right ) \right ) . \label{eq : define_s_prime_inf}\end{aligned}\ ] ] a proof of eq.([eq : cluster_integral_22 ] ) is given also in [ sec : calculation_of_cluster_integral ] . the cluster integrals ( [ eq : cluster_integral_22 ] ) agree with those derived by the tba method ( see [ sec : calculation_of_cluster_integral_with_tba_method ] for detail calculation ) . in this way , we have proved that the thermal bethe ansatz ( tba ) method by yang and yang gives the exact equation of state . taking a one - dimensional @xmath1-function bose gas as a typical example of integrable systems , we have derived the @xmath0-particle partition function . a method in this paper , referred to as the direct method , is an exact analysis of the partition function only based on the periodic boundary condition . using the explicit expression of the partition function , we have calculated the @xmath0-particle cluster integral , and proved a perfect agreement between the results of this direct method and the thermal bethe ansatz ( tba ) method . the extensions and applications of the direct method to integrable and non - integrable systems may clarify mathematical structures of the tba method . those problems are left for future studies . one of authors ( g.k ) likes to express his sincere thanks to t.sasamoto , y.komori and a.nishino for valuable comments and stimulating discussions . we prove ( [ eq : sumrule_h ] ) , that is , @xmath117 where @xmath90 and @xmath118 are defined in ( [ eq : define_f ] ) and ( [ eq : define_sigma ] ) , and @xmath119 satisfies @xmath60 we define a semiorder on a set @xmath120 , @xmath121 a sufficient condition of eq.([eq : sumrule ] ) is @xmath122 where @xmath123 we consider a function , @xmath124 and a mapping @xmath125 , @xmath126 which satisfies the following relations @xmath127 where @xmath128 are arbitrary polynomial functions , and @xmath129 is defined in ( [ eq : define_g ] ) . it is readily shown that the following relation holds , @xmath130 if we substitute @xmath131 for @xmath132 , the relation ( [ eq : sumrule_1 ] ) becomes @xmath133 eq.([eq : tmp_a ] ) is the sufficient condition ( [ eq : sumrule_suf_1 ] ) of eq.([eq : sumrule ] ) . thus , ( [ eq : sumrule_h ] ) is proved . we prove ( [ eq : jacobian_h ] ) , that is , @xmath86\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right ) , \label{eq : jacobian}\end{aligned}\ ] ] where @xmath134 , @xmath135 , @xmath136 , @xmath137 , and @xmath138\right)$ ] and @xmath139 are defined in ( [ eq : define_i ] ) and ( [ eq : define_a ] ) . only the terms @xmath140 , @xmath141 , @xmath142 and @xmath143 contain the variable @xmath144 , and the minor determinant becomes @xmath145 where @xmath146 the right hand side of ( [ eq : tmp_3 ] ) contains only the terms of @xmath144 and @xmath147 when we regard the equation as a polynomial of @xmath144 . therefore , an exponent of a variable @xmath144 in the determinant ( [ eq : jacobian ] ) is 1 or 0 . we define a set , @xmath148 while @xmath66 ( [ eq : define_theta_tilde ] ) does not include @xmath67 which contains rings of connections , @xmath149 include such kind of @xmath67 . in terms of this set , the determinant is written as @xmath150\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right ) . \label{eq : determinant_1}\end{aligned}\ ] ] here @xmath151\right)$ ] is an undetermined function which is projected on a polynomial of @xmath12 . we make an assumption : @xmath151\right)$ ] is not zero on the condition of @xmath67 that there exists @xmath152 such that @xmath153 this assures that the connection pattern @xmath67 contains the rings of connections . we define a set @xmath154 we can choose @xmath155 such that @xmath156 this means that @xmath155 consists of some cluster of @xmath67 , and @xmath155 contains some rings of connections . in this case , it is clear that @xmath157\right ) & = & i^{\prime}\left(\left[\lambda_r,\theta\right]\right ) i^{\prime}\left(\left[\bar{\lambda}_r,\theta\right]\right ) l^{-n},\end{aligned}\ ] ] and that @xmath158\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } + \sum_{\lambda\in \lambda'\left(\theta\right ) , \lambda\neq\lambda_r } \left ( i^{\prime\prime}\left(\left[\lambda,\theta\right]\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right),\end{aligned}\ ] ] where @xmath159 @xmath160\right)$ ] is an undetermined function . from ( [ eq : tmp_4 ] ) , we can see that @xmath161 which indicates that row vectors of the matrix are linearly dependent ; @xmath162 is identically zero . this negates the assumption that @xmath151\right)$ ] is not zero on the condition ( [ eq : condition_a ] ) . therefore , we have @xmath163\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right ) . \label{eq : determinant_2}\end{aligned}\ ] ] a difference between ( [ eq : determinant_1 ] ) and ( [ eq : determinant_2 ] ) is the region of summation over @xmath67 . we notice that each term of the right hand side is not made from off - diagonal elements of @xmath139 ; all the terms which are made from off - diagonal elements are canceled by diagonal elements . then , we have @xmath164\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right ) \nonumber\\ & & { } + \sum_{\lambda\in \lambda'\left(\theta\right ) } \sum_{\sigma_1,\sigma_2\in \theta } i^{\left(3\right ) } \left(\left[\lambda,\theta\right],\sigma_1,\sigma_2\right ) x_{\sigma_1,\sigma_2}^2 \prod_{\left\{\sigma_3,\sigma_4\right\}\in \lambda } x_{\sigma_3,\sigma_4 } \nonumber\\ & & { } + \sum_{\lambda\in \lambda'\left(\theta\right ) , \not\in\lambda\left(\theta\right ) } \left ( i^{\left(4\right ) } \left(\left[\lambda,\theta\right]\right ) \prod_{\left\{\sigma,\sigma^{\prime}\right\}\in \lambda } x_{\sigma,\sigma^{\prime } } \right ) , \label{eq : determinant_3}\end{aligned}\ ] ] where @xmath165 and @xmath166,\sigma_1,\sigma_2\right)$ ] and @xmath167\right)$ ] are undetermined functions . therefore , we obtain @xmath168\right ) & \equiv & l^{n-\sharp \lambda } \left ( \prod_{\sigma\in \theta } \sharp \sigma^{n\left(\sigma,\lambda\right)-1 } \right ) \left ( \prod_{\sigma^{\prime } \in g\left(\lambda,\theta\right ) } \sharp \sigma^{\prime } \right ) \nonumber\\ & = & i\left(\left[\lambda,\theta\right]\right ) , \label{eq : tmp_proof_b_2}\end{aligned}\ ] ] and ( [ eq : determinant_2 ] ) with ( [ eq : tmp_proof_b_2 ] ) proves ( [ eq : jacobian ] ) . first , we prove ( [ eq : cluster_integral_21 ] ) . substituting the expression of @xmath25 in ( [ eq : partition_function ] ) into ( [ eq : cluster_integral ] ) yields @xmath169\right ) s\left(\left[\lambda^{\prime},\theta^{\prime}\right]\right ) . \label{eq : cluster_integral_1}\end{aligned}\ ] ] @xmath170\right)$ ] and @xmath171\right)$ ] are defined in ( [ eq : define_j ] ) and ( [ eq : define_s ] ) . eliminating exponentially decreasing terms with respect to @xmath12 , we have @xmath172\right ) s_{\infty}\left(\left[\lambda^{\prime},\theta^{\prime}\right]\right ) . \label{eq : cluster_integral_2}\end{aligned}\ ] ] the integral @xmath173 is defined in ( [ eq : define_s_inf ] ) . we define @xmath174\right ) & \equiv & \left\ { \left[\lambda',\theta'\right ] \left| \begin{array}{c } \lambda ' = \emptyset , \quad \ # \theta'=1 , \quad \theta ' \subset\theta , \quad \xi\in\lambda , \quad \sigma\not\in\xi\\ { \rm or}\\ \lambda'\in g\left(\lambda\right ) , \quad \lambda ' \neq \emptyset , \quad \sigma\in\theta ' , \quad \xi\in\lambda ' , \quad \sigma\in\xi \end{array } \right . \right\}\end{aligned}\ ] ] in words , @xmath175\right)$ ] is a set of elements each of which consists of connection pattern @xmath152 and a set @xmath176 . here , @xmath152 is a cluster of the connection pattern @xmath67 , and @xmath176 is a subset of @xmath39 and is a set of element which is linked with connections in @xmath152 . note that @xmath175\right)$ ] contains @xmath177}$ ] when @xmath35 is not connected by @xmath67 . from the definition , it is shown that the integral @xmath178\right)$ ] is factorized into `` connected '' integrals as @xmath110\right ) & = & \prod _ { \left[\lambda^{\prime},\theta^{\prime}\right ] \in \tilde{g}\left(\left[\lambda,\theta\right]\right ) } s_{\infty}\left(\left[\lambda^{\prime},\theta^{\prime}\right]\right ) . \label{eq : facorize_equation_s}\end{aligned}\ ] ] similarly , the coefficient @xmath170\right)$ ] is factorized : @xmath179\right ) & = & \prod _ { \left[\lambda^{\prime},\theta^{\prime}\right ] \in \tilde{g}\left(\left[\lambda,\theta\right]\right ) } \left ( \sum_{\sigma'\in\theta'}\#\sigma ' \right ) ! j_{\infty}\left(\left[\lambda^{\prime},\theta^{\prime}\right]\right ) . \label{eq : facorize_equation_j}\end{aligned}\ ] ] due to ( [ eq : facorize_equation_s ] ) and ( [ eq : facorize_equation_j ] ) , the cluster integral ( [ eq : cluster_integral_2 ] ) can be rewritten as @xmath180 \in \tilde{g}\left(\left[\lambda',\theta'\right]\right ) } \left(\sum_{\sigma''\in\theta''}\#\sigma''\right ) ! j\left(\left[\lambda'',\theta''\right]\right ) s_{\infty}\left(\left[\lambda'',\theta''\right]\right ) . \label{eq : cluster_integral_2a}\end{aligned}\ ] ] we can show that for arbitrary function @xmath181\right)$ ] , @xmath182 \in \tilde{g}\left(\left[\lambda,\theta\right]\right ) } f\left(\left[\lambda^{\prime},\theta^{\prime}\right]\right ) & = & \sum_{\theta''\in\theta\left(n\right ) } \prod_{\sigma''\in\theta '' } \sum_{\theta'\in\theta\left(\sigma''\right ) } \sum _ { \lambda^{\prime } \in \lambda_c\left(\theta'\right ) } f\left(\left[\lambda^{\prime},\theta^{\prime}\right]\right ) . \label{eq : combinatorial_c1}\end{aligned}\ ] ] the left hand side of ( [ eq : combinatorial_c1 ] ) is a summation over all the patterns which are generated by the following process : first divide a set @xmath183 into elements of @xmath39 , then connect them with @xmath67 . the right hand side of ( [ eq : combinatorial_c1 ] ) is a summation over all the patterns which generated by the following process : first divides a set @xmath183 into sets of @xmath184 each of which is a direct sum of elements connected by a cluster @xmath152 , second divide each of the sets into elements of @xmath176 , then define connection pattern @xmath152 , which consists of one cluster , of elements in @xmath176 . fig.[fig : combi_1l ] and fig.[fig : combi_1r ] illustrate graphical representations of both sides of patterns in ( [ eq : combinatorial_c1 ] ) . due to ( [ eq : combinatorial_c1 ] ) , the cluster integral ( [ eq : cluster_integral_2a ] ) becomes @xmath185\right ) s_{\infty}\left(\left[\lambda'',\theta''\right]\right ) . \label{eq : cluster_integral_3}\end{aligned}\ ] ] we define a family of sets @xmath186 , @xmath187 where an element of @xmath188 is also a family of sets . to repeat , an element of @xmath186 is a sum of sets of which a family of sets , an element of @xmath188 , consists . for example , @xmath189 we can show that for arbitrary functions @xmath190 , @xmath191 , @xmath192 in the left hand side of ( [ eq : combinatorial_c2 ] ) , one sums up all the patterns which are generated by the following process : first divide @xmath183 into elements in @xmath184 , then make family of sets @xmath188 . each of the sets consist of some elements in @xmath184 , and @xmath193 . in the right hand side of ( [ eq : combinatorial_c2 ] ) , one sums up all the patterns which are generated by the following process : first divide a set @xmath183 into elements of @xmath39 , then divide each of elements in @xmath39 into elements of @xmath176 fig.[fig : combi_2l ] and fig.[fig : combi_2r ] illustrate graphical representations of both sides of patterns in ( [ eq : combinatorial_c2 ] ) . using a formula ( [ eq : combinatorial_c2 ] ) in ( [ eq : cluster_integral_2a ] ) , @xmath114 becomes @xmath194\right ) s_{\infty}\left(\left[\lambda'',\theta''\right]\right ) . \right ) \nonumber\\ & & { } \sum_{\tau\in\theta\left(\theta\right ) } \frac{\left(\#\tau-1\right)!\left(-1\right)^{\#\tau-1 } } { ln ! } \label{eq : cluster_integral_4}\end{aligned}\ ] ] note that we have used a relation @xmath195 . substitutions of @xmath196 into @xmath25 in ( [ eq : cluster_integral ] ) and ( [ eq : grand_partition_function ] ) give a relation , @xmath197 using this relation in ( [ eq : cluster_integral_4 ] ) , we finally obtain @xmath108\right ) s_{\infty}\left(\left[\lambda,\theta\right]\right ) , \label{eq : cluster_integral_5}\end{aligned}\ ] ] which is ( [ eq : cluster_integral_21 ] ) . note that the right hand side of eq.([eq : cluster_integral_5 ] ) does not depend on @xmath12 . therefore , we do not write @xmath198 any more . in the no - interaction limit , where @xmath199 is replaced with @xmath200 , eq.([eq : cluster_integral_5 ] ) gives @xmath201 next , we prove ( [ eq : cluster_integral_22 ] ) . we define @xmath202,\lambda'\right ) & = & [ \lambda'',\theta'']\nonumber\\ & \equiv & \left [ \left\ { \left\ { \sigma_1 , \sigma_2 \right\ } \left| \begin{array}{c } \sigma_1,\sigma_2\in\theta''\\ \sigma_1\in \sigma_3\quad \sigma_2\in \sigma_4\\ \left\ { \sigma_3 , \sigma_4 \right\ } \in\lambda,\not\in\lambda '' \end{array } \right . \right\ } , g\left(\left[\lambda',\theta\right]\right ) \right]\\ & & \lambda'\subseteq\lambda\nonumber\end{aligned}\ ] ] @xmath184 is a set of a direct sum of elements linked with one cluster in @xmath152 , and @xmath203 is a connection pattern of a set @xmath184 . a connection in @xmath203 links two elements of @xmath184 , where elements of @xmath39 , each of which is a subset of one of the two elements , are connected by @xmath67 . for example , @xmath204,\lambda_9 \right ) & = & \left[\lambda'_9,\theta'_9\right ] , \label{eq : tmp_proof_c_1}\end{aligned}\ ] ] where @xmath205 fig.[fig : proof_c1 ] and fig.[fig : proof_c2 ] illustrate @xmath206 $ ] and @xmath207 $ ] in ( [ eq : tmp_proof_c_1 ] ) . by substituting @xmath208 into @xmath199 , ( [ eq : cluster_integral_5 ] ) is changed into @xmath209\right ) \sum_{\lambda'\subseteq\lambda } s^{\prime}_{\infty } \left ( d\left(\left[\lambda,\theta\right],\lambda'\right ) \right),\nonumber\end{aligned}\ ] ] where @xmath210\right)$ ] is defined in ( [ eq : define_s_prime_inf ] ) . with the explicit expression of @xmath170\right)$ ] , we have @xmath211,\lambda'\right ) \right ) \nonumber\\ & & \prod_{\sigma\in\theta } \#\sigma^{n\left(\sigma,\lambda\right)-1 } \left(-1\right)^{\left(\#\sigma-1\right ) } \left(\#\sigma-1\right)!. \label{eq : sum_relation_11_a}\end{aligned}\ ] ] we can show that for arbitrary functions @xmath212 and @xmath213\right)$ ] , @xmath214,\lambda''\right ) \right ) \prod_{\sigma\in\theta } f_1\left(n\left(\sigma,\lambda\right),\sigma\right)\nonumber\\ & = & \sum_{\theta'\in\theta\left(n\right ) } \sum_{\lambda'\in \lambda_c\left(\theta'\right ) } f_2\left(\left[\lambda',\theta'\right]\right ) \prod_{\sigma'\in\theta ' } \sum_{\theta''\in\theta\left(\sigma'\right ) } \sum_{\lambda''\in \lambda_c\left(\theta''\right ) } \nonumber\\ & & \sum_{\sigma_1,\cdots,\sigma_{n\left(\sigma,\lambda\right)}\in\theta '' } \prod_{\sigma\in\theta '' } f_1\left(n\left(\sigma,\lambda''\right ) + \sum_{i=1}^{n\left(\sigma,\lambda\right)}\delta_{\sigma,\sigma_i},\sigma\right ) . \label{eq : combinatorial_c3}\end{aligned}\ ] ] the meaning of the left hand side of ( [ eq : combinatorial_c3 ] ) is to sum up all the patterns generated by the following process : first , one divides a set into elements of @xmath39 , second defines a connection pattern @xmath67 , then makes @xmath176 and @xmath152 to join elements of @xmath39 together which are connected with @xmath203 . the meaning of the right hand side of ( [ eq : combinatorial_c3 ] ) is to sum up all the patterns generated by the following process : first one divides a set into elements of @xmath176 , second makes a connection pattern @xmath152 , third divides each elements of the set @xmath176 into elements of @xmath184 , fourth defines connection pattern @xmath203 on elements of @xmath184 each of which is a subset of an element in @xmath176 , then connects elements each of which is a subset of one of two elements in @xmath176 linked with a connection pattern @xmath152 . fig.[fig : combi_3l ] and fig.[fig : combi_3r ] show graphical representations of both sides of patterns in ( [ eq : combinatorial_c3 ] ) . ( [ eq : combinatorial_c3 ] ) is similar to ( [ eq : combinatorial_c2 ] ) , in the sense that if @xmath212 and @xmath213\right)$ ] in eq.([eq : combinatorial_c3 ] ) are independent of @xmath99 and @xmath67 ; there is no connection pattern dependence , then eq.([eq : combinatorial_c3 ] ) becomes ( [ eq : combinatorial_c2 ] ) . by use of ( [ eq : combinatorial_c3 ] ) , we rewrite ( [ eq : sum_relation_11_a ] ) as @xmath215\right ) \prod_{\sigma\in\theta } \sum_{\theta'\in \theta\left(\sigma\right ) } \sum_{\lambda'\in \lambda_c\left(\theta'\right ) } \nonumber\\ & & { } \sum_{\sigma'_1,\cdots,\sigma'_{n\left(\lambda,\theta\right)}\in\theta ' } \prod_{\sigma'\in\theta ' } \sharp\sigma^{\prime n\left(\sigma',\lambda'\right)+\sum_{i=1}^{n\left(\sigma,\lambda\right)}\delta_{\sigma',\sigma'_i}-1 } \left(-1\right)^{\#\sigma'-1 } \left(\sharp \sigma ' -1\right)!. \nonumber\end{aligned}\ ] ] since ; @xmath216 @xmath114 becomes @xmath217\right ) \frac{1}{\left(n-1\right ) ! } \prod_{\sigma\in\theta } \sharp \sigma^{n\left(\sigma,\lambda\right ) } \nonumber\\ & & { } \sum_{\theta^{\prime}\in \theta\left(\sigma\right ) } \sum _ { \lambda^{\prime } \in \lambda_c\left(\theta^{\prime}\right ) } \left ( \prod_{\sigma^{\prime}\in \theta^{\prime } } \sharp { \sigma^{\prime}}^{n\left(\sigma^{\prime},\lambda^{\prime}\right)-1 } \left(-1\right)^{\#\sigma^{\prime}-1 } \left(\sharp \sigma^{\prime } -1\right ) ! \label{eq : sum_relation_11_b}\end{aligned}\ ] ] using the relation([eq : sum_relation_2 ] ) , we arrive at ( [ eq : cluster_integral_22 ] ) , @xmath115\right ) s^{\prime}_{\infty}\left(\left[\lambda,\theta\right]\right),\end{aligned}\ ] ] where @xmath218\right)$ ] is defined in ( [ eq : define_j_prime ] ) . a result of the bosonic formulation of the tba @xcite is @xmath219 where @xmath220 we define a function , @xmath221 and prove that @xmath222,k,\left\{0\right\}\right ) \label{eq : expansion_of_self_enargy } , \quad\quad \\ f_0\left(k\right)&=&1 , \label{eq : expansion_of_self_enargy_0}\end{aligned}\ ] ] where for @xmath223 , @xmath224 , @xmath225,k,\sigma\right ) & \equiv & \int\prod_{\sigma'\in \theta,\sigma'\neq\sigma}\frac{dk_{\sigma'}}{2\pi } \left ( \prod_{\left\{\sigma',\sigma''\right\}\in\lambda } k\left(k_{\sigma'-k_\sigma''}\right ) \right ) \nonumber\\&&{}\times \exp \left ( -\beta\sum_{\sigma'\in\theta,\sigma'\neq\sigma } \#\sigma ' k^2_{\sigma ' } \right).\end{aligned}\ ] ] from eq.([eq : self_enargy ] ) , a recursive relation for @xmath226 is obtained as @xmath227 substituting the expression of @xmath226 in ( [ eq : expansion_of_self_enargy ] ) into @xmath228 in ( [ eq : f_recusive_formula ] ) , we obtain @xmath229,q_\sigma,\left\{0\right\}\right ) . \label{eq : f_recusive_formula_e-}\end{aligned}\ ] ] for non - integer argument of @xmath46 , ( [ eq : f_recusive_formula_e- ] ) becomes @xmath230,q_\sigma,\left\{n\right\}\right ) . \label{eq : f_recusive_formula_e}\end{aligned}\ ] ] we can show that @xmath231,\left\{n\right\}\right ) \prod_{\sigma'\in\theta''',\sigma'\neq\left\{n\right\ } } f_3\left(n\left(\sigma',\lambda'''\right),\sigma'\right ) \nonumber\\ & = & \sum_{\theta\in\theta\left(n\right ) } \sum _ { \lambda\in\lambda_c\left(\theta\right ) } \sum_{\sigma\in\theta } \#\sigma^{n\left(\sigma,\lambda\right ) } f_1\left(\#\sigma\right ) f_2\left(\left[\lambda,\theta\right],\sigma\right ) \prod_{\sigma'\in\theta,\sigma'\neq\sigma } f\left(n\left(\sigma',\lambda\right),\sigma'\right ) \label{eq : combinatorial_d4}\end{aligned}\ ] ] where @xmath232 and @xmath233 are arbitrary functions , and @xmath213\right)$ ] is a function which satisfies @xmath234,\sigma\right ) & = & \prod_{i}f_2\left(\left[\lambda_i,\theta_i\right],\sigma_i\right)\end{aligned}\ ] ] @xmath235 the meaning of the right hand side of ( [ eq : combinatorial_d4 ] ) is to sum up all the patterns generated by the following process : first , divide a set @xmath183 into sets of @xmath39 , second define a connection pattern @xmath67 which consists of one cluster , then choose one set @xmath35 in @xmath39 . the meaning of the left hand side of ( [ eq : combinatorial_d4 ] ) is to sum up all the patterns generated by the following process : first , divide @xmath183 into sets of @xmath184 each of which contain one element @xmath99 in the @xmath35 chosen in r.h.s . and elements of which some sets in @xmath39 consists . the @xmath35 chosen in r.h.s . is not one of these sets , and these sets are connected with each other by connections in @xmath67 and any sets are connected with the @xmath35 by connections in @xmath67 . second , divide each of sets in @xmath184 into sets of @xmath176 and @xmath236 . then , define a connection pattern @xmath203 of sets in @xmath237 and @xmath236 . the connection pattern @xmath203 consists of one cluster . fig.[fig : combi_3l ] and fig.[fig : combi_3r ] illustrate graphical representations of both sides of patterns in ( [ eq : combinatorial_c3 ] ) . note that if one chooses a r.h.s . pattern , there exist the corresponding @xmath238 l.h.s . patterns . applying the relation ( [ eq : combinatorial_d4 ] ) to ( [ eq : f_recusive_formula_e ] ) , we have @xmath239,q_\sigma,\sigma'\right ) \nonumber\\ & & { } \prod_{\sigma''\in\theta',\sigma''\neq\sigma ' } \left(\#\sigma''-1\right ) ! \#\sigma^{\prime\prime n\left(\sigma'',\lambda'\right)-1}. % % % % % % % % % % % \nonumber\\ & = & \frac{1}{n ! } \sum_{\theta\in\theta\left(n\right ) } \prod_{\sigma\in\theta } \sum_{\theta^{\prime}\in\theta\left(\sigma\right ) } \sum_{\lambda'\in\lambda_c\left(\theta'\right ) } \sum_{\sigma'\in\theta ' } \nonumber\\ & & { } \int\frac{dq_{\sigma'}}{2\pi}k\left(k - q_{\sigma'}\right ) \#\sigma^{\prime n\left(\sigma',\lambda'\right ) } \left(\#\sigma'-1\right ) ! \exp\left(-\beta \#\sigma ' q^2_{\sigma'}\right ) \tilde{s}\left(\left[\lambda',\theta'\right],q_{\sigma'},\sigma'\right ) \nonumber\\ & & { } \prod_{\sigma''\in\theta',\sigma''\neq\sigma ' } \left(\#\sigma''-1\right ) ! \#\sigma^{\prime\prime n\left(\sigma'',\lambda'\right)-1}. % % % % % % % % % % % \nonumber\\ & = & \frac{1}{n ! } \sum_{\theta\in\theta\left(n\right ) } \sum_{\lambda\in\lambda\left(\theta\right ) } \prod_{\left[\lambda',\theta'\right]\in\tilde{g}\left(\left[\lambda,\theta\right]\right ) } \sum_{\sigma'\in\theta ' } \nonumber\\ & & { } \int\frac{dq_{\sigma'}}{2\pi}k\left(k - q_{\sigma'}\right ) \#\sigma^{\prime n\left(\sigma',\lambda'\right ) } \left(\#\sigma'-1\right ) ! \exp\left(-\beta \#\sigma ' q^2_{\sigma'}\right ) \tilde{s}\left(\left[\lambda',\theta'\right],q_{\sigma'},\sigma'\right ) \nonumber\\ & & { } \prod_{\sigma''\in\theta',\sigma''\neq\sigma ' } \left(\#\sigma''-1\right ) ! \#\sigma^{\prime\prime n\left(\sigma'',\lambda'\right)-1 } \label{eq : tmp_proof_d_1}\end{aligned}\ ] ] the second equality is due to a replacement of integration variables from @xmath240 to @xmath241 , and the relation ( [ eq : combinatorial_c1 ] ) is used in the third equality . by definitions , the following relation can be shown , @xmath242 , k , \left\{0\right\ } \right ) & = & \prod_{\left[\lambda_i,\theta_i\right]\in\tilde{g}\left(\left[\lambda,\theta\right]\right ) } \int\frac{dq_{\sigma_i}}{2\pi}k\left(k - q_{\sigma_i}\right ) \exp\left(-\beta \#\sigma ' q^2_{\sigma_i}\right ) \tilde{s}\left(\left[\lambda_i,\theta_i\right],q_{\sigma_i},\sigma_i\right ) , \label{eq : tmp_proof_d_2}\end{aligned}\ ] ] @xmath243 using a relation ( [ eq : tmp_proof_d_2 ] ) in ( [ eq : tmp_proof_d_1 ] ) , we obtain @xmath244,k,\left\{0\right\}\right ) . \label{eq : f_recusive_formula_e_2}\end{aligned}\ ] ] we see that ( [ eq : f_recusive_formula_e_2 ] ) is equal to ( [ eq : expansion_of_self_enargy ] ) . therefore , ( [ eq : expansion_of_self_enargy ] ) is recursively proved . from ( [ eq : integrant_of_grand_partition_function ] ) , we have @xmath245 substituting the expression of @xmath246 in ( [ eq : expansion_of_self_enargy ] ) into @xmath247 in ( [ eq : b_recusive_formula ] ) gives @xmath248,k,\left\{0\right\}\right ) . \label{eq : tmp_2}\end{aligned}\ ] ] using the relation ( [ eq : combinatorial_d4 ] ) , we have @xmath249,k,\sigma\right ) \nonumber\\ & & { } \prod_{\sigma'\in\theta,\sigma ' \neq \sigma } \left(\#\sigma'-1\right ) ! \#\sigma^{\prime n \left(\sigma^{\prime},\lambda\right)-1}. \label{eq : tmp_proof_d_3}\end{aligned}\ ] ] since the following relation holds , @xmath250 \right ) & = & \int\frac{dq_{\sigma}}{2\pi}k\left(k - q_{\sigma}\right ) \exp\left(-\beta \#\sigma ' q^2_{\sigma}\right ) \tilde{s}\left(\left[\lambda,\theta\right],q_{\sigma},\sigma\right ) , \label{eq : tmp_proof_d_4 } \\ & & \lambda \in\lambda_c\left(\theta\right),\quad \sigma\in\theta , \nonumber\end{aligned}\ ] ] we arrive at @xmath251\right)\\ & = & \frac{1}{\left(n-1\right ) ! } \sum_{\theta\in\theta\left(n\right ) } \sum_{\lambda\in\lambda\left(\theta\right ) } \left ( \prod_{\sigma\in\theta } \left(\#\sigma-1\right ) ! \#\sigma^{n\left(\sigma,\lambda\right)-1 } \right ) s^{\prime}_{\infty}\left(\left[\lambda,\theta\right]\right),\end{aligned}\ ] ] where @xmath252 is defined in ( [ eq : define_s_prime_inf ] ) . this expression of the cluster integral @xmath114 is the same as ( [ eq : cluster_integral_22 ] ) . 9 e. lieb and w. liniger , exact analysis of an interacting bose gas . i.the general solution and the ground state , phys . 1963;130:1605 - 1616 . c. n. yang and c. p. yang , thermodynamics of a one - dimensional system of bosons with repulsive delta - function interaction , j. math . 1969;10:1115 - 1122 . bethe , zur theorie der metalle , z. physik . 1931;71:205 - 226 . thacker , statistical mechanics of ( 1 + 1)-dimensional quantum field theory at finite density and temperature , phys . d. 1977;16:2515 - 2525 . m. wadati , bosonic formulation of bethe ansatz method , j.phys.soc.jpn . 1985;54:3727 - 3733 . k.huang : statistical mechanics , john wiley & sons , inc . new york , 1963 , 14.3 . g.kato , m. wadati : explicit calculation of the partition function of a one - dimensional @xmath1-function bose gas , chaos solitons & fractals(in press ) .

this paper proposes a frequency domain approach to test the hypothesis that a complex - valued vector time series is proper , i.e. , for testing whether the vector time series is uncorrelated with its complex conjugate . if the hypothesis is rejected , frequency bands causing the rejection will be identified and might usefully be related to known properties of the physical processes . the test needs the associated spectral matrix which can be estimated by multitaper methods using , say , tapers . standard asymptotic distributions for the test statistic are of no use since they would require but , as increases so does resolution bandwidth which causes spectral blurring . in many analyses is necessarily kept small , and hence our efforts are directed at practical and accurate methodology for hypothesis testing for small our generalized likelihood ratio statistic combined with exact cumulant matching gives very accurate rejection percentages and outperforms other methods . we also prove that the statistic on which the test is based is comprised of canonical coherencies arising from our complex - valued vector time series . our methodology is demonstrated on ocean current data collected at different depths in the labrador sea . overall this work extends results on propriety testing for complex - valued vectors to the complex - valued vector time series setting . generalized likelihood ratio test ( glrt ) , multichannel signal , spectral analysis .

1605.05910

there has long been an interest in time series motions on the complex plane : the rotary analysis method decomposes such motions into counter - rotating components which have proved particularly useful in the study of geophysical flows influenced by the rotation of the earth @xcite . let a complex - valued @xmath3-vector - valued discrete time series be denoted @xmath4 this has as @xmath5-th element , ( @xmath6 the column vector @xmath7^t.$ ] a length-@xmath8 realization of @xmath9 namely @xmath10 has @xmath11 in this paper we assume the @xmath3 processes are jointly second - order stationary . we propose a frequency domain approach to testing the hypothesis that a complex - valued @xmath3-vector - valued time series is proper , i.e. , for testing whether the vector time series @xmath9 is uncorrelated with its complex conjugate @xmath12 if we denote the covariance sequence between these terms by @xmath13 then propriety corresponds to @xmath14 or @xmath15 over the nyquist frequency range , where @xmath16 is the fourier transform of @xmath13 . otherwise the time series is said to be improper ; the practical importance and occurrence of improper processes is discussed in , e.g. , @xcite , @xcite and @xcite . the relevance of propriety for two - component complex - valued series ( @xmath17 can be found in @xcite . because the series are complex , two types of cross - covariance can be defined : that between the two series , known as the inner cross - covariance @xcite , and that beween one series and the complex conjugate of the other , known as the outer cross - covariance @xcite . if the vector time series is proper then the outer cross - covariance is everywhere zero . in this paper we take as an example a six - component complex - valued ocean current time series recorded in the labrador sea . frequency domain analysis is particularly useful in a scientific setting : if the hypothesis is rejected , frequency bands causing the rejection can be identified and quite possibly related to known properties of the physical processes . analogous tests applicable to complex - valued random vectors rather than time series have been descibed by , e.g. , @xcite and @xcite . however , we need to consider new methodology suitable for very limited degrees of freedom . our test uses the associated spectral matrix which can be estimated by multitaper methods using , say , @xmath0 tapers . standard asymptotic distributions for the test statistic are of no use since they would require @xmath1 but , as @xmath0 increases so does resolution bandwidth which causes spectral blurring . in many analyses @xmath0 is necessarily kept small , and hence our efforts are directed at practical and accurate methodology for hypothesis testing for small @xmath2 our generalized likelihood ratio statistic combined with exact cumulant matching gives very accurate rejection percentages and outperforms competitor methods . for the scalar case , @xmath18 a parametric hypothesis test for propriety of complex time series is given in @xcite , @xcite . this is based on the series being well - modelled by a matrn process in @xcite or complex autoregressive process of order one in @xcite , and utilises the @xmath19 distribution for the test statistic , an asymptotic result . this is in contrast to our approach which ( i ) is suitable for @xmath20 ( ii ) is nonparametric , so does not rely on a good fit to a parametric model , and ( iii ) develops a suitable non - asymptotic distribution for the test statistic . our test statistic is comprised of canonical coherencies arising from the complex - valued vector time series , analogous to the situation for complex - valued random vectors . canonical analysis of real - valued vector time series has been extensively studied and utilised ( e.g. , @xcite ) , mostly in the context of parametric autoregressive moving - average ( arma ) models . miyata @xcite looked at real - valued vector time series , and developed canonical correlations through linear functions of discrete vector fourier transforms of two sets of time series . rather than work with the fourier transforms , which are sample values , we instead work with the orthogonal processes underlying the complex - valued vector time series , and whose variances and cross - covariances correspond exactly to the spectral components . we are thus able to define population as well as sample canonical coherencies for complex - valued vector time series . our methodology is demonstrated on ocean current data collected at different depths in the labrador sea . following some background in section [ sec : backg ] on complex - valued time series , and the statistical properties of their spectral matrix estimators under the gaussian stationary assumption for @xmath21 the contributions of this paper are as follows : 1 . in section [ sec : cancoh ] we formally derive the canonical coherencies for @xmath22 and @xmath23 and show in section [ sec : testpropriety ] how a test statistic @xmath24 for testing @xmath15 arises from the sample canonical coherencies . 2 . after giving further research context in section [ sec : context ] , we carefully study the statistical properties of @xmath25 in section [ sec : basicp ] , concentrating on the small @xmath0 case . we show that box s scaled chi - square approximation is exact for @xmath26 but not for @xmath27 and we derive the cumulants of @xmath28 3 . in section [ sec : other ] we show that for @xmath29 and small @xmath0 matching the first three cumulants of @xmath30 exactly to a scaled @xmath31 distribution performs at least as well as competitor methods . a simulation study is given in section [ sec : simulationt ] which supports the use of the scaled @xmath31 approximation for @xmath30 for the complex - valued vector time series setting . a data analysis using 6-vector valued oceanographic time series is given in section [ sec : realdata ] which shows that when propriety is rejected , the frequency domain approach usefully shows which frequency bands cause the rejection , which may be linked to the physical processes involved . 5 . in section [ sec : othervc ] we show how our use of canonical coherencies in the complex - valued setting is quite different to an existing approach in the literature derived for real - valued processes , even though there are some structural features in common . we consider a complex - valued _ @xmath3-vector - valued discrete time stochastic process _ @xmath32 whose @xmath5th element , @xmath33 is the column vector @xmath7^t,$ ] and without loss of generality take each component process to have zero mean . the sample interval is @xmath34 and the nyquist frequency is @xmath35 we assume the @xmath3 processes are _ jointly _ second - order stationary ( sos ) , i.e. , @xmath36 and @xmath37 @xmath38 are functions of @xmath39 only . note that @xmath40 the covariance between one process and the complex conjugate of the other . a matrix autocovariance sequence is then given by @xmath41 where superscript @xmath42 denotes hermitian ( complex - conjugate ) transpose . we define @xmath43 and a matrix cross - relation sequence follows as @xmath44 with @xmath45 from their definitions we see that @xmath46 we assume @xmath47 and @xmath48 for @xmath49 which means that the fourier transforms @xmath50 and @xmath51 for @xmath52 exist and are bounded and continuous . in fact for @xmath53 the corresponding matrices are defined as @xmath54 we note that @xmath55 a result which will prove useful later . the covariance stationarity means that there exists @xcite an orthogonal process @xmath56 such that @xmath57 where @xmath58 if @xmath14 or @xmath59 then the process @xmath60 is said to be _ proper_. equivalently we see that if @xmath9 is uncorrelated with its complex conjugate @xmath61 then the vector - valued process is proper . this paper considers the problem of testing that the vector process is proper . based on the naming convention adopted in @xcite for complex - valued vectors , an alternative would be to call the component processes ` jointly proper . ' let @xmath62 with @xmath63 and @xmath64 real - valued , for @xmath65 where @xmath66^{t}=[x_{1,t},\ldots , x_{p , t},y_{1,t},\ldots , y_{p , t}]^{t}$ ] is a real @xmath67-dimensional vector - valued gaussian stationary process . then if @xmath68,\ ] ] we see that @xmath69 = \left [ \begin{matrix } { \bz}_t\\ { \bz}_t^ * \end{matrix}\right]=\bu_t,\ ] ] where @xmath70^{t}=[z_{1,t},\ldots , z_{p , t},z_{1,t}^*,\ldots , z_{p , t}^*]^{t}$ ] is a real @xmath67-dimensional vector - valued gaussian stationary process . the spectral matrix for @xmath71 is given by @xmath72 the spectral matrix for @xmath73 is @xmath74 and has the form @xmath75 the matrix @xmath76 can be written in the alternative covariance matrix form @xmath77 where @xmath78^t.\ ] ] given a length-@xmath8 sample @xmath79 , form @xmath80 using a suitable set of @xmath0 length-@xmath8 orthonormal data taper sequences @xmath81 , and compute @xmath82 in this work we use sine tapers ( e.g. , @xcite ) . as @xmath83 , with the number of degrees of freedom , @xmath0 fixed , and with the given taper properties , @xmath84 are proper , independent and identically distributed random vectors such that @xmath85 for @xmath86 ( e.g. , @xcite ) . as @xmath87 , as @xmath83 , with @xmath0 fixed , @xmath88 are also a set of proper , independent and identically distributed random vectors each of which are distributed as @xmath89 the probability density function ( pdf ) of @xmath90 a proper gaussian vector in @xmath91 is given by @xcite @xmath92^{-1}\exp\left\{-{\bj}_{\mbu , k}^{h}(f){\bs}^{-1}_{\mbu}(f){\bj}_{\mbu , k}(f)\right\}.\ ] ] the independence of @xmath90 s allows us to write the joint pdf of @xmath93 as the product of their marginal densities given by ( [ eqn : complexnorm2p ] ) . so the likelihood function , @xmath94 of @xmath95 given @xmath96 is given by @xmath97^{-k } \exp\left\{-\sum_{k=0}^{k-1}{\bj}_{\mbu , k}^{h}(f){\bs}^{-1}_{\mbu}(f){\bj}_{\mbu , k}(f)\right\}. % & \equiv & { \pi^{-2kp}}[\det({\bs}_{\mbu}(f))]^{-k}\exp\left\{-{k}\tr[{\bs}^{-1}_{\mbu}(f)\hat{{\bs}}_{\mbu}(f)]\right\},\ ] ] now @xmath98 is the sample covariance matrix of @xmath99 , i.e. , @xmath100 noting that the argument of @xmath101 in ( [ eqn : gjorig ] ) is scalar , and so is equal to its trace , and recalling the linearity and cyclicity of the trace operator , we can write @xmath102^{-k}\exp\left\{-{k}\tr\{{\bs}^{-1}_{\mbu}(f)\hat{{\bs}}_{\mbu}(f)\}\right\},\ ] ] where dependence of @xmath103 on its arguments is suppressed for convenience . for a finite value of @xmath8 , @xmath99 are proper random variables with @xmath104 where @xmath105 $ ] is the extent of the spectral window induced by tapering @xcite . for sine tapers @xmath106,\ ] ] ( e.g. , @xcite ) . therefore , in practice , we have to restrict interest to frequencies in the range @xmath107 . the structure of the testing problem is related to measures of coherence between vector - valued processes , and so we next turn our attention to the idea of canonical coherence . consider the cross - correlation of complex - valued deterministic matrix sequence @xmath108 with the time series @xmath22 to give @xmath109 @xmath110 likewise we define the cross - correlation of complex - valued deterministic matrix sequence @xmath111 with the time series @xmath112 to give @xmath113 @xmath114 component - wise we have @xmath115 so , for @xmath116 , @xmath117 the spectral representation theorem allows us to write @xmath118 and @xmath119 , @xmath65 as @xmath120 substituting the spectral representation for @xmath121 in the first term of ( [ eqn : zjt ] ) , we get @xmath122 where @xmath123 . proceeding in analogous fashion , and using the fact that the orthogonal process in a spectral representation is unique @xcite , we obtain @xmath124 so @xmath125 for @xmath126 a similar procedure gives @xmath127 and @xmath128 the usual definition of the ( magnitude squared ) coherencies @xmath129 between series @xmath130 and @xmath131 is @xmath132 it should be emphasized that throughout we use the usual definition of coherence as a magnitude squared quantity , basically a squared correlation coefficient . in vector notation , @xmath133 where @xmath134.$ ] consider @xmath135 . this can be written @xmath136^{1/2 } [ \bb_j^h(f ) \bs^t_{\mbz}(-f)\dif f \bb_j(f)]^{1/2}}\\ & & \!\!\!\!\!\!\!\!\!\!\!= \frac{|\ba_j^h(f)\br_{\mbz}(f)\bb_j(f)| } { [ \ba_j^h(f ) \bs_{\mbz}(f ) \ba_j(f)]^{1/2 } [ \bb_j^h(f ) \bs^t_{\mbz}(-f ) \bb_j(f)]^{1/2}}.\end{aligned}\ ] ] suppose we choose @xmath137 and @xmath138 so that @xmath139 then @xmath140 @xmath141 define @xmath142 so that @xmath143 [ def : one ] the first definition of the canonical coherence problem under the standardization in ( [ eq : standardize ] ) is as follows . find @xmath144 and @xmath145 such that @xmath146 is maximized . next find @xmath147 and @xmath148 such that @xmath149 is maximized , subject to @xmath150 being uncorrelated with @xmath151 in general , at step @xmath152 for @xmath153 @xmath154 and @xmath155 are found such that @xmath156 is maximized subject to @xmath157 being uncorrelated with @xmath158 for @xmath159 the problem can be defined in a different but equivalent way @xcite . the second definition of the canonical coherence problem under the standardization in ( [ eq : standardize ] ) is as follows . choose @xmath137 and @xmath138 such that all partial sums over the @xmath160 are maximized , i.e. , @xmath161 the canonical coherencies @xmath162 and @xmath154 and @xmath155 for @xmath163 solving ( [ eq : defineprob ] ) are eigenvalues and eigenvectors defined as follows : @xmath164 moreover we have that as a result , @xmath165 see appendix [ app : lem1 ] . from lemma [ lem : one ] the optimal @xmath154 and @xmath155 give rise to the @xmath152th pair of canonical series via ( [ eq : firstcan ] ) and ( [ eq : secondcan ] ) . results ( [ eq : corra ] ) and ( [ eq : corrb ] ) ensure that the uncorrelated requirements in definition [ def : one ] hold . the glr test statistic for @xmath166 for any @xmath107 , is given by ratio of the likelihood function ( [ eqn : gj ] ) with @xmath167 constrained to have zero off - diagonal blocks ( @xmath168 ) to the likelihood function with @xmath167 unconstrained , i.e. , @xmath169 the unconstrained maximum likelihood estimate of the covariance matrix @xmath167 is given by the corresponding sample covariance matrix @xmath98 in ( [ eqn : sufest ] ) , thus maximum likelihood estimate of @xmath167 under the constraint @xmath168 is , @xmath170 from ( [ eqn : gj ] ) , ( [ eqn : glr ] ) it follows that @xmath171 is @xmath172^{-1}\exp\left\{-\tr\{\breve{{\bs}}^{-1}_{\mbu}(f)\hat{{\bs}}_{\mbu}(f)\}\right\ } } { [ \det\ { \hat{{\bs}}_{\mbu}(f)\}]^{-1}\exp\left\{-\tr\{\hat{{\bs}}^{-1}_{\mbu}(f)\hat{{\bs}}_{\mbu}(f)\}\right\ } } \nonumber \\ % \left-\bi_{2p})\right\}\right]\\ & = & \det\{\breve{{\bs}}^{-1}_{\mbu}(f)\hat{{\bs}}_{\mbu}(f)\}\nonumber\\ & \times&\exp\left\{-\tr\{\breve{{\bs}}^{-1}_{\mbu}(f)\hat{{\bs}}_{\mbu}(f)-\bi_{2p}\}\right\}.\end{aligned}\ ] ] the result ( [ eqn : gj ] ) is valid for @xmath173 but from ( [ eq : rposneg ] ) we see that if @xmath174 for @xmath175 then it is also @xmath176 for @xmath177 hence in practice we need only concern ourselves with the positive frequency range @xmath178 and calculate @xmath24 over this interval . from ( [ eqn : sufest ] ) and ( [ eqn : sufestro ] ) we see that @xmath179 so that the @xmath101 term is unity . thus ( [ eqn : preveqn ] ) becomes @xmath180 starting with ( [ eqn : allu ] ) and using ( [ eqn : sufestro ] ) we also have that @xmath181 now , the glr test may be based on any of the above equivalent forms for @xmath182 form ( [ eqn : glrformtwo ] ) , unlike other formulations does not involve computation of either @xmath183 or @xmath184 . by definition of the glr test statistic ( [ eqn : glr ] ) , we shall reject the null hypothesis of @xmath185 , for small values of @xmath182 for a given size @xmath186 , the rule is to reject @xmath187 iff @xmath188 where pr@xmath189 . here we have used the more precise notation @xmath190 which emphasizes the dependence of the glr test on ( i ) the sample size @xmath8 , ( ii ) the number of tapers @xmath0 ( also the number of complex degrees of freedom ) , and ( iii ) dimension @xmath3 of the complex time series . now @xmath191 apply @xmath192 to @xmath193 so that @xmath194 and therefore @xmath195 so @xmath196^h\}\\ & = & { \bl}(f ) \br_{\mbz}(f)\dif f { \bl}^t(-f ) = { \mathbf{0 } } , \end{aligned}\ ] ] i.e. , @xmath168 is invariant to the linear transformation @xmath197 . so the decision rule for our glr test must be likewise invariant . note that under this transformation , @xmath198^t\\ & = & \left [ \begin{matrix } \bl(f ) & { \bf 0}\\ { \bf 0 } & \bl^*(-f ) \end{matrix}\right ] \mbu(f)\\ & { \displaystyle \equaldef } & \bq(f ) \mbu(f),\end{aligned}\ ] ] so that we require invariance under the group action @xmath199 . under the null hypothesis @xmath76 takes the form in ( [ eqn : sufestro ] ) so that @xmath200 is @xmath201,\ ] ] and the choice @xmath202 ( which exists for @xmath203 positive definite ) renders the matrix equal to @xmath204 this means that under the null hypothesis we can always replace @xmath76 by @xmath205 without loss of generality . from lemma [ lem : one ] we know that the eigenvalues @xmath206 of @xmath207 are canonical coherencies which are invariant under the group action specified above ; moreover , the corresponding empirical or sample canonical coherencies are maximal invariant and the glr statistic which requires this invariance must be a function of them . let @xmath208 be the sample versions of the canonical coherencies @xmath206 between @xmath209 and @xmath210 they are the sample eigenvalues of @xmath211 . then from ( [ eqn : glrform ] ) it follows that for @xmath212 , @xmath213 testing @xmath185 is the same as testing the independence of two complex gaussian @xmath3-vectors , namely @xmath209 and @xmath214 ( see ( [ eq : defdu ] ) ) . the glr test based on ( [ eqn : glrformtwo ] ) falls in the class of multiple independence tests in multivariate statistics theory . some distributional results for the complex case were given in @xcite but did not include the case of interest here , namely two @xmath3-vectors . a later paper @xcite gave the exact distribution of a power of @xmath24 but this involves an infinite sum with very complicated components ; small @xmath0 approximations were not discussed . other relevant results can be found in @xcite and @xcite , and these are discussed in detail in section [ subsec : beta ] . the statistic @xmath24 is the frequency - domain time series analogue to those used in @xcite and @xcite to examine independence between a gaussian random vector and its complex conjugate . in @xcite a complex formulation was maintained but only an asymptotic approach to testing was considered . in @xcite a real - valued representation of the problem was used and box s scaled chi - square method was used to improve on the asymptotic critical values . in the rest of this paper we adopt the complex formulation , derive box s refinement , but also improve on it for @xmath29 by _ exactly _ matching the first three cumulants to a scaled @xmath31-distribution . ( we point out that box s refinement is exact for @xmath215 ) this latter @xmath31-method is very simple to implement practically , involving only the first three polygamma functions . we emphasize that our efforts are directed at practical and accurate methodology for small @xmath2 this is important in a time series setting where as @xmath0 increases so does resolution bandwidth which potentially causes spectral blurring . in many analyses @xmath0 must necessarily be kept small . in the remainder of this paper we will always assume any frequency under consideration to lie in the interval @xmath216 the application of wilk s theorem @xcite gives that under @xmath217 as @xmath218 , @xmath219 where @xmath220 denotes convergence in distribution and @xmath221 denotes the chi - square distribution with @xmath222 degrees of freedom . here @xmath222 is the difference between the number of free real parameters under @xmath223 and @xmath224 . comparing @xmath225 in ( [ eqn : sufestro ] ) ( for @xmath226 ) and @xmath76 in ( [ eqn : suf ] ) ( for @xmath227 ) we note that @xmath228 follows directly from @xmath229 so that there is only an additional @xmath230 degrees of freedom , i.e. , those contributed by @xmath231 hence we have @xmath232 while ( [ eqn : wilks ] ) is a very useful and convenient result when the exact distribution of the glr test statistic is analytically intractable , @xmath0 here denotes the number of tapers used for multitaper spectral estimation and not the sample size @xmath8 . for a given value of @xmath8 , @xmath0 could be around @xmath233 or less . since ( [ eqn : wilks ] ) is an asymptotic result , @xmath0 must be sufficiently large to expect a reasonable @xmath221 approximation to @xmath234 since @xmath0 may not be large in a time series setting , a small-@xmath0 approximation to the distribution of the test statistic under the null hypothesis is imperative . since @xmath235 are gaussian distributed random vectors , from ( [ eqn : sufest ] ) it follows that @xmath236 i.e. , @xmath137 is distributed as a @xmath67-dimensional complex wishart distribution with @xmath0 complex degrees of freedom and mean @xmath237 given the form of @xmath238 , we partition @xmath137 analogously in terms of sub - matrices as @xmath239 then the glr test statistic in ( [ eqn : glrformtwo ] ) can be expressed as @xmath240 [ lemma : one ] the @xmath241th moment of @xmath242 namely @xmath243 is given by @xmath244-j - p+1)}{\prod_{j=1}^{p}\gamma(k[1+r]-j+1)}.\ ] ] this is given in appendix [ app : lemmaone ] . a random variable @xmath245 is said to be of box - type ( * ? ? ? ( 70 ) ) if for all @xmath246 @xmath247^{r}\frac{\prod_{i=1}^{m}\gamma(a_{i}[1+r]+\vartheta_{i } ) } { \prod_{j=1}^{l}\gamma(b_{j}[1+r]+\zeta_{j})},\ ] ] where @xmath248 , and the constant term @xmath249 is @xmath250 so that it s zeroth moment is unity . we see that @xmath251 is a random variable of box - type with @xmath252 and @xmath253 is @xmath254 the moment generating function for @xmath255 is given by ( with @xmath256 suppressed ) , @xmath257 so using ( [ eqn : moment ] ) , @xmath258-j - p+1)}{\gamma(k[1 - 2s]-j+1)}.\ ] ] the gamma functions will be valid if @xmath259 for all @xmath163 which requires @xmath260 the cumulants @xmath261 of @xmath262 can be easily obtained from the cumulant generating function by successively differentiating @xmath263 and setting @xmath264 notice that the requirement @xmath265 corresponds to @xmath266 when @xmath264 then , for @xmath267 @xmath268 so that @xmath261 is @xmath269^i\sum_{j=1}^{p } \!\!\left[\psi^{(i-1)}(k - j - p+1)-\psi^{(i-1)}(k - j+1)\right].\ ] ] here for @xmath270 , @xmath271/{\dif x}$ ] is the digamma function , while for @xmath272 and 3 , @xmath273 and @xmath274 are the trigamma and tetragamma functions respectively ; these are all ` polygamma functions . ' @xmath275 is the mean , @xmath276 is the variance , @xmath277 is the skewness and @xmath278 is the excess kurtosis . box @xcite provides a scaled chi - squared approximation for @xmath262 of the form @xmath279 the constant @xmath280 is chosen so that the cumulants of @xmath281 match those of @xmath30 up to an error of order @xmath282 the degrees of freedom @xmath283 associated with the chi - square approximation for @xmath30 is given by box @xcite @xmath284\\ & = & -2\left[\sum_{i=1}^{p}(1-i - p)-\sum_{j=1}^{p}(1-j)\right]\\ & = & -2\left[-\sum_{i=1}^{p}i-\sum_{i=1}^{p}p+\sum_{j=1}^{p}j\right]=2p^{2}=\nu,\end{aligned}\ ] ] as expected . the scaling factor @xmath285 is a constant determined as follows @xcite . define @xmath286\ ] ] where @xmath287 is the bernoulli polynomial of degree @xmath288 and order unity , with @xmath289 subsequently , let @xmath290 and @xmath291 , then @xmath285 is chosen according to the following rule : @xmath292 using ( [ eq : bern ] ) we find that @xmath293 it is straightforward to see that @xmath294 for all @xmath295 combinations , implying that @xmath296 , giving box s finite sample approximation as @xmath297 ( this agrees with ( [ eqn : wilks ] ) asymptotically as @xmath218 for a fixed dimension @xmath298 ) for the case @xmath26 we note that @xmath24 in ( [ eqn : firstone ] ) becomes @xmath299 where @xmath300 is the ` conjugate coherence , ' i.e. , the ordinary coherence between @xmath301 and @xmath302 ( e.g. , @xcite ) . then @xmath303 under the null hypothesis it is known that @xmath304 i.e. , coherence has the @xmath305 distribution . it then follows readily that @xmath30 has pdf @xmath306},\ ] ] so that @xmath307 and box s approximation ( [ eqn : boxapprox ] ) is in fact _ exact _ for the case @xmath215 when @xmath26 we note that @xmath308 for small values of @xmath0 , matching cumulants of @xmath30 up to an error of order @xmath309 could be problematic for @xmath29 @xcite . this leads us to consider other approaches . [ lemma : two ] under the null hypothesis the distribution of @xmath24 can be expressed as a product of independent beta random variables : @xmath310 where @xmath311 independently . this is given in appendix [ app : lemmatwo ] . if @xmath312 ( [ eq : prodbeta ] ) gives @xmath313 as it should since @xmath314 and ( [ eq : cohdist ] ) holds . in a different context gupta @xcite developed the distribution of the product of @xmath3 independent beta distributions : a likelihood ratio criterion for testing a hypothesis about regression coefficients in a multivariate normal setting takes the form @xmath315 under the corresponding null hypothesis , with @xmath316 and @xmath317 independently distributed as @xmath318 for integer parameters @xmath319 and covariance matrix @xmath320 then @xmath321 has the three - parameter complex @xmath322 distribution @xmath323 which is distributed as a product of @xmath3 beta variables with @xmath324 so setting gupta s parameters @xmath325 and @xmath326 to @xmath327 and @xmath3 , respectively , shows that @xmath24 has the three - parameter complex @xmath322 distribution @xmath328 this helps only a little because there are no simple expressions for this distribution s pdf or quantiles etc . however , by using convolution techniques gupta did obtain some exact results for the case @xmath329 in fact it turns out that for @xmath330 the right - side of ( [ eqn : boxapprox ] ) can be improved to @xmath331 where @xmath332 is an exact ( tabulated ) correction factor and @xmath333 is the @xmath334 point of the chi - square distribution with @xmath335 degrees of freedom . for example for @xmath336 and @xmath337 the factors are @xmath338 ( * ? ? ? * table 1 ) . the work of gupta was extended as part of @xcite who produced tables of approximate correction factors for the right - side of ( [ eqn : boxapprox ] ) for @xmath339 so that @xmath30 is compared to @xmath340 setting their parameters @xmath288 and @xmath341 to @xmath327 and @xmath3 respectively , shows that for example for @xmath342 and @xmath337 the factors are @xmath343 ( * ? ? ? * table 7 ) . the effect of these correction factors will be discussed shortly . the result ( [ eq : prodbeta ] ) is very nice , and quantiles of @xmath24 could be found through , say , successive convolution techniques , but this is very complicated see @xcite who develop this approach for a related statistic . the look - up tables of @xcite and @xcite are not convenient and so we now develop a simple and fast method for approximating the percentage points of the distribution of @xmath28 box @xcite considered using the very flexible pearson system for approximating the distribution of likelihood ratios . box @xcite introduced a discriminant @xmath344 such that if @xmath345 a pearson type vi should be fitted ; this corresponds to @xmath346 for @xmath347 with @xmath348 we always found @xmath345 using ( [ eq : cums ] ) . ( note @xmath26 is excluded since @xmath349 in that case . ) box @xcite considered distributions of the form @xmath350 i.e. , a scaled @xmath31 distribution ( pearson type vi ) with parameters @xmath351 and suggested matching cumulants _ we have chosen to match the first three cumulants of the form ( [ eq : cums ] ) _ exactly _ ; the parameters of @xmath352 are related to the cumulants via @xcite @xmath353 then to carry out the test @xmath30 would be compared to @xmath354 where @xmath355 is the @xmath334 point of the @xmath31 distribution with parameters @xmath356 given by ( [ eq : fparams ] ) . for some combinations of @xmath357 the asymptotic result ( [ eqn : wilks ] ) is compared to box s basic approximation ( [ eqn : boxapprox ] ) , the adjusted box method ( [ eq : correctedtwo ] ) , ( [ eq : corrected ] ) and the scaled @xmath31 method ( [ eq : fmethod ] ) in table [ tab : percentiles ] which gives the @xmath358 and @xmath359 points of the distribution of @xmath30 according to the four approaches . there is very good agreement between the adjusted box method and the scaled @xmath31 method , the latter being quick and simple to compute . box s basic approximation is a massive improvement on the asymptotic result . for @xmath330 the adjusted box approximation due to @xcite is exact and we see that the scaled @xmath31 approximation is therefore very accurate . other combinations of @xmath3 and small @xmath0 lead to similar results . the agreement of the scaled @xmath31 approximation with the previous historically tabulated results ( adjusted box approximation ) leads us to the following recommendation . @xmath360 in view of the discusssions and results above , the following is recommended for a given choice of @xmath361 * if @xmath312 reject @xmath223 if @xmath362 this test is distributionally exact . * if @xmath363 reject @xmath223 if @xmath364 the accuracy of the scaled @xmath31 approximation for our time series test ( [ eq : tssetting ] ) is now confirmed by simulation . ( a ) @xmath365 and ( b ) @xmath366 the vertical dotted line marks the frequency @xmath367 , width=326,height=230 ] for @xmath368 we will show that using the scaled @xmath31 approximation test where we reject @xmath226 if ( [ eqn : frule ] ) holds brings about a worthwhile accuracy improvement over box s approximation test where we reject @xmath226 if @xmath369 to be able to do this we need to simulate from a model such that @xmath76 in ( [ eqn : suf ] ) has @xmath370 for some frequency range . we can proceed as follows . we know @xcite that any complex second - order stationary scalar process ( assumed zero mean here ) , whether proper or improper , can be written as the output of a widely linear filter driven by proper white noise , i.e. , @xmath371 where @xmath372 and @xmath373 are sequence of complex constants , and @xmath374 is proper white noise for which @xmath375 for @xmath376 where @xmath377 is the kronecker delta . for simulation purposes it is convenient to set @xmath378 then @xcite @xmath379 where @xmath380 is the frequency response function of @xmath372 given by @xmath381 and @xmath382 is the frequency response function of @xmath383 for @xmath368 we generate processes @xmath384 such that @xmath385 where the @xmath67 processes @xmath386 are all independent of each other . the filter @xmath372 was chosen to be low - pass with a frequency transition zone @xmath387.$ ] the filter @xmath373 was of ` hilbert - type ' or all - pass in the frequency zone @xmath388.$ ] thus @xmath380 is real and symmetric while @xmath382 is imaginary and skew - symmetric . according to ( [ eq : wl2 ] ) , if using just these two filters , the resulting @xmath389 is zero for @xmath390.$ ] however , the filter @xmath391 was chosen to be high - pass above @xmath392 and therefore generates non - zero @xmath389 values at these high frequencies . the resulting @xmath365 and @xmath389 are shown in fig . [ fig : szrz ] . the matrix @xmath203 is thus of the form @xmath393 with frequency dependence as shown in fig . [ fig : szrz](a ) while @xmath229 is of the form @xmath394 with frequency dependence as shown in fig . [ fig : szrz](b ) . we can thus simulate from this model to evaluate our hypothesis tests , knowing that for frequencies where @xmath370 in fact @xmath395 and thus ( [ eqn : glrformtwo ] ) is well - defined . @xmath396 sample results are shown in table [ tab : one ] for @xmath397 and @xmath398 so here @xmath399 and 8 are indeed small . here @xmath400 but smaller time series lengths such as 128 produced very similar results . shown are rejection percentages for @xmath226 over @xmath401 independent repetitions . the nominal rates are shown in the second column . the first three columns of rejection percentages are for frequencies where @xmath402 ( @xmath226 is true ) and the latter two are for frequencies where @xmath403 ( @xmath226 is false ) see fig . [ fig : szrz](b ) . the top line of each entry is for box s @xmath19 approximation ( [ eqn : rule3 ] ) and the lower line is for the @xmath31 approximation of ( [ eqn : frule ] ) . we see that , proportionately , the latter has a much more accurate rejection rate than box s approximation when @xmath226 is true , but is slightly less accurate when @xmath226 is false . here we apply our results to ocean current speed and direction time series recorded at a mooring in the labrador sea @xcite . we associate the eastward ( zonal ) measurement of current speed with @xmath404 and the northward ( meridional ) measurement with @xmath405 and thus obtain the complex - valued series from ( [ eq : constructz ] ) . series were recorded at six depths , ( 110 , 760 , 1260 , 1760 , 2510 and 3476 m ) . the series are labelled 1 to 6 with increasing depth . we used @xmath406 observations for the @xmath407-vector - valued complex time series , with a sampling interval of @xmath408hr . in the spectral analysis @xmath409 sine tapers were applied . since @xmath410 in ( [ eq : wndef ] ) is @xmath411c / hr , the validity range @xmath412 for our statistical results for a finite-@xmath8 sample is given by @xmath413c / hr . there was no evidence to reject the gaussian assumption for this data set @xcite . of great interest to oceanographers are deep ocean motions well away from boundaries , especially in the internal wave frequency band . we pay special attention to low frequencies @xmath414 , $ ] in the internal wave band and near to the semi - diurnal tidal frequency . the so - called ` inertial frequency ' is approximately @xmath415c / hr for this latitude and purely clockwise rotation occurs at the inertial frequency in the northern hemisphere , making a band centred around the inertial frequency particularly interesting to study for such complex - valued processes . the dominant semi - diurnal tide at around @xmath416c / hr was estimated and removed to avoid spectral leakage affecting estimation near the inertial frequency . for this data @xmath417^t.$ ] in order to use different depth - contiguous sets of series we shall use the shorthand @xmath418^t$ ] with @xmath419 the degree of polarization of a single random vector measures the spread amongst the eigenvalues of its covariance matrix . a random vector is completely polarized / unpolarized if all of its energy is concentrated in one direction / equally distributed amongst all dimensions . this idea can be extended to the correlation between two random vectors by defining the correlation spread @xcite which provides a single , normalized measure of how much of the overall correlation is concentrated in a few coefficients , i.e. , correlation is contained in a low dimensional subspace . using the analogous definition to @xcite in our context we have _ coherence spread _ defined by @xmath420 if only one canonical coherence is non - zero , then @xmath421 whereas if all canonical coherences are equal , @xmath422 we note that if for a given @xmath256 , @xmath423 , the likelihood ratio test statistic @xmath424 i.e. achieves its minimum value , implying @xmath425 of course , for @xmath426 we are not able to conclude anything . in practice , we can only obtain an estimate @xmath427 where the @xmath206 are replaced by the @xmath428 and therefore , the hypothesis test must be used to check for @xmath429 fig . [ fig : coherencespread ] displays @xmath427 for vector time series ( a ) @xmath430 , ( b ) @xmath431 , ( c ) @xmath432 , ( d)@xmath433 and ( e ) @xmath434 . an immediate observation is that the coherence spread estimate for @xmath430 is highly erratic , with many values close to one . this is in contrast to all other plots where the spread ranges from @xmath435 gradually decreasing in range as we consider time series at increasing depths . a notable feature of ( b ) @xmath431 is the broader peaks around @xmath436 and @xmath437 and we see how the spread changes as we go from ( b ) @xmath431 to ( c ) @xmath432 with the broader peaks at @xmath438 and @xmath437 remaining intact whereas the one at @xmath439 shrinks from its value of @xmath440 to @xmath441 ; the sharper peaks at @xmath442 and @xmath443 disappear and a new peak appears at @xmath444 which persists in both ( d ) @xmath433 and ( e ) @xmath434 . we have thus seen how an additional series ( depth ) notably changes the concentration level of the overall coherence at some frequencies while disturbing it much less at others . for time series vectors ( left to right , top to bottom ) : ( a ) @xmath430 , ( b ) @xmath431 , ( c ) @xmath445 ( d ) @xmath433 and ( e ) @xmath434 . , width=364,height=288 ] as defined in section [ subsec : defproper ] the process @xmath22 is proper when @xmath446 our test for @xmath447 is valid , and may be carried out , for any @xmath448 we test the same sets of time series for propriety and the results are displayed in fig . [ fig : hypt_cum ] . the solid line shows the test statistic @xmath30 and the dotted line shows the critical value for each case . the test rejects @xmath226 at frequencies where @xmath30 exceeds the critical value ( thick line portions ) . the dashed line is the semi - diurnal tidal frequency . the coherence spread for @xmath430 ( first subplot in fig . [ fig : coherencespread ] ) takes the maximum value of @xmath449 at @xmath450 very close to the inertial frequency , and fig . [ fig : hypt_cum ] ( a ) shows that our test rejects @xmath226 around this frequency very clearly . the band of frequencies around @xmath451 is most prominent with rejection also clearly visible at frequencies @xmath452 and @xmath453 . for @xmath431 , the test rejects @xmath226 for almost the same set of low frequencies with rejection also at a higher frequency around @xmath454 . results for @xmath432 are very similar to that for @xmath431 , the main difference being that a small frequency band near @xmath455 also rejects @xmath456 in general , we see that as other series ( deeper in the ocean ) are considered , @xmath226 is rejected , not only at low frequencies but also due to some additional higher frequencies , but less definitively so . importantly then , fig . [ fig : hypt_cum ] shows _ which _ frequency bands cause propriety to be rejected . ( solid ) and the critical value ( dotted line ) for ( left to right , top to bottom ) : ( a ) @xmath430 , ( b ) @xmath431 , ( c ) @xmath445 ( d ) @xmath433 and ( e ) @xmath434 . the test rejects the null hypothesis of propriety at frequencies where @xmath30 exceeds the critical value ( thick line portions ) . the dashed line is the semi - diurnal tidal frequency . , width=364,height=288 ] from lemma [ lem : one ] , one measure for vector coherence is the sum of all the canonical coherencies : @xmath457 levikov and sokolov @xcite looked for a coefficient of coherence in the case of two real - valued vector random processes . in our paper , for the vector @xmath66^{t},$ ] we consider @xmath458 and @xmath459 to be two geometrically related vector components and combine them to form a complex - valued vector time series . levikov and sokolov did not consider the two processes to be related in such a way and treated them simply as two vector process . they did , however , make use of the frequency domain and derived the quantity @xmath460,\ ] ] where @xmath461 taking the trace of this quantity we get @xmath462 this is of the same form as ( [ eqn : secondstat ] ) only now using the components of the partition in ( [ eqn : svf ] ) ; it will be the sum of all the canonical coherencies between @xmath463 and @xmath464 where @xmath465 now @xmath466 and @xmath467 the spectral representation gives @xmath468 and @xmath469 so we can write @xmath470 = { \bt } \left [ \begin{matrix } \dif \bz_\bx(f)\\ \dif \bz_\by(f ) \end{matrix } \right],\ ] ] where @xmath471 is given in ( [ eq : deft ] ) . we know that affine transformations of @xmath463 and of @xmath464 will not change the canonical coherencies ; however , ( [ eq : whattransform ] ) does not represent affine transforms of @xmath463 and of @xmath464 since a mixing is involved . hence the quantities ( [ eqn : secondstat ] ) and ( [ eq : thirdstat ] ) will in general be different . indeed for the example of section [ sec : simulationt ] the value of ( [ eq : thirdstat ] ) is zero over @xmath472 because @xmath473 we have developed a frequency domain approach to test for propriety of complex - valued vector time series . for propriety of @xmath22 we require @xmath446 we can carry out the test @xmath447 for any @xmath448 most importantly for the vector case @xmath474 we have justified use of the rule that @xmath226 is rejected if @xmath475 there is no assumption that @xmath0 is large , and indeed this would rarely be expected in practice . we have shown in detail how the statistic @xmath24 arises by consideration of canonical coherencies for complex - valued vector time series . when propriety is invalid , the frequency domain approach has the scientific advantage of showing which frequency bands are causing rejection , likely allowing linkage to known or hypothesized properties of the physical processes involved . given ( [ eq : standardize ] ) , since @xmath203 and @xmath476 are positive - definite ( hermitian ) covariance matrices , we have solutions @xmath477 where @xmath478 are unitary . then @xmath479 we now make use of the weak majorization result @xcite let @xmath480^t.$ ] then @xmath481 are the singular values of @xmath482 and @xmath483}(f)\geq \sigma_{[2]}(f)\geq \cdots \geq \sigma_{[p]}(f),$ ] ( a descending size order ) . hence the solution to ( [ eq : defineprob ] ) is found by making @xmath482 diagonal . from ( [ eq : svd ] ) , we thus choose @xmath484 and @xmath485 to diagonalize @xmath486 i.e. , @xmath484 and @xmath485 are determined by singular value decomposition of @xmath487 giving@xmath488 where @xmath489 denotes a diagonal matrix with @xmath152th element @xmath490 in descending size order . now multiply through on the left by @xmath491 and on the right by @xmath484 to obtain @xmath492 which , using ( [ eq : aeq ] ) , can be written @xmath493 so that @xmath494 and @xmath206 are the eigenvalues , and @xmath154 are the eigenvectors of the @xmath495 matrix @xmath496 as required . note that this matrix is the product of the two hermitian matrices @xmath497 and @xmath498 similarly , @xmath499 multiply through on the left by @xmath500 and on the right by @xmath501 and use ( [ eq : beq ] ) to obtain @xmath502 so that , as required @xmath503 @xmath206 are the eigenvalues , and @xmath155 are the eigenvectors of the @xmath495 matrix @xmath504 notice that the matrices @xmath505 and @xmath506 are just cyclic permutations of each other . finally , @xmath507 and the solution of the optimization problem makes @xmath482 diagonal . hence , @xmath508 to simplify notation we drop explicit frequency dependence consider the distribution of @xmath509 given by ( [ eqn : formtouse ] ) , _ under the null hypothesis_. we have @xmath510^r}{[\det\{\ba_{11}\}\det\{\ba_{22}\}]^r } } } g({\ba } ) \dif \ba,\ ] ] where @xmath511 is the pdf for the complex wishart distribution . as explained in the text under the null hypothesis we can take @xmath512 to be @xmath513 because of invariance under the group action . we can thus replace ( [ eqn : acwishart ] ) by @xmath514 and using @xcite we know that for @xmath515 @xmath516^{k-2p } \e^{-\tr\{\ba\}},\ ] ] where @xmath517 is a constant defined by @xmath518 so @xmath519 takes the form @xmath520^{k-2p+r}}{[\det\{\ba_{11}\}\det\{\ba_{22}\}]^r}}}\e^{-\tr\{\ba\ } } \dif \ba\\ & & \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\ ! = { { \frac{c(k,2p)}{c(k+r,2p ) } } } \int \cdots\int { \textstyle{\frac{1}{[\det\{\ba_{11}\}\det\{\ba_{22}\}]^r}}}\\ & & \times\left[c(k+r,2p)[\det\{\ba\}]^{k-2p+r}\e^{-\tr\{\ba\}}\right ] \dif \ba.\end{aligned}\ ] ] the integration is w.r.t . @xmath521 the term in the square brackets above is the pdf for the @xmath522 distribution . the integral of this density with respect to the elements in @xmath523 and @xmath524 must give the marginal density of @xmath525 which is the product @xmath526 since @xmath527 and @xmath528 are independent under the null hypothesis . carrying out the integration and using ( [ eq : defg ] ) and ( [ eq : prodg ] ) we obtain @xmath529^r}}}\nonumber\\ & & \!\!\!\!\!\!\!\!\!\!\ ! \times\prod_{j=1}^2 c(k+r , p)[\det\{\ba_{jj}\}]^{k+r - p}{\e}^{-\tr\{\ba_{jj}\ } } { \dif } { \ba}_{jj}\nonumber\\ & & \!\!\!\!\!\!\!\!\!\!\!\ ! = { { \frac{c(k,2p)}{c(k+r,2p)}}}\prod_{j=1}^2 \int \cdots\int c(k+r , p)[\det\{\ba_{jj}\}]^{k - p } \nonumber\\ & & \times\quad{\e}^{-\tr\{\ba_{jj}\ } } { \dif } { \ba}_{jj } \nonumber\\ & & = { { \frac{c(k,2p)}{c(k+r,2p)}}}\prod_{j=1}^2 \frac{c(k+r , p)}{c(k , p)}.\label{eq : allthecs}\end{aligned}\ ] ] using ( [ eq : defcs ] ) and ( [ eq : allthecs ] ) we get @xmath530 ^ 2 } { \prod_{j=1}^{2p } \gamma(k\!+\!1-j ) [ \prod_{j=1}^p \gamma(k\!+\!r+1-j)]^2}.\ ] ] this agrees with ( * ? ? ? ( 2.6 ) ) which appears without reference or proof . now @xmath531 so if we let @xmath532 then @xmath533 so @xmath534 + 1-j ) \prod_{j=1}^p \gamma(k\!+\!1\!-\!j ) } { \prod_{j=1}^{2p } \gamma(k\!+\!1\!-\!j ) \prod_{j=1}^p \gamma(k[1+r]+1-j)}\\ & & \times \left [ \frac { \prod_{j=1}^p \gamma(k\!+\!1\!-\!j ) } { \prod_{j=1}^p \gamma(k[1+r]+1-j ) } \right]\\ & & = \frac{\prod_{j=1}^{p } \gamma(k[1+r]+1-j - p ) } { \prod_{j=1}^p \gamma(k+1-j - p)}\\ & & \times \left [ \frac { \prod_{j=1}^p \gamma(k\!+\!1\!-\!j ) } { \prod_{j=1}^p \gamma(k[1+r]+1-j ) } \right]\\ & & = \left [ \frac { \prod_{j=1}^p \gamma(k\!+\!1\!-\!j ) } { \prod_{j=1}^p \gamma(k+1-j - p ) } \right]\\ & & \times\frac{\prod_{j=1}^{p } \gamma(k[1+r]+1-j - p ) } { \prod_{j=1}^p \gamma(k[1+r]+1-j)},\end{aligned}\ ] ] which is ( [ eqn : moment ] ) . under the null hypothesis the @xmath241th moment of @xmath24 is @xmath535 to see this start with ( [ eq : etr ] ) and proceeed in analogous vein to the last part of the proof of lemma [ lemma : one ] ; since we are continuing to look at @xmath536 the step @xmath537 is not made . note that when @xmath538 the critical gamma function argument is still positive : @xmath539 since @xmath266 with @xmath540 a real scalar random variable @xmath541 is said to have a ( type-1 ) beta distribution , @xmath542 if the pdf is @xmath543 the @xmath241th moment for this distribution is @xmath544 comparing ( [ eq : firstbeta ] ) and ( [ eq : secondbeta ] ) we see for a fixed @xmath152 that @xmath545 and @xmath546 which gives the required result . the authors are grateful to jon lilly for the labrador sea data . c. fang , p. r. krishnaiah and b. n. nagarsenker , `` asymptotic distributions of the likelihood ratio test statistics for covariance structures of the complex multivariate normal distributions , '' _ j. multivariate analysis _ , vol . 12 , pp . 597611 , 1982 . a. k. gupta , `` on a test for reality of the covariance matrix in a complex gaussian distribution , '' _ j. statist . computation and simulation , _ vol . 2 , pp . 333 - 342 , 1973 . r. krishnaiah , , j. c. lee and t. c. chang , `` the distributions of the likelihood ratio statistics for tests of certain covariance structures of complex multivariate normal populations , '' _ biometrika _ , vol . 63 , pp . 543549 , 1976 . p. r. krishnaiah , , j. c. lee and t. c. chang , `` likelihood ratio tests on covariance matrices and mean vectors of complex multivariate normal populations and their applications in time series , '' technical report tr-83 - 03 , center for multivariate analysis , univ . pittsburgh , 1983 . s. levikov and s. sokolov , `` coefficient of coherence in the case of two vector random processes , '' _ deep - sea research part i : oceanographic research papers _ , vol . 44 , pp . 13291338 , 1997 . j. m. lilly , p. b. rhines , m. visbeck , r. davis , j. r. lazier , f. schott and d. farmer , `` observing deep convection in the labrador sea during winter 1994/95 , '' _ j. phys . oceanogr . , _ vol . 29 , 206598 , 1999 . j. m. lilly and p. b. rhines , `` coherent eddies in the labrador sea observed from a mooring , '' _ oceanogr . , _ vol . 32 , 58598 , 2002 . c. n. k. mooers , `` a technique for the cross spectrum analysis of pairs of complex - valued time series , with emphasis on properties of polarized components and rotational invariants , '' _ deep - sea research _ , vol . 20 , pp . 11291141 , 1973 . w. min and r. s. tsay , `` on canonical analysis of multivariate time series , '' _ statistica sinica _ , vol . 15 , pp . 303323 , 2005 . m. miyata , `` complex generalization of canonical correlation and its application to a sea - level study , '' _ j. marine research _ 28 , pp . 202214 , 1970 . j. navarro - moreno , m. d. estudillo - martnez , r. m. fernndez - alcal & j. c. ruiz - molina , `` estimation of improper complex - valued random signals in colored noise by using the hilbert space theory , '' _ ieee trans . inf . theory _ 55 , pp . 28592867 , 2009 . a. m. sykulski , s. c. olhede , j. m. lilly & j. j. early , `` on parametric modelling and inference for complex - valued time series , '' arxiv preprint arxiv:1306.5993v2 , 2015 . available : http://arxiv.org / abs/1306.5993 . h. van haren and c. millot , `` rectilinear and circular inertial motions in the western mediterranean sea , '' _ deep - sea research part i _ , vol . 51 , pp . 144155 , 2004 . a. t. walden , `` rotary components , random ellipses and polarization : a statistical perspective , '' _ phil . , vol . 371 , doi:10.1098/rsta.2011.0554 , 2013 .

we investigate , analytically and numerically , the emergence of spatio - temporal order in nonequilibrium scalar field theories . the onset of order is triggered by destabilizing interactions ( dis ) , which instantaneously change the interacting potential from a single to a double - well , tunable to be either degenerate ( sdw ) or nondegenerate ( adw ) . for the sdw case , we observe the emergence of spatio - temporal coherent structures known as oscillons . we show that this emergence is initially synchronized , the result of parametric amplification of the relevant oscillon modes . we also discuss how these ordered structures act as bottlenecks for equipartition . for adw potentials , we show how the same parametric amplification mechanism may trigger the rapid decay of a metastable state . for a range of temperatures , the decay rates associated with this resonant nucleation can be orders of magnitude larger than those computed by homogeneous nucleation , with time - scales given by a simple power law ,^b$ ] , where depends weakly on the temperature and is the free - energy barrier of a critical fluctuation .

cond-mat0310157

the emergence of spatio - temporal ordered structures in nonlinear systems is an ideal laboratory for investigating the trend toward complexification observed in nature at the physical , chemical , and biological level @xcite . it is known that localized , or spatially - bound , order emerges in systems which are out of thermodynamic equilibrium , at the expense of growing overall disorder @xcite . examples can be found in hydrodynamics , in networks of chemical reactions @xcite , in phase transitions and critical phenomena @xcite , and in living organisms @xcite . for these localized ordered structures to survive , they must interact with an external environment , which maintains nonequilibrium conditions : in general , an isolated nonlinear system will eventually reach equilibrium , maximizing its entropy to the detriment of localized ordered structures . in other words , if localized ordered structures emerge during the approach to equilibrium of a closed system , they will eventually disappear as energy becomes increasingly equipartitioned amongst the system s many degrees of freedom . the emergence of localized ordered structures is thus a nonequilibrium phenomenon , usually corresponding to an unequal partitioning of energy : if they are long - lived , they will serve as bottlenecks to equipartition . [ an exception occurs for 1d field theories with sdw , where configurations such as kink - antikink pairs correspond to the equilibrium . ] following the pioneering work of fermi , pasta , and ulam in the early 1950s @xcite , several recent works have investigated the possibility that discrete kink - like structures may act as bottlenecks for equipartition in chains of nonlinear oscillators @xcite . in field theory and cosmology , most of the interest on ordered configurations has focused on topological ( kinks , vortices , monopoles ) @xcite or nontopological ( nontopological solitons , q - balls ) @xcite _ static _ solutions of the equations of motion . those configurations can be boosted towards each other with a certain velocity and made to interact , to model nonlinear composite particle interactions . a well - known example are 1-dimensional breathers , time - dependent bound states that have been shown to result from interactions of kink - antikink pairs @xcite . however , this is not surprising : any two or more static solitonic configurations with short - range attractive interactions can be prepared to generate spatially - bound time - dependent structures @xcite . this is to be contrasted with the spatially - bound , time - dependent structures of interest here , that emerge _ spontaneously _ during the nonlinear evolution of a system . we have recently found such structures in 2 + 1-dimensional simulations of nonlinear scalar field models @xcite . such structures were identified with oscillons , long - lived _ time - dependent _ localized field configurations that have been previously found deterministically in a variety of physical systems , ranging from field theory models with amplitude - dependent nonlinearities @xcite , vertically vibrated grains @xcite , and stellar interiors @xcite . in this paper , we will examine the emergence of spatially - bound , time - dependent field configurations during the approach to equilibrium of ( 2 + 1)-dimensional scalar field models with typical double - well interactions . these interactions are controlled by a tunable external parameter , which can be interpreted as an extra interaction that may be switched on or off at will . in the parlance of ising magnetic systems , the external interaction may be a spatially - homogeneous magnetic field , @xmath3 . we will investigate both the symmetric double - well ( sdw ) extending previous results of ref . @xcite as well as provide new results for the asymmetric case ( adw ) . the interest in examining both cases stems from the fact that the ensuing dynamics will be quite different : while for the sdw the degeneracy in free energy density dictates that no critical nucleation is possible , this will not be the case for the adw . in fact , for the sdw we observe the synchronous emergence of long - lived localized configurations ( oscillons ) , which eventually disappear as the system reaches equipartition @xcite . we will argue that these configurations act as bottlenecks for equipartition , concentrating the field s energy in a narrow band of long - wavelength modes . for the adw , and with the system initially quenched to the metastable state , the synchronous emergence of oscillons may trigger the nucleation of a critical droplet of the lower free - energy phase , which grows to complete the phase transformation . we will show that , for a range of temperatures , the time - scales associated with this process are orders of magnitude faster than the typical statistical homogeneous nucleation , where a system _ adiabatically _ cooled into its metastable phase decays to its lowest free energy phase with typically exponentially - large time scales @xcite . in fact , the present treatment , with an `` instantaneous '' quench ( faster than other typical time - scales in the system ) , and the arbitrarily slow cooling implicit of homogeneous nucleation calculations , represent the two extreme cases of a whole spectrum of possible cooling rates . since we will argue that the emergence of coherent structures is due to the resonant amplification of parametric excitations of certain field modes , we will refer to this nucleation mechanism as _ resonant nucleation . _ anticipating one of our results , for field models described by amplitude - dependent scalar nonlinearities , resonant nucleation of spatio - temporal order will be relevant whenever the cooling time scale @xmath4 is faster than the equilibration time - scale of the longest wavelength in the system , @xmath5 . [ the superscript `` 0 '' is a reminder that often the longest wavelength is associated with the zero mode of the system , which has the slowest relaxation time - scale . ] we note that we will always work far from the critical point where , of course , @xmath6 . for our recent work on the approach to criticality please see ref . @xcite . the paper is organized as follows . in the next section we discuss the general properties of the model ( a general ginzburg - landau model ) , the initial conditions , and the numerical implementation of the dynamics . in section 3 we present a mean field ( homogeneous hartree ) approximation adequate to describe the system in the presence of small fluctuations about equilibrium . the fact that the homogeneous hartree approximation breaks down as nonperturbative excitations become progressively more important will provide us with useful information about the emergence of ordered structures . in section 4 we argue that the emergence of spatio - temporal order is a consequence of parametric resonance , induced by the nonlinear coupling between the field s zero mode the forcing agent and smaller-@xmath7 modes . we also provide entropic arguments in support of the statement that the emergence of coherent spatio - temporal structures delays equipartition of the energy and , thus , the approach to equilibrium . in section 5 , we extend our investigation to adws . for a range of temperatures [ or initial energies , see below ] , we measure the time scales for the decay of the metastable state as a function of barrier height , contrasting them with those computed from homogeneous nucleation theory . we identify two possible mechanisms for the decay : for small asymmetries , a critical nucleus is produced by the coalescence of two or more oscillons , while for larger asymmetries a single oscillon evolves into a critical nucleus . we show that , for a range of temperatures associated with the presence of oscillons in the system , the time scale associated with the decay obeys a simple power law , @xmath8^b$ ] , where @xmath2 is the free - energy barrier of a critical fluctuation , and the exponent @xmath1 satisfying @xmath9 . we conclude in section 6 with a summary of our results and outlook for future work . we start by introducing the model we will work with and its statistical and numerical implementation . we consider a ( 2 + 1)-dimensional real scalar field ( or scalar order parameter ) @xmath10 evolving under the influence of an on - site potential @xmath11 . if the system does not interact with an external bath , the continuum hamiltonian is conserved and gives the total energy of the system , for a given field configuration , @xmath12=\int d\,^2x\left[\frac{1}{2}(\partial_{t}\phi)^2+\frac{1}{2}(\nabla\phi)^2+v(\phi)\right ] , \label{h}\ ] ] where @xmath13 is the potential energy density . the parameters @xmath14,@xmath15 , and @xmath16 are positive definite and temperature independent . before we go any further , a note on our choice of potential . one could as easily have chosen a typical ginzburg - landau ( gl ) potential , @xmath17 , with @xmath3 an external homogeneous magnetic field . the gl potential can be transformed into the potential of eq . [ v ] by a field shift , @xmath18 , where @xmath19 is the solution of the cubic equation @xmath20 , @xmath21 , and @xmath22 . we chose the potential of eq . [ v ] for convenience . our control parameter is @xmath15 . we note that when @xmath23 the potential is an anharmonic single well . in this case , there are no spinodal instabilities . ( by spinodal instabilities we mean @xmath24 for certain values of @xmath25 , which may result in the unstable growth of modes . ) when @xmath26 ( in scaled units , see below ) the potential can assume a typical double - well shape [ sdw for equality and adw for @xmath27 , and spinodal instabilities are possible . we will thus refer to the interactions modeled by @xmath15 as destabilizing interactions ( dis ) . changing @xmath15 from an initial value of zero to a finite positive value can also be thought of as an `` instantaneous '' quench : the system , initially prepared in a single well , is `` tossed '' into a double - well potential within a time - scale much faster than any equilibration time - scale for longer wavelengths in the system , those with @xmath28 , where @xmath7 is the mode s wave number . we will say more about this later on . it is helpful to introduce the dimensionless variables @xmath29 , @xmath30 , @xmath31 , and @xmath32 ( henceforth dropping the primes ) . when @xmath33 the two minima are degenerate ( sdw ) . if @xmath34 , the @xmath35 symmetry is explicitly broken . we will only be concerned here with the cases modeled by @xmath36 . for @xmath37 the minimum at @xmath38 becomes the local ( or metastable ) minimum ( adw ) and @xmath39 is the global minimum . a statistical description of the state of the system is given by a set of time - dependent observables @xmath40 , each one calculated as an ensemble average over @xmath41 realizations , or microstates . for example , @xmath42 in most cases , an observable within a particular microstate is itself an average over the area of the system , such as the average value of the field @xmath43 for the particular microstate @xmath44 . most of the observables considered will have this `` double '' averaging . such is the case for @xmath45 , @xmath46 ( the field s conjugate momentum ) , and the corresponding hierarchy of correlation functions . it will be noted when this is not the case , particularly when quantities are no longer assumed to be translationally invariant , such as those describing localized configurations . we prepare the field such that it is in thermal equilibrium at temperature @xmath47 in the single - well ( @xmath23 ) potential at some time prior to the quench at @xmath48 , when @xmath15 is set to a nonzero value . when one is faced with a nonlinear equation of motion such as eq . [ lang ] below , the thermal probability distribution for @xmath49 can not be obtained exactly . to generate the initial conditions for each of the microstates in the ensemble , we instead couple the field in the single - well potential to a heat bath via a generalized langevin equation with additive noise , @xmath50 where the viscosity coefficient @xmath51 is related to the stochastic force of zero mean @xmath52 by the fluctuation - dissipation relation , @xmath53 eq . [ fd ] defines the two - point correlation function of the stochastic force , assumed here to be markovian . since this force will only be used during the preparation of the initial thermal state after the quench the coupling with the bath is turned off ( @xmath54 ) this assumption is sufficient . [ however , it would be quite interesting to examine the effects of multiplicative noise in the nonequilibrium dynamics of this system when the contact with the thermal bath is kept throughout its evolution . [ the role of multiplicative noise on the onset of synchronization in 1 + 1-dimensional fields has been recently investigated by munz and pastor - satorras @xcite . ] we note that even though we are preparing each microstate in an initial thermal state , there is a one to one correspondence between the temperature and the average energy of the field . in fact , we can refer to the ensemble prepared as above as a microcanonical ensemble , since each microstate has energy @xmath55 , where @xmath56 . such numerical accuracy in energy matches that obtained with other methods used preparing the initial state , including the popular monte carlo sampling . it simply expresses the fact that , for large enough lattices that approach the thermodynamic limit , there is an equivalence between canonical and microcanonical ensembles @xcite . the full evolution of the scalar field for each microstate is simulated using a staggered leapfrog integration of the equation of motion . it is implemented on a square lattice with periodic boundary conditions . integrations are performed with @xmath57 and @xmath58 with 1024 lattice sites per side . the field is then defined over an area with sides of length @xmath59 . the leapfrog scheme replaces second - order derivatives with finite - difference expressions that are accurate to @xmath60 . the lattice spacing and system size are chosen so that computations can be carried out in a timely manner , while the numerical values of any observables and their corresponding continuum values are in good agreement . for example , after the quench , the energy in this closed system is conserved by more than 1 part in @xmath61 . results are independent of lattice size , provided large enough lattices are used . finally , all of the observables are then averaged over an ensemble of @xmath62 microstates . we start our investigation using a familiar perturbative method , the leading order or homogeneous hartree approximation ( hha ) . an excellent account can be found in @xcite , where successive truncations within the hierarchy of correlation functions were used to simplify the equation of motion , all the while capturing the essential aspects of the dynamics . we will borrow many of the same methods and notation found in this reference . the hha assumes that the statistical distribution of the fluctuations in the field and its momentum remain gaussian throughout the evolution of the system . we will see that this is indeed an excellent approximation in the following cases : i ) just prior to the quench ( that is , the initial state of the field in a single well ) for all temperatures considered ; ii ) at all times after the quench when the temperature is low ; iii ) at very early times after the quench when the temperature is high . we start by considering the system before the quench is implemented ( @xmath23 ) , that is , the dynamics with the potential @xmath63 @xcite . a simple way to implement the hha in this situation is to factor the cubic term in eq . [ lang ] as @xmath64 , where @xmath65 is the translationally invariant mean - square variance of the field . since the field is in thermal equilibrium , @xmath65 is time - independent . in this approximation , the equation of motion becomes linear , @xmath66 the fourier transform of the field is a complex quantity given by @xmath67 the corresponding equation of motion for @xmath68 in @xmath7-space is [ with @xmath69 @xmath70 with the effective frequency squared @xmath71 we have introduced the hartree mass @xmath72 . notice that when there are fluctuations in this field ( @xmath73 ) the effective hartree mass is greater than the `` bare '' mass ( @xmath74 ) given by the potential in eq . [ v ] . in this approximation , the modes evolve independently of one another , just as they would in a linear truncation of the equation of motion , only now each interacts with the homogeneous mean - field background . a higher - order approximation is needed , however , to implement direct scattering between different modes . if the system is in thermal equilibrium , the momentum modes in @xmath7-space will satisfy @xmath75 it then follows from eq . [ eomk ] that the corresponding power spectrum for the field is @xmath76 to obtain the equilibrium quantity @xmath77 , one integrates eq . [ phimode ] over the entire @xmath7-space and divides by the area . ( equivalently , one can integrate the corresponding two - point correlation function over @xmath78-space to obtain @xmath65 , since parseval s theorem guarantees that the two quantities are the same . ) one must be careful here , since the lattice results depend logarithmically on the ultraviolet cutoff ( lattice - spacing ) . we refer the reader to ref . @xcite , where a method is provided to handle such dependence perturbatively . for various temperatures . the continuous lines are the theoretical predictions given by eq . [ p_pi].,width=288,height=288 ] for various temperatures . the continuous lines are the approximations given by eq . [ p_phi].,width=288,height=288 ] in figures [ dist_pi ] and [ dist_phi ] we show the probability distributions for the initial state of the field and its momentum for various temperatures . these distributions are obtained numerically by collecting the values of @xmath79 and @xmath49 of every microstate into bins of width @xmath80 , and then normalizing so that the sum over each distribution is unity . not all data points are shown , making these figures more readable . also shown are the gaussian probability distributions ( solid lines ) that are predicted from the boltzmann - gibbs formalism , @xmath81 the fitting parameter @xmath82 is obtained by comparing the area ( and ensemble)-averaged two - point corrrelation function of @xmath25 . the variance in figure [ dist_phi ] , @xmath65 , can be fitted linearly , @xmath83 . the agreement between the numerical results and the theoretical predictions confirms the accuracy of the lattice dynamics and the applicability of the hha for low enough temperatures . the probability distribution of @xmath79 is exact , while that of @xmath49 is indeed close to gaussian for the temperatures considered . we also show , in figure [ spec_phi ] , the power spectrum of the field modes , @xmath84 , and compare it with the prediction from eq . [ phimode ] , valid for the hha . for temperatures @xmath47=0.1 , 0.2 , and 0.4 . the solid lines are the approximations given by eq . [ phimode].,width=288,height=288 ] in this subsection we examine the dynamics of the field after the quench . we will limit our investigation to the case where @xmath85 at @xmath48 , that is , the potential is switched from a single well to a sdw . the dynamics for adw potentials will be examined in section 5 . at @xmath48 the coupling to the bath is set to zero , and the dynamics is thus conservative thereafter . the system is closed , and can be described by a microcanonical ensemble of energy @xmath86 , where @xmath87 . as we remarked before , for large enough systems in equilibrium , as is the case here for @xmath88 , a description with fixed @xmath89 is equivalent to one with fixed @xmath47 . we will use this initial temperature to characterize the system . it is interesting to note that in the thermodynamic limit , an instantaneous quench to _ any _ positive value of @xmath15 will not change the energy of the system . the field is initially symmetric in its distribution and the quench introduces an odd term into the hamiltonian . instead , the energy is redistributed between its kinetic , gradient , and potential parts , and as a result the final equilibrium temperature will necessarily differ from the initial temperature . in this investigation , however , this change is always less than @xmath90 percent . for open systems , where the system remains in contact with a heat bath , we refer the reader to ref . @xcite . due to the initial thermal distribution , defined entirely by the temperature @xmath47 , the sudden introduction of the cubic term into the potential actually has a minor effect on most of the field . in general , there is little net change in the forces present throughout the system , since the curvature of the single- and double - well potentials are the same at @xmath38 and the majority of the field is indeed close to this value ( see figure 2 ) . that is , the linear equation of motion remains unchanged for the small - amplitude fluctuations that represent the peak of the field s probability distribution . the few locations where the field is more responsive to the perturbation correspond to regions with relatively large absolute values of @xmath25 . these values define the tails of the probability distribution . the new nonlinear term present in the potential breaks the symmetry in the equation of motion , and this second - order effect causes larger - amplitude field fluctuations with negative ( positive ) values to experience a larger ( smaller ) restoring force than before . overall , a net positive force is imposed on the symmetrically prepared field and , as a consequence , its average starts at @xmath91 and then begins to move toward @xmath92 , the maximum of the potential barrier between the two states . we will see that @xmath45 is initially driven more out of equilibrium than all of the fluctuations about it . this is not surprising , as longer wavelength modes have longer relaxation time - scales @xcite . it is natural , then , to distinguish the dynamics of the two from each other . first , define @xmath93 and let @xmath94 represent the fluctuations about this average . then , @xmath95 where @xmath96 by definition . in the hha , one assumes that these fluctuations and their momenta remain gaussian in their distribution . one then obtains an equation of motion for @xmath97 , @xmath98 where we have introduced the effective potential for @xmath99 , @xmath100\phi_{\rm ave } + \frac{1}{2}\,m_{\rm h}^2(t)\,\phi_{\rm ave}^2 -\frac{\alpha_{\rm c}}{3}\,\phi_{\rm ave}^3+\frac{1}{8}\,\phi_{\rm ave}^4,\ ] ] that depends explicitly on the mean - square variance of the fluctuations through the time - dependent hartree mass @xmath101 . , just after the quench , for various initial temperatures @xmath47 . the two vertical lines correspond to the inflection points of the bare ( @xmath102 ) potential . they are also the turning points for the potential at @xmath103 . at @xmath104 the turning point is @xmath92.,width=288,height=288 ] @xmath105 at @xmath48 is shown in figure [ v_eff ] for various temperatures . for sufficiently low initial temperatures ( how low will become clear soon ) , @xmath97 will oscillate about the lhs minimum of the effective potential , @xmath106 . it will then begin to relinquish its energy to higher modes through nonlinear scattering , and its amplitude will decay . it is possible , then , to approximately bound this region of oscillation from the set of turning points for @xmath107 [ with @xmath108 retaining its @xmath48 value ] , @xmath109 these values give the location of @xmath97 where @xmath110 , and @xmath111 vanishes . when @xmath112 there are four such turning points because of the symmetry about @xmath92 . the primary region of interest is when @xmath113 , when fluctuations begin to probe the spinodal region of @xmath114 . by using eq . [ tp ] , we can predict that @xmath97 will probe this region when @xmath115 , or @xmath103 . clearly , this is also where the leading - order hartree approximation fails . it is interesting to consider also the case when @xmath116 , or @xmath117 . there are only two turning points for @xmath99 , since @xmath105 is now a single - well potential symmetric about @xmath118 . the fluctuations are large enough that the local maximum in the bare potential has little influence in the dynamics , and @xmath119 . the fluctuations actually probe beyond both minima of the potential . this case is similar to the phenomena of symmetry restoration in continuous phase transitions , and so in this context @xmath117 would be considered the critical temperature , @xmath120 . for all higher temperatures , @xmath119 @xcite . in this work , we are only concerned with temperatures well below @xmath120 , in fact below @xmath121 , where @xmath122 ( see figure 4 ) . in order to further quantify the dynamics in the hha , we derive the set of differential equations for the time evolution of the coupled two - point correlation functions . we define the translationally invariant quantities @xmath123 upon taking the time - derivative of each expression , the equations of motion can then equivalently be written in @xmath7-space . they are @xmath124 where @xmath125 $ ] . these equations , coupled with @xmath126 give the full dynamics of @xmath127 and @xmath97 and their corresponding two - point correlation functions @xmath128 and @xmath129 in the hha . these two groups of expressions are likewise coupled through the quantities @xmath130 and @xmath97 . the initial conditions are given by eqs . [ pimode ] , [ phimode ] , and also @xmath131 . these five first - order equations were solved numerically using a fourth - order runge - kutta algorithm . the same lattice spacing from the leap - frog algorithm was used . the calculation of @xmath132 is obtained by numerically averaging the radial function @xmath133 over the disk in @xmath7-space with radius @xmath134 , bounded by @xmath135 . there is a slight discrepancy between the area of rectangular phase space used in the two - dimensional numerical integrations and the circular phase space used in this method . in light of the previous discussion regarding the dependence of the correlation functions on the lattice spacing , this difference could possibly introduce error in the calculations of @xmath132 . to compensate for the smaller area used in the hha , the initial conditions for @xmath136 , given by eq . [ phimode ] , are multiplied by a normalization factor of 1.041 so that the radial integration gives the expected average @xmath137 at @xmath48 . the results will confirm the validity of this normalization method . compared to that obtained through the hartree approximation , for various initial temperatures . shown are @xmath138 ( short dash ) , @xmath139 ( long dash ) , and @xmath140 ( solid ) , as well as the corresponding hartree approximations ( dotted).,width=288,height=288 ] shown in figure [ phiavehartree ] is the numerical time evolution of @xmath97 compared to that obtained through the hha , for various initial temperatures . the agreement is clearly excellent at lower temperatures , where @xmath141 undergoes almost - linear oscillations about the minimum of @xmath105 : there is very little scattering between modes . at @xmath142 scattering begins to take place , allowing for a slow transfer of energy from low-@xmath7 modes to higher-@xmath7 modes not accounted for in the hha . we will soon show that this is also the temperature where the synchronous emergence of spatio - temporal order begins to take place . notice also the dependence of the oscillation frequencies on the initial thermalization temperature . compared to that obtained through the hartree approximation , for various initial temperatures . shown are @xmath138 ( short dash ) , @xmath139 ( long dash ) , and @xmath140 ( solid ) , as well as the corresponding hartree approximations ( dotted).,width=288,height=288 ] compared to that obtained through the hartree approximation , for various initial temperatures . shown are @xmath138 ( short dash ) , @xmath139 ( long dash ) , and @xmath140 ( solid ) , as well as the corresponding hartree approximations ( dotted).,width=288,height=288 ] figures [ corrpihartree ] and [ corrphihartree ] compare the numerical time - evolution of @xmath143 and @xmath144 with the corresponding hha predictions . the quantities are normalized to unity at @xmath48 through @xmath145 and @xmath83 , respectively . the agreement is again excellent at lower temperatures . at @xmath140 , however , it is evident that energy is being fed into the fluctuations of the field . the full numerical evolution depicts the amplitude of oscillation of @xmath143 decreasing at a rate larger than that predicted by the hha . similarly , the amplitude of oscillation of @xmath144 increases at a larger rate . again , this discrepancy is due primarily from the approximation s inability to account for direct scattering between modes . for various values of @xmath7 . also shown are the results from the corresponding hartree approximation ( dotted lines ) . the initial temperature is @xmath146 . the amplitudes decrease for increasing @xmath7 . shown , from the top down , are @xmath7=0.75 , 1.25 , 2.0 , and 4.75.,width=384,height=384 ] in figure [ speckhartree ] we show the transfer of energy between different @xmath7-modes . displayed is the time evolution of @xmath147 for indicative values of @xmath7 . the initial thermalization temperature is @xmath146 , corresponding to the value in which the majority of the field probes the spinodal region ( cf . section iv ) . this is well above the region of temperatures in which the hha is accurate . from the figure it is clear that scattering between low @xmath7 modes modes is taking place . the approximation is only accurate for shorter wavelength modes , due to the fact that they remain in equilibrium even after the quench . the breakdown of the hha at longer wavelengths implies that important nonlinear effects are strongly influencing the behavior of the field . in the next section we investigate the emerging dynamics of these modes . the results of the last section should have made it clear that at high temperatures nonlinear scattering leads to an efficient energy transfer between modes . this energy transfer will eventually lead to the thermalization of the field in the sdw potential . as predicted by the hha , at @xmath148 there is consistent energy transfer from the zero - mode to these higher-@xmath7 modes . at these initial temperatures , @xmath97 undergoes damped oscillations about its effective minimum . for @xmath146 , the amplitudes of these oscillations are large enough that @xmath97 just crosses the inflection point of the bare potential : roughly half of the field probes the spinodal region . ( solid line ) and @xmath149 ( dashed line ) for @xmath146.,width=288,height=288 ] figure [ corrandave ] illustrates the transfer of energy from @xmath97 to the fluctuations @xmath144 for @xmath146 . the quantity @xmath150 is introduced so that a visual comparison can be made between these two components of the field . notice the correlation between the decaying amplitude of oscillation for @xmath97 and the increasing amplitude of the fluctuations . at @xmath151 most of this energy transfer is complete , after which time the two quantities display smaller oscillations that are almost exactly out of phase with each other . for earlier times , @xmath152 , a highly nonlinear scattering process occurs , as the zero - mode of the field pumps its energy into its neighboring small-@xmath7 modes . figure [ spec60 ] shows the numerical time - evolution of @xmath147 for @xmath153 and @xmath152 , at @xmath146 . first , notice that growth occurs mainly for modes with @xmath154 . their maximum amplitude reaches a value that is up to 15 times greater than at @xmath48 . the oscillations then begin to decay and , for @xmath155 ( discussed later ) , the amplitudes of the modes take on a constant value , albeit still many times larger than their initial value . it is also apparent that for the time interval shown modes with @xmath156 are nearly time - independent , maintaining their initial thermal spectrum given by eq . [ phimode ] . this is true of all large-@xmath7 modes present in the numerical simulation , up to @xmath157 , which also indicates that the lattice spacing used has no effect on the emerging dynamics of the system . .,width=288,height=288 ] we would now like to search for possible spatio - temporal ordering during the field s approach to equilibrium . however , due to the noisy initial conditions , it is quite difficult to extract information about any emergent pattern , be it spatial and/or temporal . an added advantage to having an analytical expression for the initial thermal spectrum for all temperatures of interest ( cf . . [ phimode ] ) should be clear : a wiener filter is extremely successful at removing undesirable noise whenever its spectrum is known , all the while maximally preserving the nonperturbative fluctuations that emerge @xcite . with the filtered field , one can catalog and track the location of all local extrema in the system at each instant in time and for all initial temperatures . to distinguish the large - amplitude fluctuations from the long - wavelength thermal noise that is also passed by the filter , only the values of these extrema that are greater than the characteristic amplitude of the noise , @xmath158 , are considered . with ample sorting , a library is compiled containing the description of every large - amplitude fluctuation present during the entire evolution of the system ( the maximum time for the simulations is @xmath159 ) . for each fluctuation the following attributes are recorded : its core location @xmath160 , nucleation time @xmath161 , core amplitude @xmath162 ( measured at @xmath160 ) , radius @xmath163 ( measured at half maximum ) , its total number of oscillations @xmath164 , the period of each oscillation @xmath165 , and its lifetime @xmath166 . each of these attributes is time - dependent , with the exception of @xmath164 and @xmath166 . almost every large - amplitude fluctuation oscillates anharmonically about the background value @xmath97 , which is recognized by the fact that most of them reappear at approximately the same location with regular frequency . therefore , in general it is possible to establish the relationship @xmath167 , where @xmath168 is the average period of oscillation . the set of observables is further reduced by averaging @xmath169 , @xmath162 , and @xmath163 over their corresponding values at the maximum of each oscillation for the lifetime of the fluctuation . other than the nucleation time @xmath161 , each attribute is now time - independent . hereafter , we will omit the @xmath170 notation and assume that these values are indeed averages . with this information , a probability distribution function for the large - amplitude fluctuations is obtained for each initial temperature . this function , denoted as @xmath171 , can be used to calculate averages over any of its variables . similarly , one can integrate this function over a subset of variables to obtain a reduced distribution . in all cases , @xmath172 is normalized so that a numerical integration over all variables evaluates to unity . the bin widths of each variable ( as any one variable of @xmath172 will be visualized as a histogram ) will be specified below . ) . the darkest regions correspond to the most abundant fluctuations ( probability @xmath173 ) , with each lighter region then descending in intervals @xmath174.,width=288,height=288 ] for the moment , we focus our attention on the characteristics of fluctuations early in the evolution of the field , @xmath175 , and at an initial temperature @xmath146 . the quantity @xmath176 gives the reduced distribution function for the core amplitude and lifetime , and is shown in figure [ disttauphi ] . the bin widths are @xmath177 and @xmath178 . the darkest regions in the figure correspond to the most probable fluctuations . particularly noticeable is that fluctuations with core amplitudes @xmath179 and lifetimes @xmath180 dominate the distribution . one can also see that fluctuations with small amplitudes @xmath181 have mostly short lifetimes . there is a group of fluctuations , however , that have lifetimes @xmath182 and thus remain present in the system for substantially long periods of time . in addition , their core amplitudes are confined to a narrower region @xmath183 . also marked on the horizontal axes is the location @xmath184 , which corresponds to the difference between the local maximum and the left minimum of the effective potential for @xmath146 , given by eq . it follows that these long - lived oscillatory fluctuations have large enough amplitudes that their cores probe a narrow region centered about @xmath92 . in fact , the large majority of long - lived fluctuations probe beyond the inflection point of @xmath105 at @xmath185 . they are clearly nonperturbative in nature and would not be seen within the hha . in figure [ oscsnap ] we show a sequence of snapshots of the filtered field within the time interval @xmath186 . there are several features to be noted . [ clearly , they are more striking with a full animation . we invite the reader to view simulations in ref . @xcite . ] first , large - amplitude fluctuations seem to be of similar size ( spatially - bound configurations ) and spread throughout the lattice . second , they emerge in near synchrony . as we will see below , this global ordering persists for the first few oscillation cycles . third , as can be seen from figure [ disttauphi ] , these configurations can be quite long - lived . . simulations can be viewed in ref . @xcite.,width=384,height=768 ] we identify these spatially - bound , long - lived configurations with oscillons , which have been originally found as deterministic solutions of models with amplitude - dependent nonlinearities @xcite . in these studies , the field was prepared with a bubble - like profile ( typically a gaussian or a tanh ) , and for large enough initial radius and amplitude , observed to fall into an oscillon configuration , characterized by a rapid oscillation of its core , as shown in figure [ oscsnap ] . using harmonic perturbation analysis about an oscillon solution , it has been shown that such configurations require a minimum radius ( in 2 spatial dimensions for the sdw potential used here ) @xcite , @xmath187 we will now extend this result to the general adw potential . first , we parameterize the radially - symmetric oscillon configuration by a gaussian @xcite @xmath188,~\ ] ] where @xmath189 gives the time behavior at the core of the configuration and @xmath163 is its radial dispertion , assumed here to be constant . ( this assumption is equivalent to reducing the dynamics to one degree of freedom ; it has been shown to provide results in excellent agreement with numerical simulations @xcite . ) with this ansatz , a lagrangian can be obtained and , from it , the equation of motion for @xmath189 . linear perturbations about the solution , @xmath190 , oscillate with frequency @xmath191 note that @xmath192 has a minimum at @xmath193 . there is an unstable bifurcation at @xmath194 for the sdw , @xmath195 and the instability is given by eq . [ rc ] . oscillons are only possible for values of @xmath163 larger than @xmath196 . larger asymmetries imply smaller values of @xmath196 . in order to strengthen our claim that oscillons are nucleated during the nonequilibrium evolution of @xmath10 , the inset in figure [ dist ] shows the distribution functions @xmath197 and @xmath198 for the radii and period of oscillation , respectively . to make a comparison with deterministic oscillons , only fluctuations with @xmath199 ( i.e. those that probe beyond the local maximum of the potential ) and @xmath200 are considered that is , these are the lower bounds placed on the corresponding integrations over @xmath172 . the bin widths are @xmath201 . the fitted curves are gaussian functions with centers @xmath202 and @xmath203 and widths @xmath204 and @xmath205 , respectively . to complete the comparison , consider a deterministic oscillon evolving in a time - independent effective potential that is representative of this case . from figure [ corrandave ] a good approximation would be to take @xmath206 , so that @xmath207 . the effective mass can thus be obtained from the curvature of the left - hand well , given by eq . [ veff ] , and evaluates to @xmath208 . finally , from eq . [ rc ] , we find that the effective minimum radius for which the deterministic oscillon remains stable is @xmath209 . comparing this result with figure [ dist ] we see that , indeed , the emerging oscillons satisfy this requirement remarkably well . [ note that a proper comparison between oscillons that evolve in different potentials ( i.e. for fields at different initial temperatures ) always requires a suitable scaling of @xmath25 , @xmath210 , and @xmath211 . ] . also shown ( inset ) are the probability distributions ( solid lines ) for the radius @xmath163 and period @xmath212 of these oscillons . the fitted curves ( dashed lines ) are gaussian functions.,width=384,height=384 ] also shown in figure [ dist ] is the distribution of nucleation times for the emerging oscillons . this quantity gives the number of oscillons nucleated between @xmath211 and @xmath213 , with @xmath214 . it is obtained by calculating the reduced distribution function @xmath215 integrated over all the remaining variables . the same lower limits are used for the integrals over @xmath162 and @xmath166 , so that only long - lived large - amplitude fluctuations are considered . if there are @xmath216 such fluctuations , then the final expression for the number of oscillons nucleated as a function of time is @xmath217 the sharp peaks in @xmath41 at early times , @xmath218 , correspond to the synchronous emergence of oscillons . for @xmath155 , this global ordering gives way to only local ordering , in which oscillons emerge at arbitrary times with similar probability . even this local emergence disappears after approximately @xmath219 ( more on this later ) . notice the correlation between the synchronous nucleation of oscillons and the energy loss from the zero - mode ( c.f . figure [ corrandave ] ) at early times , @xmath220 . this phenomenon was observed for temperatures within @xmath221 ( see below ) . below this temperature range the hha is valid and no coherent structure emerges . above this temperature range the field separates into large , slowly evolving thin - walled domains , signalling the approach to criticality . in order to understand the mechanism behind the observed synchronous emergence , we decompose the field as in eq . [ sepphi ] . the linearized equation satisfied by the fourier modes of the fluctuations is @xmath222 where @xmath223 . \label{omegaparares}\ ] ] in this approximation , the fluctuations obey a linear equation of motion with a time - dependent frequency determined by the value of @xmath97 . equations of this type , generalized mathieu equations , are known to exhibit parametric resonance , which can lead to exponential amplification ( @xmath224 ) in the oscillations of @xmath225 at certain wavelengths . they have been the focus of much recent attention in studies of reheating of the universe after an inflationary expansion phase @xcite . to verify that this is the mechanism behind the synchronous amplification of oscillon modes early in the field s evolution , we use in eq . [ omegaparares ] the anharmonic solution for @xmath97 that would be obtained if the effective potential remained time - independent . that is , we require that @xmath97 oscillates in the effective potential as determined immediately after the quench ( i.e. , without energy loss ) , just as the parametric driving force in generalized mathieu equations oscillates with constant amplitude . in this sense , we are considering the initial response of the fluctuations to the oscillations of @xmath97 , without accounting for the subsequent transfer of energy between the two . only with this approach can one achieve a constant exponential amplification of @xmath226 and thus a good measure of resonance . even though this approximation only strictly holds for early times , the results obtained will confirm its validity in describing the essential physics behind the observed synchronous emergence . for the bottom - most contour and increasing in increments of @xmath227.,width=288,height=288 ] figure [ parares ] shows lines of constant exponential amplification rate @xmath228 of the fluctuations @xmath225 for various @xmath7 and @xmath47 . at low temperatures , @xmath229 , no modes are ever amplified . as the temperature is increased , so is the amplitude of oscillation in @xmath97 , gradually causing the band @xmath230 to resonate . it is helpful at this point to consider what modes comprise a typical oscillon . to do so , we use a gaussian ansatz for the oscillon with a fixed radius @xmath163 , @xmath231 @xcite . the fourier transform is itself a gaussian with a characteristic radius @xmath232 . thus , a gaussian in real space is generally comprised of the band @xmath233 in @xmath7-space . earlier [ cf . figure [ dist ] ] it was found that the average radius of the emergent oscillons is @xmath202 at an initial temperature @xmath146 , which in this approximation corresponds to the band @xmath234 . this agrees very well with what is seen in figure [ parares ] . although in this linear approximation the entire band does not resonate until higher temperatures , it is expected that in the full equation of motion the nonlinear coupling between modes will broaden the resonance band at lower temperatures . we have observed that the synchronous emergence of oscillons occurs for initial temperatures @xmath235 . the lower bound is directly related to the nonlinear scattering between modes , which is absent at low temperatures . for temperatures above @xmath236 , the appearance of large domains precludes the identification of oscillons . this is typical of the approach to criticality , where large spatial correlations in the field become more probable , destroying any local ordering . thus , the observed quench - induced synchronous emergence of order is limited to temperatures sufficiently below @xmath120 . in the introduction , we mentioned that long - lived locally - ordered structures may serve as bottlenecks to equipartition , as they slow down the distribution of energy between the system s various degrees of freedom . below , we present evidence that oscillons do precisely that . a more formal proof is left for future work . a good observable to follow is the time evolution of @xmath237 for different modes , which should become flat when thermal equilibrium is achieved and the equipartition theorem is satisfied . in highly dissipative systems ( such as those governed by a langevin equation of motion with a large viscosity coefficient ) , energy exchange between each mode and the thermal bath tends to dominate the dynamics . in the nonequilibrium closed system considered here , however , energy is only exchanged between the modes themselves , and the thermalization process turns out to be quite different . for various modes ( see text ) . the data is boxcar averaged over a window @xmath238 . also shown ( inset ) is the time evolution of @xmath127 , which attains equilibrium noticeably earlier than any of the higher modes shown.,width=288,height=288 ] in figure [ thermal ] we show the time evolution of @xmath237 for the values @xmath7=0.25 , 0.5 , 0.75 , 1.0 , 1.25 , and 1.5 ( top to bottom lines , respectively ) . the curves are time - averaged over a window @xmath238 so that the large oscillatory behavior at early times ( @xmath218 ) is suppressed . also shown in the inset is the time evolution of @xmath239 , which exhibits the thermalization trend of the zero mode . @xmath127 displays damped oscillatory motion before reaching its equilibrium value at @xmath240 . this is roughly the time in which the mode @xmath241 reaches its maximum average kinetic energy . the mode @xmath242 also increases from @xmath48 to a maximum value at about the same time before decaying . for modes @xmath243 , however , the general trend is to slowly absorb energy as they asymptotically approach their equilibrium values from below on time - scales that can be much _ longer _ than the lower-@xmath7 modes . there are two clear trends : small modes ( with @xmath244 ) grow very quickly and then relax to their final equilibrium values , while larger modes ( with @xmath245 ) slowly approach theirs . thus , it is reasonable to conclude that most of the zero - mode s initial energy goes into its neighboring modes , those responsible for oscillon - like configurations [ cf . figure [ parares ] ] , and only later on it is shared amongst larger @xmath7 modes . this seems to suggest that there is a _ two - step energy cascade _ , a fast one at early times to small-@xmath7 modes , followed by a slower one to larger @xmath7 modes . the early emergence of oscillons is consistent with this scenario . we can further quantify this argument by introducing a measure of the entropy of the system based on the partitioning of the kinetic energy , @xmath246 : @xmath247 where @xmath248 is the probability distribution of the kinetic energy in @xmath7-space . @xmath249 attains its maximum value when equipartition is satisfied , and @xmath250 where @xmath251 is the area of this space . this occurs both at the initial thermalization @xmath252 and final equilibrium states , since in these cases all modes carry the same fractional kinetic energy . at these times we recover the thermodynamic entropy of the system , @xmath253 , which is a measure of the number of accessible states in the system . [ the numerical implementation gives @xmath254 , where @xmath41 is the number of lattice points , or degrees of freedom . ] in figure [ s ] we show the change of @xmath249 from the initial state , @xmath255 , for the system at @xmath146 . at late times ( @xmath256 ) , we have found that the system equilibrates exponentially in a time - scale @xmath257 . at early times , the localization of energy at lower @xmath7 modes , corresponding to the synchronous emergence of oscillons , prolongs this approach to equipartition . the inset of figure [ s ] shows the large oscillations in the early evolution of @xmath249 ( dotted line ) . also shown ( solid line ) is the average between successive peaks of @xmath249 , with a plateau at @xmath258 that coincides with the maximum oscillon presence in the system . this is to be contrasted with an exponential approach to equilibrium . thus , we see evidence that oscillon configurations indeed serve as early bottlenecks to equipartition , temporarily suppressing the cascading of energy from low-@xmath7 modes to higher modes . from the initial state for closed systems at @xmath146 . the exponential approach to equilibrium is clear at late times . the inset illustrates the role of oscillons as a bottleneck to equipartition.,width=288,height=288 ] we will now investigate the dynamics with the adw potential , that is , for values of @xmath259 . homogeneous nucleation predicts that , had the system been prepared in the metastable state at @xmath38 at some temperature @xmath47 , its decay rate per unit area would be controlled by the arrhenius factor , @xmath260 $ ] , where @xmath261 is the free - energy barrier for the nucleation of a critical nucleus or bounce @xcite . ( we set @xmath262 . ) it is often overlooked that this result is quite sensitive to how the initial state is prepared . it assumes that the bounce is nucleated within a homogeneous background , by adopting a perturbative gaussian approximation for the evaluation of the partition function . in other words , it assumes that no large - amplitude perturbations are present in the system @xcite . it is also well - known that there are discrepancies between the theoretical prediction from homogeneous nucleation and a large array of experimental results concerning nucleation rates , from he@xmath263 @xcite and nematic liquid crystals @xcite to martensitic materials and polycrystals @xcite to fluids @xcite and numerical simulations @xcite . more often than not , theoretical estimates greatly underestimate nucleation rates . surely , many of these discrepancies are related to limited knowledge of the nonequilibrium processes leading to nucleation at the atomic or mesoscopic scale for various materials . however , we propose here that in certain cases the problem is rooted on the detailed set up of the initial metastable state . in particular , the quench we propose in this work , which may be mimicked in experimental situations , generates an instability capable of greatly accelerating the decay of the metastable state . in fact , we have observe that , for a range of temperatures , the nucleation rate per unit area is not that of homogeneous nucleation , @xmath264 $ ] , but a power law , @xmath265 , where @xmath266 is the critical nucleus free energy and the exponent @xmath1 is weakly temperature - dependent . rn stands for resonant nucleation . the power law applies precisely within the temperature range where we observe the synchronous emergence of oscillons at early times , @xmath267 , as we show next . the change in @xmath15 will not modify the basic parametric amplification mechanism discussed above in the context of the sdw : for a range of temperatures , oscillons are still nucleated in synchrony at early times . the difference is that , for @xmath37 , the system must reach the global free - energy minimum at @xmath268 . [ of course , for large enough temperatures , @xmath269 , and/or small energy barriers the system will go straight into the global minimum , a cross - over decay . we are not interested in these cases . ] in the usual treatment , the critical nucleation is described as the saddle - point in field - configuration space , the unstable direction dictated by the negative eigenvalue associated with the growth of the critical nucleus . this is given by the solution of the equation @xmath270 , with appropriate boundary conditions , which determines the 2d bounce @xcite . in the absence of forcing , which is the case for homogeneous nucleation , the system will randomly probe phase space until it eventually `` finds '' the saddle point . the presence of quench - induced large - amplitude fluctuations in the field will drastically accelerate its decay . in figure [ decay - phi ] we show the evolution of the order parameter @xmath107 as a function of time for several values of asymmetry , @xmath271 for @xmath146 . not surprisingly , as @xmath272 , the field remains longer in the metastable state , since the nucleation energy barrier @xmath273 at @xmath274 . however , a quick glance at the time axis shows the fast decay time - scale , of order @xmath275 . for comparison , for @xmath276 , homogeneous nucleation would predict nucleation time - scales of order @xmath277 ( in dimensionless units ) , respectively . [ the nucleation barriers are , @xmath278 and @xmath279 . ] clearly , while for smaller asymmetries @xmath97 displays similar oscillatory behavior to the sdw case before transitioning to the global minimum , as @xmath15 is increased ( asymmetry in potential is increased ) the number of oscillations decreases sharply . at @xmath146 for several values of the asymmetry . ( from left to right , @xmath280 . ) each curve is an ensemble average over 100 runs.,width=288,height=288 ] in figure [ powerlaw ] we show the log - log plot of the ensemble - averaged nucleation time - scales as a function of the nucleation barrier , or critical droplet free energy , @xmath281 for different temperatures , @xmath282 and @xmath283 . the error bars are computed from the dispersion of the measured time scales within the ensemble . the best fit is a power law , @xmath284 , with @xmath285 for @xmath286 and @xmath146 , and @xmath287 for @xmath140 . this simple power law holds for the same range of temperatures where we have observed the synchronous emergence of oscillons . [ the range is similar for both sdw and adw . ] it is not surprising that the exponent @xmath1 is larger for @xmath140 , where the synchronous emergence of oscillons is marginal : at lower temperatures , the oscillations of the field s zero mode have lower amplitude and oscillons become rarer . in fact , as we remarked before , we have not observed any for @xmath288 for the sdw . for lower temperatures , we should thus expect a smooth transition into the exponential time - scales of homogeneous nucleation . we intend to analyse this transition and to obtain an analytical explanation for the power law in a forthcoming publication . as a first step in this direction , we present below what we believe is the mechanism by which the transition completes for different nucleation barriers . for resonant nucleation as a function of critical nucleation free - energy barrier @xmath281 at @xmath140 , @xmath289 , and @xmath283 . the best fit is a power - law with exponent @xmath290 for @xmath286 and @xmath283 and @xmath291 for @xmath140.,width=288,height=288 ] first , note that the error bars in figure [ powerlaw ] increase with nucleation barrier . this spread in time scales is related to the dynamics of the system . for @xmath292 the radius of the nucleation bubble diverges , @xmath293 . clearly , due to the disparity between the power law and the exponential time scales , the mechanism by which the system approaches the global minimum is not through a random search in configuration space as is the case for homogeneous nucleation . instead , we argue that oscillons will act as seeds for the nucleation of a critical fluctuation . the way in which this happens depends on the nucleation barrier . if @xmath294 , the critical nucleus will have a radius much larger than a typical oscillon [ cf . figure [ dist ] ] ; it will appear as two or more oscillons coalesce . as @xmath15 is increased further , the radius of the critical nucleus decreases , approaching that of an oscillon . in this case , a single oscillon may become the critical nucleus and promote the decay of the metastable state . this explains the small number of oscillations on @xmath97 as @xmath15 is increased [ cf . figure [ decay - phi ] ] . to corroborate our argument , in figure [ radius ] we contrast the critical nucleation radius with that of oscillons for different values of energy barrier at @xmath146 . the average oscillon radius [ cf . figure [ dist ] ] is @xmath295 , and is identified by the dashed line , with the dotted lines denoting its dispersion . the critical nucleus radius , @xmath296 , is equal to @xmath297 for @xmath298 , or @xmath299 . thus , for @xmath300 one can not distinguish between a single oscillon and a critical bubble . since , as we have seen , oscillons appear very early on , the decay happens quite fast . for @xmath301 , approximately , the critical nucleus is the result of the coalescence of two or more oscillons . they diffuse through the lattice and scatter , forming bound states , somewhat as in kink - antikink breathers in 1d field theory @xcite . we emphasize that these values are only approximate ; they do not include the increase in coalescence rates due to diffusive motions on the lattice or an enhancement on the unstable growth of oscillons due to perturbations . we expect that the one - oscillon decay will ensue for smaller values of @xmath15 . in @xcite the reader can see a few representative simulations . clearly , these arguments can be refined . in particular , the details of oscillon coalescence remain unexplored . however , they encapsulate the basic mechanism by which metastable states decay via resonant nucleation , as seen in numerical simulations . . the dashed line shows the average radius for nucleated oscillons , while the dotted lines denote the dispersion of its value.,width=288,height=288 ] we have investigated the nonequilibrium dynamics of a double - well model with a 2d scalar order parameter . the system was initially prepared in thermal equilibrium at a fixed temperature in a anharmonic single - well potential , @xmath302 we then turned on a cubic interaction with tunable strength so that the resulting double - well potential can be symmetric ( sdw ) or asymmetric ( adw ) . the turning - on of the interaction can be interpreted as a `` quench , '' as it occurs within a time - scale faster than any in the system . the system is thus tossed into a double - well potential and , through its nonlinear interactions , redistributes its energy until it reaches equipartition . for the sdw case , we have observed that this approach to equilibrium occurs as a two - step energy cascade : quickly at first , as the zero - mode relinquishes its energy to nearby ( in @xmath7-space ) long - wavelength modes , and slower later , as the energy is partitioned with the short - wavelength modes . although we have adopted the language of nonequilibrium field theory , we remark that one could equally well treat the system as a microcanonical ensemble with fixed energy @xmath89 . we showed that , for a range of initial temperatures , the initial stages of the dynamics are dominated by damped oscillations of the field s zero mode , which , via parametric resonance , excite the modes associated with coherent structures named oscillons . we also showed that the initial emergence of oscillons occurs at the breakdown of the homogeneous hartree approximation . furthermore , this emergence occurs in synchrony , characterizing a global emergence phenomenon . as oscillons are long - lived coherent structures , they act as bottlenecks for equipartition . the observed synchronous emergence dies away with the damping of the zero - mode oscillations . oscillons then appear randomly throughout the lattice , until such a time when there is nt enough energy in long - wavelength modes . we then investigated the dynamics in adws , where the final equilibrium state is at the global minimum of the potential . the system was thermalized as for the sdw , in its metastable well . also as before , the quenching induces damped oscillations of the field s zero mode and oscillons emerge . we observed that , approximately for the same range of temperatures where oscillons emerged in the sdw case , the decay rate was greatly enhanced , when compared to homogeneous nucleation results . in fact , we obtained an excellent fit to a _ power law decay _ time scale , @xmath303^b$ ] , where @xmath261 is the energy of the critical nucleus or bounce and @xmath1 is a numerical exponent , @xmath304 , for the temperatures we investigated . we argued that there were two main mechanisms for what we called resonant nucleation : for small asymmetries , the critical nucleus appears as two or more oscillons coalesce . for larger asymmetries , a single oscillon grows to become the critical nucleus , promoting the decay of the metastable state . the greatly accelerated decay is due to the resonant oscillations of the zero mode , induced by the quenching . it is thus plausible that such behavior may be observed in several related systems with a scalar order parameter . the emergence of oscillons in granular materials appears due to the sinusoidal vibrations of their container @xcite ; this is mimicked here by the oscillation of the field s zero mode , which corresponds to the periodic oscillation of the whole lattice . as for granular materials , patterns emerge above a certain amplitude . we do not claim to have modelled the behavior of granular materials here , but believe to have captured some of the essential physics . also , since the model here falls in the ising universality class , we expect that similar qualitative behavior should occur , for example , in ferromagnets and possibly binary liquids and metal alloys , if long wavelength oscillations can be induced by the quenching process . j. d. gunton , m. san miguel , and p. s. sahni , in _ phase transitions and critical phenomena _ , ed . c. domb and j. l. lebowitz , v. 8 ( academic press , london , 1983 ) ; j. d. gunton , j. stat . phys . * 95 * , 903 ( 1999 ) . here is an incomplete list of references : g. n. felder and l. kofman , phys . d * 63 * , 103503 ( 2001 ) ; p. b. greene and l. kofman , phys . b * 448 * , 6 ( 1999 ) ; p. b. greene , l. kofman , a. d. linde and a. a. starobinsky , phys . d * 56 * , 6175 ( 1997 ) ; d. boyanovsky , h. j. de vega and r. holman , arxiv : hep - ph/9701304 ; d. boyanovsky , m. dattanasio , h. j.de vega , r. holman and d. s. lee , phys . d * 52 * , 6805 ( 1995 ) ; d. boyanovsky , h. j. de vega and r. holman , arxiv : hep - ph/9903534 .

this paper focuses on the polarized profiles of resonance scattering lines that form in magnetized disks . optically thin lines from keplerian planar disks are considered . model line profiles are calculated for simple field topologies of axial fields ( i.e. , vertical to the disk plane ) and toroidal fields ( i.e. , purely azimuthal ) . a scheme for discerning field strengths and geometries in disks is developed based on stokes q - u diagrams for the run of polarization across line profiles that are doppler broadened by the disk rotation . a discussion of the hanle effect for magnetized disks in which the magnetorotational instability ( mri ) is operating is also presented . given that the mri has a tendency to mix the vector field orientation , it may be difficult to detect the disk fields with the longitudinal zeeman effect , since the amplitude of the circularly polarized signal scales with the net magnetic flux in the direction of the observer . the hanle effect does not suffer from this impediment , and so a multi - line analysis could be used to constrain field strengths in disks dominated by the mri .

1010.1979

spectropolarimetry continues to be a valuable technique for a broad range of astrophysical studies ( e.g. , adamson 2005 ; bastian 2010 ) , including applications for circumstellar envelopes . advances in technology and access to larger telescopes means an ever growing body of high quality spectropolarimetric data . it is therefore important that the arsenal of diagnostic methods and theoretical models in different astrophysical scenarios keep pace . this paper represents the fifth in a series devoted toward developing the hanle effect as tool for measuring magnetic fields in circumstellar media from resonance line scattering polarization . the observational requirements for the effects examined in this series are ambitious : high signal - to - noise ( s / n ) data and high spectral resolving power . however , these demands are being met , as illustrated by harrington & kuhn ( 2009a ) in a spectropolarimetric survey of circumstellar disks at h@xmath0 . there are numerous effects that can influence the polarization across resolved lines . a number of researchers have investigated the effects of line opacity for polarization from thomson scattering ( wood , brown , & fox 1993 ; harries 2000 ; vink , harries , drew 2005 ; wang & wheeler 2008 ; hole , kasen , & nordsieck 2010 ) . scattering by resonance lines can generate polarization similar to dipole scattering ( e.g. , ignace 2000a ) . with high s / n and high spectral resolution , harrington & kuhn ( 2009b ) have identified a new polarizing effect for lines that coincides with line absorption . an explanation for this previously unobserved effect in stars is discussed by kuhn ( 2007 ) and is attributed to the same dichroic processes detailed by trujillo bueno & landi deglinnocenti ( 1997 ) for interpreting polarizations in some solar spectral lines . generally associated with circular polarization , the zeeman effect has received acute attention of late as a result of techniques that co - add many lines ( donati 1997 ) . the method has been used successfully in many studies ( see the review of donati & landstreet 2009 ) . in relation to massive stars , the technique has led to the detection of magnetism in several stars ( e.g. , donati 2002 , 2006a , 2006b ; neiner 2003 ; grunhut 2009 ) . in fact , there exists a large collaborative effort called the mimes collaboration ( e.g. , wade 2009 ) to increase the sample of massive stars with well - studied magnetic fields . another method that has been used productively in studies of solar magnetic fields involves the hanle effect ( hanle 1924 ) . this is a weak zeeman effect that operates when the zeeman splitting is roughly comparable to the natural line broadening . monographs that deal with atomic physics and polarized radiative transfer provide excellent treatments of the hanle effect , including stenflo ( 1994 ) and landi deglinnocenti & landolfi ( 2004 ) . this contribution extends a series of theoretical papers that has enlarged the scope of its general use with circumstellar envelopes . building on work developed in the solar physics community , and using expressions for resonance line scattering with the hanle effect ( e.g. , from stenflo 1994 ) , ignace , nordsieck , & cassinelli ( 1997 ; paper i ) provided an introduction of the hanle effect for use in studies of circumstellar envelopes . ignace , cassinelli , & nordsieck ( 1999 , paper ii ) considered its use in simplified models of disks with constant radial expansion or solid body rotation as pedagogic examples . both of these papers approximated the star as a point source of illumination . ignace ( 2001 ; paper iii ) then incorporated the finite source depolarization factor of cassinelli , nordsieck , & murison ( 1987 ) into the treatment of the source functions for the hanle effect in circumstellar envelopes . finally , ignace , nordsieck , & cassinelli ( 2004 , paper iv ) calculated the hanle effect in p cygni wind lines using an approximate treatment for line optical depth effects . the focus of this fifth paper in the series is the hanle effect in polarized lines from keplerian disks , of relevance for accretion disks and be star disks ( e.g. , cranmer 2009 ) . as in previous papers of this series , the sobolev approximation ( sobolev 1960 ) for optically thin scattering lines is adopted for exploring the run of polarization with velocity shift across line profiles . this paper adopts the notation laid out in paper ii . the results of paper ii are expanded for keplerian rotation with the inclusion of finite source depolarization ( ala paper iii ) and stellar occultation . resonance line scattering polarization from such disks were explored in ignace ( 2000a ) for unmagnetized disks without consideration of the hanle effect . drawing on these previous works , methods for computing the polarization of line profiles will be briefly reviewed in section [ sec : lines ] . in section [ sec : hanle ] , results for the hanle effect from disk lines are described . there are three applications that will be considered : purely axial fields ( i.e. , normal to the disk plane ) , purely toroidal fields ( i.e. , azimuthal in the disk plane ) , and field topologies that can arise from the magnetorotational instability , or mri ( balbus & hawley 1991 ) . conclusions of this study and observational prospects are presented in section [ sec : conc ] . appendices detail analytic derivations for special cases of the hanle effect in keplerian disks . the focus of this paper is thin scattering lines from a planar circumstellar disk in which the gas obeys circular keplerian motion . to describe this geometry and the line scattering polarization , a number of coordinates will be needed , which are introduced here . * a cartesian coordinate system for the star is assigned to be @xmath1 . the @xmath2 axis is the rotation axis of the star and disk . * another one for the observer is @xmath3 . the observer is located at large positive distance along the @xmath4 axis . * spherical coordinates in the star frame are @xmath5 . cylindrical coordinates are identified by @xmath6 . * cylindrical coordinates in the observer frame are @xmath7 . * spherical angular coordinates defined in a frame of the local magnetic field at any point will be @xmath8 . the latter system is needed to evaluate the hanle effect . * the @xmath4 axis is taken to be inclined to the @xmath2 axis by an angle @xmath9 . * the circumstellar disk assumed to be axisymmetric and exists entirely in the equatorial plane of the star , and so it is located at @xmath10 . the scattering angle between a point in the disk to the observer is signified by @xmath11 . figure [ fig1 ] shows a spherical triangle used in relating the observer and star coordinate systems . the sobolev approximation is employed to describe the polarimetric line profiles . the sobolev approach relies on identifying `` isovelocity zones '' , which are the locus of points that share the same doppler shift for a distant observer . the doppler shift in frequency of any point in the scattering volume is determined by @xmath12 where @xmath13 is the frequency of the line transition , and @xmath14 is the line - of - sight velocity shift given by @xmath15 a keplerian disk follows a velocity profile of the form @xmath16 where @xmath17 is the innermost radius of the disk with tangential speed @xmath18 at that location . it is convenient to define dimensionless variables for distances and velocities . for the normalized cylindrical radius in the planar disk , @xmath19 is used . the line - of - sight velocity shift then becomes @xmath20 a normalized velocity is also introduced with @xmath21 . the solution for the isovelocity zones in terms of @xmath22 is thus given by @xmath23 this describes a loop path that starts at @xmath17 at an azimuth @xmath24 on the front side of the disk , where @xmath25 , and then ends again at @xmath17 and @xmath26 . the loop extends to a maximum distance at azimuth @xmath27 with a value of @xmath28 . the system is left - right mirror symmetric , with redshifted velocities on one side and blueshifted ones on the other . the convention here is that the disk rotates counterclockwise as seen from above ( i.e. , `` prograde '' ) . an example is displayed in figure [ fig2 ] . with the arrow toward the observer as the reference for @xmath29 , redshifted velocities come from the interval @xmath30 , and blueshifted ones come from @xmath31 . for optically thin scattering , the observed flux of scattered light at a given velocity shift in the line is determined by a volume integral along the associated isovelocity loop . to describe the polarization , a stokes vector prescription is adopted , with standard @xmath32 notation . referring back to figure [ fig1 ] , an edge - on disk with @xmath33 would yield intensities with @xmath34 and @xmath35 . in this case `` north '' is chosen to be along the direction of @xmath36 , and @xmath34 corresponds to oscillations of the electric vectors of the light being preferentially parallel to north . for unresolved disks the corresponding observed fluxes are @xmath37 , where it is assumed that the stokes - v flux is zero for the problem at hand . generally following the notation of paper ii and using results from ignace ( 2000a ) for keplerian disks , the disk is assumed to have a power - law surface number density given by @xmath38 for power - law exponent @xmath39 and inner value @xmath40 at @xmath41 . note that in the case of the be stars , values of @xmath39 range from 34 ( e.g. , waters 1986 ; lee , osaki , & saio 1991 ; porter 1999 ; jones , sigut , & porter 2008 ) , and @xmath42 will be adopted in example cases . implicit is that the lines of interest form from the dominant ion . temperature and ionization variations or different disk densities could be included in the model , but the intent of this contribution is to highlight the line polarization and hanle effect , and so a simple power - law density is adopted as a baseline for the analysis . the stokes fluxes for thin lines from this disk are given by @xmath43 the denominator in the integrand arises from the sobolev approximation , with @xmath44 . the factor @xmath45 will more conveniently be written as @xmath46 in what follows . the vector @xmath47 represents the `` scattering factor '' with @xmath48 . it is this factor that incorporates the dipole scattering of resonance line polarization , the impact of magnetic fields via the hanle effect , and the finite depolarization factor . finally , the scaling factors outside the integral are from equations ( 24 ) and ( 25 ) of paper ii , with @xmath49 where @xmath50 is the frequency integrated line cross - section , and @xmath51 where @xmath52 is the specific luminosity of the star and @xmath53 is the distance to the star from earth . before progressing to a consideration of the hanle effect , there is value in formulating the solution for the line scattering polarization in the zero field case . this is because the hanle effect primarily _ modifies _ the scattering polarization , meaning that a good understanding of the non - magnetic case is a necessary reference for understanding the hanle effect . many papers have dealt with line formation from keplerian disks . this work draws in particular on the work of huang ( 1961 ) and rybicki & hummer ( 1983 ) for planar disks , and also ignace ( 2000a ) for scattering polarization without the hanle effect . there are two main cases that are explored here : the simplistic case of illumination by a point source and then the more realistic case involving the effects of the finite size of the illuminating star . analytic and semi - analytic solutions are described in the appendices . the value of the point source approximation is that it gives context for how finite stellar size effects modify the line scattering polarization . the only portion of the point source approach that is reproduced here from appendix [ sec : iso ] is the expression for _ isotropic _ scattering of starlight by a planar disk in the sobolev approximation . for isotropic scattering of light from a point star , one has @xmath54 . only the stokes - i flux survives in equation ( [ eq : flux ] ) . it is convenient to introduce a change of variable for computing the integration of that equation , with @xmath55 . then the flux becomes @xmath56 where @xmath57 . the appearance of the factor of 2 arises because the integration for the front half of the loop ( from @xmath24 to @xmath27 ) is the same as for the back half . appendix a details analytic cases for the above case . the addition of polarization and finite source effects amount to inserting multiplicative `` weighting '' functions that modify the above integrand that describes the emission contribution as a function of location along a loop . there are several finite source effects that can be included , such as finite star depolarization ( cassinelli 1987 ) , stellar occultation effects ( e.g. , fox & brown 1991 ) , limb darkening effects ( brown , carlaw , & cassinelli 1989 ) , and absorption of starlight by the disk ( ignace 2000a ) . only two of these effects will be considered here : finite star depolarization and stellar occultation of the disk . limb darkening could be included ; however , limb darkening mainly `` softens '' the finite star depolarization factor , making it slightly less severe . thus the inclusion of limb darkening does not seem to be a pressing issue for illustrating the hanle effect in disks . absorption by the disk does affect the shape of the stokes - i profile ; however , in the current treatment it does not influence @xmath47 which determines the line scattering polarization . consequently , absorption affects the continuum level of direct starlight at the line , but not the scattered stokes fluxes , , or . on the other hand , a photospheric absorption line would alter the line shapes of the scattered flux profiles ; however , results presented here will be in terms of ratios of scattered fluxes , @xmath58 and @xmath59 , for which the influence of a photospheric line will cancel . these ratios and should be considered as polarimetric `` efficiencies '' . they are not generally what an observer would actually measure , because the denominator involves only the scattered light of the stokes - i flux . instead an observer would normally measure , , and @xmath60 , the latter being the sum of the direct starlight ( first term ) and the scattered light ( second term ) . in this case @xmath61 , and likewise for the stokes - u polarization , assuming that the scattered flux in the line is small compared to the direct flux by the star itself . as a result , @xmath62 . it is worth noting that the and `` efficiency '' profiles are independent of the line optical depth in the thin limit , since , , and scale linearly with . equation [ eq : hizero ] gives the scattering function components of in the case of no magnetic field with the star treated as a point star of illumination . modifications to those functions that allow for scattering by a finite sized and uniformly bright central star were determined in paper iii . for the case of zero magnetic field , the scattering function is shown explicitly with factors arising from the finite stellar size . these factors also appear in the scattering function with the hanle effect in the same positions as without the hanle effect . the vector components of are : @xmath63 \\ h_q & = & \frac{3}{8}e_1\,{\mbox{$\mu_\ast$}}\,\left[\sin^2 i \right . \\ & & \left . i \right)\,\cos 2\varphi \right ] \\ h_u & = & \frac{3}{8}e_1\,{\mbox{$\mu_\ast$}}\,\cos i\,\sin 2\varphi , \end{array } \right . \label{eq : finite}\ ] ] where @xmath64 with the dilution factor @xmath65 , and @xmath66 is the finite depolarization factor when there is no limb darkening . the factors @xmath67 and @xmath68 represent corrections to the scattering function that account for the effects of finite star size in relation to the incident radiation field at a scattering point around the star . scattering by resonance lines can be approximated as part dipole and part isotropic . the parameter @xmath69 is a factor with values from 0 to 1 that represents the extent to which a resonance line scatters like a dipole radiator ( see chandrasekhar 1960 ) . only the dipole portion of the scattered light contributes to observable polarization . in the solar literature , it is more common to use the notation @xmath70 for the fraction of scattered light that is dipole - like ( e.g. , stenflo 1978 ) . accounting for stellar occultation breaks the back - front symmetry of the isovelocity loops . the front loop still has lower and upper limits of @xmath71 and 1 , but the back half has limits of @xmath71 and @xmath72 . here @xmath73 is associated with the minimum value of @xmath74 at the back - projected limb of the star corresponding to a maximum azimuth of @xmath75 . an expression for @xmath76 is given in ignace ( 2000a ) , expressed here as @xmath77 this expression provides a cubic relation in @xmath73 as a function of fixed viewing inclination @xmath9 and velocity shift . numerical results for polarized profiles without the hanle effect are shown in figure [ fig3 ] . this figure and the ones that will follow use @xmath42 and @xmath78 . figure [ fig3 ] should be compared to figure [ figapp1 ] for the point star case . for point star illumination , the maximum polarization occurs at the line wings and @xmath79 by symmetry . this changes dramatically when the effects of the finite stellar size are included . in figure [ fig3 ] , each panel has 5 curves color - coded for different viewing inclinations : @xmath80 is red , @xmath81 is green , @xmath82 is deep blue , @xmath83 is light blue , and @xmath84 is magenta . the values correspond to viewing inclination angles of @xmath85 and @xmath86 , respectively . the upper left panel displays profiles that are normalized to have unit area . in the lower right panel for the polarization , the finite depolarization factor shifts peak polarization to lower velocity shifts as compared to the point star case , and remains symmetric about line center . stellar occultation leads to the survival of a small polarization as shown in the lower right panel . the upper right panel is a - plot across the respective line profiles . note that the scales in and differ by about an order of magnitude . these small loops signify position angle rotations across the observed polarized line profiles , with polarization position angle @xmath87 given by @xmath88 . we next turn our attention to the hanle effect . the results of this section will prove valuable in following how the hanle effect modifies the run of polarization across the line profile . the hanle effect applies to resonance line scattering and can be interpreted in semi - classical terms as a precession of an oscillating emitter that occurs over the radiative lifetime of the line emission . an angular quantity can be defined to represent the effective amount of precession , as given by @xmath89 where @xmath90 is the angular larmor frequency , @xmath91 is the einstein radiative rate for the transition of interest , @xmath92 is the land factor of the upper level , @xmath93 is the magnetic field strength in gauss , and is the hanle field sensitivity defined as @xmath94 with @xmath95 the radiative rate normalized to @xmath96 hz . the hanle effect manifests itself in the scattering physics with the appearance of cosine and sine functions of the angle ( and a related quantity @xmath97 ) within the scattering functions . note that with no magnetic field , @xmath98 and @xmath99 . the limit of a high larmor frequency yields @xmath100 . in this latter case , the hanle effect is said to be `` saturated '' . there are a few rules of thumb to help understand the hanle effect in polarized lines . first , modifications to the polarized profile will depend on the orientation of the field in relation to both the direction of incoming radiation and outgoing radiation ( i.e. , the observer ) . second there is no hanle effect when the incoming radiation field is symmetric about the field direction . these first two `` rules '' indicate that there is no hanle effect for radial magnetic fields ( assuming a spherically symmetric star and no diffuse radiation ) and that only non - radial components of the field contribute to modifying the polarization . however , the third point is that in spite of this , the angle is sensitive to the _ total _ field strength : both radial and non - radial components , at the site of scattering . fourth , and finally , the hanle effect is sensitive only to the field topology in the saturated limit , not the field strength . with these rules of thumb , consideration of the hanle effect in three particular cases follow : a purely axial field , a purely toroidal field , and last a scenario involving fields as arising from the operation of mri in disks . in this case an axial magnetic field is envisioned as threading the disk perpendicular to the equatorial plane , hence @xmath101 . as for the field distribution throughout the disk , the following is adopted as a conservative limiting case : @xmath102 and so the hanle angle is determined by @xmath103 where @xmath104 for @xmath105 a field strength at the inner disk radius . it is important to note that if @xmath106 , then the hanle effect is weak everywhere in the disk . if @xmath107 , then the inner disk will be in the saturated limit ; however , there will be a transition to a weak hanle effect at some radius in disk , characteristically where @xmath108 . as a general rule , for a given magnetized disk , different lines will have different values of , and thus will be sensitive to different aspects of the field at different locations in the disk . expressions for in the point star limit are given in appendix [ app : axial ] . revisions to those expressions for finite star depolarization and stellar occultation is the same as in equation ( [ eq : finite ] ) . results are shown in figure [ fig4 ] . profiles of and are displayed at left and right , respectively . panels from top to bottom are now for different viewing inclinations of @xmath109 and 1.0 . different colors are for different values of @xmath110 with @xmath111 ( red ) , 0.1 ( green ) , 1.0 ( dark blue ) , 10 ( light blue ) , and 100 ( violet ) . the stokes - q profiles are symmetric about line center , whereas the u profiles evolve from weak and anti - symmetric , owing to stellar occultation , to strong and mostly symmetric , to a null profile at the strongly saturated limit . in the point star case , the saturated limit yields @xmath112 and @xmath113 for all velocity shifts . with finite source effects , is mostly flat throughout the central portion of the profile , but goes to zero in the wings . in the presence of limb darkening , the profile would become more flattened toward the wings . figure [ fig5 ] displays - curves across the line profile . note that the two axes have different scales : variations in are smaller than for . the colored curves are the same cases as shown in figure [ fig4 ] . viewing inclinations are indicated . for the edge - on case of @xmath33 , the curves are all degenerate with @xmath112 . the red loop is for a very weak field and is seen to be top - bottom symmetric in this space . as the field is increased , and profiles become individually more nearly symmetric . in the - space , this results in curves with only small loops , for example the modest field case with the dark blue curve . as the disk enters the saturated limit , drops to zero , becomes somewhat flat - topped in appearance ( again , except at the line wings ) , which degenerates mainly to a point in the - space for most velocity shifts , with extension to zero polarization only for the line wings . a completely azimuthal field configuration has also been considered . as in the axial field case , the toroidal field strength is assumed to decrease inversely proportional to @xmath114 , and so equation ( [ eq : tanalt ] ) remains valid . this means that differences in the model polarized profiles between the axial and toroidal field configurations arise because the hanle effect is sensitive to the vector field orientation . as with previous considerations , some analytic results are derivable in the point star limit , and these are detailed in appendix [ app : toroidal ] for a disk with a toroidal field . with a toroidal field , the geometry associated with determining the hanle effect is more complex than in axial field case . geometrical relationships between the various angles defining the scattering problem with a toroidal field in a disk were given in the appendix of paper ii and will not be repeated here . results for the calculation of polarized profiles , including the effects of the star s finite extent , are shown in figure [ fig6 ] . this figure is presented in the same manner as figure [ fig4 ] for the axial field case : is at left and at right ; the panels are for different values of @xmath115 ; the colors are for different field strengths characterized by @xmath110 . note the marked differences in the line polarization between the toroidal case and that of an axial field . the stokes - u profiles are always antisymmetric . like the axial case , profiles are symmetric , but the behavior is quite different . for example the edge - on disk case at lower left indicates that the profile polarization completely changes sign , from positive definite at every velocity shift when @xmath98 to everywhere negative in the saturated limit . this implies a @xmath86 rotation in position angle between these two limiting cases for the entire profile . for intermediate field strengths , there is a position angle rotation across the profile that occurs at different velocity shifts . in the axial field case , there is never a position angle rotation for an edge - on disk . for the - shapes , differences in the polarizations are especially clear , with results for the toroidal field in figure [ fig7 ] to be compared against those for an axial field in figure [ fig5 ] . what is the source of these differences between the axial field and the toroidal one ? the field orientation with respect to the viewer is clearly the key to the interpretation . for the axial case , the field at every point in the disk has some component of the magnetic vector directed toward the observer ( if seen at @xmath116 ) or away from the observer ( if seen at @xmath117 ) . thus , an axial field leads to a net projected magnetic flux of one sign or the other , which is true for every isovelocity loop . the toroidal field is manifestly different , since one side of the disk has components toward the observer and the other side has them away . for an axisymmetric toroidal field , the projected net magnetic flux is identically zero for the entire disk , but is oppositely signed in isovelocity loops for blueshifted velocities versus redshifted ones . in terms of the semi - classical precession description , flipping the field by @xmath118 amounts to a precession in the opposite direction . in the axial field case , the precession of the radiating oscillator is uniform entirely clockwise ( cw ) or counterclockwise ( ccw ) . both cw and ccw precessions occur for a toroidal field configuration , with one precession occurring in half the line profile , and the opposite for the other half . to illustrate this effect , figure [ fig8 ] shows line profile results for a toroidal field that now goes in the other direction as compared to figure [ fig6 ] . the profiles are nearly the same , but the profiles are nearly mirror images of one another . slight differences are a result of the stellar occultation , which does _ not _ flip when the field orientation is reversed . a large literature has developed over the last twenty years in relation to the magnetorotational instability ( mri ; balbus & hawley 1991 , 1998 ) . there is an interesting history about this effect ( e.g. , balbus 2003 ) . the instability can be illustrated through an analogy to two masses in a differentially rotating disk that are slightly offset from each other along a radius . these two masses are connected by a spring , to represent the effect of an axial magnetic field . the end result is that the two masses on different orbits seek to increase their displacement from one another . coupling by the spring leads to a runaway situation ensues . key for the context of magnetic diagnostics is how this instability impacts the field topology and strength throughout the disk . the mass - spring analogy above has built into it the assumption of flux freezing . the separation of the masses along with the differential rotation would appear to evolve the axial field through the disk into a toroidal one . different researchers have studied the operation of the mri in accretion disks through both semi - analytic work and numerical simulations ( e.g. , balbus & hawley 1991 , 1992 ; hawley & balbus 1991 , 1992 ; hawley , gammie , & balbus 1995 ; hawley 2000 ; fromang & stone 2009 ; lesur & longaretti 2009 ; maheswaran & cassinelli 2009 ) . the goals of the mri simulations are to understand better the physics of angular momentum transport through disks and disk structure . however , this paper seeks better insight into whether and how disk magnetism might be directly detected , with a focus on the zeeman and hanle effects for spectral lines . ignace & gayley ( 2008 ) reported on a simplistic calculation of the zeeman effect and the hanle effect for a keplerian disk with a purely toroidal field . here , application of the hanle effect to keplerian disks has been developed in a more complete way . but before discussing its application to the mri scenario , it is worth commenting on the conclusions of ignace & gayley for the use of the zeeman effect . the mri leads to a field topology that consists of a toroidal component and a randomized component . the toroidal components exists in mixed polarity , by which it is meant that some sectors run cw and others run ccw . for the weak longitudinal zeeman effect , the stokes - v flux scales with the net magnetic flux associated with a spatial resolution element ( e.g. , mathys 2002 ) . with bulk motions frequency ( or wavelength or velocity ) resolution ultimately maps to geometrical zones at the source ( an example of this involving the sobolev approximation was detailed by gayley & ignace 2010 for spherical winds ) . the key point is that if both the randomized field and the polarity of the toroidal component changes on small scales , then the flux of circularly polarized light arising from the zeeman effect will be strongly suppressed owing to little net magnetic flux . the issue of variations of the field on `` small scales '' must be evaluated with care . in the sobolev approximation for purely keplerian rotation , we have seen that the isovelocity zones are `` loop '' structures . these loops extend out to large radius for low velocity shifts , but they can be quite small at high velocity shifts . the highest velocity shifts degenerate to a pair of points for either limb of the star at the projected stellar equator ! more importantly , every isovelocity zone intersects with the innermost radius of the disk . ( if the disk extended down to the star , this would be the photospheric radius . ) a steep power - law density ensures that the bulk of emission or scattered light comes from inner disk radii . as a result , the zeeman effect would be most sensitive to a magnetic field at these locations , and thus `` small scales '' refers to turnovers in the field direction that are small compared to the inner radius of the disk . consequently , the detection of the zeeman effect from a disk with a given spectral resolution and density distribution constrains the characteristic spatial wavelengths at which the field turns over with radius from the star , azimuth around the star , and vertical height through the disk . a detection of the zeeman effect in a disk has previously been reported by donati ( 2005 ) in the case of fu ori . this is thought to be an accretion disk with a disk wind , as evidenced by some lines showing p cygni absorption . however , the circular polarization profile is argued as being associated with the innermost region of the rotating disk where the field is of kilogauss strengths . the detection implies a net magnetic flux per spectral resolution element , and thus sets limits on the turnover ( or `` tangledness '' ) length scale for the disk field , if indeed the mri is operating in this case . it is interesting to consider signals that could result with the hanle effect . the hanle effect can operate in regions where magnetic fields are `` tangled '' or randomized . this means that spatial averages of @xmath119 tend toward zero although @xmath120 does not , such as is the case for the mri mechanism . an extensive literature exists for the hanle effect with random fields in applications to solar studies ( e.g. , frisch 2009 , and references therein ) . here i simply want to outline some of the limiting behavior in applications to disks . consider a keplerian disk that contains everywhere a truly randomized magnetic field . if the field is weak at all locations , meaning that @xmath121 , then of course a line profile results as in the case of no hanle effect . but if the field is strong , such that the denser regions of the disk are largely saturated , then the behavior is much different . with different field orientations , one expects that will yield a null profile , by symmetry considerations . however , the profile is different . as a specific example , consider the resulting line from an edge - on disk . without a field , the polarization at line center would be zero , owing to forward scattering of unpolarized starlight . with a randomized field , the polarization will still tend toward zero at line center . in the point star limit , some net polarization is expected to survive in the line wings . this polarization will be significantly reduced in comparison to the zero field case . if @xmath69 were unity , the polarization at the line wings would be 100% , since the scattering geometry is @xmath86 . in analogy with considerations of scattering polarization off the solar limb , a reduction in polarization by a factor of 5 for isotropically distributed fields should be expected ( see stenflo 1982 ) . now consider the introduction of a sustained toroidal field component . of course toroidal fields were considered in the previous section . now however , the toroidal field has polarity flips within the disk @xmath122 is alternately cw or ccw at essentially random points within the isovelocity zones . what this means is that there is a sign change in the direction of larmor precession in the classical picture of a harmonic oscillator . the effect of this is to drive the signal to zero faster than if the toroidal field had one sense of polarity . in the saturated limit , the profile is _ unchanged _ , because the surviving polarized signal does not depend on polarity at all . this is quite different from the zeeman effect that is sensitive to the net magnetic flux . for the zeeman effect , the circular polarization will be suppressed when @xmath122 switches polarity on small spatial scales . polarized line profile shapes from magnetized keplerian disks have been calculated under a number of simplifying assumptions : the disk is geometrically thin ; the scattering lines are optically thin ; primarily simple fields were considered ( axial or toroidal ) ; and no account was taken of photospheric absorption lines . on the other hand , the model profiles do include finite source depolarization and the effects of stellar occultation . the presentation of results focused on the polarimetric `` efficiencies '' and with a description of how to identify the occurrence of the hanle effect in scattering lines from disks . in addition , a discussion was presented for the hanle effect from a magnetized disk in which the mri mechanism is operating . one of the main conclusions from this work is that axial and toroidal fields in disks are easily distinguishable through an analysis of - figures for the run of stokes polarizations across line profiles . although a profile does exist even without the hanle effect , owing to stellar occultation , its amplitude is quite small . the strongest profiles result when much of the inner disk , where most of the scattered light is produced , has values of @xmath123 of order a few . if the inner disk is mostly in the saturated limit of the hanle effect , the profile becomes a null profile for both axial and toroidal fields . one should bear in mind that the polarized efficiencies are upper limits to the polarizations that would actually be measured , since the efficiencies are with reference to the scattered light only and do not take account of the direct starlight from the system . since this starlight is expected to be largely unpolarized , its contribution acts to `` dilute '' the polarization substantially below the efficiency levels reported here . for example , if the line is relatively weak at 20% of the continuum level , then the expected measured polarizations would have fractional values at about 1/5 of the efficiency values , resulting in line polarizations of around 1%2% for and 0.5% or less for . as percent polarizations , such values would appear to be easily measurable , but in practice there are several challenges . first , a spectral resolution yielding several points across the polarized profile is needed . circumstellar disks have rotation speeds of order 500 km s@xmath124 . a requisite velocity resolution of perhaps 50 km s@xmath124 would then be needed , implying a resolving power of @xmath125 . harrington & kuhn ( 2009a ) have demonstrated that such resolving powers can be achieved in spectropolarimetry ; however , the next requirement is that of finding suitable scattering lines . the very interesting effects seen in the sample of harrington & kuhn ( 2009a ) exploits a relatively new effect of enhanced polarization that is coincident with regions of higher line absorption , a consequence of optical pumping effects . for the hanle effect , or even for non - magnetic resonance scattering , lines that are predominantly scattering are needed . for hot star disks , such as the disks of be stars , resonance scattering lines are generally to be found at uv wavelengths . this requires space - borne spectropolarimeters . although the wisconsin ultraviolet photo - polarimeter experiment ( wuppe ) obtained exciting new results from uv polarimetry , its resolving power was only about 200 ( nordsieck 1994 ) . a new uv spectropolarimeter called the far - ultraviolet spectropolarimeter ( fusp ) is a sounding rocket payload that will have a resolving power of about 1800 ( nordsieck 1998 ) . although possibly two low for circumstellar disks , it will be suitable for studying scattering line polarizations from high velocity stellar wind sources . as a matter of practical analysis , how should spectropolarimetric data best be managed to measure a hanle effect ? plotting the velocity shifted polarizations in the - space appears to be most promising . the sequencing of the analysis for scattering lines from disk sources might proceed as follows : * subtract off the continuum polarization in the vicinity of the line of interest . this will counter the effects of both interstellar polarization and any other broadband source polarization , such as may arise from thomson scattering in the disk . * plot the q and u line _ fluxes _ , not fractional or percent polarizations that would be derived through normalization by the total i - flux . the reason for plotting polarized fluxes is that many common uv resonance lines are doublets ( e.g. , nv , siiv , and civ uv resonance doublets ) . the shorter wavelength component ( `` blue '' ) has @xmath78 , and the long wavelength component ( `` red '' ) has @xmath126 ( e.g. , tab . 1 of paper ii ) . if the lines are thin , then the polarized flux from the red line will not be influenced by the blue one . however , if the lines are sufficiently closely spaced , then will be a blend , and normalization by that blend would artificially skew the q - u figure shape , making interpretation more difficult . if the doublet components are well separated , then relative polarizations could also be used in what follows . * determine whether the resultant figure for the line polarization shows any axis of symmetry . if so , then either ( a ) there is no hanle effect , or ( b ) there is a hanle effect with a dominant toroidal component . for an axis of symmetry , a rotation of the figure from observer q - u axes to a source set of axes q-u could be accomplished with a mueller rotation matrix . after rectifying the figure in this way , the relative amplitude of should be compared to . if @xmath127 , then there is likely little hanle effect or the disk is in the saturated limit of the hanle effect . if @xmath128 , then a hanle effect is required with @xmath129 few in the disk where the bulk of scattered light is produced this means @xmath130 . * if there is no symmetry axis to the q - u figure , then a hanle effect involving an axial field is most likely the culprit . recall that a radial field could be present , but this would give no hanle effect by itself . * for identifying a field distribution arising from mri , things are more complicated . if the hanle effect is in operation , then the stokes - u flux is likely driven to zero , even if @xmath131 . a net profile should survive ; however , it may not be symmetric . using an oversimplification , each isovelocity zone can be considered to have a different effective @xmath110 value . this is already the case in the pure axial or toroidal field case that is axisymmetric , but the variation of the effective @xmath110 with velocity shift is monotonic ( in a flux - weighted sense ) . with mri , one expects non - monotonicity from fluctuations of the field strength . this amounts to the introduction of amplitude fluctuations across the polarized line . there are complications to the above approach . for example , in the case of the be star disks , there is substantial evidence for one - armed spiral density wave patterns that make the disk density non - axisymmetric ( e.g. , okazaki 1997 ; papaloizou & savonije 2006 ) . for thin thomson scattering , there is little observational consequence of this effect which is mainly a redistribution of disk scatterers in a point antisymmetric pattern ( e.g. , ignace 2000b ) . however , with resolved line profiles , the non - axisymmetric density distribution will produce asymmetry in the profile and will be a new source of net signal , also asymmetric . certainly it will be important to model the profile self - consistently along with the polarized profiles . note that there may be some concern about the choice adopted for the radial distribution of the field strength through the disk . the choice of @xmath132 is perhaps the most shallow distribution that one could expect . a steeper gradient of the field strength will restrict the operation of the hanle effect to a more restricted range of radii in the disk . if the interval in radius where @xmath133 becomes narrow in relation to where most of the scattered light is produced , then the hanle effect will be irrelevant for the observed polarization . what are the next steps in formulating better diagnostics of the density and magnetic field structure in disks ? most of work considered in this series of papers has focused on optically thin lines in an attempt to gain a better understanding of how the hanle effect influences line polarizations that form in circumstellar media . in paper iv the issue of optical depth effects for p cygni lines from stellar winds were considered by treating regions of high optical as contributing no polarization at all . although simplistic , insight was gained into how optical depth effects provide an additional spatial `` filter '' in terms of where bulk of line polarization will be produced . naturally , rigorous radiative transfer and a more realistic disk model ( i.e. , not simply planar ) is badly needed to extend the considerations of this paper . even with thin line scattering , it would be useful to explore how non - axisymmetric disk models , such as the one - armed spiral density wave pattern for be disks , will modify the line polarization . moreover , winds driven off magnetized disks ( e.g. , knigl 1989 ; proga , stone , & drew 1998 ; proga , stone , & kallman 2000 ) have been ignored entirely in this work . these are all new calculations that will need to be considered in the future . the author is indebted to the anonymous referee whose comments have improved this manuscript . appreciation is expressed to ken gayley , joe cassinelli , and gary henson for engaging conversations about line polarizations and the zeeman and hanle effects . this work was supported by a grant from the national science foundation ( ast-0807664 ) . in these appendices , a number of instructional cases are considered for the polarized line profiles shapes from a keplerian disk when the illuminating star is treated as a point source . this means that both stellar occultation and the finite star depolarization factor are ignored . a consequence of this approximation is that a non - zero profile can _ only _ result from the hanle effect . before considering polarized line profiles , stokes - i profile shapes are derived for the case of isotropic scattering . these solutions form the base emissivity function from which the polarized lines are constructed . isotropic scattering corresponds to @xmath126 , and it means there is no polarization from resonance line scattering . of course , that also means there is no hanle effect , regardless of the field strength . even though there is no hanle effect , the isotropic case is useful to explore as a reference for the production of the stokes - i line shape . the integrand for the line emission as a function of velocity shift in the observed line represents the contributions by the disk density and the sobolev effect for the profile shape . allowing for @xmath134 and the hanle effect simply represents new weighting functions for non - isotropic scattering that multiply the integrand from the isotropic case . the flux of line emission at normalized doppler shift @xmath135 is @xmath136 where the factor of 2 arises from the back - front symmetry of the integration along the isovelocity zone . as a reminder , @xmath55 and @xmath137 . again the preceding expression is only valid in the point star approximation . the power law exponent @xmath39 is from the surface density distribution that is assumed to be a power law of the form @xmath138 . this formulation leads to symmetric double - peaked line profile shapes for @xmath139 . larger values of @xmath39 result in profiles that have more pronounced double - horns at greater velocity shifts from line center . for integer values of @xmath39 , the integral is analytic , and solutions for @xmath140 and @xmath141 are given here by way of example . for @xmath140 the result is @xmath142 and for @xmath141 , @xmath143 note that increasing values of @xmath39 also result in profile shapes of lower amplitude . in fact it is possible to solve for the integrated light from a resonance scattering line in some special cases . defining @xmath144 a few selected results are @xmath145 , @xmath146 , @xmath147 , and @xmath148 . it would appear that with @xmath149 , different values of @xmath150 , and thus lines of different brightness levels , could result . this seems counterintuitive if lines characterized by different values of @xmath39 have the same optical depth . however , @xmath151 depends only on the density at the inner portion of the disk @xmath40 , not the value of @xmath39 . thus when the line is optically thin , the appropriate optical depth to use is one that is angle averaged , similar in spirit to @xmath152 in brown & mclean ( 1977 ) for optically thin electron scattering . hence use of a new optical depth parameter @xmath153 would ensure that lines of different @xmath39 values with _ isotropic _ scattering will have the same total line emission ( i.e. , `` area under the curve '' ) even though they have different profile shapes . for resonance line scattering with @xmath134 but with @xmath98 everywhere , the vector scattering function @xmath154 greatly simplifies . following paper ii , we have that @xmath155 , @xmath156 , @xmath157 , @xmath158 , and @xmath159 . the components of the phase function become @xmath160 \label{eq : hizero } \\ h_q & = & \slantfrac{3}{8}e_1\,\left[\sin^2 i -\left(1+\cos^2 i \right)\,\cos 2\varphi \right ] \label{eq : hqzero } \\ h_u & = & \slantfrac{3}{8}e_1\,\cos i\,\sin 2\varphi . \label{eq : huzero } \end{array } \right.\ ] ] the stokes flux of scattered light is @xmath161 note that care must be taken in dealing with terms that are odd and even in @xmath29 between angles of 0 and @xmath162 . for example , @xmath163 is odd around the loop , ensuring that @xmath164 . accounting for the odd / even terms , and using the fact that @xmath165 , solutions for the scattered flux in stokes - i and stokes - q are : @xmath166 note at the extrema of the line wings , @xmath167 and @xmath168 , and @xmath169 always . the resultant polarization across the line is entirely in stokes - q when the observer s reference axes are aligned with the symmetry axis of the star . with @xmath78 , figure [ figapp1 ] shows a plot of @xmath170 for @xmath140 ( left ) and @xmath141 ( right ) at different viewing inclination angles of @xmath171 and @xmath86 along the top panels . the profiles are normalized with respect to the total emission produced if the line had been isotropically scattering . at bottom is the relative fractional polarization @xmath172 for @xmath140 and @xmath141 for the same viewing inclinations . note that at the edges of the line , @xmath173 for @xmath78 , independent of the viewing inclination , as expected under the point star approximation . the goal here is to illustrate the polarimetric `` efficiency '' . the actual measured fractional polarization would be much smaller owing to dilution by direct starlight . these efficiencty curves are relatively smooth functions of velocity shift . this smoothness is partly due to the fact that dipole scattering is a fairly slowly varying function of location around the disk and also because isovelocity loops sample a range of scattering angles . for an axial magnetic field with @xmath175 , we have that @xmath176 and @xmath177 . the scattering phase functions are quite similar to the zero field case , except that now the hanle effect appears in factors in the functions @xmath178 and @xmath179 . using paper ii , the phase functions are given by @xmath180 \\ h_q & = & \slantfrac{3}{8}e_1\,\left[\sin^2 i -\left(1+\cos^2 i \right)\,c(\varpi,\varphi ) \right ] \\ h_u & = & \slantfrac{3}{4}e_1\,\cos i\,d(\varpi,\varphi ) . \end{array } \right . \label{eq : axial}\ ] ] where @xmath181 where @xmath182 and @xmath183 solutions for the vector stokes flux is no longer analytic . with the hanle effect , @xmath184 except for @xmath98 or in the saturated limit . in the latter case of @xmath185 everywhere , an analytic solution can be obtained , which is given by @xmath186 note that in this limit , the relative polarization becomes @xmath187 which is a constant across the profile and a function of viewing inclination only . this means that the polarized profile is flat - topped . figure [ figapp2 ] displays profiles for the axial field case with @xmath78 at a fixed value of @xmath188 but with different field values at the inner disk radius of @xmath41 , of @xmath189 to achieve a large dynamic range in hanle ratios throughout the disk . as @xmath110 increases , the location where @xmath190 moves outward to @xmath191 . the upper left panel in figure [ fig8 ] shows normalized profiles of @xmath170 ; lower panels show the fractional polarizations and ; and upper right shows the - shapes . the color sequencing is the same as in figure [ fig5 ] . in large part the effect of an axial field is to rotate and foreshorten the q - u segments relative to the zero field case . for a toroidal magnetic field with @xmath193 , the spherical geometry for the scattering problem is moderately complex . it is not possible to write down simple complete expressions for the phase scattering functions . but as demonstrated in paper ii , there are still some special cases that are analytic . for example in the saturated limit , @xmath194 , and the @xmath195 and @xmath196 fluxes become examples of scattering and polarized profiles are shown in figure [ figapp3 ] as well as q - u plots across the polarized profiles . in the previous section , @xmath199 was used for an axial field to give a significant signal in . for this toroidal field case , an edge - on disk with @xmath200 was used . the style of this figure is the same as figure [ figapp2 ] . the a major distinctive in relation to a disk with an axial field is that a toroidal field leads to profiles that are antisymmetric instead of symmetric . adamson , a. , aspin , c. , davis , c. , fujiyoshi , t. ( eds ) 2005 , astronomical polarimetry : current status and future directions , asp conf . # 343 balbus , s. a. 2003 , araa , 41 , 555 balbus , s. a. , hawley , j. f. 1991 , apj , 376 , 214 balbus , s. a. , hawley , j. f. 1992 , apj , 400 , 610 balbus , s. a. , hawley , j. f. 1998 , revs . of mod . physics , 70 , 1 bastian , p. ( ed ) 2010 , astronomical polarimetry 2008-science from small to large telescopes , asp conf . ser . , to appear brown , j. c. , mclean , i. s. 1977 , a&a , 57 , 141 brown , j. c. , carlaw , v. a. , cassinelli , j. p. 1989 , apj , 344 , 341 cassinelli , j. p. , nordsieck , k. h. , murison , m. a. , 1987 , apj , 317 , 290 chandrasekhar , s. 1960 , radiative transfer ( new york : dover ) cranmer , s. r. 2009 , apj , 701 , 396 donati , j .- f . , babel , j. , harries , t. j. , howarth , i. d. , petit , p. , semel , m. 2002 , mnras , 333 , 55 donati , j .- f . , howarth , i. d. , jardine , m. m. , petit , p. , catala , c. , landstreet , j. d. , 2006a , mnras , 370 , 629 donati , j .- f . , howarth , i. d. , bouret , j .- c . , petit , p. , catala , c. , landstreet , j. 2006b , mnras , 365 , l6 donati , j .- f . , landstreet , j. d. , 2009 , araa , 47 , 333 donati , j. f. , paletou , f. , bouvier , j. , ferreira , j. , 2005 , nature , 438 , 466 donati , j .- f . , semel , m. , carter , b. d. , rees , d. e. , collier cameron , a. , 1997 , mnras , 291 , 658 fox , g. k. , brown , j. c. 1991 , apj , 375 , 300 frisch , h. , anusha , l. s. , sampoorna , m. , nagendra , k. n. 2009 , a&a , 501 , 335 fromang , s. , stone , j. m. 2009 , a&a , 507 , 19 gayley , k. g. , ignace , r. 2010 , apj , 708 , 615 grunhut , j. h. , wade , g. a. , marcolino , w. l. f. , petit , v. , henrichs , h. f. , 2009 , mnras , 400 , l94 hanle , w. 1924 , z. phys . , 30 , 93 harrington , d. m. , kuhn , j. r. , 2009a , apjs , 180 , 138 harrington , d. m. , kuhn , j. r. , 2009b , apj , 695 , 238 harries , t. j. 2000 , mnras , 315 , 722 hawley , j. f. , 2000 , apj , 528 , 462 hawley , j. f. , balbus , s. a. 1991 , apj , 376 , 223 hawley , j. f. , balbus , s. a. 1992 , apj , 400 , 595 hawley , j. f. , gammie , c. f. , balbus , s. a. 1995 , apj , 440 , 742 hole , k. t. , kasen , d. , nordsieck , k. h. 2010 , apj , 720 , 1500 huang , s .- s . 1961 , apj , 133 , 130 ignace , r. , 2000a , a&a , 363 , 1106 ignace , r. , 2000b , in the be phenomenon in early - type stars , m. a. smith , h. f. henrichs ( eds ) , iau colloq . # 175 , asp , conf . 214 , 452 ignace , r. , 2001 , apj , 547 , 393 ( paper iii ) ignace , r. , cassinelli , j. p. , & nordsieck , k. h. , 1999 , apj , 520 , 335 ( paper ii ) ignace , r. , gayley , k. g. , 2008 , in clumping in hot - star winds , ( eds ) w .- r . hamann , a. feldmeier , l. m. oskinova , ignace , r. , nordsieck , k. h. , & cassinelli , j. p. , 1997 , apj , 486 , 550 ( paper i ) ignace , r. , nordsieck , k. h. , & cassinelli , j. p. , 2004 , apj , 609 , 1018 ( paper iv ) jones , c. e. , sigut , t. a. a. , porter , j. m. 2008 , mnras , 386 , 1922 knigl , a. 1989 , apj , 342 , 208 kuhn , j. r. , berdyungina , s. v. , fluri , d. m. , harrington , d. m. , stenflo , j. o. 2007 , apj , 668 , l63 landi deglinnocenti , e. , landolfi , m. 2004 , polarization in spectral lines , astrophysics and space library , v. 307 ( kluwer : dordrecht ) lee , u. , osaki , y. , saio , h. 1991 , mnras , 250 , 432 lesur , g. , longaretti , p .- y . 2009 , a&a , 504 , 309,101 maheswaran , m. , cassinelli , j. p. 2009 , mnras , 394 , 415 mathys , g. 2002 , in astrophysical spectropolarimetry , ( eds . ) j. trujilo - bueno , f. moreno insertis , f. sanchez , 101 neiner , c. , geers , v. c. , henrichs , h. f. , floquet , m. , fremat , y. , hubert , a .- m . , 2003 , a&a , 406 , 1019 nordsieck , k. h. , marcum , p. , jaehnig , k. p. , & michalski , d. e. 1994 , proc spie , 2010 , 28 nordsieck , k. h. 1998 , in ultraviolet - optical space astronomy beyond hst ( asp conf series ) , in prep . okazaki , a. t. 1997 , a&a , 318 , 548 papaloizou , j. c. b. , savonije , g. j. 2006 , a&a , 456 , 1097 porter , j. m. 1999 , a&a , 348 , 512 proga , d. , stone , j. m. , drew , j. e. 1998 , mnras , 295 , 595 proga , d. , stone , j. m. , kallman , t. r. 2000 , apj , 543 , 686 rybicki , g. b. , hummer , d. g. 1983 , apj , 274 , 380 sobolev , v. 1960 , moving envelopes of stars ( mass . : harvard univ . ) stenflo , j. o. 1978 , a&a , 66 , 241 stenflo , j. o. 1982 , solar phys . , 32 , 41 stenflo , j. o. 1994 , solar magnetic fields ( dordrecht : kluwer ) trujillo bueno , j. , landi deglinnocenti , e. , 1997 , apj , 482 , l183 vink , j. s. , harries , t. j. , drew , j. e. 2005 , a&a , 430 , 213 wade , g. a. , alecian , e. , bohlender , d. a. , bouret , j .- c . , grunhut , j. h. , henrichs , h. , 2009 , in cosmic magnetic fields : from planets , to stars and galaxies , iau symp . # 259 , 333 waters , l. b. f. m. 1986 , a&a , 162 , 121 wang , l. , wheeler , j. c. 2008 , araa , 46 , 433 wood , k. , brown , j. c. , & fox , g. k. 1993 , a&a , 271 , 492

we summarize the multiwavelength properties of x - ray sources detected in the 80 ks xmm - newton observation of the groth - westphal strip , a contiguous strip of 28 hst wide - field planetary camera 2 ( wfpc2 ) images . among the x - ray sources detected in the xmm - newton field of view , 23 are within the wfpc2 fields . ten spectroscopic redshifts are available from the deep extragalactic evolutionary probe ( deep ) and canada - france redshift survey ( cfrs ) projects . four of these show broad mg ii emission and can be classified as type 1 agns . two of those without any broad lines , nevertheless , have [ nev ] emission which is an unambiguous signature of agn activity . one is a narrow - line seyfert 1 and the other a type 2 agn . as a followup , we have made near - infrared ( nir ) spectroscopic observations using the ohs / cisco spectrometer for five of the x - ray sources for which we found no indication of an agn activity in the optical spectrum . we have detected h+[nii ] emission in four of them . a broad h component and/or a large [ nii]/h ratio is seen , suggestive of agn activity . nineteen sources have been detected in the band and four of these are extremely red objects ( eros ; ) . the optical counterparts for the majority of the x - ray sources are bulge - dominated . the color of these bulge - dominated hosts are indeed consistent with evolving elliptical galaxies , while contaminations from star formation / agn seems to be present in their color . assuming that the known local relations among the bulge luminosity , central velocity dispersion , and the mass of the central blackhole still hold at , we compare the agn luminosity with the eddington luminosity of the central blackhole mass . the agn bolometric luminosity to eddington luminosity ratio ranges from 0.3 to 10% .

astro-ph0402617

deep x - ray images of a patch of the sky show numerous x - ray sources , which mainly consist of a mixture of absorbed and unabsorbed active galactic nuclei ( agns ) . it is now recognized that these agns make up the bulk of what has been called the `` x - ray background '' . while normal galaxies , whose x - ray emission is probably dominated by the integration of x - ray binaries , start to emerge @xcite at the faintest fluxes , the dominant x - ray source population in the deep chandra and xmm - newton surveys comes from agn activities . thus multiwavelength studies of these x - ray sources are key to understanding the detailed history and physical conditions of the formation and the growth of the supermassive blackholes ( smbhs ) , which are now known to reside in the centers of almost all galaxies with a bulge @xcite . the region known as the `` groth - westphal strip '' ( gws ) , consists of hubble space telescope ( hst ) wide - field planetary camera 2 ( wfpc2 ) medium - deep images of a strip of 28 contiguous fields@xcite . it is a particularly useful field for extensive multiwavelength studies . numerous on - going and future followup projects have been/ are being conducted on and around this field . existing morphological information from the original wfpc2 observations , combined with the deep extragalactic evolutionary probe ( deep ) , canada - france redshift survey ( cfrs)lilly / cfrs/ ] and the ongoing deep 2 redshift surveys , is providing us with the first clues to the nature of the x - ray source counterparts . because of the contamination from starlight in the host galaxy , optical searches for faint agns in deep survey fields such as gws need elaborate efforts . such attempts have been made by searching for an unresolved component at the centers of galaxies , searching for variable nuclei , and by selecting those with ultraviolet - excess cores @xcite . these surveys reveal up to 10% of galaxies as agn candidates with nuclei extending as faint as m@xmath7@xmath8 15 . optical searches , however , are less sensitive to agns obscured by dust around the active nucleus . on the other hand , x - ray surveys for agns are not hindered by the luminosity of the underlying host galaxy . in particular , x - ray surveys with hard band ( @xmath9 kev ) sensitivity are also sensitive to the obscured agns . in view of this , we have obtained an 80 ks exposure the northeast part of the gws with xmm - newton . this field also has the advantage of a low column density of neutral gas in our galaxy , corresponding to @xmath10 @xmath11 @xcite . a quick look view and the preliminary log n - log s relations from these data have been presented in @xcite and @xcite . in this paper , we present the nature of the 23 x - ray sources in the gws , where morphological properties from the wfpc2 images and some redshifts are available from the deep / cfrs redshift surveys . we further make supplemental near infrared spectroscopic observation for some of these x - ray sources for which we did not observe agn signatures in the optical spectra . @xmath2 band photometry of the x - ray sources are also presented . the scope of this paper is as follows . in section [ sec : xobs ] , we describe the x - ray data and analysis , including source detection and spectral analysis . in section [ sec : opt_ir ] , we explain the source of the optical and near infrared ( nir ) data . the optical and nir nature of the xmm - newton sources and related statistics are presented . in [ sec : disc ] , we discuss the overall results on these x - ray sources and black hole mass and the relationship with the bulge luminosity . we also comment on selected individual sources . a summary is given in [ sec : sum ] . throughout this paper , we use @xmath12 , @xmath13 , and @xmath14 . unless otherwise noted , @xmath15 is the 2 - 10 kev rest - frame luminosity in units of @xmath16 calculated using the cosmological parameters shown above . the field has been observed with xmm - newton as part of a guaranteed time program ( pi : griffiths ) . after cleaning for background flares , we have obtained a total of 69 ks and 81 ks of exposure for epic pn and mos respectively . the log of observation is shown in table [ tab : log ] . cll rev / obsid & cleaned exp . & date + 0113/0127921001 & 51.(pn)/56.(mos ) & 21-jul.-2000 + + 0114/0127921101 & 18.(pn)/25.(mos ) & 23-jul.-2000 + /0127921201 & & + firstly we have generated an x - ray image for each of the pn , mos1 , and mos2 data from xmm - newton revolutions 0113 and 0114 . we have determined the positions of the counterpart x - ray sources by gaussian fits for two broad line agns detected in the canada - france redshift survey @xcite and a few objects in the medium deep survey ( mds ) database @xcite with trivial x - ray counterparts . based on the positions of the optical counterparts , we have determined the linear transformation coefficients of the x - ray source positions . during this process , we did not use x - ray sources which fell in the gaps of the ccds . using the coefficients , we then aligned each cleaned epic event file . using the aligned events , we made summed images in three energy bands , corresponding to photon energies of 0.5 - 2 , 2 - 4.5 and 4.5 - 10 kev . we also made a 2 - 10 kev image . in making the images , we have used pattern@xmath174 ( 12 ) for the pn ( mos ) data and applied the standard screening criteria . in order to avoid the instrumental line emission feature which dominates the background in 7.33 - 9.05 kev , the channel energy range corresponding to this energy range has been excluded from the pn data . also exposure maps have been created for each energy band using the task `` eexpmap '' . the mos exposure maps in each band have been multiplied by the ratio of the mos and pn efficiencies in the given band , so that the source counts from the summed ( pn+2moss ) image divided by the summed exposure map gives count rates which can be converted to fluxes using the pn response . the slight shift due to the alignment process ( at most 2@xmath18 ) is negligible in the usage of the exposure map . the final source detection has been made to the sum of all pn and mos data for the three bands . these three images were simultaneously fed into the sas software eboxdetect in the `` local '' mode . for each band , we have created the `` background map '' by cubic spline fitting the source excluded image ( divided by the exposure map ) using the sources detected in the first step . the next step is to use these `` background '' maps to re - run the eboxdetect in the `` map '' mode , i.e. , search for the excess over the background map . the final step is to make a multi - source maximum - likelihood fit over the background map using the positions of the sources in the previous step as a starting model . this has been made using the sas task `` emldetect '' . the images in the soft ( 0.5 - 2 kev ) , medium ( 2 - 4.5 kev ) , and ultra - hard ( 4.5 - 10 kev ) bands have been simultaneously fit . the procedure also calculates the source detection likelihood for each band as well as for the sum of these . we accepted sources with a likelihood parameter of @xmath19 where @xmath20 is the probability that the source exists . we also made a separate source detection to the 2 - 10 kev image to obtain fluxes of the sources in a more traditional energy band . one of the caveats is that cubic spline fits generated using the sas procedure `` esplinemap '' sometimes generate spurious wavy structures depending on the choice of the number of spline nodes ( the esplinemap parameter _ nsplinenodes _ ) . this is probably because some structures in the input image are not well represented by the spline function for the given number of nodes . in order to obtain the best background map , we have manually varied the number of spline nodes until a satisfactory background map is obtained . while this strategy involves a certain subjectiveness , we have checked the goodness of this strategy with extensive simulations . the basic results of this analysis are presented in @xcite . while the source detection should be improved for further quantitative analysis @xcite , our current source detection scheme provides a sufficiently complete list of x - ray sources for the purpose of this paper which is focused on the multiwavelength properties of the x - ray sources within a limited area . these have been verified by visual comparisons of the images and the detected source lists . the detected source counts have been converted into `` effective '' count rates by dividing by the value of the exposure map smoothed with a @xmath21 gaussian . in this paper , we concentrate on the sources which fall within the hst wfpc2 fields of view . sixteen of the original 28 gws fields overlap with the xmm field of view . additionally , there is another wfpc2 field in this general field ( pi lilly ) . we excluded from this paper a few x - ray sources located close to the edge of the wfpc2 field if their optical counterparts were not included in the mds database . table [ tab : srcs ] lists the x - ray sources within the wfpc2 fovs detected using the above procedure . in this table , the fluxes have been obtained from the count rate assuming a power - law or an absorbed power - law spectrum , depending on the hardness ratios from detected count rates in the 0.5 - 2 and 2 - 10 kev bands . if the hardness ratio corresponds to an effective photon index of @xmath22 , a power - law spectrum is used . for a harder spectrum , an absorbed spectrum with fixed @xmath23 and varying intrinsic absorption column density ( @xmath24 ) placed at @xmath25 ( a typical redshift of the sources ) as long as the value of @xmath24 is matched to the observed hardness ratio , the count rate to flux conversion factor is insensitive to the assumed redshift . ccrrrrrrr x8 & xmmgws j141734.8 + 522809 & 2.3 & 1000 & 2.32@xmath260.10 & 3.03@xmath260.24 & 2581 & 2397 & 340 + x9 & xmmgws j141700.6 + 521918 & 2.3 & 1083 & 1.63@xmath260.06 & 1.16@xmath260.12 & 3005 & 3025 & 224 + x10 & xmmgws j141651.3 + 522046 & 2.3 & 1082 & 1.46@xmath260.06 & 2.17@xmath260.17 & 2844 & 2500 & 463 + x11 & xmmgws j141642.4 + 521813 & 2.4 & 910 & 1.21@xmath260.06 & 4.01@xmath260.23 & 2335 & 1490 & 835 + x20 & xmmgws j141741.7 + 522822 & 2.4 & 635 & 0.93@xmath260.05 & 2.35@xmath260.19 & 1463 & 1087 & 371 + x22 & xmmgws j141653.2 + 522104 & 2.4 & 681 & 0.80@xmath260.04 & 1.82@xmath260.16 & 1286 & 962 & 326 + x26 & xmmgws j141715.0 + 522311 & 2.4 & 443 & 0.47@xmath260.03 & 0.82@xmath260.10 & 810 & 681 & 132 + x28 & xmmgws j141704.1 + 522139 & 2.4 & 568 & 0.29@xmath260.03 & 3.45@xmath260.20 & 1049 & 202 & 896 + x33 & xmmgws j141816.3 + 522940 & 2.6 & 409 & 1.04@xmath260.07 & 1.06@xmath260.22 & 631 & 614 & 30 + x40 & xmmgws j141653.8 + 522042 & 2.7 & 187 & 0.27@xmath260.03 & @xmath270.4 & 134 & 151 & 6 + x44 & xmmgws j141729.9 + 522748 & 2.6 & 241 & 0.31@xmath260.03 & 0.72@xmath260.12 & 307 & 224 & 65 + x46 & xmmgws j141745.6 + 522801 & 2.7 & 204 & 0.35@xmath260.04 & 0.58@xmath260.14 & 245 & 233 & 28 + x52 & xmmgws j141746.0 + 523031 & 2.8 & 179 & 0.16@xmath260.03 & 2.01@xmath260.23 & 189 & 44 & 177 + x55 & xmmgws j141633.5 + 521502 & 3.1 & 152 & 0.32@xmath260.04 & 1.21@xmath260.23 & 160 & 93 & 44 + x61 & xmmgws j141758.7 + 523138 & 3.1 & 135 & 0.32@xmath260.04 & 0.94@xmath260.19 & 134 & 108 & 39 + x66 & xmmgws j141649.8 + 521810 & 3.0 & 112 & 0.06@xmath260.02 & 1.15@xmath260.16 & 107 & 10 & 102 + x69 & xmmgws j141749.1 + 522811 & 3.1 & 129 & 0.23@xmath260.03 & 0.41@xmath260.13 & 121 & 111 & 13 + x83 & xmmgws j141639.7 + 521502 & 4.3 & 73 & 0.16@xmath260.03 & @xmath270.7 & 33 & 31 & 2 + x118 & xmmgws j141753.1 + 522840 & 3.8 & 73 & 0.13@xmath260.03 & @xmath270.7 & 47 & 41 & 6 + x125 & xmmgws j141756.9 + 523125 & 4.2 & 68 & 0.16@xmath260.03 & 0.54@xmath260.17 & 44 & 35 & 13 + x130 & xmmgws j141754.3 + 523125 & 4.8 & 52 & 0.10@xmath260.03 & 0.51@xmath260.17 & 22 & 14 & 12 + x133 & xmmgws j141720.0 + 522459 & 4.1 & 53 & 0.03@xmath260.02 & 0.33@xmath260.08 & 23 & 3 & 23 + x146 & xmmgws j141745.8 + 522950 & 3.9 & 61 & @xmath270.02 & 1.13@xmath260.21 & 31 & 0 & 53 + the columns of table [ tab : srcs ] are explained below : x - no . : : : the internal xmm source i d ( hereafter referred to as the x - number ) , which is used to identify the xmm - newton sources throughout this paper ( with or without the preceding `` x '' ) . name : : : the source name according to the iau registry in the format xmmgws jhhmmss.s+ddmmss . this also gives the detected xmm source position . note that the last digits are truncated .. pos . err : : : positional error ( 90% encircled ) in arcseconds . see below . xmm counts : : : the detected counts in the total ( 0.5 - 10 kev ) band . @xmath28 : : : the soft ( 0.5 - 2 kev ) band flux , in units of @xmath29 . @xmath30 : : : the hard ( 2 - 10 kev ) band flux , in the same units . @xmath31 : : : the source existence likelihood ( see above ) in the total band . @xmath32 : : : the source existence likelihood in the soft band . @xmath33 : : : the source existence likelihood in the hard band . while we treated sources with @xmath34 for the total band as detected and included them in table [ tab : srcs ] , in each of the soft and hard bands , the @xmath35 values are smaller . in the soft and hard bands , we list a nominal flux and 1@xmath36 error if @xmath37 and a 3@xmath36 upper limit for a lower significance . fig . [ fig : xmmfield ] shows the summed full - band ( 0.5 - 10 kev ) xmm - newton image smoothed with a @xmath38 gaussian for display . the hst wfpc2 fovs are overlaid and the positions of the sources in table [ tab : srcs ] are marked with crosses with the x - number designations . there is a 200 ks chandra acis - i observation ( pi nandra ) covering the northeastern part of the field and the data came to public archive immediately after the observation . we have also generated a source list from the chandra data . we have aligned the chandra events in the same way as the xmm - newton data . within 7.4 arcminutes of the chandra pointing center , the chandra data provide much more accurate x - ray source positions and we used them for unambiguous identification of the x - ray source counterparts as shown below . in this paper , we choose not to discuss x - ray sources detected in the chandra data but not detected in the xmm - newton data , because they are fainter than those in this paper and the current optical data and redshift information are very limited . the uncertainties of the source positions have been determined by the combination of the empirically determined systematic errors and the statistical errors . the former has been determined by the rms residuals of the alignment procedure described above . these are @xmath39 for the xmm data and @xmath40 for the chandra data along each of the x and y axes . for fainter sources and/or sources with large off - axis angles , the statistical error of the source position ( @xmath41 ) becomes comparable to @xmath42 . since the @xmath41 is much smaller than @xmath42 for brighter sources , including those used for the alignment process , @xmath41 and @xmath42 are independent of each other . thus we can estimate the 90% encircle radius ( @xmath43 ) by @xmath44,\ ] ] where 2.15 is the conversion factor between 1@xmath36 and 90% encircle radius assuming a two - dimensional gaussian probability distribution . the xmm x - ray sources with improved positions from the chandra data are summarized in table [ tab : chanpos ] . clll xmm & + x - no . & ra(@xmath45 ) & dec(@xmath45 ) & @xmath46 + x8 & 214.39514 & 52.46958 & 0.7 + x20 & 214.42441 & 52.47316 & 0.7 + x26 & 214.31275 & 52.38685 & 0.8 + x44 & 214.37476 & 52.46324 & 0.7 + x46 & 214.44027 & 52.46722 & 0.7 + x52 & 214.44140 & 52.50901 & 0.7 + x61 & 214.49547 & 52.52754 & 0.8 + x69 & 214.45498 & 52.46987 & 0.8 + x125 & 214.48674 & 52.52354 & 0.8 + x130 & 214.47583 & 52.52329 & 0.8 + x133 & 214.33364 & 52.41676 & 0.8 + x146 & 214.43923 & 52.49754 & 0.8 + as a preview of the x - ray spectral properties of the detected sources , we plot a hardness ratio diagram in fig . [ fig : hr ] . the hardness ratio between the 2 - 4.5 kev and 4.5 - 10 kev is plotted against that between 0.5 - 2 kev and 2 - 4.5 kev . the typical 1@xmath36 errors of the hardness ratios are 0.050.2 ; errors exceeding 0.4 are shown . we overplot grids showing the loci of absorbed power - law spectra for varying values of the photon index @xmath47 and intrinsic absorption @xmath48 for z=0.7 and z=2 respectively as described in the figure caption ( galactic @xmath49 @xmath11 has been included ) . key optical / ir properties ( sect . [ sec : opt_ir ] ) of the x - ray sources are indicated . for bright sources , ( xmm counts @xmath50 300 ) , x - ray pulse - height spectral analysis gives further constraints on the spectral parameters . for the spectral analysis , we have used the data from the first observation in table [ tab : log ] ( rev / obsid = 0113/0127921001 ) , which contain mostly good intervals free from background flares . the extraction of the spectra has been made using the sas analysis package v5.4 and we have used the standard screening criteria with pattern@xmath51 and pattern@xmath52 for pn and mos respectively as before . the calibration files used to generate the calibrated data products and response files for the spectral analysis ( using the sas task `` epproc '' and `` emproc '' ) were up to date as of aug . we extracted the source spectra using a circular region with a typical radius of 25@xmath18 , depending on the source count . the background spectra have been extracted using an annular region surrounding the source with inner and outer radii of 30@xmath18 and 60@xmath18 respectively . we have further jointly analyzed two co - added mos spectra and the pn data . for sources with @xmath53 counts , we could reasonably constrain the parameters with a @xmath54 fit to the background subtracted spectra rebinned to have at least 20 counts per bin . for fainter sources , we have utilized the xspec implementation of the c - statistics @xcite to the spectra with much finer binnings . because the c - statistics can not be applied to background - subracted data , the background and ( source+background ) spectra have been jointly fit . the background has been approximated by one or two power - laws and a broad gaussian . this background model was then scaled by the source and background extraction areas and added to the source model . the result was then fitted to the source spectrum without background subtraction . ccccccc x8 & 1.223 & 2.2(2.0;2.5 ) & @xmath271.3 & 1.6 & 44.2 & 22./38 + x9 & 0.5 * & 2.5(2.3;2.9 ) & @xmath270.07 & 1.0 & ... & 20./47 + x10 & 0.801 & 2.2(2.0;2.8 ) & @xmath270.3 & 1.4 & 43.7 & 7./29 + x11 & 1 * & 1.7(1.4;2.2 ) & 1.6(0.7;2.5 ) & 3.5 & ... & 16./31 + x20 & 1.148 & 2.0(1.6;2.7 ) & 1.8(0.7;3.3 ) & 1.9 & 44.2 & 9./19 + x22 & 0.983 & 1.4(1.2;1.7 ) & @xmath270.5 & 2.7 & 44.2 & 11./20 + x28 & 0.76p & 1.2(0.8;1.7 ) & 3.0(1.7;5.3 ) & 4.1 & 43.8 & 62 . + the source spectrum is fitted with a power - law , galactic absorption ( fixed at @xmath10 @xmath11 ) , and an intrinsic absorption at the source redshift . for the sources without known redshifts , we have assumed a redshift for determining the value of @xmath48 . the results have been summarized in table [ tab : spec ] . see comments to this table for details . this field has been targeted by a number of past and future survey projects in many wavelength , starting with the 28 contiguous hst wfpc observations @xcite . the main source of the optical data summarized in this paper comes from the hst mds @xcite database and the deep survey . we have also made supplemental nir spectroscopic observations for several of the x - ray sources . infrared ( @xmath2 ) photometry of these sources from the emir - cosmos project@xcite are also included . the basic results are summarized in table [ tab : opt - ir ] . cccccccccccccc x8 & u26x7:0004 & 20.6 & 20.2 & s , s & 083_5407 & 17.4 & 1.22 & 3 & d & agn1 & * mgii * & ... & 44.1 + x9 & u26xd:0007 & 20.4 & 19.4 & 0.28,0.24 & 144_2774 & 16.7 & ... & 0 & ... & ... & ... & ... & ... + x10 & u26xd:0011 & 21.4 & 19.8 & s,0.25 & 142_4838 & 16.9 & 0.808 & 5 & d & nls1 & [ nev ] & ... & 43.8 + x11 & u26xf:0001 & 18.1 & 17.7 & s , s & 173_7369 & 15.9 & ... & 0 & ... & ... & ... & ... + x20 & u2ay1:0025 & 23.4 & 21.5 & b , b & 083_5273 & 16.5 & 1.148 & 3 & d & agn1 & * mgii * & ... & 44.2 + x22 & u26xd:0034 & 22.7 & 21.5 & b , b & 142_2752 & 18.3 & 0.983 & 3 & o & agn - ir & ... & * h@xmath1*,[nii ] & 43.9 + x26 & u26xb:0010 & 21.2 & 20.9 & s , s & 123_2458 & 19.4 & 1.263 & 3 & d & agn1 & * mgii * & ... & 43.8 + x28 & u26xc:0019 & 22.8 & 20.9 & b , b & 133_1016 & 17.6 & 0.757 & -1 & i & ... & ... & ... & 44.0 + x33 & u2iy4:0004 & ... & 19.5 & ... , s & none & ... & 1.603 & 4 & c & agn1 & * mgii * & ... & 44.3 + x40 & u26xd:0062 ? & 23.1 & 23.3 & s , s & 142_2530 & @xmath5521.5 & ... & 0 & ... & ... & ... & ... & ... + x44 & u26x7:0120 & 25.3 & 23.7 & g , b & 082_5240 & 19.6 & ... & 0 & ... & ... & ... & ... & ... + x46 & u2ay1:0019 & 21.4 & 20.7 & ( d , d ) & 074_2638 & 18.9 & 0.432 & 4 & d & ... & lex & ... & 42.6 + x52 & u26x6:0011 & 22.9 & 21.9 & b,0.41 & 062_2060 & 17.5 & 0.985 & 4 & d & agn - ir & lex & * h@xmath1*,[nii ] & 44.0 + x55 & u26xh:0177 & 24.7 & 23.7 & d , b & 184_2148 & @xmath5520.9 & 0.0000 & 0 & ... & ... & ... & ... & ... + x61 & u26x4:0021 & 23.0 & 21.6 & 0.51,b & 053_4446 & 18.5 & 0.637 & 1 & c & ... & lex & ... & 43.2 + x66 & u26xf:0024 & 22.4 & 20.1 & 0.22,b & 164_6109 & 17.6 & 0.808 & 3 & d & agn - ir & lex & * h@xmath1*,[nii ] & 43.6 + x69 & u2ay1:0092 & 24.6 & 22.8 & b , b & 074_6236 & 18.7 & 0.995 & 3 & o & agn - ir & ... & h@xmath1,[nii ] & 43.3 + x83 & u26xh:0001 & 14.8 & 15.6 & s , b & 184_7960 & 13.4 & 0.0000 & 0 & ... & star & ... & ... & ... + x118 & u26x6:0004 & 20.6 & 20.4 & 0.26,0.30 & 064_2873 & 17.6 & 0.719 & -1 & b & ... & ... & ... & 42.8 + x125 & u26x5:0056 & 24.0 & 22.7 & s , s & 053_3424 & 19.3 & 1.554 & -1 & b & ... & ... & ... & 43.9 + x130 & u26x5:0160 & 25.2 & 24.8 & g , d & 052_0940 & @xmath5521.0 & 0.0000 & 0 & ... & ... & ... & ... & ... + x133 & u26x9:0067 & 24.1 & 22.8 & s , s & 113_6354 & 19.9 & 0.704 & -1 & b & ... & ... & ... & 42.9 + x146 & u26x6:0010 & 22.8 & 21.0 & 0.12,0.21 & 073_7749 & 17.9 & 0.873 & 3 & d & agn2-opt & [ nev ] & ... & 43.8 + fig . [ fig : poststamp ] shows the hst wfpc2 f814w images ( with the exception of x8 , x20 , and x44 , where the f606w images are shown ) with @xmath43 error circles for xmm and chandra ( where applicable ) source positions . because each wfpc2 image may have absolute astrometric error of @xmath56 , in a few cases , the x - ray sources in one wfpc2 fov show systematic offsets from their counterparts . in those cases , we have shifted the wfpc2 image by matching the positions of one of the x - ray sources with its unambiguous counterpart in making overlays . in fig . [ fig : poststamp ] , the overlays for x8 , x20 , and x44 have been tuned using x8 , those for x61 , x125 , and x130 using x61 , those for x52 , x146 , and x118 using x146 , and those for x69 , x46 and x20 using x69 . in any case , a translational shift without a rotation gives a satisfactory correction , where all the unambiguous counterparts ( including chandra sources without xmm detections ) are within the chandra error circles . note that the drawn error circles ( @xmath43 ) may be an overestimate in these cases . all of the x - ray source counterparts have an entry in the mds database , where the results of the disk - bulge decomposition are provided . the entries from the mds database , including the mds - id , magnitudes in f814w ( or @xmath57 ) and f606w ( or @xmath58 ) filters , and the bulge - to - total brightness ratios ( b / t ) ( or other morphological classifications ) have been shown in table [ tab : opt - ir ] in columns ( 2),(3),(4 ) , and ( 5 ) respectively . the first and second entries in column ( 5 ) ( separated by a comma ) are for the f606w and f814w bands respectively . if an entry in column ( 5 ) shows a number between 0 and 1 , it means that the host galaxy has been deconvolved with an exponential disk+bulge model and the bulge - to - total light ratio is shown . other possible entries in this column are `` b '' ( pure bulge ) , `` d '' ( pure disk ) , `` s '' ( unresolved point source ) , and `` g '' ( galaxy ; emission is extended , but could not discriminate between the exponential disk and bulge models ) . optical spectra are available for 8 of our x - ray sources through the deep survey and two from the cfrs . key agn classification emission lines are shown in column ( 12 ) of table [ tab : opt - ir ] . broad permitted lines are indicated in boldface . in summary , four x - ray sources show broad mg ii emission lines and we classify these as type 1 seyfert / qsos . they are identified with `` agn1 '' in column ( 12 ) of table [ tab : opt - ir ] . other objects with optical spectra show only narrow permitted and/or forbidden lines . two of these show [ nev ] emission , which is an unambiguous indicator of an agn activity . one of these ( x10 ) is a narrow - line seyfert 1 galaxy ( `` nls1 '' in column [ 11 ] ) as discussed in sect . [ sec : indi ] . we tentatively classify the other object ( x146 ) as a type 2 agn and classify it as `` agn2-opt '' . the other four objects do not show signs of agn activity in their optical spectra in that they show neither broad permitted lines nor high excitation lines . these are classified as `` lex '' ( low - excitation emission line objects ) in column ( 12 ) . we have made nir spectroscopic observations for a subset of the x - ray sources with no optical spectroscopic indication of an agn activity . these include objects with optical spectra which show only low excitation lines as well as those without optical spectra . observations were made using the cisco spectrometer with the oh suppressor ( cisco / ohs ) @xcite attached to the nasmyth focus of the 8.2 m subaru telescope on march 22 - 24 , 2003 . the jh - grism with a @xmath59 slit width was used . the spectral resolution of this configuration is @xmath60 . the primary purpose of the nir spectroscopy was to detect the redshifted h@xmath61+[nii ] @xmath62 complex and investigate the agn nature of these x - ray sources . the ratio of the strengths of the [ nii ] doublet and h@xmath1 is a good discriminator between the agn and starburst origins of the ionizing radiation . typical seyfert 2 galaxies have @xmath63\lambda 6583/h\alpha ) > } $ ] -0.25 @xcite . we also intended to search for broad components in h@xmath1 . since a dust cloud is more transparent to redder light , moderately obscured agn activities are much more visible in the longer wavelength lines such as h@xmath1 rather than shorter wavelength lines such as h@xmath64 or mg ii . in fact , there have been a number of reported cases where x - ray source counterparts show broad h@xmath1 lines , even though they only show narrow lines in the shorter wavelength optical spectra @xcite . we did not include those with z@xmath270.7 ( x46 , x61 ) in our nir spectroscopy targets , because the h@xmath1+[nii ] line complex does not fall within the wavelength range observable by cisco / ohs jh - grism . also in order to make spectroscopic observation feasible within a reasonable exposure time , we limited our spectroscopic observation to those with an estimated h - band magitude of @xmath65 @xcite , where h band magnitudes were estimated by @xmath66 @xcite . five from the 23 xmm - newton sources were observed during our observing run . the resulting spectra , reduced following the procedure by @xcite , are shown in fig [ fig : ohs ] . preliminary flux calibrations have been made using a nearby standard star observed immediately after the target . we have detected the h@xmath1+[nii ] complex in 4 of the 5 observed x - ray sources . in all four cases , the spectra suggest a possible broad h@xmath1 component ( fwhm @xmath67 ) and/or strong [ nii ] lines . this is indicated in column ( 13 ) of table [ tab : opt - ir ] and the sources are classified as `` agn - ir '' in column ( 11 ) . because of the weak signal - to - noise ratio and difficulty in determining the background , careful analysis is required to make any quantitative conclusions . this will be the topic of a separate paper that will describe the details of the observations , data reduction , and quantitative analyses for the subaru cisco - ohs spectra of the sources shown in this paper and a few additional sources observed during the observation run . the @xmath2 photometry was provided by the emir - cosmos project @xcite . the gws was imaged with the near - infrared camera ingrid on the william herschel telescope in the spanish observatorio del roque de los muchachos during two runs in 2000 april and 2001 june . these data are fully described in @xcite . atmospheric conditions were photometric , with seeing ranging from @xmath68 to @xmath69 fwhm in 2000 april , and @xmath70-@xmath71 fwhm in 2001 june . source extraction and photometry of @xmath2-band data was done using sextractor @xcite . zeropoint uncertainty is 0.03mag . sextractor best magnitudes are provided with errors including photon noise . the resulting @xmath2 magnitudes are listed under column ( 7 ) of table [ tab : opt - ir ] . redshifts , redshift quality flag and the reference source of the redshift are shown in columns ( 8) , ( 9 ) , and ( 10 ) of table 5 . the meanings of the redshift quality flags are : -1 : photometric redshift only , 0 : no redshift information , 1 - 5 : as defined by the deep project @xcite or the cfrs @xcite confidence class if the redshift source is cfrs . the redshift of x22 has been solely determined by the h@xmath1+[nii ] complex from the subaru ohs data . since this feature is unambiguous , it has been assigned the quality flag of 3 . for those with redshift information , we have calculated the x - ray luminosity of the source and listed the results in column ( 13 ) . the luminosities have been calculated from the 2 - 10 kev rest - frame fluxes given in table [ tab : srcs ] and the k - corrections have been made using the spectral assumptions described in sect . [ sec : xobs ] . the luminosities have been corrected for possible intrinsic absorption using the same spectral assumption . because of the difference in methods and exact time selections , the luminosities listed here can be slightly different from those in the full spectral analysis listed in table [ tab : spec ] . the 23 sources detected in our 80 ks xmm - newton observation of the gws are representative of the x - ray sources that contribute most to the `` cosmic x - ray background '' and many of them represent the regime which marks the peak of accretion onto smbhs in centers of galaxies . a dominant population in this field consists of agns with @xmath72 44 at @xmath6 . only a few of them show signs of agn activity in their optical spectra . subaru ohs nir spectroscopy of four of the x - ray sources with no previous optical signature of agns revealed h@xmath1+[nii ] emission lines showing hints of broad h@xmath1 and/or stronger narrow [ nii ] lines indicative of agn activity . the host galaxies of the x - ray sources tend to be bulge - dominated and four are extremely - red objects ( eros ) ( or very red objects ; vros ) ( @xmath73 ) . also one object ( x130 ) has an upper limit @xmath74 , which is consistent with being an ero . [ fig : col ] shows @xmath5 and @xmath75 colors of the x - ray sources in our sample as a function of redshift . we have excluded x33 ( no @xmath2 and @xmath58 photometry available),x46 ( x - ray source is off - nucleus , see below ) , and x83 ( galactic star ) from the plot . those without redshift information are plotted left of z=0 . for reference , we plot k- and evolution - corrected galaxy colors for elliptical ( labeled as e1 in fig . [ fig : col ] ) , sa and sc galaxies from @xcite . model e1 corresponds to poggianti s model `` e '' , which has a star - formation rate with e - folding time of 1 gyr . we have neglected the difference of cosmological parameters used by us and@xcite , which corresponds to a @xmath76 difference in the age of the universe . in plotting these color tracks , we have converted the @xmath77 and @xmath78 magnitudes to @xmath57 and @xmath58 using the formulae given by @xcite . we also plot the colors of an elliptical galaxy model , which we have calculated for our filters and cosmology from the evolving seds by @xcite ( labeled as e2 ) . the plotted model is for a passively evolving galaxy after a short burst at the formation epoch of @xmath79 and @xmath80 at @xmath81 . the difference between @xmath82 and @xmath2 magnitudes have been neglected , following @xcite . [ fig : col ] shows that the @xmath75 color of the x - ray sources with bulge - dominated hosts indeed traces that of elliptical galaxies and , in particular , those of eros with redshifts are roughly consistent with the passively evolving elliptical model ( e2 ) . the @xmath83 color , which is more sensitive to the contaminations from star formation and agn activities , shows a scatter towards bluer colors from the elliptical regime . the reddest one ( x20 ) , with @xmath84 may be contributed by a dust enshrouded agn or a starburst . point - like sources tend to be distributed towards bluer ( qso ) colors , although the scatter is large . the scatter may be contributed to by unresolved host galaxies , intrinsic scatter in qso colors and/or reddening by intrinsic dust absorption . the hst imaging of this field has allowed us to find a significant population of x - ray sources at @xmath6 whose counterparts have a resolved bulge component , either as part of a disk+bulge structure or a pure bulge . in view of the relationship found between bulge mass and the central blackhole mass in nearby galaxies @xcite , it is interesting to make a first - order estimation of blackhole mass ( @xmath85 ) from the bulge - component of the host galaxy and compare it with the x - ray luminosity . in the rough estimation below , we take the approach of @xcite by first converting the bulge luminosity to the central velocity dispersion ( @xmath36 ) of the bulge stellar component using an empirical relation . then we use the tight @xmath86 @xcite correlation to obtain the estimated blackhole mass . we use the f814w k - correction for the e galaxy in fig . 18(d ) of @xcite to calculated the f814w absolute magnitude ( @xmath87 ) . we also assume an early - type galaxy color of @xmath88 @xcite . we then use the relations , @xmath89 @xcite to obtain the estimated blackhole mass . because the f814w band corresponds to the @xmath90 band at the source rest frame of @xmath6 , the combination of the k - correction and the magnitude conversion to @xmath91 is insensitive to the assumed galaxy spectral energy distribution . [ fig : mbhvslx ] shows the estimated @xmath92 versus @xmath15 ( absorption corrected , see sect . [ sec : z_and_l ] ) for 10 x - ray sources in the sample which have a resolved bulge component in the hst wfpc2 f814w image ( mds ) . it is interesting to estimate the eddington ratio @xmath93 for these x - ray sources , where @xmath94 is the bolometric luminosity of the agn and @xmath95 is the eddington luminosity corresponding to the blackhole mass . writing @xmath96 , where @xmath97 @xcite , we overplot three lines showing @xmath98 and @xmath99 in fig . [ fig : mbhvslx ] . fig . [ fig : mbhvslx ] shows that the estimated blackhole masses for the 10 agns range from @xmath100 @xmath101 and the eddington ratios from 0.3%-10% . these results have interesting implications on how agn evolve and the current stage of evolution for the agn represented here ( @xmath6 , @xmath102 ) . this is indeed a characteristic redshift and luminosity marking the peak of the accretion history of the universe . the result that these agn are typically radiating at a few percent may have important implications on the accretion history and formation of smbhs . if this is a typical eddington ratio throughout the agn phase of these objects , the growth of the blackhole occurs on a timescale of @xmath103 a few @xmath104 yrs , where @xmath105 yrs is the salpeter timescale , that is the timescale at which the mass of an object accreting at the eddington luminosity grows by a factor of @xmath106 ( for a radiative efficiency @xmath107 of a standard accretion disk model ) . this scenario has difficulty in that the timescale of a few @xmath108 yrs may be too long , while the number density of agn at @xmath109 decreases beyond @xmath6 @xcite . alternatively , it is possible that these agn have gone through a brief luminous phase in the past with near - eddington accretion rates . these may have been observed as more luminous qsos ( @xmath110 ) at @xmath111 . yet another more exotic possibility is that they are just accreting with a low radiative efficiency ( @xmath112 ) , allowing much less time for the smbh to grow . we note , however , that there are a number of caveats in interpreting these results and drawing conclusions relating the x - ray agn evolution and growth of the smbh . firstly , the current mds database shows the analysis for stellar ( point - like ) images or galaxies ( pure bulge , pure disk or bulge+disk decomposition ) , but only limited analysis @xcite has been completed for decomposing point - like ( stellar ) nuclei from the host galaxy ( stellar+bulge+disk , stellar+bulge or stellar+disk ) . thus we select against those with strong agn components ( or with large @xmath113 ) , which are likely to be listed as `` stellar '' in the mds database or the bulge luminosity in the database may be contaminated by the central agn component . this situation should be improved in the future , where the hst images are analyzed with point - like nucleus+host galaxy decomposition . extending this study to other deep fields with x - ray and hst ( wfpc2 as well as acs ) data , including the extended chandra deep field - south ( e - cdfs ) and the cosmos field , will be a next logical step . secondly , a much more fundamental limitation is that we have assumed the local relations in eq . [ eq : mass ] are still valid at @xmath6 . this assumption is not guaranteed to be valid . x8 : : : this is a typical type 1 qso at z=1.22 with a broad mg ii line . x10 : : : the deep optical spectrum shows many narrow emission lines including high - excitation lines like [ nev]@xmath114 , which are unambiguous indicators of agn activity . permitted lines ( mg ii , h@xmath64 ) are also narrow . since h@xmath115/[oiii]@xmath116 , these lines are not dominated by a seyfert 2 . it is either a seyfert 2 whose emission lines are heavily contaminated by starburst activity or a narrow - line seyfert 1 galaxy ( nls1 ) @xcite , where h@xmath64 is contributed to from the ( narrow end of the ) broad line region . because our x - ray spectral analysis shows no x - ray absorption ( @xmath117 ; see table [ tab : spec ] ) , the nls1 interpretation is more plausible . x11 : : : the optical counterpart has a very bright stellar ( point - like ) morphology and there is no optical spectrum available for this source to discriminate between a galactic star and a qso . however , its x - ray spectrum is inconsistent with a thermal plasma , having a significant residual in the soft part . also its x - ray to optical flux ratio @xmath118 ( @xmath119 is measured in 2 - 10 kev ) is well within the agn regime ( roughly between -1 and 1 ; see e.g. @xcite ) . thus it is most likely to be a qso . x20 : : : thanks to the chandra position , we can identify the x - ray source with the brightest bulge - dominated galaxy at @xmath120 , among four candidates apparently interacting with one another indicated by tidal bridges ( see fig . [ fig : poststamp ] ) . it is an interesting case where galaxy interactions are possibly feeding the agn activity at this early stage of the universe . it has a qso luminosity ( @xmath121 ) , but the optical image is dominated by bulge component of the host galaxy . the deep spectrum shows a broad mg ii line . the x - ray spectrum shows an absorption of @xmath122 . this is an example of optical type-1 x - ray type 2 agn . this is also an ero ( @xmath84 ) and a sub - mm source detected in a deep scuba survey @xcite . x22 : : : no previous optical / ir spectroscopic observations existed for this source . our subaru ohs / cisco observation detected h@xmath1 and @xmath123\lambda\lambda$ ] 6548,6583 emission lines , giving z=0.983 . based on the sign of broad h@xmath1 and strong @xmath124 , high luminosity @xmath125 . while we mark it as an agn - ir , it may well be a type 1 agn . x28 : : : this source is the most conspicuous hard x - ray source in the field with an intrinsic absorption of @xmath126 [ @xmath11 ] . it is a bulge - dominated galaxy with no optical / ir spectroscopy and a photometric redshift of @xmath128 @xcite based on its v i color alone . this is most likely a seyfert 2 based on the x - ray properties . x46 : : : the position of the source has been determined with chandra and the relative alignment of the wfpc2 image and x - ray sources has been achieved the three other cxo sources in the field . the x - ray source counterpart is identified with a hot spot just off the patchy irregular starforming galaxy deep gss 074_2638 ( mds u2ay1:0019 ) at z=0.432 . if we assume that the x - ray source is at the same redshift as this irregular galaxy , the luminosity would be @xmath129 . this is too luminous for an ultraluminous x - ray source ( ulx ) . we also note that there is a nearby edge - on disk galaxy ( deep gss 074_2237 , z=0.156 ) . the x - ray source is 4@xmath18 away ( projected distance of 10 kpc ) from its nucleus and located towards the direction perpendicular to the disk . therefore the x - ray source could be a ulx associated with the halo of the galaxy . however , even if the x - ray source was at the redshift of this disk galaxy , its luminosity would still be @xmath130 , again well above the ulx regime . one possibility is that this irregular galaxy is undergoing a merging process and the x - ray source is at the nucleus of one of the merging galaxies , or it may simply be a background qso . x52 : : : the deep spectrum show no broad lines and the location of mgii shows only absorption . other features include [ oii]@xmath131 and [ neiii]@xmath132 . the presence of [ nev]@xmath133 is suggested but uncertain . we observed this object with subaru ohs and found a moderately broad h@xmath1 ( fwhm @xmath67 ) and a strong [ nii ] doublet . [ fig : ohs ] ) . the fact that there is no broad mgii line but a moderately broad h@xmath1 suggests that the agn is obscured by a dust cloud . this is consistent with the hard color of this object , with the second largest hr(2 - 4.5 kev/0.5 - 2 kev ) in the sample . [ fig : hr ] . this is an ero ( @xmath134 ) . x66 : : : we find no indication of an agn in the deep spectrum . like x52 , our subaru ohs observation revealed possible agn activity through the detection of a moderately broad @xmath135 ( fwhm @xmath67 ) line and a strong [ nii ] doublet . this is an obscured agn similar to x52 and has a hardest hr(2 - 4.5 kev/0.5 - 2 kev ) . there is also a hint of [ neiii]@xmath132 , a line which is often stronger in agn than starforming galaxies . x69 : : : a spectroscopic redshift of @xmath136 has been determined by our subaru observation in close agreement with the photometric redshift determined by @xcite of @xmath137 . the subaru ohs spectrum seems to indicate either strong [ nii]@xmath138 or broad h@xmath1 suggestive of agn activity . this is an ero ( @xmath139 ) . x83 : : : because of the low x - ray to optical flux ratio ( @xmath140 ) for this bright optical object ( f606w=14.8 ) , this is certainly a galactic star . x125 : : : although the mds database shows that it is a point source , its color is consistent with an elliptical galaxy at the photometric redshift of @xmath141 @xcite . the optical counterpart is probably dominated by the host galaxy . x146 : : : this object is detected only in the hard ( 2 - 8 kev ) x - ray band . the deep optical spectrum clearly shows [ nev]@xmath133 emission and strong [ neiii]@xmath132 . this is a typical seyfert 2 galaxy with absorbed x - ray spectrum 1 . we summarize the properties of x - ray sources in the gws detected with xmm - newton and discuss their morphology , optical and nir spectra . ten out of the 23 x - ray sources have spectroscopic information available in the optical band through the deep and cfrs projects . four of them have broad emission lines and thus can be classified as type 1 agn / qsos . two others have high excitation lines indicative of narrow line region of active galactic nuclei . others do not reveal agn signatures in their optical spectra . 3 . we have obtained nir spectra for five of the sources which did not reveal agn signatures in their optical spectra using the cisco / ohs spectrograph on the subaru telescope . in four of the five , we detect broad h@xmath1 and/or strong [ nii ] lines suggestive of agn activity . the x - ray hardness ratios and spectroscopic analysis show that a major part of the observed x - ray sources consists of a mixture of absorbed and unabsorbed agns . one of these agns ( x20 ) has an absorbed x - ray spectrum with a broad mgii emission line and is another example of an optical type 1/x - ray type 2 agn . the @xmath5 and @xmath142 colors of the x - ray sources with bulge - dominated hosts well trace those of evolving elliptical galaxies . sources with stellar morphologies show bluer colors with some scatter , and consistent with agn(qso ) with a host galaxy contamination and/or dust absorption . ten x - ray sources with redshift information have a resolved bulge . based on a number of assumptions concerning the bulge luminosity - blackhole mass relation , we estimate that these agn have luminosities ranging from 0.3 - 10% of the eddington luminosity , suggesting that these massive blackholes have previously gone through massive accretion ( maybe as luminous qsos ) or that their lifetimes are relatively long ( @xmath143 yrs ) . the authors are indebted to the efforts of the deep team for their observations , analysis , and archiving . this work has been supported by the nasa grant nag5 - 10875 to tm ( long - term space astrophysics ) , nasa grant nag5 - 3651 to reg ( xmm - newton mission scientist support ) . akiyama , m. , ueda , y. , & ohta , k. 2002 , , 567 , 42 alexander , d. m. , vignali , c. , bauer , f. e. , brandt , w. n. , hornschemeier , a. e. , garmire , g. p. , & schneider , d. p. 2002 , , 123 , 1149 aller , m. c. & richstone , d. 2002 , , 124 , 3035 balcells , m. 2003 , revista mexicana de astronoma y astrofsica , 16 , 69 baldi , a. , molendi , s. , comastri , a. , fiore , f. , matt , g. , & vignali , c. 2002 , , 564 , 190 barger , a. j. , cowie , l. l. , mushotzky , r. f. , & richards , e. a. 2001 , , 121 , 662 beck - winchatz , b. & anderson , f. 1999 , aj 117 , 2582 bertin , e. & arnouts , s. 1996 , , 117 , 393 brunner , b.j . 1999 , phd thesis , johns hopkins university cash , w. 1979 , , 228 , 939 cowie , l. l. , songaila , a. , & barger , a. j. 1999 , , 118 , 603 cristbal - hornillos , d. , balcells , m. , prieto , m. , guzm ' an , r. , gallego , j. , cardiel , n. , serrano , ' a. , & pell ' o , r. 2003 , , 595 , 71 dickey , j. m. & lockman , f. j. 1990 , , 28 , 215 elvis , m. et al . 1994 , , 95 , 1 fukugita , m. , shimasaku , k. , & ichikawa , t. 1995 , , 107 , 945 gonzalez , a. h. , williams , k. a. , bullock , j. s. , kolatt , t. s. , & primack , j. r. 2000 , , 528 , 145 guzmn , r. 2003,revista mexicana de astronoma y astrofsica , 16 , 209 groth , e. et al . 1994 , baas 185 , 5309 hornschemeier , a. e. et al . 2001 , , 554 , 742 hornschemeier , a. e. et al . 2003 , , 126 , 575 i m , m. et al . 2002 , , 571 , 136 iwamuro , f. , motohara , k. , maihara , t. , hata , r. , & harashima , t. 2001 , , 53 , 355 kodama , t. & arimoto , n. 1997 , , 320 , 41 lehmann , i. et al . 2001 , , 371 , 833 lilly , s. j. , hammer , f. , le fevre , o. , & crampton , d. 1995 , , 455 , 75 lowenthal , j. d. et al . 1997 , , 481 , 673 magorrian , j. et al . 1998 , , 115 , 2285 merritt , d. & ferrarese , l. 2001 , in the central kiloparsec of starbursts and agn , ed . j. h. knapen et al . ( san francisco : asp ) , 335 miyaji , t. , griffiths , r. e. 2001 in galaxy clusters and the high redshift universe observed in x - rays : proceedings of the xxist moriond astrophysics meeting , eds . d. neumann et al . ( cd - rom ) miyaji , t. & griffiths , r. e. 2002 , , 564 , l5 miyaji , t. , griffiths , r. e. , lumb , d. , sarajedini , v. , & siddiqui , h. 2003 , astronomische nachrichten , 324 , 24 motohara , k. et al . 2002 , , 54 , 315 osterbrock , d. e. & pogge , r. w. 1985 , , 297 , 166 poggianti , b. m. 1997 , , 122 , 399 ratnatunga , k. u. , griffiths , r. e. , & ostrander , e. j. 1999 , , 118 , 86 sarajedini , v. l. , green , r. f. , griffiths , r. e. , & ratnatunga , k. 1999 , , 121 , 417 sarajedini , v. l. , gilland , r. l. , kasm , c. 2003 , , in press ( astro - ph/0308383 ) schmidt , m. et al . 1998 , , 329 , 495 tanaka , i. , yamada , t. , arag ' on - salamanca , a. , kodama , t. , miyaji , t. , ohta , k. , & arimoto , n. 2000 , , 528 , 123 tremaine , s. et al . 2002 , , 574 , 740 ueda , y. , akiyama , m. , ohta , k. , & miyaji , t. 2003 , , 598 , 886 valtchanov , i. , pierre , m. , & gastaud , r. 2001 , , 370 , 689 veilleux , s. & osterbrock , d. e. 1987 , , 63 , 295 waskett , t. j. et al . 2003 , , 341 , 1217 wilkes , b. j. 2000 in allen s astrophysical quantities , fourth edition , ed . cox ( new york : springer ) , p585

we investigate the problem of factorization of large numbers on a quantum computer which we imagine to be realized within a linear ion trap . we derive upper bounds on the size of the numbers that can be factorized on such a quantum computer . these upper bounds are independent of the power of the applied laser . we investigate two possible ways to implement qubits , in metastable optical transitions and in zeeman sublevels of a stable ground state , and show that in both cases the numbers that can be factorized are not large enough to be of practical interest . we also investigate the effect of quantum error correction on our estimates and show that in realistic systems the impact of quantum error correction is much smaller than expected . again no number of practical interest can be factorized . 23.9 cm 16.5 cm 0.cm -1.5 cm

quant-ph9610015

since shor s discovery @xcite of an algorithm that allows the factorization of a large number by a quantum computer in polynomial time instead of an exponential time as in classical computing , interest in the practical realization of a quantum computer has been much enhanced @xcite . recent advances in the preparation and manipulation of single ions as well as the engineering of pre - selected cavity light fields have made quantum optics that field of physics which promises the first experimental realization of a quantum computer . several proposals ( reviewed in @xcite ) for possible experimental implementations have been made relying on nuclear spins , quantum dots @xcite , cavity qed @xcite and on ions in linear traps @xcite . the realization of a quantum computer in a linear trap in particular was regarded as very promising as it was thought that decoherence could be suppressed sufficiently to preserve the superpositions necessary for quantum computation . indeed , a single quantum gate in such an ion trap was recently realized by monroe et al @xcite . nevertheless , the error rate in this experiment was too high to allow the realization of extended quantum networks . this experiment was not solely limited by fundamental processes but rather by technical difficulties and one aim of future experiments is to reduce these technical problems to come closer to the fundamental limits , such that at least small networks could be realized . however , there remains the question whether overcoming technical problems will be sufficient to realize practically useful computations such as factorization of big numbers on a quantum computer in a linear ion trap . in fact one has to investigate more thoroughly those limitations that are due to processes of more fundamental nature than for example laser linewidths , pulse lengths etc . general considerations revealed that decoherence will lead to an exponential decrease in the probability to obtain the correct result of a calculation @xcite . however , these considerations did not state how fast this exponential decay would be and therefore left open the question to whether factorization will be possible or not . an analysis of this problem was first carried out in refs . @xcite where the impact of spontaneous emission on the problem of factorization of a large number by a quantum computer was considered . it was shown that for realistic laser powers , factorization is limited to small numbers ( at least if no sophisticated error correction methods are implemented ) . subsequent investigations have appeared to be less pessimistic , but we note that these did not consider spontaneous emission but instead concentrated on other effects such as phonon decoherence @xcite or the influence of the ion separation @xcite . in refs . @xcite limits were obtained that still depended on the power of the applied lasers and which therefore leave open the possibility of improving the maximal bitsize @xmath0 of the largest factorizable number by using high power lasers . extremely high intensities , however , will lead to ionization or , more importantly , to a breakdown of the two - level - approximation as was mentioned briefly in @xcite . in section ii we will use the breakdown of the two - level - approximation to derive new estimates for the bitsize @xmath0 of the factorizable number . the new feature in these estimates is their _ independence _ on the power of the applied laser . we derive intensity independent limits of @xmath0 not only for qubits stored in a metastable optical transition ( see fig . [ fig1 ] ) but also for qubits stored in the zeeman sublevels of an ion which are then manipulated using detuned raman transitions ( see fig . [ fig2 ] ) . on the basis of experimental parameters for several ions we give estimates for @xmath1 and we show that spontaneous emission already imposes strong limitations to @xmath0 . we then conclude that factorization without the use of efficient error correction methods is limited to almost trivial numbers if spontaneous emission is present , which in reality is inevitable . in section iii we therefore proceed to investigate the degree to which such quantum error correction methods are able to improve on these results . it is generally believed that quantum error correction is able to increase the number of possible operations substantially , e.g. a single error correcting code should allow to perform a number of operations which is approximately the square of the number of allowed operations without error correction . we analyze critically this idea in real atomic systems , especially taking into account the fact that due to the breakdown of the two - level approximation spontaneous transitions may leak population out of those levels that represent the qubit in the ion . it will be difficult , if not impossible , to bring this population back into the qubit . this means that after such an event the quantum error correction code will fail to reconstruct the state , as the reconstruction procedure works only if the entire population remains in the qubit states . we take this possibility into account and obtain estimates that show that the efficiency of quantum error correction is smaller than expected , although the application of quantum error correction allows factorization of somewhat larger numbers than without the use of quantum error correction . in section iv we will then discuss the prospects of factorization and other applications of quantum computers in view of the results of the present paper . we conclude that spontaneous emission considerations preclude many applications within the present model of quantum computation . before we come to the derivation of limits imposed on quantum computation by spontaneous emission we briefly describe the ion trap implementation of the quantum computer @xcite on which we base our considerations . several ions of mass @xmath2 are trapped in a linear ion trap such that they are lined up and are well separated , ie the next neighbour distance is many wavelengths of the lasers used to manipulate the ions . this is necessary to be able to address ions separately and can lead to further restrictions on quantum computation in this model @xcite . for a schematic picture see fig . the motional degrees of freedom of the ions , and especially their collective center - of - mass ( com ) motion with frequency @xmath3 , are cooled to the ground state . in order to be able to implement two - bit gates in this scheme we use the com mode as a bus which allows us to create entanglement between different ions . this is achieved with an interaction that creates a phonon in the com mode when we deexcite an ion and annihilates a phonon in the com mode when an ion is excited . this interaction is generated by a laser which is detuned from the transition frequency by @xmath4 . in that case the hamilton operator in the lamb - dicke limit , the rwa and a suitable interaction picture is given by @xcite @xmath5\ ; , \label{1}\ ] ] where @xmath6 is the lamb - dicke parameter , @xmath7 is the rabi frequency on the @xmath8 transition where levels @xmath9 and @xmath10 represent the corresponding logical values of the qubit . the denominator @xmath11 originates from the fact that we are considering the com mode which has an effective mass @xmath12 because all @xmath13 ions are oscillating in the trap potential . the trap has to contain @xmath13 ions , as this is the number of ions that is required to implement shor s algorithm to factorize an @xmath0 bit number @xcite ( actually the correct number is @xmath14 but for simplicity we drop the @xmath15 ) . it is further assumed that coupling to other vibrational levels can be neglected . this assumption is a reasonable first approximation as the closest lying vibrational mode has frequency @xmath16 independent of the number of ions in the trap . nevertheless , this approximation breaks down when the rabi frequency becomes too large , ie . if @xmath17 @xcite . for our considerations of the influence of spontaneous decay we neglect this effect although it may well become important in longer calculations . using the hamilton operator eq . ( [ 1 ] ) it can be shown that it is possible to construct a cnot gate with only four @xmath18-rotations , and to construct the more involved toffoli gate @xcite with six @xmath18-rotations @xcite . in the following we will define an elementary time step in terms of the computation time that is required to perform a cnot gate . all other performance times can be reexpressed in units ofthat of the cnot gate . the implementation of shor s algorithm also requires a number of one bit operations , ie . operations that leave the com mode unaffected during their performance . however , these operations can be performed much faster than two - bit gates , because @xmath19 and also because fewer individual laser pulses are required . therefore their contribution to calculation time and decoherence is small and will be neglected in the following . after this short discussion of the ion trap model we will now derive the intensity independent estimates for @xmath0 , first for a qubit which is stored in a metastable two - level transition ( see fig . [ fig1 ] ) . later in this section we will also consider the case of a qubit stored in zeeman sublevels manipulated by strongly detuned raman pulses ( see fig . [ fig2 ] ) . first we assume that the two - level approximation holds and derive an expression for the computation time that is required to factorize an @xmath0 bit number . as the next step we then include in our calculation possible extraneous levels and their spontaneous decay . as stated above , a cnot gate can be implemented by four @xmath18-pulses according to the hamilton operator eq . ( [ 1 ] ) . this requires an elapsed time @xmath20 which we will call the elementary time step . it is known @xcite that @xmath21 of these elementary time steps are required to implement shor s algorithm using elementary gates such as one bit gates , cnot gates , toffoli gates and fredkin gates . therefore the total computation time to complete a factorization will be @xmath22 which assumes that the gates are executed one after the other with zero time delay between two gates . the value of the constant @xmath23 very much depends on the actual practical implementation of shor s algorithm . the implementation given in @xcite requires @xmath24 cnot gates , @xmath24 toffoli gates and @xmath25 fredkin which result in @xmath26 . we neglect here all contributions of order @xmath27 which give significant corrections only for small @xmath0 . if the quantum computation is to give a useful answer , no spontaneous emission is allowed to occur during the whole calculation time @xmath28 , because an emission usually alters the wavefunction of the quantum computer completely . we will illustrate this point for a quantum computer which performs a discrete fourier transformation ( dft ) @xcite . qualitatively similar results were obtained in ref . @xcite for phase errors . spontaneous emission in fact leads to two sources of errors . one , quite obviously , arises from the actual spontaneous emissions in the quantum computer . the other one is less obvious and is due to the conditional time evolution when no spontaneous emission has taken place . this time evolution differs from the unit operation because the failure to detect a photon provides us with information about the system which is reflected in a change of the wavefunction @xcite . first we illustrate the case of an unstable quantum computer which performs a dft on a function which is evaluated at @xmath29 points . the resulting square modulus of the wavefunction of the quantum computer is compared to the exact result obtained from a absolutely stable quantum computer . the function on which we perform the dft is given by @xmath30 for @xmath31 . we have implemented the hamilton operators ( in the lamb - dicke limit ) for all the necessary quantum gates in a linear ion trap @xcite to realize this dft . in addition to the coherent time evolution we also take into account possible spontaneous emissions from the upper levels of the ions but we neglect all other sources of loss . to calculate the time evolution of the quantum computer , we then use the quantum jump approach @xcite to simulate the time evolution of a _ single _ quantum computer . the simulation runs as follows : we generate a random number and compare this random number after each time step with the squared norm of the wavefunction of the quantum computer ( the norm of the wavefunction decreases as it represents the probability that no photon has been emitted ) . if the squared norm of the wavefunction is smaller than the random number , an emission is deemed to have occured . we then generate another random number and continue our simulation . two results of our simulations are shown in figs . [ fig5 ] and [ fig6 ] . in fig . [ fig5 ] one emission has taken place during the calculation time of the quantum computer . if we compare the resulting wavefunction with the correct wavefunction , we observe a marked difference between the two . in fig . [ fig6 ] we show the wavefunction of an unstable quantum computer which has not suffered a spontaneous emission during the calculation of the dft . we clearly see that even when no spontaneous emission has taken place , the wavefunction of the quantum computer differs substantially from the correct result . this difference becomes stronger and stronger the larger the ratio between the computation time @xmath28 and the spontaneous lifetime @xmath32 of the quantum computer becomes . therefore the wavefunction of the quantum computer will be sufficiently close to the correct result only if the whole computation is finished in a time @xmath28 that is much shorter than the spontaneous lifetime @xmath32 of the quantum computer . therefore we need to determine the spontaneous lifetime of the quantum computer . it is reasonable to assume that each ion has an average excitation of @xmath33 during the whole calculation . the reason for this is that the quantum computer is in a superposition of very many states each of which represents a string of logical states @xmath9 and @xmath10 . the probability to find an atom excited will on average be @xmath33 . in addition during the computation each ion will suffer many @xmath34 rotations which again leads to an average population of @xmath33 . as the ions are separated by many wavelengths it is reasonable to assume that each ion decays independently of all others with a decay rate of @xmath35 . as the quantum computer consists of @xmath13 ions the spontaneous lifetime of our quantum computer is @xmath36 therefore we obtain the condition @xmath37 and from that @xmath38 we can now insert eq . ( [ 6 ] ) into eq . ( [ 3 ] ) and obtain @xmath39 where we have isolated the fraction @xmath40 because it can be expressed in a form that is independent of the actual type of transition we are considering ( electric dipole , quadrupole , ... ) . we find @xmath41 which can be derived along similar lines as in the usual quantum mechanical derivation of the relation between the einstein @xmath42 and @xmath43 coefficient . if we want to reduce the estimated computation time @xmath28 we have to have a ratio @xmath44 which is as large as possible . as the transition frequency is given by the ion we can adjust only the field strength of the laser . however , there is an upper limit to this field strength which is approximately given by the field strength between an electron and the proton in a hydrogen atom in its ground state . the value is @xmath45 where @xmath46 . this would yield a maximum value of @xmath47 for eq . ( [ 8 ] ) . however , one should note that this field strength is sufficient to destroy the ion within one optical cycle by tunnel ionization @xcite and represents therefore a very hypothetical upper limit . in fact we will see later in this section that , not surprisingly , other approximations such as the two - level approximation break down much earlier . we may use eq . ( [ 7 ] ) to obtain a first estimate for an upper limit on @xmath0 by requiring that @xmath28 has to be smaller than the decoherence time @xmath48 due to all other decohering processes that may occur , such as collisions , stray fields , phonon losses , laser phase fluctuations , . nevertheless , from a fundamental point of view the situation is not yet entirely satisfying , as in principle higher laser powers and technical improvements concerning the auxiliary decoherence effects could change the estimates quite drastically . therefore we will now proceed to show that the inclusion of spontaneous emission from auxiliary levels into the dynamics will lead to an estimate for the upper limit of @xmath0 which is entirely _ intensity independent_. to model the influence of extraneous levels on the decoherence rate due to spontaneous emissions we assume that there is one other level @xmath15 present which couples to both levels @xmath9 and @xmath10 by the same laser that drives the @xmath8 transition . as the contribution of even further detuned levels decreases rapidly with the detuning ( it can be shown that the infinitely many states in the atom give only a finite contribution ) we restrict our treatment to one additional level . as the laser is strongly detuned from the @xmath49 transitions the population in level @xmath15 will be very small and we can safely assume that the contributions from levels @xmath9 and @xmath10 to the population of level @xmath15 add up . the population in level @xmath15 will then be @xmath50 where @xmath51 is the rabi frequency on the @xmath49 transition and @xmath52 the detuning on that transition . in eq . ( [ 12 ] ) we have not taken into account that for very large detuning the rwa is not very good any more . however , the contribution of all possible atomic levels should be approximated well by eq . ( [ 12 ] ) . we have used the notation @xmath53 because the rabi frequency on the @xmath49 transition now depends on the type of the @xmath54 transition . there are two cases . \a ) if the @xmath8 transition is an electric quadrupole transition ( e2 ) then for the excitation of phonons it is necessary to place the ion at the antinode of the electric field of the laser . then the same laser on an electric dipole transition ( e1 ) will , in leading order , leave the com mode untouched . this implies that @xmath55 \b ) if the @xmath8 transition is an electric octupole transition ( e3 ) then for the excitation of phonons it is necessary to place the ion at the node of the electric field of the laser . then the same laser on an electric dipole transition ( e1 ) will also excite phonons in the com mode . this implies that @xmath56 where @xmath57 is the lamb - dicke parameter on the @xmath49 transition . we will have to distinguish between the two possibilities a ) and b ) , the first of which occurs for example in @xmath58,@xmath59 and @xmath60 while the second one occurs in @xmath61 ( an ion which might allow factorization of very large numbers according to estimates which do not include spontaneous emission @xcite ) . we first deal with case a ) and then state the result for case b ) . in case a ) we obtain @xmath62 as in the pure two - level case , we want to be sure that no spontaneous emission will take place during the whole calculation , neither from level @xmath10 nor from level @xmath15 . this implies the condition @xmath63 we now use eq . ( [ 7 ] ) and @xmath64 which derives from eq . ( [ 8 ] ) and where @xmath65 is the decay rate of level @xmath15 on the @xmath54 transition and @xmath66 the corresponding transition frequency . inserting eqs . ( [ 7 ] ) and ( [ 15 ] ) into eq . ( [ 13 ] ) we obtain @xmath67 where we have assumed that @xmath68 , ie . level @xmath10 only decays back into state @xmath9 . for case b ) we obtain a slightly different result which nevertheless has a very similar structure . we find @xmath69 where we have used the fact that the lamb - dicke parameter depends on the transition frequency so that @xmath70 it is important to note that the expressions eqs . ( [ 16 ] ) and ( [ 17 ] ) are _ independent _ of the laser power as long as @xmath71 . this independence has its origin in the fact that increasing the laser power gives rise to two competing effects . it decreases the computation time @xmath28 given in eq . ( [ 7 ] ) and therefore decreases the probability for an emission from level @xmath15 . on the other hand an increased laser power increases the population in level @xmath15 which increases the probability for a spontaneous emission . both effects cancel , leading to the intensity independent results eqs . ( [ 16 ] ) and ( [ 17 ] ) . in table 1 we give values for the bounds eq . ( [ 16 ] ) and ( [ 17 ] ) for realistic ions . the values for @xmath72 and @xmath73 were calculated using eq . ( [ 16 ] ) as the qubit transition is quadrupole allowed , while @xmath74 with an octupole allowed qubit transition is an example of case b ) and is therefore calculated according to eq . ( [ 17 ] ) . we calculate the values for @xmath0 assuming @xmath75 , which is the most optimistic choice , and for two values of the lamb - dicke parameter : an optimistic @xmath76 and the more realistic @xmath77 @xcite . we see that the numbers that may be factorized on a quantum computer using metastable optical transitions are very small even for the optimistic choice @xmath76 . only the @xmath74 ion gives results from which one may hope to factorize at least small numbers . however , it should be realized that it is extremely difficult to drive the hyperstable qubit transition in @xmath74 sufficiently quickly to finish the calculation in a reasonable time . if we assume a ratio @xmath78 where @xmath79 then from eq . ( [ 3 ] ) we find for @xmath76 the value @xmath80 for a @xmath81 bit number . this is so long that it is very unlikely that one can isolate the system during that time from all other decoherence sources . therefore the practical limit to @xmath0 for @xmath74 is probably smaller than the one given in table 1 . so far we have dealt with the case where the qubit is stored in a metastable optical transition . this method certainly has the disadvantage that at practically all times the quantum computer has a mean excitation of about @xmath33 per qubit . therefore , even if a qubit does not take part in a quantum gate operation , the qubit will decohere due to spontaneous emissions from the excited state of the qubit . to circumvent this problem one would like to store the qubits in zeeman sublevels of a stable ground state . in that case stored qubits do not suffer any decoherence due to spontaneous emissions . however , how do we manipulate the qubit ? it is unfortunately not practical to drive the qubit with a microwave field directly because the very long wavelength of this radiation does not allow us to address single ions as they can not be separated far enough in a linear trap . therefore we have to use a different method of manipulating the qubit . this method uses the presence of another level @xmath15 ( see fig . [ fig2 ] ) which is coupled to the two qubit levels @xmath9 and @xmath10 by two lasers of rabi frequency @xmath82 and @xmath83 . each laser is strongly detuned from its transition , and its rabi frequency is much smaller than the detuning . however , it is assumed that both detunings are equal , ie . the two photon detuning vanishes while the one photon detuning @xmath84 is large . the advantage of this method is that due to the strong detuning @xmath84 the population in level @xmath15 is small , so that spontaneous emissions from that level are rare . a hamilton operator that implements an interaction analogous to the one generated by eq . ( [ 1 ] ) , ie the qubit is excited while a phonon in the com mode is deexcited and vice versa , is given by @xmath85 + \frac{\hbar\eta\omega_{12}}{2\sqrt{5l } } \left [ |2\rangle\langle 1| a + |1\rangle\langle 2| a^{\dagger}\right ] \ ; , \label{19}\ ] ] where again the lamb - dicke limit , the rwa and a suitable interaction picture is used . to make sure that the raman transitions have the effect as full rabi oscillations between levels @xmath9 and @xmath10 the condition @xmath86 has to be satisfied which we will assume in the following . the condition eq . ( [ 20 ] ) can be obtained from the solution of the time evolution @xmath87 which is valid for @xmath88 and for an initial state @xmath89 . to obtain @xmath90 for some @xmath91 eq . ( [ 20 ] ) has to be satisfied . before we continue we have to realize that it is indeed enough to consider the time evolution eqs . ( [ 21]-[23 ] ) and the corresponding time evolution for the initial condition @xmath92 . to see this we realize that to a good approximation the coherences between levels @xmath9 and @xmath10 are uniformly distributed . averaging the initial state over this distribution then yields a mixed state as an initial state in which both states @xmath9 and @xmath10 have equal weight . therefore calculating the time evolution for both possible initial states separately and averaging over the two results leads to the final result . again in the raman pulse implementation of qubits a cnot gate can be implemented with four @xmath18 pulses . from the solution eqs . ( [ 21]-[23 ] ) we see that the effective rabi frequency of the qubit transition is given by @xmath93 therefore the performance of a cnot gate requires the time @xmath94 again the probability for a spontaneous emission from level @xmath15 has to be small , ie . @xmath95 this condition immediately yields @xmath96 which is already intensity independent . from eq . ( [ 27 ] ) one could conclude that one is able to factorize gigantic numbers by making the ratio @xmath97 sufficiently large , eg . @xmath98 and @xmath99 would lead to @xmath100 . but again this reasoning is invalid because we have neglected the fact that the two - level approximation breaks down because in the ion other levels , which we model by a level @xmath101 , also couple to levels @xmath9 and @xmath10 via the raman pulses . this additional coupling is very important because it potentially involves levels that may have @xmath102 . in addition the presence of level @xmath101 changes the effective rabi frequency in the qubit and may therefore lead to additional errors . finally there is a limit to the size of @xmath84 from the nature of the qubit quantum numbers , as usually states @xmath9 and @xmath10 are hyperfine levels which require a nuclear spin flip which is precluded for very large @xmath84 . to achieve a more restrictive , but again intensity independent , upper bound for @xmath0 than the one given in eq . ( [ 27 ] ) we now take into account the presence of one more level @xmath101 which acquires population during the execution of the quantum gate . one can give a complete analysis of the full four level system ; however , this is very tedious , although not complicated in principle . in this paper we will restrict ourselves to two limiting cases where the analysis is simpler and already reveals the basic physics . we avoid an intermediate regime which is also not of practical interest as the precise control of the qubits will be difficult . if we denote the rabi frequencies on the @xmath103 transition by @xmath104 and the one photon detuning as @xmath105 we can distinguish two cases , one of which reduces to the analysis given above , the other one requiring further investigation . first consider the case where @xmath106 in that case the analysis given above , neglecting level @xmath101 , was not correct as the influence of level @xmath101 on the dynamics of the qubit is actually larger than that of level @xmath15 . for example the proper effective rabi flopping frequency of the qubit is then given by @xmath107 instead of @xmath108 . due to relation eq . ( [ 8 ] ) we can conclude from eq . ( [ 28 ] ) that @xmath109 and therefore @xmath110 in fact what one has to do is to redo the analysis leading to eq . ( [ 27 ] ) but replacing level @xmath15 by level @xmath101 . the other case is given by @xmath111 which means that indeed level @xmath15 determines the dynamics of the system . in that case we immediately find @xmath112 where @xmath105 and @xmath113 is more or less determined because the lasers are extremely far detuned from the @xmath114 transitions . this could already serve as a strong limit for @xmath0 , however , it is of a different structure compared to the intensity independent limit eqs . ( [ 16]-[17 ] ) for the two - level case . it is interesting to note that it is possible to construct a limit which has a form completely analogous to eqs . ( [ 16]-[17 ] ) only with slightly different exponents . for this derivation we assume that level @xmath15 influences the qubit dynamics the most , ie . ( [ 31 ] ) is satisfied . this assumption itself together with @xmath115 ( detuned raman pulses ) yields a lower limit to the computation time @xmath116 we will see from this expression for the computation time that any system with an exceedingly high lifetime of level @xmath15 is practically useless for quantum computation because the time required to perform a computation is extremely high . again of course it is important to distinguish between the cases a ) and b ) for the different types of qubit transitions ( electric quadrupole , octupole , .. ) as we have done when we investigated the case of a two - level system as a qubit . we will treat case a ) explicitly in the following while we only state the result for case b ) . to calculate the influence of level @xmath101 to the decoherence we first have to find out the average population in that level . for this let us realize that during the implementation of a cnot a full @xmath117-rotation will be performed . that means that on average the population in both levels @xmath9 and @xmath10 is equal . therefore the much slower coupling to level @xmath101 sees an averaged population in both levels with no average coherence . we can therefore average over two contributions to @xmath118 which originate from the initial states @xmath119 and @xmath120 and we obtain @xmath121 where @xmath57 is the lamb - dicke parameter for the @xmath122 transition . ( [ 33 ] ) contains an additional factor @xmath123 compared to eq . ( [ 23 ] ) . this is due to the fact that level @xmath101 is strongly decaying and that therefore @xmath124 . note also that the roles of the lasers on the @xmath114 transitions has been reversed compared to their role on the @xmath49 transitions ; the phonon is now destroyed by an excitation from level @xmath9 and not from level @xmath10 anymore . this is due to the fact that @xmath114 transitions are dipole allowed ( e1 ) while the @xmath49 transition is quadrupole allowed ( e2 ) . again an emission from level @xmath101 should not occur during the whole calculation which means @xmath125 now we have to find out which term in the brackets of eq . ( [ 34 ] ) is dominant . because we need @xmath126 , ( [ 20 ] ) , and because we are in the lamb - dicke limit , ie . @xmath127 , the laser on the @xmath128 transition has to be weaker than the one on the @xmath129 transition . but again @xmath130 and therefore @xmath131 . this implies @xmath132 where we have inserted the decay coefficients @xmath133 from level @xmath134 to level @xmath135 to be able to apply eq . ( [ 8 ] ) which then yields @xmath136 where @xmath137 is the electric field of the laser on the @xmath49 transition . we now define the constant @xmath138 and resolve eq . ( [ 36 ] ) for @xmath0 to obtain @xmath139 note that @xmath140 as argued below eq . ( [ 34 ] ) and that the ratio of the decay coefficients in @xmath141 will be of the order of @xmath10 . therefore @xmath141 is much smaller than @xmath10 . in fact , because of @xmath126 , is a good approximation to assume @xmath142 ( with @xmath143 defined below eq . ( [ 39 ] ) ) which then yields @xmath144 following very similar lines we obtain for case b ) @xmath145 where now @xmath146 is of the order of @xmath10 . both limits ( [ 38]-[39 ] ) are analogous in their form to the estimates eqs . ( [ 16 ] ) and ( [ 17 ] ) if we neglect in eqs . ( [ 16 ] - [ 17 ] ) the contribution from the upper level @xmath10 coupling to level @xmath101 . in table 2 we present some values for @xmath0 calculated from eqs . ( [ 38a ] ) and ( [ 39 ] ) for the same ions as in table 1 . again we assume that @xmath147 ( an optimistic choice ) and observe that the resulting limits for the numbers that can be factorized on a quantum computer are very small , especially if @xmath77 . nevertheless one could think that transitions in @xmath148 and @xmath149 are promising . however , two effects appear which lead to further limitations . one is the fact that eqs . ( [ 38a ] ) and ( [ 39 ] ) are not necessarily more restrictive than the equivalent of eq . ( [ 27 ] ) for level @xmath101 as we have assumed eq . ( [ 32 ] ) which does not need to be true . in fact simply applying eq . ( [ 27 ] ) where @xmath150 is replaced by @xmath113 yields for @xmath151 @xmath152 , for @xmath148 @xmath153 and for @xmath149 @xmath154 which are somewhat smaller than the values for those ions in table 2 . secondly and more importantly the promising results for @xmath148 and @xmath149 are illusionary because a very stable transition is also very slow . from eq . ( [ 32a ] ) we can easily calculate that the factorization time using @xmath148 and assuming @xmath76 is @xmath155 which gives for @xmath156 a value of @xmath157 during which no decohering event of any kind is allowed to happen . this is a very long time and it will be difficult to isolate the quantum computer from the environment . for @xmath149 the numbers are even more devastating . we obtain @xmath158 which indicates for @xmath159 that @xmath160 , a completely unrealistic number . this again shows very clearly that the solution to the problem of decoherence can not be to employ extremely stable transitions for qubits but to find methods that enable the quantum computer to cope with a certain level of decoherence , ie we need a form of quantum error correction . in principle it is indeed possible to implement quantum error correction in quantum computers but it remains the question inasmuch these methods will in fact improve the prospects of quantum computation . in the next section we will address this problem and derive limitations to quantum factoring including the use of quantum error correction methods . in the preceding section we derived intensity independent upper bounds for the numbers that can be factorized on a quantum computer . the results that we obtained are not very promising as they rule out the possibility to factorize large numbers ( numbers with @xmath161 can be factorized on a classical computer ) on a quantum computer . however , in these estimates we have yet to include the possibility of quantum error correction @xcite and fault - tolerant quantum computation @xcite . these methods have the ability to change drastically the probability that the calculation becomes corrupted by an emission . for example , consider a quantum error correction code that can correct a single general error . if the probability to suffer an error is @xmath162 then the probability to suffer two errors , an event that can not be corrected by the code , is @xmath163 . therefore , we have reduced the probability for a fatal error from @xmath164 to @xmath163 and it is then possible to perform quadratically as many computational steps as possible without quantum error correction . higher order correcting codes promise even higher gains in possible computational steps . however , this simple argument neglects two effects that counteract the expected benefits of quantum error correction codes . one reason is the quite obvious fact that the implementation of quantum error correction and fault - tolerant quantum computation requires substantial overheads in additional qubits for encoding the qubits and additional quantum gate operations to implement quantum logic operations on the encoded quantum bits . this overhead can be quite substantial even for simple quantum error correction codes . the other effect is less obvious but can have an even stronger impact on the practical efficiency of quantum error correction codes . the problem is that quantum error correction codes loose their ability to correct for decohering events if population leaves the qubit as a result of the decohering event . therefore we have to hope that all population will remain within the system representing the qubit . however , as we have seen in the analysis of section ii this will not be the case for a realistic implementation of a qubit . the reason is simply that the two - level approximation is never perfectly satisfied . if we excite an extraneous level then an emission from this level does not necessarily bring the population back into the qubit . an error of this kind can not be dealt with by a quantum error correction code . one would have to find ways to repump the lost population back into the qubit . this , however , is quite difficult if not impossible because of the sheer multitude of possible levels where the spontaneous emission may lead the population to . in this section we will first consider the case of quantum computation using two - level systems to represent the qubits . subsequently we will then consider the representation of qubits in zeeman sublevels using far detuned raman pulses for their manipulation . before we start with the discussion of the two - level system case , we first define two constants , @xmath165 and @xmath166 , describing the necessary overheads in the number of qubits and the number of additional quantum gate operations . * the number of required qubits rises by a factor of @xmath165 as all qubits will have to be encoded and will be kept encoded during the calculation . repeated decoding and encoding would raise the number of operations substantially and lead to unnecessary errors . * the number of operations required to implement one quantum logical operation on encoded qubits rises by a factor of @xmath166 because we have to perform more than one quantum gate to implement one quantum logic operation on the encoded qubits . in addition we also have to perform quantum error corrections periodically . the value of the factors @xmath165 and @xmath166 depends heavily on the quantum error correction code that is used . if one wants to protect a single qubit against one general error , then it is known that @xmath167 is the minimal possible value @xcite . the number @xmath166 for this code is uncertain in general for the most optimal codes but is at least @xmath168 for a cnot gate @xcite and probably much more for other gates such as a toffoli gate . for the following analysis we assume an error correction code capable of correcting one error . a similar calculation will then yield the result for codes that can correct @xmath169 errors and we simply state these results . first we need to find the time required to perform a cnot operation on encoded quantum bits . the computer contains @xmath165 times more qubits , ie . qubits , and @xmath166 times more operations are required . therefore @xmath171 the spontaneous lifetime of the quantum computer is then @xmath172 as we now have the ability to correct for errors during the calculation , we no longer need to require that there is practically no spontaneous emission during the whole calculation . instead we ask for the probability @xmath173 to have one error during @xmath174 logical operations of our algorithm to be small compared to @xmath10 . we find @xmath175 if we perform a fault - tolerant error correction after these n operations then this correction will fail with probability @xmath176 , the probability to have suffered two errors , because the code is not designed to correct for two errors . therefore the probability that the whole computation fails is given by @xmath177 this immediately implies that @xmath178 is the optimal choice unless we take the @xmath174 dependence of @xmath166 into account which is however quite difficult as @xmath166 is not known very well . for the following we will assume @xmath178 and we obtain @xmath179 from eq . ( [ 44 ] ) we easily deduce the total computation time @xmath180 again eqs . ( [ 44 ] ) and ( [ 45 ] ) are dependent on the laser power driving the system . to eliminate this intensity dependence we now have to take other levels into account as we did in section ii . again we have to distinguish the two cases a ) ( qubit transition is quadrupole allowed ) and b ) ( qubit transition is octupole allowed ) depending on the nature of the qubit transition . we first treat case a ) and then state the result for case b ) . to keep the resulting expressions more transparent we neglect the contribution of the upper qubit level @xmath10 to the violation of the two - level approximation . the generalization to this case is easy.we then find for the population in the extraneous level @xmath15 @xmath181 now we want to know the probability @xmath182 that an emission from level @xmath15 which leads to population _ outside _ the qubit transition occurs during the whole computation . it is very important to note that it is this probability @xmath182 that has to be much smaller than unity because even a single emission leading out of the qubit can not be corrected by the quantum error correction code . we find @xmath183 this can be solved for @xmath0 and with eq . ( [ 8 ] ) we obtain @xmath184 with an analogous calculation we obtain for case b ) @xmath185 the estimates eqs . ( [ 48 ] ) and ( [ 49 ] ) are valid for a quantum error correction code that is able to correct for one error . one can quite easily extend these results to quantum error correction codes that are able to correct for @xmath169 errors . we state only the main results for the cases a ) and b ) as they can be derived easily along the lines that lead us to eqs . ( [ 48 ] ) and ( [ 49 ] ) . for case a ) we obtain the computation time @xmath186 now we take into account other levels outside the qubit transition . a spontaneous emission that leads from these levels out of the qubit system can not be corrected for by the quantum error correction code . therefore we obtain @xmath187 for case b ) we obtain the result @xmath188 again we can now discuss numerical values for eqs . ( [ 48 ] ) and ( [ 49 ] ) obtained from the atomic data of real ions . in table 3 we give the results for eqs . ( [ 48 ] ) and ( [ 49 ] ) . we see that although the estimates have improved they still restrict factorization to very small numbers especially for a lamb - dicke parameter @xmath77 where we obtain upper limits that restrict the numbers that are possible to factorize to trivial sizes . again @xmath149 seems to be very promising but again we should note that it is very hard to actually drive the this transition sufficiently quickly so that the computation times quickly reach astronomical values . for @xmath78 we obtain for @xmath159 a value of @xmath189 , a value which is extremely high . for @xmath148 we obtain for @xmath159 , @xmath78 and @xmath76 the value @xmath190 . now we perform the same analysis including quantum error correction codes for the case where the qubits is stored in zeeman sublevels . of course the analysis here runs along similar lines as the one for the two - level system and that of section ii . therefore we only state the final result for the cases a ) and b ) . we present the results for an error correcting code which is able to correct @xmath169 errors . we only treat the case where the extraneous levels give a small correction to the time evolution of the system , ie we assume the regime of eq . ( [ 31 ] ) . for the computation time we obtain the lower limit @xmath191 for both cases a ) and b ) . for case a ) we obtain the estimate @xmath192 where @xmath193 @xmath137 denotes the electric field strength of the laser on the @xmath194 transition . note that again we have to take into account that @xmath141 contains a hidden dependence on @xmath195 and @xmath165 . this dependence is due to the fact that we have to satisfy the condition @xmath196 therefore @xmath197 ( @xmath143 is defined in ( [ 57 ] ) ) and we obtain @xmath198 for case b ) we obtain a very similar result . we find @xmath199 where @xmath200 assuming that @xmath201 we then obtain the values given in table 4 which again seem to be promising if we assume @xmath76 . for the more realistic values of @xmath77 , however , the estimate reduces substantially . in the discussion of table 2 we already pointed out that high values for the estimates on @xmath0 alone do not suffice to allow a practical implementation of quantum factoring . important also is that the computation time be sufficiently short . this restriction again excludes @xmath149 immediately because for @xmath202 and @xmath159 we obtain @xmath203days . it also makes @xmath148 a very unlikely candidate for successful factorization as for @xmath156 we obtain with the same parameters as for @xmath149 the value @xmath204 . in this paper we have discussed the constraints imposed by spontaneous emission onto factorization of big numbers using a quantum computer . in section ii we have investigated the possibility of factorization without the use of quantum error correction . we were able to derive limits to the bitsize of the numbers that can be factorized on a quantum computer . these limits are _ independent _ of the intensity of the laser that is used to implement quantum gates . this intensity independence was achieved by taking into account the failure of the two - level approximation even for modest laser powers . the result of these limits is that without the use of quantum error correction the factorization even of small numbers will not be possible . therefore we then investigated in section iii whether the implementation of quantum error correction can improve the prospects of factorization on a quantum computer . it turned out that quantum error correction in realistic atomic systems is much less effective than in the ideal case of closed two - level systems because the failure of the two - level approximation can lead to a leakage of population out of the system which can not be corrected by a quantum error correction code . therefore even high order quantum error correction codes are susceptible to single errors that lead to loss of population out of the system which corrupts their performance . nevertheless the upper limits in this case are higher than without quantum error correction . however , the total computation time using quantum error correction , and even in the case without quantum error correction , tends to be very large because a very stable transition can not be driven very fast . we find lower bounds for the computation times and conclude that even for moderate numbers it will be difficult to isolate the system from the environment such that no decohering event , other than a spontaneous emission , can take place . at this point other decoherence effects have to be taken into account ( see eg . however , we think that even the sole inclusion of spontaneous emission as a decohering effect already shows that the practical application of quantum computers in factorization of big numbers will be highly unlikely within the present models of quantum computation . the considerations of this paper also apply to other possible algorithms for quantum computers if they also require @xmath205 operations . although the conclusions of this paper are rather pessimistic with regard to the practical application of quantum computers for actual computations we would like to stress that there are applications such as for example in precision spectroscopy @xcite which require much less operations , @xmath206 , and for which therefore the above presented estimates will be much more optimistic . we believe that in this area interesting applications of the concept of quantum computing will be found . however , unrealistic hopes that have been raised in the past about the practicality of quantum computers have to be moderated . we would like to thank c. monroe and j.i . cirac for alerting our attention to the method of employing raman transitions to drive qubits . we acknowledge discussions with d.f.v . james and r.j . hughes and thank them for providing us with the experimental data for the ions . this work was supported by a european community network , the uk engineering and physical sciences research council and by a feodor - lynen grant of the alexander von humboldt foundation + 99 t. andersen and g. srensen , 1973 systematic trends in atomic transition probabilities in neutral and singly - ionized zinc , cadmium and mercury , journal of quantitative spectroscopy and radiative transfer * 13 * , 369 . augst , s. , strickland , d. , meyerhofer , d.d . , chin , s.l . & eberly , j.h . , 1989 tunnel ionization of noble - gases in a high - intensity laser field , phys . lett . * 63 * , 2212 - 2215 and references therein barenco , a. , deutsch , d. , ekert , a. & josza , r. , 1995 conditional quantum dynamics and logic gates , phys . lett . * 74 * , 4083 - 4086 barenco , a. , bennett , c.h . , cleve , r. , divincenzo , d.p . , margolus , n. , shor , p. , sleator , t. , smolin , j.a . & weinfurter , h. , 1995 elementary gates for quantum computation , phys . a * 52 * , 3457 - 3467 barenco , a. , quantum physics and computers , contemp . phys . * 37 * , 375 - 395 ( 1996 ) barenco , a. , , ekert , a. , suominen , k.a . & p. torm , 1996 approximate quantum fourier transform , phys . rev . a * 54 * , 139 - 146 bashkin , s. , & stoner , j.o . , 1978 _ atomic energy - level and grotian diagrams _ vol ii , sulfur i to titanium xxii p. 360 - 361 , ( north holland , amsterdam ) bell , a.s . , gill , p. , klein , h.a . , levick a.p . & rowley , w.r.c.,1992 , precision measurement of the @xmath207 @xmath208 interval in trapped yb+ , j. mod . opt . * 39 * , 381 - 387 bergquist , j.c . , wineland , d.j . , itano , w.i . , hemmati , h. , daniel , h .- u . & leuchs , g. , 1985 energy and radiative lifetime of the @xmath209 state in hg ii by doppler free two photon laser spectroscopy , phys . * 55 * , 1567 - 1570 calderbank , a.r . & shor , p.w . , 1996 good quantum error - correcting codes exist , phys . rev . a * 54 * , 1098 - 1105 carmichael , h.j . , 1993 , _ an open systems approach to quantum optics _ , lecture notes in physics , ( springer , berlin ) . cirac , j.i . & and zoller , p. , 1995 quantum computation with cold trapped ions , phys . lett . * 74 * , 4091 - 4094 cirac , j.i . , pellizzari , t. & zoller , p. , 1996 enforcing coherent evolution in dissipative quantum dynamics , submitted to science dalibard , j. , castin , y. & molmer , k. , 1992 wave - function approach to dissipative processes in quantum optics , phys . 68 * , 580 - 583 divincenzo , d.p . & shor , p.w . , 1996 fault - tolerant error correction with efficient quantum codes , in prl * 77 * , september ekert , a. , & josza , r. , 1996 rev . phys . * 68 * , 733 - 753 ekert , a. , & macchiavello , c. 1996 quantum error correction for communication , phys . lett . * 77 * , 2585 - 2588 eriksen , p. & poulsen , o. , 1980 lifetime measurements of the @xmath210 and @xmath211 levels in hg(ii ) , journal of quantitative spectroscopy and radiative transfer * 23 * , 599 fawcett , b.c . & wilson , 1991 m. , computed oscillator strengths , lande g values and lifetimes in yb+ , atomic data and nuclear data tables * 47 * , 241 - 317 gallagher , a. , 1967 oscillator strengths of ca ii , sr ii and ba ii , phys . rev . * 157 * , 24 garg , a. , 1996 decoherence in ion - trap computers , phys . * 77 * , 964 - 967 gill , p. , klein , h.a . , roberts , m. , rowley , w.r.c . & taylor , p. , 1995 absolute measurement of the @xmath212 @xmath213 interval and the search for the @xmath214 @xmath215 transition in laser - cooled trapped yb+ , _ proc . 5th symposium on frequency standards and metrology _ ( woods hole , ma ) gosselin , r.n . , pinnington , e.h . & ansbacher , w. , 1988 measurement of the lifetime of the 4p levels in caii using laser excitation of a fast beam , phys . rev . a * 38 * , 4887 - 4890 hegerfeldt , g.c . , and wilser , t. s. , 1991 , _ proceedings of the ii . international wigner symposium _ , edited by h. d. doebner , w. scherer , and f. schroeck ( world scientific , singapore ) . hughes , r.j . , james , d.f.v . , knill , e.h . , laflamme , r. & petschek , a.g . , 1996 , decoherence bounds on quantum computation with trapped ions , phys . 77 * james , d.f.v . & hughes , r.j . , 1996 global limits on quantum computation with cold trapped ions i : two level qubits , preprint knight , p.l . & garraway , b.m . , 1996 , _ quantum dynamics of simple systems . proceedings of the forty fourth scottish universities summer school in physics stirling _ edited by g - l . oppo , s.m . barnett , e. riis and m. wilkinson ( institute of physics publishing , bristol ) knill , e. , & laflamme , r. , 1996 , a theory of quantum error - correcting code , lanl e - print quant - ph/9604015 monroe , c. , meekhof , d.m . , king , b.e . , itano , w.m . & wineland , d.j . , 1995 demonstration of a fundamental quantum logic gate , phys . rev . lett * 75 * , 4714 - 4717 palma , g.m . , suominen , k .- a . & ekert , a.k . , 1996 quantum computers and dissipation , proc . * a*452 , 567 - 584 plenio , m.b . & knight , p.l . , 1996 _ proceedings of the 2nd international symposium on fundamental problems in quantum physics _ , edited by m. ferrero and a. van der merwe ( kluwer , dordrecht ) plenio , m.b . , and knight , p.l . , 1996 realistic lower bounds for the factorization time of large numbers on a quantum computer , phys . a * 53 * , 2986 - 2990 plenio , m.b . , and knight , p.l . , 1996 the quantum jump approach to dissipative dynamics in quantum optics , in preparation for rev . plenio , m.b . , vedral , v. & knight , p.l . , 1996 computers and communication in the quantum world , phys . world * 9 * , 19 - 20 plenio , m.b . , vedral , v. & knight , p.l . , 1996 conditional generation of error syndromes in fault - tolerant error correction , lanl e - print quant - ph/9608028 sauter , th . , blatt , r. , neuhauser , w. & toschek , p.e . , 1986 quantum jumps observed in the fluorescence of a single ion , opt . comm . * 60 * , 287 - 292 sauter , th . , neuhauser , w. , blatt , r. & toschek , p.e . , 1986 observation of quantum jumps , phys . rev . * 57 * , 1696 - 1698 shor , p. w. , 1994 in _ proceedings of the 35th annual symposium on the foundations of computer science , los alamitos , _ ca ( ieee computer society press , new york ) , p. 124 shor , p.w . , 1995 scheme for reducing decoherence in quantum computer memory , phys . a * 52 * , r2493 - 2496 shor , p.w . , 1996 fault - tolerant quantum computation lanl e - print quant - ph/9605011 sleator , t . & weinfurter , h. , 1995 realizable universal quantum logic gates , phys . lett . * 74 * , 4087 - 4090 steane , a.m. , 1996 error - correcting codes in quantum - theory , phys . lett , * 77 * , 793 - 796 steane , a.m. , 1996 multiple particle interference and quantum error correction , to appear in proc . a vedral , v. , barenco , a. & ekert , a. , 1996 quantum networks for elementary arithmetic operations , phys . a * 54 * , 147 - 153 wineland , d.j . , bollinger , j.j . , itano , w.m . & moore , f.l . , 1992 spin squeezing and reduced quantum noise in spectroscopy , phys . a * 46 * , r6797 - 6800 wiese , w.l . , smith , m.w . & miles , b.m . , 1969 _ atomic transition probabilitie , vol ii sodium through calcium _ p.251 , ( u.s . government printing office , washington ) a two - level system representing a qubit . the levels @xmath9 and @xmath10 represent the two possible logical values . unlike in classical computing coherent superpositions between the logical values @xmath9 and @xmath10 are possible . the upper level may decay spontaneously to the ground state with a decay rate @xmath35 and the system can be driven by a laser with rabi frequency @xmath7 . the two lower levels @xmath9 and @xmath10 of the @xmath216 system , are zeeman sublevels of a stable ground state and represent the logical values @xmath9 and @xmath10 of the qubit . they are coupled via raman pulses that are strongly detuned from the intermediate level @xmath15 . the two - photon detuning is assumed to be zero while the one photon detuning @xmath84 is much larger than the associated rabi frequencies on the @xmath49 transition is @xmath217 . schematic picture of the excitation of several ions in a linear ion trap . the translational degrees of freedom of the ions are assumed to be cooled to their respective ground states . to implement quantum gates , standing wave fields interact with the ions and thereby change the inner state of the ions as well as the state of the center - of - mass mode ( which leads to entanglement ) . schematic level scheme envisaged for quantum computation , in which the @xmath8 transition represents the qubit . it is driven by a laser of rabi frequency @xmath7 . level @xmath10 has a spontaneous decay rate @xmath35 which is small . the laser which is resonant with the @xmath8 transition inevitably couples level @xmath9 also to other non - resonant levels as for example level @xmath15 . the rabi frequency on that transition is then @xmath82 and the decay rate is @xmath218 is usually much larger than @xmath219 . the effective rabi frequency on the @xmath128 transition is very small as the laser is detuned by @xmath220 . results of a discrete fourier transform ( dft ) of a function @xmath221 with @xmath31 . the solid line is the result for a quantum computer with stable qubits and represents the correct result . the dashed line shows the result of the same computation using a quantum computer with unstable qubits , one of which has suffered a spontaneous emission during the calculation . the results clearly differ and show the impact of a single spontaneous emission on a quantum computation . for the parameters chosen on average the quantum computer will suffer one emission per dft , ie @xmath222 in this case . the same quantum computation as in fig . the solid line again represents the result using a quantum computer with stable qubits , while the dashed line shows the result using a quantum computer with unstable qubits . this time , however , the unstable quantum computer does not suffer an emission during the whole calculation . again the results differ illustrating the impact of the conditional time evolution between spontaneous emissions . for several possible systems the upper limit on the bitsize @xmath0 of the number @xmath174 that can be factorized on a quantum computer is calculated . a qubit is stored in a metastable optical transition . the atomic levels which are abbreviated in fig . [ fig1 ] by @xmath223 and @xmath15 are given . the atomic data are inserted into eqs . ( [ 16 ] ) and ( [ 17 ] ) and the result is given in the last row of the table . the atomic data are inserted into eqs . ( [ 16 ] ) and ( [ 17 ] ) and the result is given in the last row of the table . the atomic data are derived from ( a ) = @xcite , ( b ) = @xcite , ( c ) = @xcite , ( d ) = calculated using@xcite and @xcite , ( e ) = @xcite , ( f ) = @xcite,(g ) = @xcite , ( h ) = @xcite , ( i ) = @xcite , ( j ) = @xcite . for several possible systems the upper limit on the bitsize @xmath0 of the number @xmath174 that can be factorized on a quantum computer is calculated . a qubit is stored in two zeeman sublevels of a stable ground state . the atomic levels which are abbreviated in fig . [ fig7 ] by @xmath224 and @xmath101 are given . the atomic data are inserted into eq . ( [ 21 ] ) and the result is given in the last rows of the table . the @xmath216 system as in fig . 2 but now the coupling to an extraneous levels @xmath101 is taken into account . the lasers performing the raman pulses also couples the two lower states to level @xmath101 which may subsequently decay spontaneously . for several possible systems the upper limit on the bitsize @xmath0 of the number @xmath174 that can be factorized on a quantum computer using quantum error correction is calculated . a qubit is stored in a metastable optical transition.the atomic levels which are abbreviated in fig . [ fig7 ] by @xmath223 and @xmath15 are given . the atomic data are inserted into eq . ( [ 21 ] ) and the result is given in the last rows of the table . for several possible systems the upper limit on the bitsize @xmath0 of the number @xmath174 that can be factorized on a quantum computer is calculated . a qubit is stored in two zeeman sublevels of a stable ground state . the atomic levels which are abbreviated in fig . [ fig7 ] by @xmath224 and @xmath101 are given . the atomic data are inserted into eq . ( [ 21 ] ) and the result is given in the last rows of the table . ( 100,75 ) ( 40,0)(1,0)30 ( 40,0.3)(1,0)30 ( 40,30)(1,0)30 ( 40,30.3)(1,0)30 ( 55,3)(0,1)26 ( 55,27)(0,-1)26 ( 53,27.5)(0,-1.75)14(0,0)[c]@xmath225 ( 52.8,27.5)(0,-1.75)14(0,0)[c]@xmath225 ( 52.9,6)(0,-1)5 ( 47.5,15)(0,0)[c]@xmath219 ( 60,15)(0,0)[c]@xmath7 ( 55,-5)(0,0)[c]@xmath9 ( 55,34)(0,0)[c]@xmath10 ( 50,40 ) ( 10,-10)(1,0)30 ( 10,-9.7)(1,0)30 ( 5,-10)(5,5)[bl]0 ( 25,32)(1,0)30 ( 25,32.3)(1,0)30 ( 20,32)(5,5)[bl]2 ( 40,-6)(1,0)30 ( 40,-5.7)(1,0)30 ( 75,-6)(5,5)[bl]1 ( 24.8,-7)(1,3)10 ( 34,21)(-1,-3)10 ( 20,7)(5,5)[bl]@xmath82 ( 57,-4)(-1,3)9 ( 48.2,22)(1,-3)9 ( 56,7)(5,5)[bl]@xmath83 ( 25,24)(5.4,0)6(1,0)3 ( 25,24.3)(5.4,0)6(1,0)3 ( 60,23.5)(5,5)[bl]@xmath226 ( 67,26)(5,5)[bl]@xmath84 ( 100,50 ) ( 0,0)(1,0)10 ( 0,0)(0,1)5 ( 0,5)(1,0)10 ( 10,2.5)(10,5)[r ] ( 25,2.5)(15,0)8 ( 25,-3)(15,0)8(1,0)5 ( 25,-3)(15,0)8(-1,0)5 ( 23,20)(15,0)8(0,1)10 ( 27,20)(15,0)8(0,1)10 ( 23,30)(15,0)8(1,0)4 ( 23,20)(15,0)8(1,0)1 ( 26,20)(15,0)8(1,0)1 ( 39.5,29)(.2,0)5(0,-1)22 ( 99.5,29)(.2,0)5(0,-1)22 ( 49,35)(5,5)[bl]standing wave laser fields ( 145,2.5)(10,5)[l ] ( 145,0)(1,0)10 ( 155,0)(0,1)5 ( 155,5)(-1,0)10 ( 100,75 ) ( 10,0)(1,0)60 ( 10,0.3)(1,0)60 ( 40,30)(1,0)30 ( 40,30.3)(1,0)30 ( 10,60)(1,0)30 ( 10,60.3)(1,0)30 ( 55,3)(0,1)26 ( 55,27)(0,-1)26 ( 26,3)(0,1)56 ( 26,57)(0,-1)56 ( 23,57.5)(0,-1.75)32(0,0)[c]@xmath225 ( 22.8,57.5)(0,-1.75)32(0,0)[c]@xmath225 ( 22.9,6)(0,-1)5 ( 53,27.5)(0,-1.75)14(0,0)[c]@xmath225 ( 52.8,27.5)(0,-1.75)14(0,0)[c]@xmath225 ( 52.9,6)(0,-1)5 ( 47.5,15)(0,0)[c]@xmath219 ( 17.5,30)(0,0)[c]@xmath150 ( 31,30)(0,0)[c]@xmath82 ( 60,15)(0,0)[c]@xmath7 ( -.5,60)(0,0)[c]@xmath227 ( 80,30)(0,0)[c]@xmath228 ( 40,-5)(0,0)[c]@xmath9 ( 25,64)(0,0)[c]@xmath15 ( 55,34)(0,0)[c]@xmath10

we consider the electronic transport in multi - terminal mesoscopic networks of weakly disordered metallic wires . after a brief description of the classical transport , we analyze the weak localisation ( wl ) correction to the four - terminal resistances , which involves an integration of the cooperon over the wires with proper weights . we provide an interpretation of these weights in terms of classical transport properties . we illustrate the formalism on examples and show that weak localisation to four - terminal conductances may become large in some situations . in a second part , we study the correlations of four - terminal resistances and show that integration of diffuson and cooperon inside the network involves the same weights as the wl . the formulae are applied to multiconnected wire geometries . 73.23.-b , 73.20.fz , 72.15.rn

1506.08224

classical laws of transport on electrical networks have been established by gustav kirchhoff in 1845 . they rely on three fundamental hypotheses . two of them are energy and charge conservation leading respectively to the so - called voltage and current laws . the third one is ohm s law which states that the current @xmath0 along a wire @xmath1 of the network is proportional to the voltage drop between the vertices @xmath2 and @xmath3 connected by this wire : @xmath4 @xmath5 being the length of the wire , @xmath6 its cross section and @xmath7 the drude conductivity . we now understand that this third hypothesis relies on the assumption of diffusive motion of the charge carriers at the microscopic level , within a classical description . in particular , it is not appropriate to describe quantum effects ( like aharanov - bohm effect ) or non - diffusive regime ( ballistic regime , quantum hall effect ) . after the fundamental breakthrough proposed by rolf landauer to describe the electrical conductance as a transmission coefficient , a generalization of laws of transport beyond classical transport was highly desirable . it has been formalized in a beautiful work by markus bttiker ( for reviews , see refs . @xcite ) . if we restrict ourselves to the regime of linear transport , a convenient description is to start by introducing the conductance matrix @xmath8 which relates the currents at the contacts ( also called terminals ) of the circuit to the values of the voltage at these terminals @xmath9 this relation is completely general in the limit of linear transport . the two first kirchhoff s laws imply : @xmath10 expressing the conservation of current , and the invariance of the current distribution against a global shift of the potentials ( gauge invariance ) . as a generalization of landauer s formula , elements of the conductance matrix are related to transmission coefficients @xmath11 where the factor @xmath12 stands for the spin degeneracy . at this level , the formalism is completely general and no hypothesis is made on the nature of electrical transport which is totally encoded in the transmission coefficients @xmath13 . these coefficients are related to the scattering matrix which may be determined explicitly within specific models . for example , in the regime of the integer quantum hall effect , the current is carried by edge states which makes the problem effectively one - dimensional and allows for a simple construction of the scattering matrix @xcite . more generally , the scattering matrix may be constructed efficiently by assuming strictly one - dimensional character , like in a network of strictly one - dimensional ( 1d ) wires @xcite . another powerful approach applies to devices in which the electron dynamics can be considered as ergodic , leading to a random matrix formulation of quantum scattering @xcite . in the present article , we consider the case of metallic samples made of weakly disordered wires such that the electron dynamics is _ diffusive _ , as it is the case in narrow metallic wires deposited on a substrate @xcite ( see also the recent experiments @xcite ) , or wires etched in a two - dimensional - electron gas @xcite . is the ratio of the voltage between two contacts @xmath2 and @xmath3 and the current injected at contact @xmath14 and collected at contact @xmath15.__,scaledwidth=35.0% ] bttiker emphasized the importance of the measurement process when a quantum circuit is connected to the outside macroscopic world @xcite . the measured resistance is not only a property of the system itself but also depends on the way it is connected to the outside world . moreover , although the concept of conductance is natural from a theoretical point of view , experiments most frequently deal with voltage measurements : current is injected and collected at two specific contacts and voltages are measured at any pair on contacts playing the role of _ voltage probes _ [ fig:4tr ] ) . therefore , the relevant quantities characterizing the response of the device are the four - terminal resistances defined as @xmath16 by appropriate inversion of relation ( [ eq : idev ] ) , bttiker could relate the four - terminal resistances to the elements of the conductance matrix , therefore to the transmission coefficients @xcite : @xmath17 where @xmath18 is any minor of the dimensionless conductance matrix . the expression is valid when all indices are different . in this paper , we are interested in transport properties of mesoscopic diffusive wires , where quantum interferences lead to small deviations to ohm s law . this is the so - called _ weak - localisation regime_. for classical transport , simple application of ohm s law leads to the expression of the transmission coefficients , which amounts to classical addition of resistances and conductances ( kirchhoff ) . they are expressed in terms of the elements of a matrix which encodes the conductances of all the links of the network , defined below in eq .. for a single wire of length @xmath19 , it is well - known that the weak - localisation correction to the classical transmission coefficient can be written as @xcite @xmath20 where @xmath21 is an average over disorder configurations . the so - called cooperon @xmath22 measures the contribution of interfering closed electronic diffusive trajectories . we have shown that in a network of diffusive wires , this simple relation generalizes to @xcite @xmath23 where @xmath24 labels all the wires of the network . @xmath25 is the localisation length . ] in the infinitely long wire with @xmath26 conducting channels @xcite , @xmath27 the elastic mean free path and @xmath28 a dimensionless parameter of order unity , which will be given below . from the knowledge of the quantum corrections , we will show in this paper that the quantum correction to the classical four - terminal resistance is @xmath29 this expression is quite simple since the weights attached to each wire have a simple interpretation : they express the sensitivity of the classical four - terminal resistance when the resistance of this wire is modified . similarly we have found convenient expressions for the correlation functions of the transmission coefficients , from which one can obtain the correlation functions of the four - terminal conductances . like the weak - localisation correction , these expressions involve contribution of all the wires , which are weighted by similar factors @xmath30 , eqs . ( [ eq:4trcorrel1],[eq:4trcorrel2],[eq:4trcorrel3],[eq:4trcorrel4 ] ) . these equations , which are , with eq . , the main results of the article , will be illustrated by several examples in simple devices . before going specifically to the analysis of quantum transport in networks of quasi one - dimensional weakly disordered wires , we close the section with some general remarks as the concept of four - terminal resistance ( 4tr ) , of which bttiker has been one of the main promotors , has been extremely fruitful in mesoscopic physics . let us mention few directions : * in the early developments of the landauer - bttiker approach , the concept of four - terminal resistance has helped clarifying the question of contact resistance . the concept of contact resistance is the `` mesoscopic version '' of the electric resistance appearing when the electronic fluid is injected from a macroscopic conductor into a small hole ( known as sharvin resistance @xcite ) . the role of contact resistances has been nicely explained in several papers of bttiker @xcite ( see also chapter 5 of @xcite ) . * a fundamental aspect of quantum transport concerns the symmetry of transport coefficients : symmetry with respect to current sources and voltage probes exchange , and symmetry with respect to the reversal of the magnetic field @xmath31 . extending these ideas to coherent conductors , bttiker has demonstrated the symmetry relation for the 4trs @xcite @xmath32 * the concept of 4tr provides an illuminating description of the integer quantum hall effect @xcite from the edge state picture introduced by halperin @xcite , as it allows to compute straightforwardly the longitudinal and hall resistances ( see also bttiker s beautiful review article @xcite ) . furthermore , this framework permits to analyze in simple terms other more subbtle effects , like the scattering between edge states at opposite boundaries due to a constriction , leading to so - called anomalous hall effect @xcite , or by impurities @xcite , the description of transport in a hall cross @xcite , etc . * another issue which has been put forward by bttiker with others , and which will be central in the present article , concerns the _ nonlocal _ nature of quantum transport , and the influence of voltage probes on the transport properties of a coherent conductor . motivated by a set of experiments in multiconnected metallic wires @xcite ( see also whasburn and webb s review @xcite ) , various authors have analyzed the role of voltage probes in devices made of disordered wires by various approaches : maekawa , isawa and ebisawa @xcite , bttiker @xcite , divincenzo , kane and lee @xcite , chandrasekhar , prober and santhanam @xcite , hershfield and ambegaokar @xcite . a more general discussion of nonlocality of weak localisation in networks of metallic wires was made possible within the theory developed by us in ref . @xcite . our formalism has allowed us to study how altshuler - aronov - spivak oscillations of the magnetoconductance are affected by the network geometry @xcite ( this was beautifully demonstrated by earlier experiments in arrays of lithium rings by bishop , dolan and licini @xcite ) , or the role of electronic interactions @xcite . the outline of the paper is as follows : in the next section we introduce a classical description of electronic transport in networks of metallic wires . section [ sec : wl ] is devoted to the analysis of the weak localisation correction , as a warm up exercice preparing the more complicate study of four - terminal resistance correlations presented in section [ sec : fluc ] . the main formulae , eqs . , etc . , are illustrated on simple cases . section [ sec : conclu ] closes the paper with some concluding remarks . a typical network is represented in fig . [ fig : net4 t ] . in this section , we introduce a _ specific notation in order to distinguish between internal vertices and vertices corresponding to reservoirs , by labelling these latter with a prime_. . primed vertices correspond to reservoirs for which @xmath33.__,scaledwidth=40.0% ] before going to the discussion of quantum transport , let us analyze the classical transport in the network . for this purpose we start by solving the poisson equation @xmath34 for the electrostatic potential @xmath35 inside the network . boundary conditions are @xmath36 for @xmath37 . introducing the coordinate @xmath38 measuring the distance along the wire @xmath39 ( from @xmath2 to @xmath3 ) , the potential varies linearly as @xmath40 . the ( classical ) current density is given by fick s law @xmath41 , where @xmath42 is the diffusion constant . the density in excess @xmath43 is related to the potential through the effective ( screened ) coulomb interaction @xmath44 , leading to @xmath45 , where @xmath46 is the drude conductivity and @xmath47 the density of states at fermi energy . the current in the wire @xmath39 is given by , hence current conservation at each `` internal '' vertex @xmath2 , @xmath48 , may be rewritten as @xmath49 , where the matrix is defined as @xmath50 @xmath51 is the adjacency matrix element , equals to @xmath52 if a wire connects the two vertices and @xmath53 otherwise ; thus it constraints the sum in to run over vertices neighbours of @xmath2 . the parameters @xmath54 have been introduced for convenience for the following and describe connection to reservoirs ( @xmath55 for an internal vertex and @xmath56 for a vertex in a reservoir , cf . [ fig : net4 t ] ) . up to a factor @xmath57 , the matrix @xmath58 thus simply gathers all the wire conductances @xmath59 , where @xmath6 is the cross section of the wires . if we split the vector gathering the electrostatic potentials at the vertices into two parts related to internal and external vertices , @xmath60 , we may write @xmath61 , i.e. , the inverse of the matrix is simply the inverse of the block @xmath62 related to internal vertices . all other matrix elements are zero , @xmath63 for any reservoir @xmath64 . ] @xmath65 where the sum runs over the reservoirs . it will be convenient for the following to introduce the notation @xmath66 which represents the so - called diffuson , measured at the two vertices : @xmath67 the diffuson is solution of the diffusion equation @xmath68 with dirichlet boundary conditions at the reservoirs : @xmath69 for all reservoirs @xmath70 ( for details , cf . appendix of ref . using the linearity of the diffuson on the wires and that it vanishes at the reservoirs , we write @xmath71 for @xmath72 ( the reservoir is at @xmath73 ) . hence , we can rewrite the coefficients in as @xmath74 , where @xmath75 is any position in the wire @xmath76 . in the following we will prefer to write the slope of the diffuson on the wire @xmath77 as @xmath78 , where @xmath79 denotes that the argument is taken at a distance @xmath27 of the reservoir @xmath80 , @xmath27 being the elastic mean free path , i.e. the smallest length scale of the problem . the diffuson at a distance @xmath27 naturally appears in the diagrammatic calculation of the classical transport coefficients @xcite . the relation with this reference s notations will be clear by using below this expression for the diffuson s slope . since the vertex @xmath14 may be any point of the network , we may rewrite as @xmath81 the expression is only valid when @xmath38 is at distance larger than @xmath27 from the reservoirs . integrating the current density @xmath82 across the section @xmath6 of the wire @xmath39 , we obtain the current under the form : @xmath83 where we have introduced @xmath84 the quantity @xmath85 is a generalised conductance matrix relating the _ external _ potentials to the _ internal _ currents . they obviously satisfy the symmetry property @xmath86 . although the physical interpretation was not provided in ref . @xcite , the explicit expression of in terms of the matrix @xmath58 was given : @xmath87 if we consider the case of a wire connected to a reservoir , @xmath88 , eq . coincides with the usual relation between currents and voltage in the terminals : @xmath89 the conductance matrix is obviously related to the generalised conductances by @xmath90 the expression of the classical conductance matrix in terms of the matrix @xmath58 may be deduced by setting @xmath88 in : we recover the expression of ref . @xcite @xmath91 we have used @xmath92 , where @xmath12 is the spin degeneracy , @xmath26 the number of conducting channels and @xmath93 involves the volume of the @xmath94-dimensional sphere of unit radius ( thus @xmath95 , @xmath96 and @xmath97 ) . [ [ example ] ] example + + + + + + + as a simple illustration of eq . , we consider the ring of fig . [ fig : ring ] . from the definition , we write the internal part of the matrix @xmath58 ( i.e. the block related to vertices 1 and 2 ) : @xmath98 where @xmath99 . . leads to @xmath100 giving the expected conductance . using , the 4trs can be deduced from the conductance matrix ( see also the discussion in [ app : crft ] ) . for simple enough networks , the determination of the resistances is however more simple than that of the conductance matrix and does not require the knowledge of the latter , as the simple example analyzed just before has shown . weak localisation is a small quantum correction to transport coefficients originating from quantum interferences between time reversed electronic trajectories @xcite . the main interest in this small quantum correction to transport coefficients is that it gives a measure of the phase coherence length @xmath101 , the fundamental characterisitc length scale which sets the boundary between quantum and classical physics . it is worth stressing that there is no intrinsic definition of @xmath101 , which can only be obtained by extracting a characteristic length scale from the analysis of a physical quantity sensitive to quantum interference , such as weak localisation . a precise experimental determination of @xmath101 thus requires a perfect knowledge of the functional form of the transport coefficients as a function of the various length scales , the magnetic field , etc . the weak localisation correction to the conductance matrix elements @xmath102 is given by @xcite @xmath103 where @xmath104 is the cooperon , solution of @xmath105 @xmath106 being the covariant derivative and @xmath101 the phase coherence length . few remarks : * the effect of the magnetic field is twofold : ( _ i _ ) in the presence of loops in the network , the cooperon depends on the magnetic fluxes , which leads to altshuler - aronov - spivak oscillations @xcite ( see also @xcite ) . ( _ ii _ ) the penetration of the magnetic field in a narrow wire of width @xmath107 can be accounted for through the substitution @xmath108 , where the magnetic length is @xmath109\,\phi_0/(\mathcal{b}w)$ ] and @xmath110 the quantum flux @xcite . * the expression is of great generality : it is not only valid for a system made of quasi-1d wires ( network ) but only assume that the contacts have a quasi-1d geometry . in such a more general situation , the derivatives should simply be replaced by gradients @xmath111 . * expression is valid for @xmath112 . the direct diagrammatic calculation of the reflection probability is more difficult and involves a description of the matching between the metallic system and the contacts which goes beyond the derivation of @xcite . however the wl correction to the diagonal conductance matrix elements can always be deduced by using current conservation @xmath113 . * interestingly , we see that the contribution of each wire is weighted by the `` internal conductances '' introduced above : @xmath114 this expression shows that a uniform integration of the cooperon in the network is possible only if the distribution of the classical currents in the wires is uniform . more conveniently , we showed in ref . @xcite that these weights are related to the derivatives of the classical conductance matrix , eq . , or equivalently : @xmath115 where @xmath25 is the localisation length in the infinitely long wire . _ from now , we will drop the prime on the vertices connected to reservoirs , as there will be no possible confusion below_. an interesting consequence of the nature of the weighting factors was pointed out in ref . @xcite : since the weight @xmath116 may change in sign for certain wires , the wl correction to some transmission coefficient may become positive . such an example is shown on fig . [ fig : netposwl ] , which has @xmath117 valid for @xmath118 long wires ( @xmath119 ) . this wl correction may become positive for sufficiently large @xmath118 , as a striking illustration of the nonlocality of quantum transport . of long wires , the wl correction to the transmission @xmath120 may be become positive @xcite.__,scaledwidth=20.0% ] we have demonstrated that a relation similar to holds for the 4trs , eq . [ appendix : proof ] ) . we illustrate this formula by considering the resistances @xmath121 and @xmath122 characterizing the network of fig . [ fig : wire4tr ] . a remarkable consequence of quantum nonlocality is the possibility of large wl correction to the conductance , induced by the presence of long 1d contacts @xcite . to illustrate this idea , we analyze the wl correction to the two resistances @xmath121 and @xmath122 for the network of fig . [ fig : wire4tr ] . _ in the following we will express the resistances in units of @xmath123 . _ at this stage it is useful to recall the well - known result for wl correction to the conductance ( or resistance ) in a two - terminal measurement ( a wire of length @xmath19 between two large contacts ) . in this case the dimensionless resistance is simply @xmath124 and the cooperon satisfying dirichlet boundary condition reads @xmath125 , where @xmath126 and @xmath127 . the application of gives the result of ref . @xcite : @xmath128 \simeq\frac{l_\varphi}{l}-\left(\frac{l_\varphi}{l}\right)^2 \:,\ ] ] where the expansion corresponds to @xmath129 . the dominant term corresponds to the value of the cooperon in bulk ( i.e. inside a wire , at distance @xmath130 from the boundaries ) ; the correction is explained by the depletion of the cooperon which vanishes at the boundaries . in the coherent limit ( @xmath131 ) , one get the universal wl correction @xmath132 . note that may be obtained more directly by using the spectral determinant approach @xcite . we consider the device of fig . [ fig : wire4tr ] and consider first the two - terminal resistance . due to nonlocality , although the two arms @xmath94 and @xmath133 play no role for the classical transport , @xmath134 , they influence the quantum contributions to the transport coefficients . ( [ eq : wl4tr ] ) implies that integration of the cooperon runs over the three wires @xmath135 , @xmath136 and @xmath137 : @xmath138 where @xmath139 . the explicit expression of the integral @xmath140 over a wire in terms of the network properties can be found in ref . below , we only analyze limiting behaviours . [ [ weakly - coherent - limit - l_varphill - l_a - l_b - l_c - l_d - l_f ] ] weakly coherent limit @xmath141 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we get in this case @xmath142 up to exponentially small corrections . if we compare this expression with , we remark that the presence of the two long wire @xmath143 and @xmath144 only affects the subleading term of the wl correction . the difference correction is explained by the depletion of the cooperon at the vertices , which behaves as @xmath145 at distance @xmath38 of the vertex . the correction to the bulk result is @xmath146 . [ [ fully - coherent - limit - l_varphitoinfty ] ] fully coherent limit ( @xmath131 ) + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we only discuss two limiting cases as the general expression is rather cumbersome . when the two wires @xmath143 and @xmath144 are very long , @xmath147 , they play no role and we recover the universal result for the two - terminal wire [ limit @xmath131 of eq . ] @xmath148 . in the opposite limit @xmath149 , the phase coherence is broken at the level of the vertices due to the vicinity of the large contacts . we get the result @xmath150 , which corresponds to the addition of resistances @xmath151 for three independent wires with @xmath152 , etc . varying the length of the wires @xmath143 and @xmath144 hence allows to cross over between the fully quantum regime where the wires @xmath153 should be considered as a whole and the regime where the three wires are independent and their resistances may be added according to the classical kirchhoff law . this is an illustration of the idea introduced by bttiker to describe dephasing in a fully coherent system by introducing fictitious voltage probes @xcite . classical resistance reads @xmath154 which immediately shows that the wl correction is given by an integral of the cooperon over the wire @xmath136 only : @xmath155 is shorter than the phase coherence length and the connecting wires , the wl correction @xmath156 is large.__,scaledwidth=40.0% ] [ [ incoherent - connecting - wires - l_varphill - l_al_cl_dl_f ] ] incoherent connecting wires @xmath157 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + using again the expression of the integral of the cooperon integrated in a wire @xcite , we obtain the explicit expression @xmath158\end{aligned}\ ] ] where @xmath159 . the first term is the result for an isolated wire of length @xmath160 , eq . . the second term originates from the non vanishing value of the cooperon at the two vertices , _ i.e. _ we can interpret this term as coming from the modification of the boundary conditions for the wire @xmath136 induced by the presence of the connecting wires @xmath135 , @xmath137 , @xmath143 and @xmath144 . after a little bit of algebra , we obtain : @xmath161 @xmath162 for a long wire @xmath163 we get the small correction @xmath164 up to exponentially small corrections . the dominant term @xmath165 , coincides with the two - terminal measurement for the wire of length @xmath160 , eq . . as for @xmath166 the presence of the connecting wires manifests itself only through the coefficient of the subleading term @xmath167 , cf . or . the subleading correction is half of the one obtained for @xmath166 , eq . , as only the depletion of the cooperon at the two boundaries of the wire @xmath136 contributes . @xmath162 for a short ( coherent ) wire @xmath168 , quite remarkably , we obtain a _ wl correction to the four - terminal conductance @xmath169 in this case the presence of the connecting wires strongly affects the cooperon inside the wire , which is the reason for the large wl correction . in a wire connected to two large reservoirs , the size of the electronic trajectories contributing to the wl is bounded by the length of the wire , which leads to a saturation of @xmath170 as @xmath131 . on the contrary , in the four - terminal configuration , the electronic trajectories can explore the connecting wires @xmath135 , @xmath137 , @xmath143 and @xmath144 on large scales compared to @xmath160 , which is the physical origin for the large wl . . characterizes the wl correction to the conductance , in a four probe configuration . although the correction to the conductance may be large , we remark that the _ relative _ correction , @xmath171 is always small , as the validity of the perturbative treatment requires @xmath172 . in order to better understand the difference between the two results and , it is instructive to consider the multiterminal network of fig . [ fig : multiterm ] . the wl correction to @xmath122 is given by as well . for the calculation , the key point is that , at a vertex @xmath38 from which issue @xmath173 long wires ( longer than @xmath101 ) , the value of the cooperon is @xmath174 ( cf . @xcite and [ appendix : star ] ) . in the weakly coherent wire limit ( @xmath163 ) , the cooperon inside the wire @xmath136 is @xmath175 , except at a distance @xmath176 from the vertices . integration of the cooperon leads to @xmath177 , similar to eq . . connected to @xmath178 reservoirs by long wires . wl correction @xmath179 is controlled by electronic trajectories starting from wire @xmath136 which may explore the wires over long distances compared to @xmath160.__,scaledwidth=30.0% ] in the coherent limit ( @xmath163 ) , with connecting wires still longer than @xmath101 , the cooperon is almost uniform inside the wire , with a value @xmath180 , where @xmath178 is the effective coordination number . we get in this case the large wl correction @xmath181 , which reduces to for @xmath182 . this argument provides the interpretation of the factor @xmath183 in eq . . [ [ fully - coherent - limit - l_varphitoinfty-1 ] ] fully coherent limit @xmath131 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + coming back to the simple network of fig . [ fig : wire4tr ] , it is also interesting to consider the fully coherent limit . using again the expression of the integral @xmath184 given in @xcite , some algebra leads to @xmath185 where @xmath186 , etc . for short connecting wire @xmath187 , we recover the well known universal result @xmath188 corresponding to a coherent wire between two large contacts , as expected . in the other limit , @xmath189 , we obtain a large correction @xmath190 similar to in which @xmath191 ( i.e. the cutoff limiting the trajectories exploring the connecting wires is not the phase coherence length but the distance @xmath192 to the reservoirs ) . the possibility for large wl correction was pointed out in ref . @xcite in the rather academic situation of an isolated wire . where @xmath19 is the size of the plane . however , the relative correction is small , @xmath193 , where @xmath194 is the fermi wave vector . the same logarithmic behaviour is obtained for large planar networks ( square grids , honeycomb lattice , etc . ) leading to @xmath195 , where @xmath196 is the length of each wire @xcite . ] a more precise discussion was provided by santhanam @xcite for the case we have considered here . as we already mentioned , this has the same origin as the large voltage fluctuations due to long coherent excursions of charge carriers in the voltage probes emphasized by bttiker @xcite ( and also in ref . although the observation of large wl corrections has not been reported so far , to the best of our knowledge , large resistance fluctuations have been observed in several experiments @xcite , with the same physical origin . mesoscopic ( interference ) phenomenon are more pronounced when the system size is reduced down to a size comparable to the phase coherence length @xmath101 : the quantum contribution to the dimensionless conductance @xmath197 of a fully coherent conductor presents fluctuations @xmath198 of the same order than the average @xmath199 ( the wl ) . for this reason , the characterization of conductance fluctuations / correlations has attracted considerable attention both and experimentally @xcite and theoretically @xcite ( for reviews , see @xcite ) . the experiments are usually performed in the four - terminal configuration , which has brought the question of the role of the voltage probes @xcite . devices similar to the wire of fig . [ fig : wire4tr ] were studied in these experiments . the nonlocal nature of quantum transport is particularly striking by considering the symmetric and antisymmetric parts of the resistance @xmath200 which were shown to present different behaviours as a function of the ratio @xmath201 ( symmetrisation is done with respect to exchange of current and voltage probes or , thanks to eq . , to magnetic field reversal ) . whereas @xmath202 is a relatively flat function of @xmath201 , the symmetric resistance presents a clear crossover at @xmath203 , see fig . [ fig : benoit ] ( note that in a weakly disordered metal with a small enough magnetic field , we can ignore the classical magnetoresistance caused by the lorentz force . as a consequence @xmath204 ) . the study of nonlocality of voltage fluctuations and/or transmission probabilities in multiterminal devices was considered theoretically by maekawa _ et al . _ @xcite and bttiker @xcite ( for a three terminal device ) by different approaches . the importance of long range potential correlations was later emphasized by kane , serota and lee @xcite , which has led to reconsider the study of voltage fluctuations in an illuminating paper of kane , lee and divincenzo @xcite , and also in refs . @xcite ( note also the numerical study @xcite ) . finally we point out few theoretical works on the ring configuration @xcite relevant for the experiments aforementioned , and specifically studied in ref . @xcite . in the following we derive formulae analogous to ( [ eq : texiermontambauxprl2004],[eq : wl4tr ] ) for the correlations of transmissions and the correlations of four - terminal resistances @xmath205 we will apply our results to the analysis of the resistances @xmath206 and @xmath202 characterizing the four - terminal wire of fig . [ fig : wire4tr ] . we will show that our formalism allows to recover the results of refs . @xcite straightforwardly . expression of the conductivity correlations in simple geometries can be found at several places @xcite . in networks , the correlations of transmission coefficients @xmath207 ( i.e. conductance matrix elements ) are given by four contributions @xcite : @xmath208 \nonumber\\ & \hspace{0.25 cm } \times { \frac{\partial p_d(\underline{\mu},x')}{\partial x ' } } { \frac{\partial p_d(x',\underline{\nu})}{\partial x ' } } \\ \label{eq : tucf4 } & { \langle { t}_{\alpha\beta}{t}_{\mu\nu } \rangle}^{(4 ) } = \mbox { same as } { \langle { t}_{\alpha\beta}{t}_{\mu\nu } \rangle}^{(3 ) } \\ \nonumber & \hspace{2.5cm}\mbox { with } p_\omega^\mathrm{(d)}\to p_\omega^\mathrm{(c ) } \:,\end{aligned}\ ] ] where we used the same notation as before , @xmath209 , in order to designate the diffuson measured at a distance @xmath27 of the vertex @xmath14 . the function @xmath210 is a normalised function with @xmath211 . ] of width @xmath212 with @xmath213 . several remarks : * as for the wl , note that these expressions are of great generality , and not only valid for networks of quasi-1d wires ; they only assume contacts of quasi-1d nature . for a more general situation , one has to replace the derivatives as @xmath214 . * the diffuson @xmath215 connecting the contacts to the bulk , and providing the weights to attribute to each wire , obeys the classical diffusion equation . the cooperon and diffuson @xmath216 , which describe phase coherent properties , are solutions of the diffusion equation @xmath217 p_\omega^\mathrm{(d , c)}(x , x')=\delta(x - x')\ ] ] where the covariant derivative @xmath218 involves the vector potential @xmath219/2 $ ] for diffuson ( @xmath220 ) and cooperon ( @xmath221 ) . * the penetration of the magnetic field in the wire is taken into account through the substitution @xmath222 , as for the wl ( [ subsec:3.1 ] ) . * these expressions are based on the current conserving expressions for the conductivity correlations , given by the procedure of kane , serota and lee ( i.e. we combine results of refs . @xcite and @xcite ) . * transmission correlations are related to conductivity tensor correlations , @xmath223 . the two first contributions ( [ eq : tucf1],[eq : tucf2 ] ) , which correlate the indices of the two tranmissions are interpreted as diffusion constant correlations @xmath224 , while the two last contributions ( [ eq : tucf3],[eq : tucf4 ] ) , which do not correlate indices , are related to density of states fluctuations @xmath225 @xcite . * thermal smearing can be accounted for by introducing additional integrals @xmath226 and the frequency dependence @xmath227 in ( [ eq : tucf1],[eq : tucf2 ] ) and @xmath228\to{\mathop{\mathrm{re}}\nolimits}\left[p_{\omega}(x , x')p_{\omega}(x',x)\right]$ ] in ( [ eq : tucf3],[eq : tucf4 ] ) , where @xmath229 . the function @xmath210 is a normalised function of width @xmath230 with @xmath213 . this difference makes the two contributions ( [ eq : tucf1],[eq : tucf2 ] ) much larger than ( [ eq : tucf3],[eq : tucf4 ] ) in the high temperature regime , when thermal length @xmath231 becomes smaller than @xmath101 and the size of the system . we will however neglect this effect in the following . * the products of diffusons may be related to derivative of classical transport coefficients , as it was done for the wl . for example @xmath232 when @xmath38 belongs to the wire @xmath24 , etc . we now simplify the above expressions by neglecting the effect of thermal smearing , @xmath233 . thermal effect will be described later in [ subsec : thermalsmearing ] . we deduce the correlations for the 4trs ( cf . [ app : dem4trf ] ) : @xmath234 where @xmath235 is the localisation length for the infinitely long wire . as there will be no possible confusion , we now adopt the simpler notation @xmath236 . note that , as for the transmission correlations , only the contributions @xmath237 and @xmath238 correlate the indices in a non - trivial way . as for the wl , the contributions of each wires are weighted by classical quantities . we apply our formalism to the analysis of the 4tr correlations in a wire connected to several voltage probes , like the one represented in fig . [ fig : wire4tr ] . let us first recall the expressions of the classical resistances , that will be needed to compute the weights in eq . ( [ eq:4trcorrel1],[eq:4trcorrel2],[eq:4trcorrel3],[eq:4trcorrel4 ] ) : @xmath239 where @xmath240 is the localisation length in the infinitely long wire . as they were considered in ref . @xcite , we will study the symmetric and antisymmetic resistances @xmath241 , which exhibit remarkable behaviours . the relations between correlators are : @xmath242 we will first consider the limit of strong magnetic field ( @xmath243 or @xmath160 ) , when the cooperon contributions ( [ eq:4trcorrel2],[eq:4trcorrel4 ] ) are suppressed . the effect of a small field will be discussed in [ subsec : bfieldeffect ] . [ [ fluctuations - langle - deltamathcalr2_1234-rangle ] ] fluctuations @xmath244 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the contribution ( [ eq:4trcorrel1 ] ) is explicitly : @xmath245 ^ 2 \:.\end{aligned}\ ] ] the wire weights are all equal to @xmath246 and imply @xmath247 and @xmath248 , thus this contribution can be splitted in nine terms @xmath249 of four different types ( [ eq : termc11],[eq : termc12],[eq : termc16],[eq : termc18 ] ) . the first term involves a double integral in the central wire @xmath250{benoit1bis } } } \\ \mathcal{c}_{1,1 } & = \frac{4}{\xiloc^4 } \int_{(b)}\d x\int_{(b)}\d x'\ , [ p_d(x , x')]^2 \\\nonumber & \simeq \frac{4}{\xiloc^4}\,l_b\int\d(x - x')\ , [ p_d(x , x')]^2 \simeq \frac { l_\varphi^3l_b } { \xiloc^4 } \ : . \end{aligned}\ ] ] where we have used the expression @xmath251 $ ] valid in the bulk ( i.e. the expression obtained in an infinitely long wire ) , as the presence of the wires @xmath135 , @xmath137 , @xmath143 and @xmath144 only affects the diffuson at distance @xmath176 from the two vertices . then four terms , @xmath252 to @xmath253 , involve integration in neighbouring wires with one coordinate in the wire @xmath136 and the other in a connecting wire , like @xmath254{benoit2bis } } } \:.\ ] ] in the limit @xmath163 , integrals are dominated by @xmath38 and @xmath75 close to the vertex and we can use the expression of the diffuson for the star graph with coordination number @xmath255 given in [ appendix : star ] @xmath256 leading to holds for connecting wires much longer than @xmath101 . ] @xmath257 two terms involve integration in neighbouring long connecting wires , like @xmath258{benoitc16 } } } \:.\ ] ] when @xmath163 , we have @xmath259 . the two last contributions are of the kind @xmath260{benoit3bis}}}\ ] ] and are exponentially suppressed @xmath261 the contribution @xmath262 is simpler to discuss as it involves the weight @xmath263 leading to the constraint @xmath264 . thus we immediately get @xmath265 summing the ten terms , we get the fluctuations @xmath266 . since the first term dominates , we recover the well known expression @xcite for the conductance fluctuations @xmath267 [ [ correlations - langle - deltamathcalr_1234-deltamathcalr_3412-rangle ] ] correlations @xmath268 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the weights of the contribution @xmath269 are @xmath270 which requires once again @xmath271 . therefore @xmath272 . the contribution @xmath273 takes the same form , hence @xmath274 . the correlations are given by summing these two terms @xmath275 , which coincides exactly with the dominant term of the fluctuations . [ [ conclusion ] ] conclusion + + + + + + + + + + + + if we now gather all these results we obtain @xmath276 @xmath277 grows with @xmath160 whereas @xmath278 is independent on the distance between the two voltage probes , in agreement with the experiment ( see figure [ fig : benoit ] ) . we now discuss the case where the central wire is coherent , with long connecting wires ( @xmath279 ) . we can write that the diffuson is almost uniform inside the wire @xmath136 , equal to the diffuson at a vertex of coordination number @xmath280 . thus @xmath281 for @xmath271 . it decays exponentially over the distance @xmath101 in the connecting wires : @xmath282 $ ] for @xmath283 and @xmath284 , etc . [ appendix : star ] ) . as a result we see that the fluctuations are dominated by four terms , @xmath285 corresponding to cases like or , when @xmath38 and @xmath75 are both integrated over the long distance @xmath101 in the connecting wires : @xmath286 , i.e. the four - terminal conductance fluctuations are large @xmath287 the correlations are much smaller as they involve integrations of @xmath38 and @xmath75 in the central wire only @xmath288 , i.e. @xmath289 finally we get @xmath290 fluctuations are independent on the distance @xmath160 between the voltage probes . in the two previous paragraphs , we have reproduced the main conclusions of hershfield @xcite by simpler arguments based on the analysis of the wire weights . although the qualitative change of behaviour between @xmath277 and @xmath278 at @xmath291 agrees with the experiment of benoit _ et al . _ @xcite ( fig . [ fig : benoit ] ) , it was noticed that thermal smearing , not accounted for by hershfield , is important in the experiment . this corresponds to the case where the thermal length @xmath231 is smaller than @xmath101 . there is also a fundamental reason to consider this regime : in the low temperature regime ( @xmath292 ) and in the absence of external degrees of freedom like magnetic impurities , the decoherence is dominated by electronic interactions . as a consequence one has @xmath293 ( see @xcite and references therein ) . in this regime the contribution @xmath294 is negligible compared to @xmath295 . [ [ weakly - coherent - limit - l_tll - l_varphill - l_b ] ] weakly coherent limit @xmath296 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we should repeat the analysis of the previous subsection by adding an integration over the frequency with the thermal function . the first contribution to the resistance fluctuations reads @xmath297 at finite temperature , the diffuson involves exponentials like @xmath298 $ ] with @xmath299 where @xmath300 . when @xmath301 , the diffuson decays rapidly inside the wire . @xmath302 is dominated by the integral in the bulk of the wire , which leads to a similar calculation as for an infinite wire . when @xmath293 ( the case @xmath303 is similar to @xmath304 , and was discussed above ) , the thermal function can be considered as a broad function and simply replaced by @xmath305 . using the expression of the diffuson in bulk we get @xmath306 this term dominates the fluctuations , therefore we recover the known expression @xcite of the conductance fluctuations @xmath307 next contributions are the six terms of type and . we can again argue that integration is dominated by @xmath38 and @xmath75 close to the vertex and use the approximate form given in [ appendix : star ] : @xmath308 where @xmath309 with @xmath299 . some algebra gives @xmath310 correlations are obtained by similar arguments @xmath311 , which coincides with the dominant term of the fluctuations . going back to the quantity of interest , we should replace by @xmath312 whereas still holds , thus @xmath313 we reproduce similar conclusions as for @xmath304 , i.e. fluctuations @xmath314 growing with the distance @xmath160 between the voltage probes , while the fluctuations @xmath315 are independent on @xmath160 . [ [ coherent - limit - l_bll - l_tll - l_varphi ] ] coherent limit @xmath316 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as in the case @xmath317 analyzed previously , the fluctuations are dominated by four terms where @xmath38 and @xmath75 are integrated over two long connecting wires . it is simply given by multiplying by a factor @xmath318 , which accounts for the fact that the diffuson feels an effective coordination number @xmath280 instead of @xmath255 , hence @xmath319 . thus @xmath320 the analysis of the correlations @xmath311 is as follows : we write @xmath321 when @xmath271 , therefore @xmath322 where @xmath299 . the presence of the function @xmath210 is needed in order to cut off the contribution of large frequencies . when @xmath293 , some algebra gives @xmath323 , leading to @xmath324 quite surprisingly , we obtain a logarithmic dependence in @xmath101 reminiscent of the 2d situation , although the system is 1d . the symmetric and antisymmetric resistance fluctuations are @xmath325 finally , we discuss the magnetic field dependence . the above results are valid when the magnetic field @xmath326 is larger than the correlation field , i.e. in a diffusive wire , when @xmath327 where @xmath107 is the width of the wires and @xmath110 the flux quantum . at small magnetic field , cooperon contributions ( [ eq:4trcorrel2],[eq:4trcorrel4 ] ) must be taken into account as well . this can be done easily by noticing that , due to the wire weights , the role of fluctuations and correlations are exchanged for the two first contributions : @xmath328 while @xmath329 the substitution @xmath330 is simply achieved by doing @xmath331 , cf . [ subsec:3.1 ] ( the analysis of the correlations for different fields requires similar substitution in both the diffuson and the cooperon , cf . [ subsec:4.1 ] ) . we deduce straightforwardly the crossover functions . we only consider the weakly coherent regime , @xmath163 . [ [ regime - l_varphill - l_b - l_t ] ] regime @xmath332 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + using the previous calculations we obtain @xmath333^{-3/2 } \right ) \\ { \langle \delta\mathcal{r}^2_{a } \rangle } & \simeq \frac13\left(\frac{l_\varphi}{\xiloc}\right)^4 \left ( 1 - \left [ 1 + ( \mathcal{b}/\mathcal{b}_c)^2 \right]^{-2 } \right ) \:,\end{aligned}\ ] ] where @xmath334 . whereas the symmetric resistance fluctuations are doubled at small field , the antisymmetric resistance vanishes as @xmath335 for @xmath336 . [ [ regime - l_tll - l_varphill - l_b ] ] regime @xmath296 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in this case we obtain @xmath337^{-1/2 } \right ) \\ { \langle \delta\mathcal{r}^2_{a } \rangle } & \simeq \frac1{12}\frac{l_t^2 l_\varphi^2}{\xiloc^4 } \frac{\mathcal{b}^2}{\mathcal{b}_c^2+\mathcal{b}^2 } \:.\end{aligned}\ ] ] the antisymmetric resistance now vanishes at small field as @xmath338 for @xmath336 . we now rediscuss the experiments at the light of our results . in fig . [ fig : skocpol ] , we have reproduced the experimental data obtained by skocpol _ et al . _ @xcite for a silicon inversion - layer narrow wire . we now denote by @xmath19 the distance between the two voltage probes ( denoted @xmath160 above ) . the experimental result exhibits the behaviours obtained above : a growth of the fluctuations with the length , @xmath339 in the incoherent regime @xmath340 , eq . , and a saturation @xmath341 in the coherent regime @xmath342 , eq . . note that accounting for thermal broadening does not change this conclusion , cf . eqs . and . the two samples have lengths @xmath343 ( @xmath344 ) and @xmath345 ( @xmath346 ) . data from ref . @xcite.__,title="fig : " ] # 1#2#3#4#5 @font ( 3026,1370)(6425,-4761 ) ( 7359,-3782)(0,0)[lb ] ( 6696,-3904)(0,0)[lb ] ( 6771,-4330)(0,0)[lb ] ( 7593,-4683)(0,0)[lb ] ( 8464,-4475)(0,0)[lb ] ( 7709,-4480)(0,0)[lb ] ( 6867,-4477)(0,0)[lb ] ( 6696,-3512)(0,0)[lb ] ( 6608,-4334 ) another remarkable experiment is the one of benoit _ et al . _ @xcite , who analyzed voltage fluctuations in sb and au narrow wires . using the data given in this reference , we have plotted the fluctuations of the symmetric and antisymmetric resistances in fig . [ fig : benoit ] . the most striking outcome is the comparison of the different behaviours for the symmetric and antisymmetric resistances . ( 0,0 ) and @xmath347 . length @xmath19 varies from @xmath348 to @xmath349 . @xmath19 is the distance between the voltage probes . blue circles correspond to the symmetric resistance @xmath350 and red diamonds to the antisymmetric resistance @xmath351 . the inset shows the sb sample ; @xmath352 are the fluctuations for a wire of length @xmath101 . data from ref . @xcite.__,title="fig : " ] # 1#2#3#4#5 @font ( 3600,2138)(4201,-5475 ) ( 6018,-5397)(0,0)[lb ] ( 4471,-4646 ) ( 5027,-3848)(0,0)[lb ] ( 5026,-3608)(0,0)[lb ] let us analyze the behaviours more into details . we follow the line of ref . @xcite , where thermal broadening was not taken into account . voltage ( sample to sample ) fluctuations were measured respectively to the fluctuations for a wire of length @xmath101 , which coincides with the relative resistance fluctuations @xmath353 where @xmath352 . in the coherent limit ( @xmath342 ) , eq . gives the value @xmath354 consistent with the experimental data . in the incoherent limit ( @xmath340 ) , eq . gives @xmath355 , whereas leads to @xmath356 . although the linear behaviour with @xmath357 agrees qualitatively with the experimental data , the prefactor obtained experimentally , @xmath358 , is significantly smaller than @xmath359 . the fact that the experimental values are smaller than the theoretical predictions could be explained by the effect of thermal broadening , as it brings a reduction factor @xmath360 . a more precise analysis would nevertheless be needed . another aspect which could explain this quantitative disagreement concerns the determination of the phase coherence length , which was not obtained by a unique independent procedure in ref . @xcite . at the highest temperatures , the phase coherence length was obtained by independent wl measurements , which does not account for the fact that , in heavy metals like gold , when spin - orbit scattering is strong and magnetic impurities present , the phase coherence length involved in the weak localisation @xcite and the conductance fluctuations @xcite differ ( see also @xcite ) . the lack of a formula for the wl correction in the multiterminal wire did not allow the authors to extract @xmath101 for the lowest temperatures . a more reliable procedure would have been to obtain @xmath101 from another long wire ( @xmath130 ) made under the same conditions . since the pioneering work of markus bttiker , four - terminal resistance is now recognized as a quantity of major importance in mesoscopic physics ( ref . @xcite is indeed bttiker s most cited article , with now almost 2000 citations ) . after having briefly reviewed several aspects of four - terminal resistances , we have focused the discussion on the analysis of quantum transport in networks of weakly disordered wires , for which explicit expressions for the quantum corrections to the conductances and the four - terminal resistances were discussed . we have seen that both the weak localisation correction and the correlations involve integrals of the cooperon ( and also of the diffuson for the correlations ) inside the wires , whose contributions must be weighted by derivatives of the classical coefficients : @xmath361 for the conductances and @xmath362 for the four - terminal resistances . although we had discussed this earlier for the wl correction to the conductance @xcite , we have provided here a new interpretation of these coefficients in terms of `` _ _ generalised conductances _ _ '' @xmath363 relating the current in a wire @xmath24 _ inside _ the network to the external potential @xmath364 , precisely : @xmath365 . we have illustrated the efficiency of our formalism by considering simple examples . our determination of the four - terminal resistance correlations only involves simple calculations ( which greatly simplify the analysis of ref . @xcite in particular ) . moreover we have been able to consider the effect of thermal broadening , which was not studied so far . all the main dependences are summarized in tab . [ tab : summary1 ] and tab . [ tab : summary2 ] . .__main behaviours of the four - terminal conductance fluctuations @xmath366 and the conductance correlations @xmath367 of the four - terminal resistances . precise numerical prefactors are given in the text . _ _ [ cols="^,^,^,^",options="header " , ] a natural way to characterize the linear response of a multiterminal structure is to introduce the conductance matrix relating the voltages at the contacts to the currents : @xmath368 . the matrix elements of the conductance matrix are not independent as they must satisfy two types of constraints : ( _ i _ ) current is conserved , thus @xmath369 whatever the choice of external potentials . ( _ ii _ ) a global shift of all voltages @xmath370 does not change the current . as a consequence @xmath371 . in other terms , introducing the vector @xmath372 where @xmath373 denotes transposition , we rewrite the second condition as @xmath374 , i.e. @xmath8 is a linear map acting in the vector space @xmath375 . the current conservation rewrites @xmath376 , meaning that @xmath8 maps @xmath377 onto itself . in the subspace @xmath377 , we may invert the conductance matrix , which leads to introduce the resistance matrix @xmath378 relating the external currents to the voltages @xmath379 . the relation between the two matrices is thus @xmath380 , where @xmath381 is the projector on the vector @xmath382 , where @xmath383 is the number of contacts . we write more conveniently : @xmath384 since we will have to consider differences between potentials at various contacts , the relation @xmath385 will be useful . the four - terminal resistance provides information in the situation where current only flows through contacts @xmath14 and @xmath15 , @xmath386 , with all other currents vanishing , eq . . the voltage at contact @xmath2 is thus conveniently expressed in terms of the resistance matrix as @xmath387 and the voltage difference @xmath388 thus involves @xmath389 a useful relation is obtained by differentiating the relation ( [ convunit ] ) @xcite : @xmath390 as we have discussed in the body of the text , the quantum contributions to the transport coefficients are naturally expressed for the _ conductance _ matrix , as linear response theory provides formulae for the conductivity or the conductance . on the other hand , resistances , which are in general complicated functions of the set of all conductance matrix elements , are more easy to handle in most situations , and usually the relevant quantities in most experiment . we show in this appendix how one can go from the quantum contributions to the conductance matrix elements to the equivalent contributions to the four - terminal resistances . let us consider a general quantity , function of the conductance matrix elements @xmath391 , where @xmath392 is a short notation for a function of all matrix elements @xmath393 ; the quantity @xmath394 may be for example a four - terminal resistance . we now show how the quantum contributions to the quantity @xmath394 can be related to the quantum contributions to the conductance . for this purpose it is convenient to analyze the scaling with the number of channels @xmath26 , which is a large parameter . ( which is simply @xmath395 for a wire of length @xmath19 , with @xmath396 ) , however in a complex network of metallic wires , involving several lengths , there is no unique such parameter . ] considering the set of conductance matrix elements @xmath397 , the disorder average is given by two contributions : a classical term and the weak localisation correction : @xmath398 . other quantum contributions of interest are the mesoscopic ( sample to sample ) fluctuations , defined by @xmath399 and characterized by the correlation functions @xmath400 . as we have recalled in the text , the three quantities present the @xmath26 dependences @xmath401 we now deduce two useful properties . writing @xmath399 , the fluctuation is smaller than the average , @xmath402 , and vanishes on average by definition . as a consequence @xmath403 using this property , we can split the average into a classical part and the wl correction : @xmath404 . hence the average of the physical quantity takes the form @xmath405 we identity the second term as the weak localisation correction to @xmath394 : @xmath406 if we consider two quantities @xmath394 and @xmath31 functions of the transmissions , following the same lines , we deduce the expression for the correlation function @xmath407 @xmath162 * wl correction to the four - terminal resistances :* it is now straightforward to get the weak localisation correction to the four - terminal resistance . we deduce from the property , the average resistance and conductance do not , up to correction of order @xmath408 due to the correlations between transmissions . however the relation is obviously satisfied for a given realization of disorder : @xmath409 . ] @xmath410 . then expanding the conductances as @xmath411 , and using ( [ di4tr ] ) and the property , one get @xmath412 using ( [ eq : texmonprl04 ] ) and once again the equation ( [ di4tr ] ) we finally get . note that this result could have been guessed from the heuristic argument presented in the introduction of ref . @xcite . @xmath162 * correlations of four - terminal resistances :* the expressions for the correlation functions ( [ eq:4trcorrel1],[eq:4trcorrel2],[eq:4trcorrel3],[eq:4trcorrel4 ] ) are demonstrated by making use of the relation with the property ( with the remark closing paragraph [ subsec:4.1 ] ) . the solution of the diffusion equation @xmath413p_c(x , x')=\delta(x - x')$ ] in the star graph with @xmath414 infinite wires is useful ( fig . [ fig : star - graph ] ) . details are given in appendix d of ref . the value of the cooperon at the vertex @xmath14 is inversely proportional to the coordination number of the vertex : @xmath415 . when the two arguments belong to wires @xmath416 and @xmath417 , we get @xmath418 where @xmath419 , @xmath420 and the distance is measured from the vertex . at large distance of the vertex , we recover the result for the infinite wire @xmath421\exp\big[-\sqrt{\gamma}|x - x'|\big]$ ] . altshuler , b. l. , aronov , a. g. , 1985 . electron - electron interaction in disordered conductors . in : efros , a. l. , pollak , m. ( eds . ) , electron - electron interactions in disordered systems . north - holland , p. 1 . ferrier , m. , angers , l. , rowe , a. c. h. , guron , s. , bouchiat , h. , texier , c. , montambaux , g. , mailly , d. , 2004 . direct measurement of the phase coherence length in a gaas / gaalas square network . lett . * 93 * , 246804 . ferrier , m. , rowe , a. c. h. , guron , s. , bouchiat , h. , texier , c. , montambaux , g. , 2008 . geometrical dependence of the decoherence due to electronic interactions in a gaas / gaalas square network . lett . * 100 * , 146802 . schopfer , f. , mallet , f. , mailly , d. , texier , c. , montambaux , g. , saminadayar , l. , buerle , c. , 2007 . dimensional crossover in quantum networks : from mesoscopic to macroscopic physics . lett . * 98 * , 026807 . spivak , b. z. , zyuzin , a. . mesoscopic fluctuations of current density in disordered conductors . in : altshuler , b. l. , lee , p. a. , webb , r. a. ( eds . ) , mesoscopic phenomena in solids . north - holland , pp . 3780 . texier , c. , 2010 . dsordre , localisation et interaction transport quantique dans les rseaux mtalliques . habilitation diriger des recherches , universit paris - sud . http://tel.archives-ouvertes.fr/tel-01091550 . umbach , c. p. , van haesendonck , c. , laibowitz , r. b. , washburn , s. , webb , r. a. , 1986 . direct observation of ensemble averaging of the aharonov - bohm effect in normal metal loops . * 56 * ( 4 ) , 386 . , b. j. , willems , e. m. m. , harmans , c. j. p. m. , beenakker , c. w. j. , van houten , h. , williamson , j. g. , foxon , c. t. , harris , j. , 1989 . anomalous integer quantum hall effect in the ballistic regime with quantum point contacts . 62 * ( 10 ) , 1181 .

we study the evolution of binary stars in globular clusters using a new monte carlo approach combining a population synthesis code ( startrack ) , and a simple treatment of dynamical interactions in the dense cluster core using a new tool for computing 3-body and 4-body interactions ( fewbody ) . we find that the combination of stellar evolution and dynamical interactions ( binary single and binary binary ) leads to a rapid depletion of the binary population in the cluster core . the _ maximum _ binary fraction today in the core of a typical dense cluster like 47 tuc , assuming an initial binary fraction of 100% , is only about 510% . we show that this is in good agreement with recent _ hst _ observations of close binaries in the core of 47 tuc , provided that a realistic distribution of binary periods is used to interpret the results . our findings also have important consequences for the dynamical modeling of globular clusters , suggesting that `` realistic models '' should incorporate much larger initial binary fractions than has usually been done in the past . [ firstpage ] binaries : close binaries : general methods :-body simulations globular clusters : general globular cluster : individual ( ngc 104 , 47 tucanae ) stellar dynamics .

astro-ph0501131

binary stars play a fundamental role in the evolution of globular clusters for at least two important reasons . first , the evolution of stars in binaries , whether in a cluster or in the galactic field , can be very different from the evolution of the same stars in isolation . in a dense environment like a globular cluster , this difference is exacerbated by dynamical encounters , which affect binaries much more than single stars . second , binary stars crucially affect the dynamical evolution of globular clusters , providing ( through inelastic collisions ) the source of energy that supports them against gravothermal collapse @xcite . in the `` binary burning '' phase , a cluster can remain in quasi - thermal equilibrium with nearly constant core density and velocity dispersion for many relaxation times , in a similar way to that in which a star can maintain itself in thermal equilibrium for many kelvin - helmholtz times by burning hydrogen in its core . the binary fraction ( and the initial , primordial binary fraction in particular ) , is therefore one of the most important parameters that determine the evolution of globular clusters . however , most previous dynamical studies of globular clusters even those including binaries have neglected stellar evolution , which can significantly impact the properties and survival of binaries and hence the reservoir of energy they provide . at present , there are very few direct measurements of binary fractions in clusters . however , even early observations showed that binary fractions in globular cluster cores are smaller than in the solar neighborhood ( e.g. , * ? ? ? * ) . recent hubble space telescope ( hst ) observations have provided further constraints on the binary fractions in many globular clusters @xcite . the measured binary fractions in dense cluster cores are found to be _ very small_. as an example , the upper limit on the core binary fraction of ngc 6397 is only 5 - 7% @xcite . on the other hand , in very sparse clusters , like ngc 288 @xcite , but also in some other `` core - collapsed '' clusters , like ngc 6752 @xcite , the upper limit for the binary fraction can be as high as @xmath1 . for the _ initial _ binary fraction in globular clusters , there are of course no direct measurements . however , there are no observational or theoretical arguments suggesting that the formation of binaries and hierarchical multiples in dense stellar systems should be significantly different from other environments like open clusters , the galactic field , or star - forming regions . binary frequencies @xmath2 are found in the solar neighborhood and in open clusters . t tauri stars also have a very high binary fraction @xcite . for the range of separations between 120 and 1800 au , their binary fraction is comparable to that of main sequence stars in the solar neighborhood , while at shorter periods it is higher . furthermore , many stars are formed in systems of multiplicity 3 or higher : in the field their abundance is no less than 40% for inner periods @xmath3 days @xcite . all this suggests that , in dense stellar systems as well , most stars could be formed in binary and multiple configurations . most dynamical interactions in dense cluster cores tend to _ destroy _ binaries ( the possible exception is tidal capture , which may form binaries , but turns out to play a negligible role ; see 5.2 ) . soft binaries ( with orbital speeds lower than the cluster velocity dispersion ) can be disrupted easily by any strong encounter with another passing star or binary . even hard binaries can be destroyed in resonant binary binary encounters , which typically eject two single stars and leave only one binary remaining @xcite , or produce physical stellar collisions and mergers @xcite . in addition , many binary stellar evolution processes lead to disruptions ( e.g. , following a supernova explosion of one of the stars ) or mergers ( e.g. , following a common envelope phase ) . these evolutionary destruction processes can also be enhanced by dynamics . for example , more common envelope systems form as a result of exchange interactions @xcite , and the orbital shrinkage and the development of high eccentricities through hardening encounters may lead to the coalescence of binary components @xcite . it is therefore natural to ask whether the small binary fractions measured in old globular clusters today result from these many destruction processes , and what the _ initial binary fraction _ must have been to explain the current numbers . we address these questions in this paper by performing calculations that combine binary star evolution with a treatment of dynamical interactions in dense cluster cores . in 3 we describe in detail the method we use , following a brief overview of the theoretical background in 2 . we test our simplified dynamical model by comparing it against full monte carlo @xmath0-body simulations in 4.1 . in 4.2 we use semi - analytical estimates to predict the upper limit for the final binary fraction in dense clusters . in 4.3 we estimate the lower limit for the final binary fraction and analyse which mechanisms of binary destruction are most efficient as a function of cluster age . in 5 we present our numerical results for the evolution of the binary fraction in dense cluster cores , and we compare these results with observations . in particular , using our theoretically predicted period distribution , we re - examine observations of 47 tuc and re - derive constraints on the core binary fraction . in the final discussion ( 6 ) , we point out how our results may be helpful in interpreting observations of core binary fractions in other clusters , and we discuss the required initial conditions for simulations of clusters with binaries , as well as which methods are best suited for these simulations . there are several possible ways to approach the study of binary evolution in dense clusters . the traditional approach is to start from full @xmath0-body simulations to study the dynamics of the stellar system and introduce on top of this various simplified treatments of single and binary star evolution . this has been used for many years . unfortunately , even with the fastest special - purpose computers available today , this direct @xmath0-body approach remains extremely expensive computationally , so that previous studies have been limited to small systems like open clusters and with limited coverage of parameter space . in addition , because binaries are particularly expensive to handle computationally ( as they increase enormously the dynamic range of direct @xmath0-body simulations ) , these previous studies have also been performed with unrealistically small numbers of binaries . for example , the time required to perform just one direct @xmath0-body simulation of a cluster containing @xmath4 stars with all stars formed initially in binaries would be at least a year on the grape-6 , with some dependence on the initial binary parameters alternatively , a binary population synthesis code ( e.g. , * ? ? ? * ) , normally used to evolve large numbers of stars and binaries without dynamical interactions , can be extended by introducing a simple treatment of dynamics . in this type of approach it is often assumed that all the relevant parameters of the cluster ( e.g. , central density and velocity dispersion ) remain constant throughout each dynamical simulation , i.e. , the dynamics is assumed to take place in a _ fixed background _ cluster . many previous studies of dense stellar systems have been based on this type of approximation ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? this approach , sometimes called `` encounter rate technique '' @xcite , is computationally much less expensive than direct @xmath0-body simulations and hence allows the systematic exploration of the vast parameter space of initial conditions for clusters and their primordial binary populations . in addition , the use of sufficiently large numbers of stars and binaries makes the simulations more realistic . although obviously much less accurate in its description of the overall cluster dynamics , this method opens the possibility of studying `` star cluster ecology '' in considerably greater detail than has been possible with @xmath0-body simulations . in particular , it makes it possible to study in detail the rare but important evolutionary channels that may play a crucial role in the formation of some of the most interesting tracers of dynamical interactions in dense clusters , such as ultracompact x - ray binaries , millisecond pulsars , and cataclysmic variables @xcite . unfortunately , it is difficult to compare these two approaches , as each is based on a very different set of simplifying assumptions . there are no comprehensive studies of dense stellar systems including a self - consistent treatment of both dynamics and binary star evolution . in many recent @xmath0-body simulations for large clusters , binary stars are treated in the point mass limit and soft binaries are eliminated from the start . binary destruction can then occur only through resonant 4-body interactions . however , @xmath0-body studies of open clusters that incorporate realistic treatments of binary stellar evolution have shown that stellar evolution affects the binaries significantly , and that , even in these low - density environments , the complex interplay between binary evolution and dynamics , even for soft binaries , can play an important role in the overall cluster evolution and in determining the properties of surviving binaries @xcite . the second approach , `` binary population synthesis with dynamics '' , which we have adopted in this work , suffers from the lack of self - consistent dynamical evolution of the cluster , which is assumed to remain in a constant state of thermal equilibrium for its entire evolution . this state , where the energy production through `` binary burning '' in the core is balanced by the outer energy flux into the cluster halo , does indeed provide nearly constant conditions throughout a typical globular cluster s lifetime . the exception might be `` core - collapsed '' clusters , which may have run out of binaries and evolved to a much more centrally concentrated state . typically , the density and `` temperature '' profiles of a cluster do not change much as long as there are enough binaries remaining to provide support against gravothermal contraction . stellar interiors provide a useful analogy : as long as a star keeps enough hydrogen to burn in its core , it can remain in thermal equilibrium on the main sequence and avoid core contraction and envelope expansion . just like main - sequence stars , globular clusters can maintain a nearly constant interior structure for many thermal time - scales ( i.e. , many relaxation times ) as long as they do not run out of `` fuel '' ( binaries ) . this behavior is expected qualitatively ( see , e.g. , * ? ? ? * ) , and has now been demonstrated quantitatively in many different studies using various numerical techniques for cluster simulations . for example , the recent study by @xcite considered the evolution of idealized clusters of equal - mass stars without stellar evolution for a range of initial binary fractions . for the case of an isolated plummer model with 10% initial hard binaries they found ( see their fig . 4 ) that the core radius of this cluster can remain nearly constant ( to within a factor @xmath5 ) for many tens of half - mass relaxation times ( i.e. , more than a hubble time for most galactic globular clusters , where the half - mass relaxation time is @xmath6 yr ) . for a more realistic cluster model with 20% hard binaries initially and tidal truncation , they found that , after about @xmath7 , when the cluster is about to disrupt in the galactic tidal field , the core radius still has not varied by more than a factor @xmath5 over the entire evolution ( see their fig . 11 ) ; and over the first @xmath8 , the core radius changed by less than @xmath9 . the central velocity dispersion also does not vary much with time ( see , e.g. , * ? ? ? * fig . 1 ) . similar results have been obtained from direct @xmath0-body simulations of open clusters , where the central density and velocity dispersion also remain nearly constant in models with significant numbers of binaries @xcite . there are several binary population synthesis codes in use today . only a few of them include , in addition to detailed treatment of mass transfer phases , stellar evolution along with tidal interaction of binary components . the code of @xcite was designed to study low- and intermediate - mass stars leading to the formation of white dwarf systems . the code of ( * ? ? ? * ; * ? ? ? * in preparation ) was originally developed to investigate more massive stars progenitors of neutron star and black hole systems ( calibrated against full binary stellar evolution codes for mass - transfer phases)and was expanded recently to treat carefully binaries with white dwarfs as well . we use the latter code ( called startrack ) to follow the evolution of single and binary stars in our simulations . as mentioned earlier , we also treat all dynamical encounters explicitly by direct integration , using a recently developed numerical toolkit for small-@xmath0 gravitational dynamics that is particularly well suited to performing 3-body and 4-body integrations @xcite . our initial conditions are described by the following parameters : total number of stars @xmath0 ( single or in a binary ) , initial mass function ( imf ) , binary fraction @xmath10 , distribution of binary parameters ( period , @xmath11 , eccentricity , @xmath12 , and mass ratio , @xmath13 ) . we typically adopt standard choices used in previous population synthesis studies , which are based on available observations for stars in the field and in young star clusters ( sills et al . 2003 ) . for most of the calculations reported here , we use the following `` standard '' initial conditions : * we adopt the imf of @xcite , which can be written as a broken power law @xmath14 , where @xmath15 for @xmath16 , @xmath17 for @xmath18 , @xmath19 for @xmath20 . we assume that all stars are formed in a single burst of star formation at @xmath21 in our simulations . * we consider a wide range of stellar masses from @xmath22 to @xmath23 . ] * the binary mass ratio , @xmath24 , is assumed to be distributed uniformly in the range @xmath25 . this is in agreement with observations for @xmath26 @xcite . * the binary period , @xmath11 , is taken from a uniform distribution in @xmath27 over the range @xmath28@xmath29d . * the binary eccentricity , @xmath12 , follows a thermal distribution with probability density @xmath30 . * we reject systems where binary components would overflow their roche lobe at pericenter . the initial average stellar mass is then @xmath31 , and , with the flat mass ratio distribution , the average binary mass is @xmath32 . we evolve all stars ( single and binary ) using the population synthesis code startrack @xcite , which has recently been updated significantly ( * ? ? ? * in preparation ) . this is the only current population synthesis code that incorporates detailed treatments of all possible types of mass transfer ( mt ) episodes : stable mt ( conservative or non - conservative ) , unstable mt ( thermally or dynamically ) , and thermal time - scale mt . also included are the effects of eddington - limited mass accretion and transient behavior of accretion discs ( based on the disc instability model ) . startrack allows us to follow the evolution of binaries with a large range of stellar masses , metallicities , and star formation histories ( constant rate , sudden or exponential bursts , etc . ) . startrack also models in detail the loss of mass and angular momentum through stellar winds ( dependent on metallicity ) and gravitational radiation , asymmetric core collapse events with a realistic spectrum of compact object masses , and the effects of magnetic braking and tidal circularization on close binaries . in our simulations , we adopted the new prescription for magnetic braking given by @xcite . compared to the older prescription closer binaries lose their angular momentum at a slower rate and hence survive as binaries longer . the evolution of single stars in startrack is based on the analytic fits provided by @xcite , but includes a more realistic determination of compact object masses @xcite . we treat the evolution of stellar collision and binary merger products following the general `` rejuvenation '' prescription of @xcite . it ensures that the merger product has the same total amount of hydrogen , helium , and carbon as the two parent stars together . for some stars , the assumptions made in the treatment of the merger depend on the types of stars and the type of merger ( collision vs binary coalescence ) . for example , we assume that there is no accretion on to a neutron star during a physical collision , and that the other star , if it is unevolved , is destroyed completely ( e.g. , we do not consider the possible formation of a thorne - hytkow object ) . we treat as a `` dynamical common envelope '' ( ce ) event the outcome of a physical collision between a compact object and a red giant , applying a standard `` alpha prescription '' ( where we adopt @xmath33 ; see * ? ? ? * ) , but taking into account the initial positive energy . in particular , if this compact object is a neutron star , a compact binary containing a neutron star and a white dwarf is formed . we assume that , during the ce phase , the neutron star will accrete a significant amount of the envelope material and will become a millisecond pulsar @xcite . if the resulting mass of the neutron star exceeds the limit for a neutron star ( taken in our simulations to be @xmath34 ) , we assume that a black hole is formed . to evolve the cluster population of single stars and binaries in our code , we consider two basic time - scales . one is associated with the evolutionary changes in the stellar population , @xmath35 , and the other with the rate of encounters , @xmath36 ( see 3.4 ) . the evolution timestep @xmath35 is computed so that no more than 2% of all stars change their properties ( mass and radius ) by more than 5% . the global timestep for the cluster evolution is taken to be @xmath37 $ ] as we described in 2 , our model for the cluster dynamics is highly simplified . we adopt a simple two - zone , core - halo model for the cluster . we assume that the core number density , @xmath38 , and one - dimensional velocity dispersion , @xmath39 , as well as the half - mass relaxation time , @xmath40 , remain strictly constant throughout the evolution . while dense globular clusters of interest have @xmath41 , the core density can vary by several orders of magnitude . here we set @xmath42 for most calculations , representative of a fairly dense cluster like 47 tuc . in general , @xmath38 is the main `` knob '' that we can turn to increase or decrease the importance of dynamics . setting @xmath43 corresponds to a traditional population synthesis simulation , where all binaries and single stars evolve in isolation after a single initial burst of star formation . to model a specific cluster , we match its core mass density today , @xmath44 , central velocity dispersion , and half - mass relaxation time the escape speed from the cluster core can be estimated from observations as @xmath45 @xcite , where @xmath46 is the three - dimensional central velocity dispersion . following an interaction or a supernova explosion , any object that has acquired a recoil speed exceeding @xmath47 is removed from the simulation . acquiring a large recoil velocity in a dynamical encounter is a very efficient mechanism for ejecting low - mass objects from the cluster . we find generally that recoil to the halo does not play a significant role : the recoil velocity into the halo differs by @xmath48 from the escape velocity from the cluster , and affects only a small number of objects . for computing interactions in the core , the velocities of all objects are assumed to be distributed according to a lowered maxwellian @xcite , with @xmath49,\ ] ] where @xmath50 ( assuming energy equipartition in the core ) and @xmath47 is the escape speed . here @xmath51 is the average mass of an object in the core . in addition , we use @xmath52 to impose a cut - off for soft binaries entering the core : any binary with maximum orbital speed @xmath53 is immediately broken into two single stars @xcite . in the presence of a broad mass spectrum , the cluster core is always dominated by the most massive objects in the cluster , which tend to concentrate there via mass segregation . as stars evolve , the _ composition _ of the core will therefore change significantly over time . mass segregation in globular clusters was investigated recently in @xcite by considering both light and heavy tracers in two - component models . it was found that the characteristic mass - segregation time - scale is given by @xmath54 here @xmath55 is a constant of order unity and @xmath56 is the average mass of an object in the halo , and @xmath57 is the current mass of the object . this equation represents a diffusion process and can be applied to all stars , not just to those more massive than average . indeed , even low - mass objects may ( rarely ) diffuse into the cluster core on a long time - scale , although on average they will tend to drift outwards . however , for very light objects with masses @xmath58 ( which is typically @xmath59 at the beginning and @xmath60 after @xmath61gyr ) , this expression becomes less accurate , although it remains valid qualitatively . for a more recent discussion of mass segregation in the presence of a broad mass spectrum , see @xcite . to model mass segregation in our simulations , we adopted the time - scale given by equation ( [ t_sc ] ) , but treated the process as stochastic : the probability for an object of mass @xmath57 to enter the core after a time @xmath62 is sampled from a poisson distribution , @xmath63 this treatment ensures that all stars heavier than @xmath64 diffusing into the core will have the appropriate mass spectrum and that interactions will occur between objects drawn from the correct distribution . ( [ t_sc ] ) was derived for the restricted case of a two - component cluster without a realistic imf therefore @xmath55 is unknown by a factor of a few . we find in simulations that the final core mass is nearly proportional to @xmath65 . in order to obtain a better fit to observations for the core mass versus total cluster mass relation , we have fine - tuned eq . ( [ t_sc ] ) using data for 47 tuc , in particular the ratio of the core mass to the total mass of the cluster . for this cluster we adopted a core mass of @xmath66 and a total cluster mass of @xmath67 at present @xcite ; we also take @xmath68 yr @xcite and an age of @xmath69gyr . while the core mass can be found directly from our simulations , the total cluster mass has an uncertainty due to the imf cut - off at the low mass end in our standard cluster model . first , we found the total mass of a cluster model evolved to @xmath69gyr and the initial number of very massive stars ( defined as those producing a black hole at the end of their stellar evolution ) . we find that at 11 gyr , the cluster has @xmath70 per black hole ( or per heavy primordial star ) , when the imf extends down to 0.01 @xmath71 . this allows us to normalize our model to the real cluster mass : with this ratio we have an estimate for the total cluster mass when we use a higher cut - off for the imf ( 0.05 @xmath71 ) . this now gives us the ratio of the core mass to the total cluster mass corresponding to our simulations . we find that the best fit for 47 tuc gives @xmath72 . all objects in our simulations are allowed to have dynamical interactions only after they have entered the cluster core . we use a simple monte carlo prescription to decide which pair of objects actually have an interaction during each timestep . the cross section for an encounter between two objects , of masses @xmath73 and @xmath74 , with relative velocity at infinity @xmath75 , is computed as @xmath76 where @xmath77 is the maximum distance of closest approach that defines a significant encounter and @xmath78 is the velocity at pericenter . here the index @xmath79 ( lowercase ) reflects an individual object in the core , while @xmath80 ( uppercase ) denotes a random representative object from the _ subclass _ of objects @xmath80 . in order to more accurately determine encounter rates , at each timestep binaries and single stars in the core are divided into 100 subclasses : 10 by size ( radius for single stars or semimajor axis for binaries ) and 10 by mass . boundaries between mass - subclasses are fixed approximately as @xmath81 . subclasses by size depend on the current sizes of single and binary populations ( separately ) , with the step between subclasses taken as @xmath82 . the encounter rate for a given object @xmath79 and an object from subclass @xmath80 is @xmath83 , where @xmath84 is the number density of objects in subclass @xmath80 , and the cross section and relative velocity are defined for an average object in subclass @xmath80 . the total interaction rate for a given object @xmath79 is the sum of the interaction rates with all relevant subclasses , @xmath85 . the corresponding interaction time is @xmath86 . the actual time for an encounter @xmath87 follows a poisson distribution with mean @xmath88 . in practice , we generate a random number @xmath89 , and assume that the encounter happened if @xmath90 . the timestep is limited so that @xmath91 . we keep track separately of the time - scales @xmath92 for interactions with each subclass @xmath80 , and the corresponding @xmath93 is generated from an independent random number . if an encounter happened , it is assumed to be with an object from the subclass with the smallest @xmath94 . the actual interacting object @xmath95 from that subclass @xmath80 is randomly selected from the list of non - interacted objects in this subclass . in this paper we consider separately binary binary , binary single , and single single interactions . the cross sections and rates are calculated using @xmath96 for single single , @xmath97 for binary single and @xmath98 for binary binary . here @xmath99 is the apocenter separation of the binary ( @xmath100 is the semi - major axis and @xmath101 is the eccentricity ) , @xmath102 is the average stellar radius in the subclass @xmath80 , and @xmath103 is the average apocenter separation in the subclass @xmath80 . a single single interaction with pericenter distance @xmath104 is treated as a physical collision and assumed to lead to a merger or a dynamical ce phase . if @xmath105 , we check whether a binary could be produced via tidal capture using the approach described in @xcite . if a tidal - capture binary is formed , its eccentricity is set to zero and semi - major axis set to @xmath106 , assuming rapid circularization ( @xmath107 yr ) as predicted by the standard model described in @xcite . each dynamical interaction involving a binary is calculated using fewbody , a new numerical toolkit for simulating small-@xmath0 gravitational dynamics that provides automatic calculation termination and classification of outcomes ( for a detailed description see * ? ? ? fewbody numerically integrates the orbits of small-@xmath0 systems , and performs collisions in the sticky - star approximation . fewbody s ability to automatically classify and terminate calculations as soon as the outcome is unambiguous makes it well - suited for carrying out large sets of binary interactions , for which calculations must be as computationally efficient as possible . we have compared our simple dynamical model to recent results from fully self - consistent monte carlo simulations based on hnon s algorithm for solving the fokker - planck equation @xcite . for our test we used the results obtained for an idealized model cluster containing 20% primordial hard binaries ( binding energies in the range @xmath108 @xmath109 , where @xmath109 is the average kinetic energy of an object in the cluster ) for a king model with dimensionless central potential @xmath110 ( * ? ? ? * hereafter f03 ) . in this simulation all stars had equal masses , were treated as point masses ( no physical collisions ) , no stellar evolution was taken into account , and all binary interactions were treated using simple recipes . we used our code to perform a similar simulation : we considered a cluster consisting of equal - mass stars , and we turned off all stellar evolution and physical collisions . to fit the f03 model , we took the core mass as 8.3% of the total cluster mass ( corresponding to a king model with @xmath110 ) , and we took an average star mass of @xmath111 and @xmath112km / s . for a total cluster mass of @xmath113 these conditions imply @xmath114yr , and @xmath115 . we evolved the cluster for @xmath116 . in fig . [ test_sf ] we show our results for the core and halo binary fractions as a function of time , compared with the model of f03 . the agreement with f03 is excellent : the core binary fraction rises very quickly to @xmath117 and then remains close to this value for @xmath118 . in contrast , the halo binary fraction decreases more gradually from 20% to 10% . considering the differences between the two treatments , especially for binary interactions , this agreement is quite remarkable . in the bottom panel of fig . [ test_sf ] we show the evolution of the core radius @xmath119 . the core radius evolves in the same way as in the dynamical simulation of f03 . overall , the core radius does not change much during the entire evolution and its value is consistent with the measured values for observed globular clusters with similar parameters ( e.g. , ngc 3201 or ngc 6254 , in which the total cluster mass , the central number density , and the central velocity dispersion are similar to those in our model ) . one can estimate the final binary fraction @xmath120 in a dense environment by considering several mechanisms of binary destruction : * all soft binaries are usually destroyed during a strong encounter . * some fraction of hard binaries is destroyed through stellar evolution ( mergers or disruptions after supernova explosions ) * when a hard binary has a strong encounter with another hard binary or a single star , it can exchange its less massive component for a more massive star , shrink its orbit , or be destroyed in a collisional merger . it should be noted that in our simplified semi - analytical treatment we neglect the effect of mass segregation , which tends to increase the core binary fraction . ( this issue is discussed in more detail in section [ sec : mainrefmodel ] . ) let us consider a dense environment with number density @xmath121 , @xmath122km / s and with an average mass of @xmath123 . with our choices of initial parameters , and an average mass of @xmath124 , 40% of all primordial binaries initially are soft ( this fraction would be 50% with respect to the average mass of @xmath125 ) . we introduce @xmath126 the hardness of a binary system as @xmath127 where @xmath128 is the binary separation , @xmath129 and @xmath130 are the masses of the binary components , and @xmath131 is the average mass of a single star . binaries that have @xmath132 are termed soft , and those with @xmath133 are termed hard . to find how many hard binaries will be destroyed by stellar evolution alone , we calculated the probability of binary destruction as a function of its initial total mass and orbital period , using the binary population synthesis code ( see fig [ fate_2d ] ) . this simulation was done with @xmath134 binaries distributed initially flat in @xmath135 and @xmath136 ( in order to have better resolution for destruction rates for high mass binaries ) , where @xmath137 is the binary period in days and @xmath138 is the total binary mass in @xmath71 , and was evolved for 14 gyr . the result is striking : most of the very hard binaries , with hardness ratios @xmath139 , are destroyed by stellar evolution . the empty space near the bottom right corner of fig [ fate_2d ] reflects the absence of systems below the minimum period for binaries on the main sequence ( the period at which stars come into contact ) . for binaries with total mass @xmath140 destructions mainly occur during the first @xmath141 years of cluster evolution . binaries with period @xmath142 days are mainly destroyed through sn explosions . binaries with period @xmath3 days are destroyed mainly via mergers at the ms stage . for periods @xmath143@xmath144 days the destructions are associated with common envelope evolution and occur at later times . destructions in binaries of smaller masses are not much different from more massive binaries at small periods , but are not destroyed through sn explosions at large periods . also , the ce event in less massive binaries occurs when the donor is a less evolved giant ( at the first red giant branch ) . one may expect that destruction rates should vary smoothly ; however , binary evolution involves many qualitatively different events . in particular , an interesting fluctuation in destruction rates can be seen at @xmath145 and @xmath146 , where local destruction rates are lower than for nearby binaries . for these and nearby binaries the destruction rates are about the same for mass ratio @xmath147 but different for larger @xmath24 . for binaries of masses close to and smaller than these , most destructions for @xmath148 occur during the ce event between a wd and a giant . this ce phase is the second interaction in the binary and follows the stable mt event with the other donor . when the total binary mass is smaller , the wd mass is @xmath149 , and ce event leads to a merger . for binaries of higher total mass , the second interaction occurs between a he star or a wd and a star at the hertzsprung gap . this mt is unstable and leads to a merger . this , therefore , provides for a small local decrease in destruction rates . with our imf and considering binaries of all masses , the fraction of hard binaries destroyed during their evolution is 20% . among binaries where at least one star is more massive than @xmath123 the destruction fraction is 50% , and for those with total initial mass above @xmath111 , this fraction is closer to 60% . the fraction of hard binaries that is not destroyed but is instead softened by evolution is very small , @xmath150 . in 4.3 we will discuss in more detail how binary destruction rates change with time . therefore , even before any dynamical processes are considered for the hard binary population , we estimate that the final binary fraction can not be higher than about 50% , and for binaries with at least one star more massive than @xmath123 this upper limit becomes 30% . for relatively massive binaries , with total initial mass above @xmath111 , the upper limit for @xmath120 is only 24% . overall , this estimate already shows that ( i ) the expected final binary fraction in a dense star cluster will be low and ( ii ) stellar evolution can not be neglected when estimating binary fractions from dynamical models of dense star clusters . let us now consider the effects of dynamical interactions . the time - scale for a binary to undergo a strong encounter with another single star , the collision time , can be estimated as @xmath151 . assuming that the strong encounter occurs when the distance of closest approach @xmath152 with @xmath153 , we obtain @xmath154 here @xmath155 is the mass of an average single star in @xmath71 , @xmath156 , where @xmath157 is the relative velocity at infinity and @xmath158 , where @xmath159 is the number density . using eq . ( [ eta_def ] ) , eq . ( [ tcoll_pd ] ) can be rewritten in a more convenient form : @xmath160 it can be seen from eq . ( [ coll ] ) that the collision time for a binary with @xmath161 and @xmath162 is only @xmath163yr . overall , with our imf , @xmath164 of all hard binaries have collision times shorter than @xmath165gyr ( see also fig . [ fate_2d ] ) , and 15% have collision times as short as @xmath166gyr . in addition , most of the hard binaries that could have experienced an encounter are binaries with @xmath167 , and therefore they are from the population that is destroyed by stellar evolution to a lesser degree than binaries harder than @xmath168 . while this estimate considered only binary single encounters , binary binary encounters will further enhance the destruction rate . moreover , when the binary fraction is higher than @xmath169 , binary binary encounters dominate over binary single encounters . each binary single encounter with a hard binary can result in a hardening of this binary , an exchange of a companion with a more massive single star , or binary destruction in a physical collision . the probability of a physical collision during an encounter increases strongly as the binary becomes harder @xcite . in addition , a very hard binary can be ejected from the core if its recoil velocity exceeds the escape speed from the cluster . each of these three processes , directly or indirectly , leads to the depletion of binaries ( immediate or delayed ) : acquiring a more massive companion , as well as orbital shrinkage or the eccentricity increase , makes systems more likely be destroyed through stellar evolution . as a conservative lower limit , we assume that half of the interacting hard binaries will be destroyed as a result of an encounter ( immediately or later ) . this is clearly a lower limit , as scattering experiments show that , for hard binaries containing main - sequence stars most encounters will lead to a physical collision @xcite . taking everything into account , we conclude that 64% of all binaries will be destroyed only @xmath170 of primordial binaries can survive to the present . the expected final binary fraction is therefore @xmath171 ( for stars in the entire mass range ) , and it is 13% ( @xmath172 ) for binaries with at least one star more massive than @xmath123 , while for @xmath173 it is only 10% ( @xmath174 ) . we note again that this upper limit for the final binary fraction does not take into account several other mechanisms leading to binary destruction , such as binary binary encounters , which are more likely to cause binary destruction than binary single encounters . @xmath141 yr , which is due to supernovae.,width=317 ] , but now for the same binary population evolved in a high - density environment with @xmath175 binaries per @xmath176 . here disruptions through binary binary ( bb ) and binary single ( bs ) interactions dominate for the first @xmath177 gyr.,width=317 ] in order to verify our encounter rates , we considered the evolution of a binary population that is completely immersed into a dense environment with @xmath178 for its entire evolution . we adopted a velocity dispersion @xmath179 km / s and evolved the system for 14 gyr . this is an even simpler model of a star cluster , where all objects are effectively inside the `` core '' at all times and the effects of mass segregation and diffusion between a core and a halo are completely ignored . it allows us to study more clearly the various types of dynamical interactions as separate processes . we ran a sequence of simulations , with ( i ) only binary single encounters , with no physical collisions allowed during an encounter ; ( ii ) binary single encounters with physical collisions ; ( iii ) both binary single and binary binary encounters but still without physical collisions ; ( iv ) binary single and binary binary encounters with physical collisions ; and ( v ) all types of encounters allowed , including single single collisions . in this last case we also eliminated from the system any object that acquired a recoil speed exceeding the escape velocity @xmath180 . in [ sec : semianalytic ] we estimated that , with only binary single encounters taken into account , the final binary fraction should not exceed @xmath181 . this assumed that all objects in the core had a mass equal to the average mass @xmath123 . with the complete imf , a single star participating in the encounter can have mass higher than @xmath123 . also , a fraction of initially hard binaries can become soft during the evolution , e.g. , because of mass loss . as a result , we obtain even lower remaining binary fractions : @xmath182 for case ( i ) and @xmath183 for case ( ii ) . these numbers are in agreement with the upper limits we estimated in [ sec : semianalytic ] . when binary binary encounters are taken into account , the result is even more dramatic : we obtain @xmath184 and @xmath185 for cases ( iii ) and ( iv ) , respectively . as expected , we see that binary binary encounters further enhance the binary destruction rate . when all dynamical processes are taken into account , in case ( v ) , we obtain a final binary fraction @xmath186 . while our semi - analytical estimate of the binary fraction provided an upper limit , the result obtained in this section is clearly a lower limit : in a real cluster , not all binaries will be exposed to the high interaction rates of the core at all times . however , since more massive objects are more likely to diffuse into the core , and since their destruction rate can be higher than for objects of average mass , this lower limit is only approximate , but we may expect it to be much closer to the real value than our previous , conservative upper limit . let us now consider in detail which mechanisms of binary destruction are most important . in fig . [ dr_field ] we show the rates of binary destruction ( events per binary per gyr ) for a field population ( with no interactions ) . except for the interval between @xmath187 and @xmath188 years , when black holes and neutron stars are formed , the binary destructions are mainly coming from mergers . the power - law behavior observed for the overall destruction rate at late times is on the one hand imposed by the rate of orbital decay driven by magnetic braking , and on the other hand depends on the evolutionary time - scale for the companion to become a red giant . in fig . [ dr_core ] we show the binary destruction rates again , but for the dense environment . during the first few gyr the main destruction mechanism is the dynamical disruption of soft binaries . the rate of disruption through stellar evolution is smaller than in the field population : some massive binaries are destroyed by dynamical encounters before they evolve to a supernova explosion . however , the rate of evolutionary mergers is about the same as for the field for the first @xmath189 years . this is consistent with our estimate for the collision time of hard binaries : at this time , hard binaries have not yet been hardened significantly by dynamical encounters . after @xmath190yr the rate of binary evolutionary mergers is increased compared to the field population , as the dynamical hardening of hard binaries has now started . the rates for binary destruction through mergers and physical collisions are very similar : in most cases , if a binary is tight enough for an evolutionary merger , the most likely outcome of a dynamical interaction is a physical collision . at about 5 gyr , the rate of binary destruction through physical collisions starts dominating over dynamical disruptions : at this stage , almost no soft binaries are left , and the hard binaries are more likely to lead to physical collisions during an encounter . consider a binary with @xmath191 at this stage . from equation ( [ coll ] ) we can find that a collision time of @xmath192gyr corresponds typically to binary components of @xmath193 and @xmath22 ( assuming an average mass ratio of 0.5 and average field mass as @xmath123 ) . this binary is clearly at the lowest - mass end of our imf , and therefore by this time almost all soft binaries in the simulation have been destroyed . on the other hand , consider some more massive binary , which has evolved through about half of its main - sequence lifetime , and has component masses of @xmath194 and @xmath195 ( and still the same collision time of 5 gyr ) . the hardness of such a binary is @xmath196 , corresponding to a binary separation @xmath197 . for such a tight binary , the probability of a physical collision in an encounter is almost 100% . the total rate of binary destructions through physical collisions become comparable to that of dynamical disruptions , as can be seen from fig . [ dr_core ] . .initial conditions for all models [ cols="<,<,<,<,<,<,<",options="header " , ] with the same notations as before , the columns give : the name of the cluster ; the cluster age in gyr ; the central mass density in @xmath198 in simulations at the given age , the core binary fraction , the binary fraction for non - degenerate objects more massive than @xmath123 , and the binary fraction among white dwarfs . ; the bottom plot shows the period distribution of binaries with at least one white - dwarf component ( the shaded area shows the fraction of dynamically formed binaries ) . as before n@xmath199 is the total number of binaries and n@xmath200 is the number of binaries in each period bin.,width=317 ] we performed several simulations with parameters that attempt to match those of specific globular clusters in the galaxy ( tables [ table1 ] and [ table3 ] ) . for these models , the total core masses we compute at present match well those of king models fitted to the observed cluster parameters , and the corresponding core radii of our models ( from total core mass and density ) are all @xmath201pc , matching the observed values . in all cases the initial binary fraction is assumed to be 100% , so our results for final core binary fractions represent upper limits . as in all reference models , we predict low values for @xmath120 in all globular clusters , with smaller values in higher - density cores . we compare our derived binary fraction for m4 with the results from @xcite . using a monte carlo modelling of the binary population , they found that the best fit to the observed binaries ( in the period range from 2 days to 3 years ) is @xmath202 . this is in good agreement with our overall predicted @xmath203 , and also with our prediction for heavier main - sequence binaries , @xmath204 . in the 47 tuc model ( for which we increased the total number of stars initially to a more realistic @xmath205 ) , @xmath120 is only 8% at an age of 11 gyr and about 7% at 14 gyr ( we provide this value at different ages given the uncertainty in observationally determined values for the ages of 47 tuc ; see , e.g. , schiavon et al . 2002 ; zoccali at al . 2001 ; also note that @xmath120 does not change much over the last several gyr ) . at first glance , this may seem to conflict with observations . in particular , @xcite derive an overall binary fraction for the core of 47 tuc of about 13% , from observations of eclipsing binaries with periods in the range @xmath206@xmath207d . this estimate was based on an extrapolation assuming a period distribution flat in @xmath208 from about @xmath209d to @xmath210yr ( soft primordial binaries with @xmath211yr are assumed destroyed and short - period primordial binaries with @xmath212d are assumed to evolve toward much shorter periods through angular momentum loss by magnetic braking ) . in figure 2 we show the final period distribution of core binaries in our simulation . note that the period range of eclipsing binaries is near the peak of the distribution , while for longer periods the number of binaries drops rapidly . in particular , if we concentrate on the binaries consistent with the observed set , with primary masses @xmath213 and @xmath214 , we find that the number of systems with periods in the range @xmath207d to @xmath210yr is about 6 times smaller than would be predicted by adopting a distribution flat in @xmath208 . for the whole period range from @xmath209d to @xmath210yr we have 2.2 times fewer binary systems compared to the flat distribution . if we take into account this depletion of wider binaries when modelling the number of observed eclipsing binaries in 47 tuc , we are led to revise the observed core binary fraction from @xcite to @xmath215 , which is much closer to our theoretical prediction . we performed three additional simulations for 47 tuc , with @xmath216 , 0.5 , and 0.75 . the corresponding core binary fraction extrapolated from observations ( corrected as above ) does not vary much among these different models . its maximum value is obtained for @xmath217 , which gives a core binary fraction of about 8% , and in this case the total number of binary systems is 1.6 times less than with an assumed flat distribution . an alternative estimate of the binary fraction in the core of 47 tuc is based on observations of by dra stars @xcite . their estimated core binary fraction , which can be considered a _ lower limit _ , is approximately 0.8% , 18 times lower than the estimate based on eclipsing binaries . this estimate was based on 31 by dra binaries ( observed in the period range 0.410 d ) and 5 eclipsing binaries ( period range 416 d ) . we analysed the core binary population in our model in order to identify by dra systems and eclipsing binaries , and considering the ratio of the two . we adopted the standard definition for a by dra binary : primary mass in the range @xmath218@xmath219 ( see , e.g. , * ? ? ? * ) and period in the range @xmath220@xmath221d ( as for the observed sample in 47 tuc ) . for eclipsing binaries we adopted the observed period range 416 d with each star @xmath222 . the ratio between the number of by dra systems and the number of eclipsing binaries is found to be 5.9 , 6.7 , 7.2 and 3.5 for models with @xmath223 and 0.25 , respectively . therefore a large initial binary fraction @xmath224 is most likely . of the quantities that we explicitly assume in our cluster model to be constant , the central ( core ) density is the one with the largest dynamic range in models that provide for dynamical evolution . hence it is also the quantity that is most likely to significantly affect our final results . to test the sensitivity of our results to a changing central density , we have run our model of 47 tuc with a central density assumed to increase by a factor of 10 from @xmath21 to the present . specifically , we still match the currently observed core density while ramping up the value either exponentially or linearly in time , starting with a value 10 times smaller than at present . this could represent qualitatively the gradual core contraction observed in some @xmath0-body models of star clusters with binaries ( see * ? ? ? * and , for a steeper core contraction , see giersz & spurzem 2003 ) . these models predict a present - day core binary fraction that is only slightly higher , by about 12% . to increase the binary fraction more significantly , the time - averaged core density in the cluster would have had to be many orders of magnitude lower than what is observed today . there is no reasonable dynamical history that would produce such an unusual result . in contrast , recent @xmath0-body simulations show that the presence of just 10% hard primordial binaries leads to core radius _ expansion _ and therefore the core density might be higher in the past ( wilkinson et al . though we did not run another 47 tuc model with a central density decreasing with time , we can predict that the present - day binary fraction would then be smaller by about 12% . we have considered in detail processes of binary destruction ( and formation ) in dense stellar systems . in particular , we have shown that the present binary fraction in cluster cores should be relatively small ( @xmath225 ) . this is caused not only by dynamical encounters , but also binary stellar evolution , which is the dominant mechanism for the destruction of hard binaries . we also find that the destruction rate due to stellar evolution is enhanced significantly by dynamical hardening of binaries . we have shown that values of binary fractions for stars in different mass ranges and at different evolutionary stages can differ significantly . the fraction of dynamically formed binaries is higher when one considers stars at more advanced evolutionary stages . the binary period distribution evolves from flat in @xmath208 for loose clusters toward a sharply peaked distribution in denser clusters , even if all clusters have identical velocity dispersions and therefore identical hard - soft binary boundaries . this implies that a flat period distribution should not be assumed when deriving overall binary fractions by extrapolation from the distribution of observed binaries in a narrow period range ( e.g. , eclipsing binaries ) . we considered several models that attempted to match observed globular clusters . for those with good available data on binaries ( m4 and 47 tuc ) , we found our predicted binary fractions to be in good agreement with observations once we take into account the correct binary period distribution . the main conclusion we derive from these calculations and our semi - analytic estimates is that the currently observed binary fractions in cluster cores suggest very high initial ( primordial ) binary fractions close to 100% . in addition to their implications for the interpretation of observed binary fractions in cluster cores , our results have important consequences for the theoretical modeling of globular clusters using @xmath0-body simulations . indeed , it is clear that `` realistic '' dynamical simulations of globular cluster evolution should include large populations of primordial binaries , with initial binary fractions in the range @xmath226@xmath227 ( similar to what is usually assumed for the field ; see , e.g. , * ? ? ? this poses a particular challenge for direct @xmath0-body simulations , where the treatment of even relatively small numbers of binaries can add enormous computational costs . for this reason , current direct @xmath0-body simulations of star clusters with large initial binary fractions typically have @xmath0 too small to be considered representative of globular clusters ( see , e.g. , portegies zwart et al . 2003 ; wilkinson et al . other methods , such as monte carlo simulations , do not suffer from the same limitations , and routinely simulate clusters with reasonably large @xmath0 ( @xmath228 ) and binary fractions ( @xmath229 ) , but have not yet included advanced treatments of binary star evolution ( see , e.g. , * ? ? ? * and references therein ) . we thank d. heggie and j. hurley for very helpful discussions . this work was supported by nsf grant ast-0206276 , nasa atp grant nag5 - 12044 , and a chandra theory grant . 66 natexlab#1#1 aarseth , s. j. 2001 , new astronomy , 6 , 277 aarseth , s. j. & heggie , d. c. 1993 , asp conf . ser . 48 : the globular cluster - galaxy connection , 701 , m. d. , gilliland , r. l. , brown , t. m. , edmonds , p. d. , guhathakurta , p. , & sarajedini , a. 2001 , apj , 559 , 1060 , d. , sigurdsson , s. , & davies , m. b. 1996 , mnras , 281 , 830 , k. , kalogera , v. , & bulik , t. 2002 , apj , 572 , 407 , k. , kalogera , v. , rasio , f . , & taam , r. 2005 , in prep bellazzini , m. , fusi pecci , f. , messineo , m. , monaco , l. , & rood , r. t. 2002a , aj , 123 , 1509 , m. , fusi pecci , f. , montegriffo , p. , messineo , m. , monaco , l. , & rood , r. t. 2002b , astron . j. , 123 , 2541 benacquista , m. 2002 , living reviews in relativity , 5 , 2 bethe , h. a. & brown , g. e. 1998 , apj , 506 , 780 , j. , & tremaine , s. 1987 , galactic dynamics ( princeton , nj , princeton university press , 1987 , 747 p. ) , b. w. , & fekel , f. 1977 , astron . j. , 82 , 490 brandner , w. , alcala , j. m. , kunkel , m. , moneti , a. , & zinnecker , h. 1996 , a&a , 307 , 121 chernoff , d. f. & huang , x. 1996 , iau symp . 174 : dynamical evolution of star clusters : confrontation of theory and observations , 174 , 263 , a. m. , & bolton , a. s. 2002 , in asp conf . ser . 263 : stellar collisions , mergers and their consequences , 163 cote , p. & fischer , p. 1996 , astron . j. , 112 , 565 , p. , pryor , c. , mcclure , r. d. , fletcher , j. m. , & hesser , j. e. 1996 , astron . j. , 112 , 574 davies , m. b. 1995 , mnras , 276 , 887 davies , m. b. & benz , w. 1995 , mnras , 276 , 876 davies , m. b. 1997 , mnras , 288 , 117 di stefano , r. & rappaport , s. 1994 , apj , 437 , 733 , p. , meylan , g. , & mayor , m. 1997 , a&a , 324 , 505 , a. , & mayor , m. 1991 , a&a , 248 , 485 freire , p. c. , camilo , f. , lorimer , d. r. , lyne , a. g. , manchester , r. n. , & damico , n. 2001 , mnras , 326 , 901 fregeau , j. m. , cheung , p. , portegies zwart , s. f. , & rasio , f. a. 2004 , mnars , 352 , 1 , j. m. , g " urkan , m. a. , joshi , k. j. , & rasio , f. a. 2003 , apj , 593 , 772 , j. m. , joshi , k. j. , portegies zwart , s. f. , & rasio , f. a. 2002 , apj , 570 , 171 , c. l. , & kalogera , v. 2001 , apj , 554 , 548 , b. , goodman , j. , cohn , h. , & murphy , b. 1991 , apj , 370 , 567 giersz , m. & spurzem , r. 2003 , mnras , 343 , 781 , j. , & hut , p. 1989 , nat , 339 , 40 goldberg , d. , mazeh , t. , & latham , d. w. 2003 , apj , 591 , 397 gratton , r. g. , bragaglia , a. , carretta , e. , clementini , g. , desidera , s. , grundahl , f. , & lucatello , s. 2003 , a&a , 408 , 529 g " urkan , m. a. , freitag , m. , & rasio , f. a. 2004 , apj , 604 , 632 halbwachs , j. l. , mayor , m. , udry , s. , & arenou , f. 2003 , a&a , 397 , 159 , w. e. 1996 , astron . j. , 112 , 1487 , d. c. 1974 , celestial mechanics , 10 , 217 hills , j. g. 1984 , astron . j. , 89 , 1811 , j. g. 1990 , astron . j. , 99 , 979 hurley , j. r. & shara , m. m. 2003 , apj , 589 , 179 , j. r. , pols , o. r. , & tout , c. a. 2000 , mnras , 315 , 543 , j. r. , tout , c. a. , & pols , o. r. 2002 , mnras , 329 , 897 hut , p. , mcmillan , s. , & romani , r. w. 1992 , apj , 389 , 527 iben , i. j. & livio , m. 1993 , pasp , 105 , 1373 ivanova , n. & rasio , f. 2004a , to appear in `` binary radio pulsars , '' asp conf . series , ed . rasio & i.h . stairs , astro - ph/0405382 ivanova , n. & rasio , f. 2004b , revista mexicana de astronomia y astrofisica conference series , 20 , 67 ivanova , n. & taam , r. e. 2003 , apj , 599 , 516 joshi , k. j. , rasio , f. a. , & portegies zwart , s. 2000 , apj , 540 , 969 joshi , k. j. , nave , c. p. , & rasio , f. a. 2001 , apj , 550 , 691 , i. r. 1965 , astron . j. , 70 , 376 k " ohler , r. , leinert , c. , & zinnecker , h. 2001 , astronomische gesellschaft meeting abstracts , 18 , 24 , p. 2002 , science , 295 , 82 mcmillan , s. l. w. , mcdermott , p. n. , & taam , r. e. 1987 , apj , 318 , 261 melo , c. h. f. 2003 , a&a , 410 , 269 , s. 1983 , mnras , 203 , 1107 , s. 1985 , mnras , 215 , 171 portegies zwart , s.f . , mcmillan , s.l.w . , hut , p. , & makino , j. 2001 , mnras , 321 , 199 portegies zwart , s. f. , hut , p. , mcmillan , s. l. w. , & verbunt , f. 1997a , a&a , 328 , 130 portegies zwart , s. f. , hut , p. , mcmillan , s. l. w. , & verbunt , f. 1997b , a&a , 328 , 143 portegies zwart , s. f. , hut , p. , mcmillan , s. l. w. , & makino , j. 2004 , mnras , 351 , 473 , s. f. , & meinen , a. t. 1993 , a&a , 280 , 174 , c. , & meylan , g. 1993 , in asp conf . ser . 50 : structure and dynamics of globular clusters , 357 rasio , f. a. , pfahl , e. d. , & rappaport , s. 2000 , apjl , 532 , l47 , e. p. , & bailyn , c. d. 1997 , apj , 474 , 701 schiavon , r. p. , faber , s. m. , rose , j. a. , & castilho , b. v. 2002 , apj , 580 , 873 sigurdsson , s. & phinney , e. s. 1995 , apjs , 99 , 609 , a. et al . 2003 , new astronomy , 8 , 605 shara , m. m. & hurley , j. r. 2002 , apj , 571 , 830 smith , k. w. & bonnell , i. a. 2001 , mnras , 322 , l1 tokovinin , a. a. 1997 , astronomy letters , 23 , 727 verbunt , f. & zwaan , c. 1981 , a&a , 100 , l7 , r. f. 1985 , in iau symp . 113 : dynamics of star clusters , 541577 , m. i. et al . 2003 , mnras , 343 , 1025 , j. , leinert , c. , & k " ohler , r. 2001 , a&a , 376 , 982 zoccali , m. , et al . 2001 , apj , 553 , 733 consider the rate of three - body binary formation ( via the close approach of three single stars ) in a dense cluster core . we denote by @xmath230 the rate per star of the formation of a binary with binding energy @xmath231 . first , we consider @xmath232 the rate ( per star ) at which three objects come together within a region of size @xmath233 . the probability that a third object will be in the vicinity @xmath234 of two other interacting objects is the product of the probability of the first two bodies meeting and the probability that during this time a third object will be in the same vicinity . the rate of two - body encounters for masses @xmath129 and @xmath130 , with number density @xmath235 is the second object spends in the vicinity @xmath234 of the first object a time @xmath241 . the probability that a third object will be within the same vicinity is then @xmath242 here @xmath243 is the relative velocity of the third object with respect to the centre of mass of the first two . effectively , it reflects velocity at which the population at the vicinity @xmath234 is increasing . we do not consider here the population decrease , assuming that if the object was in the neighborhood , a three - body encounter has happened . the final result then takes the form given in ( @xcite , 8) , but with an additional mass- and hardness - dependent factor @xmath244 , @xmath245 @xmath246 it is clear that the rate is highly dependent on the masses of the participating stars and decreases steeply with increasing hardness of the binary that is formed . using the equation ( [ g3 ] ) , the time - scale for three - body binary formation can also be written as @xmath247 \ \left ( { \frac { 10 ^ 5 } { n_c}}\right ) ^2 \left ( { \frac { m_\odot } { < m>}}\right ) ^{5 } \left ( { \frac { \sigma } { 10\ { \rm km / s}}}\right ) ^{9 } \\ & { \frac { n_c } { n_2}}{\frac { n_c } { n_3 } } \left ( { \frac { < m > } { m_1}}\right ) ^{5 } \left ( { \frac { < m > } { m_2}}\right ) ^{5 } { \frac { \sigma } { v_{12 } } } \\ & \eta^5 ( 1 + 2\eta)^{-1 } \left ( 1 + { \frac { v_3 } { \sigma } } \eta^{-0.5 } \sqrt{2 \frac { m_1 m_2 } { ( m_1+m_2)<m > } } \right)^{-1 } \end{array}\ ] ] even in the core of a large , very dense cluster , with density @xmath248 and containing @xmath249 stars , no binaries with @xmath133 will be formed from @xmath250 stars in a hubble time . it has also been shown that many three - body binary formation events will lead to physical collisions for stars as small as white dwarfs @xcite . we therefore neglect all three - body interactions in our cluster simulations .

i review many - body effects on the resistivity of a multiorbital system beyond landau s fermi - liquid ( fl ) theory . landau s fl theory succeeds in describing electronic properties of some correlated electron systems at low temperatures . however , the behaviors deviating from the temperature dependence in the fl , non - fl - like behaviors , emerge near a magnetic quantum - critical point . these indicate the importance of many - body effects beyond landau s fl theory . those effects in multiorbital systems have been little understood , although their understanding is important to deduce ubiquitous properties of correlated electron systems and characteristic properties of multiorbital systems . to improve this situation , i formulate the resistivity of a multiorbital hubbard model using the extended liashberg theory and adopt this method to the inplane resistivity of quasi - two - dimensional paramagnetic ruthenates in combination with the fluctuation - exchange approximation including the current vertex corrections arising from the self - energy and maki - thompson term . the results away from and near the antiferromagnetic quantum - critical point reproduce the temperature dependence observed in srruo and srrutio , respectively . i highlight the importance of not only the momentum and the temperature dependence of the damping of a quasiparticle but also its orbital dependence in discussing the resistivity of correlated electron systems .

1503.06937

landau s fermi - liquid ( fl ) theory@xcite can describe electronic properties of some correlated electron systems at low temperatures@xcite . in this theory , low - energy excitations are described by quasiparticles ( qps ) , i.e. electrons in a self - consistent field of surrounding electrons due to electron correlation . also , the interactions between qps , described by the landau parameters , are independent of temperature . due to these two properties , the temperature dependences of physical quantities in low-@xmath4 region are governed by the temperature dependence of the fermi distribution function , and the corrections due to electron correlation , the mass enhancement and the fl correction , are independent of temperature . as a result , thermodynamic or magnetic or transport quantities show the same temperature dependences as those of the free electron system in low-@xmath4 region , and the difference is the coefficient due to the mass enhancement or the fl correction or both . for example , the specific heat is proportional to @xmath4 , and the coefficient is renormalized by the mass enhancement ; the spin susceptibility is independent of temperature , and the coefficient is renormalized by the mass enhancement and the fl correction . the original landau phenomenological theory@xcite can be justified by microscopic perturbation theory with several imposed conditions@xcite . one of the basic assumptions , the one - to - one correspondence , becomes valid if the qp damping is much smaller than temperature considered ( i.e. , the qp lifetime , the inverse of the qp damping , is very long ) . actually , the single - particle spectral function near the fermi level becomes delta - function - type in the coherent limit [ i.e. , @xmath5 for all fermi momenta ] , @xmath6^{2}+\gamma_{\alpha}^{\ast}(\boldk)^{2 } } \rightarrow z_{\alpha}(\boldk)\delta(\epsilon-\xi_{\alpha}^{\ast}(\boldk)).\label{eq : akw - fl}\end{aligned}\ ] ] here we consider only the coherent part , @xmath7 , where the self - energy is approximated as @xmath8 ; @xmath9 is @xmath10 , @xmath11 is the qp band index , @xmath12 is the qp energy , which is of the order of @xmath4 , @xmath13^{-1}$ ] is the inverse of the mass enhancement factor , and @xmath14 is the qp damping . since the delta - function - type spectral function is obtained for an exact eigenstate , the fl becomes an approximate eigenstate for momenta near the fermi level if the qp dampings at these momenta are much smaller than @xmath4 . in the fl theory the qp damping becomes small rapidly compared with decreasing temperature as a result of its @xmath15 dependence@xcite . in addition , the other basic assumption about the landau parameters becomes valid if the reducible four - point vertex functions and mass enhancement factor are independent of temperature@xcite . note , first , that the landau parameter is proportional to the product of the reducible four - point vertex function and the square root of the four mass enhancement factors@xcite ; second , that the four - point vertex function describes the multiple scattering of an electron and a hole@xcite ; third , that all reducible diagrams can be split into two parts by removing a pair of the single - particle green s functions , while all irreducible diagrams can not be done by using that removing@xcite . in contrast to successful description@xcite in the paramagnetic ( pm ) phase of sr@xmath0ruo@xmath1 , landau s fl theory fails in describing electronic properties of other pm ruthenates near a magnetic quantum - critical point ( qcp ) . for example , in sr@xmath0ru@xmath2ti@xmath3o@xmath1 , located near an antiferromagnetic ( af ) qcp@xcite , the spin susceptibility shows the curie - weiss - like temperature dependence and the inplane resistivity , @xmath16 , shows the @xmath4-linear dependence@xcite , deviating from the fl - type @xmath15 dependence ; the characteristic wave vectors of this af qcp@xcite , @xmath17 and its symmetrically equivalent ones , are the same as the wave vectors of the most strongly enhanced spin fluctuation in sr@xmath0ruo@xmath1@xcite . also , in ca@xmath18sr@xmath19ruo@xmath1 around @xmath20 , located near a ferromagnetic qcp@xcite , the curie - weiss - like temperature dependence of the spin susceptibility and @xmath21 dependence of @xmath16 are observed@xcite . since such non - fl - like behaviors are observed in other systems near a magnetic qcp@xcite or a mott transition@xcite , where many - body effects generally become very important , these experimental facts@xcite indicate the necessity of both considering how the basic assumptions of landau s fl theory are violated and discussing many - body effects beyond landau s fl theory in a more elaborate theory . so far , there are two candidates for the origin of such non - fl - like behaviors . one is bad metal@xcite due to local correlation enhanced near a mott transition . if the low - energy excitations can be satisfactorily described by using only low - order taylor series of the self - energy in terms of frequency , the coherent part of the single - particle green s function plays dominant roles in discussing electronic properties . however , if there are some non - negligible contributions from the higher - order taylor series , the incoherent part also becomes important . such case is realized in a pm metallic phase near the mott transition due to the formation of the upper and lower hubbard peaks , arising from enhanced local correlation@xcite . since perturbation theory can treat the coherent part appropriately and its treatment of the incoherent part is insufficient@xcite , perturbation theory is unsuitable in the latter case . instead , the latter case can be well described in dynamical - mean - field theory ( dmft ) since the dmft can take account of the frequency dependence of the self - energy nonperturbatively@xcite . actually , several non - fl - like behaviors are obtained in the dmft near the mott transition as a result of the @xmath4-linear unrenormalized qp damping and the temperature - dependent mass enhancement factor@xcite . [ in this paper , i define the unrenormalized qp damping as the qp damping divided by the inverse of the mass enhancement factor . ] these results indicate the importance of nonperturbative effects of local correlation near the mott transition . note that hund s metal@xcite is classified as the bad metal since the non - fl - like behaviors in the hund s metal arise from local spin fluctuation enhanced near a mott transition , although there is a crucial difference between the hund s rule coupling dependence of the critical value of the intraorbital coulomb interaction for the mott transition at half - filling and non - half - filling@xcite . the other is nearly magnetic metal@xcite due to spatial correlation enhanced near a magnetic qcp . if a system is located far away from a magnetic qcp , all scattering processes are independent of temperature . however , if the system approaches the qcp , several characteristic scattering processes of the qcp show the strong temperature - dependent enhancement@xcite . for example , in case near a stripe - type af qcp the scattering processes mediated by af spin fluctuations with @xmath22 and its symmetrically equivalent ones are strongly enhanced as decreasing temperature@xcite . such strong temperature - dependent enhancement leads to the strong temperature dependence of the reducible four - point vertex function whose momentum is characterized by the qcp . thus , the basic assumption of landau s fl theory about the landau parameters is violated only for the characteristic momenta of spin fluctuation enhanced near the qcp@xcite . also , the basic assumption about the qp damping is violated for the characteristic momenta due to the formation of hot spot , arising from enhanced spin fluctuation@xcite ; at the hot spot , the qp damping does not become much smaller than temperature considered . for example , in case near the stripe - type af qcp , the qp dampings at the momenta connected by the nesting vector @xmath22 and its symmetrically equivalent ones are more strongly enhanced than those at the other momenta due to the enhancement of the corresponding af spin fluctuations@xcite . these two violations suggest the necessity of discussing electronic properties near a magnetic qcp in the microscopic perturbation theory beyond landau s fl theory . actually , fluctuation - exchange ( flex ) approximation reproduces several non - fl - like behaviors due to the hot - spot structure of the qp damping or the curie - weiss - like temperature dependence of spin fluctuations or both@xcite . as explained in sect . 2.3 , this approximation can take account of spatial correlation beyond a mean - field theory and describe electronic properties of a metallic phase at low temperatures for moderately strong electron correlation satisfactorily@xcite . in contrast to case near the mott transition , the dmft is inappropriate to describe electronic properties near a magnetic qcp since the dmft neglects spatial correlation completely@xcite . these results indicate powerfulness of the microscopic perturbation theory and the importance of temperature - dependent spatial correlation near a magnetic qcp . with the backgrounds explained above , i studied electronic structure and magnetic and transport properties@xcite of ruthenates near and away from the af qcp in the flex approximation with current vertex corrections ( cvcs ) arising from the self - energy and maki - thompson ( mt)@xcite term for the @xmath23-orbital hubbard model on a square lattice and succeeded in reproducing several experimental results of sr@xmath0ru@xmath2ti@xmath3o@xmath1@xcite and sr@xmath0ruo@xmath1@xcite . thus , the non - fl - like behaviors@xcite in sr@xmath0ru@xmath2ti@xmath3o@xmath1 can be understood as the nearly magnetic meal near the af qcp . moreover , since the results away from the af qcp about the orbital dependence of the mass enhancement are in better agreement with the experiment@xcite in sr@xmath0ruo@xmath1 than those of the dmft@xcite , electronic properties of ruthenates at low temperatures except a few cases@xcite near the mott transition may be better described in the microscopic perturbation theory than in the dmft . in this paper , i review part of the above previous study@xcite and show some new results . in sect . 2 , i explain the microscopic theory@xcite used for analyzing @xmath16 of some quasi - two - dimensional ( quasi-@xmath24d ) pm ruthenates . in sect . 3 , i show the results about many - body effects on @xmath16 of the ruthenates near and away from the af qcp in the flex approximation with the cvcs arising from the self - energy and mt term and compare these results with the results obtained in other cases where the cvcs are more simplified . in sect . 4 , i summarize the results and draw some conclusions . due to the limit of space , i do not consider the aslamasov - larkin ( al ) cvc@xcite , which is the other cvc in the flex approximation ; its detailed derivation and effects are going to be discussed elsewhere@xcite . in this section , i explain an effective model of some quasi-@xmath24d ruthenates , briefly review the formal derivation of the resistivity of a multiorbital hubbard model in a pm metallic phase , and formulate the microscopic perturbation theory used to calculate the resistivity . the more detailed explanations about those derivation and formulation are going to be given elsewhere@xcite . in the following , i use the unit @xmath25 , set the coordinates @xmath26 , @xmath27 , and @xmath28 in the directions of the ru@xmath29o bonds of a ruo@xmath30 octahedral , and label the @xmath31 , @xmath32 , and @xmath33 orbitals as @xmath34 , @xmath24 , and @xmath35 , respectively . i explain an effective model to describe electronic properties of some quasi-@xmath24d ruthenates without the rotation or the tilting of ruo@xmath30 octahedra@xcite . before introducing the effective model , i briefly explain several basic electronic properties . some ruthenates whose crystal structures are @xmath36-type such as sr@xmath0ruo@xmath1@xcite or @xmath37-type such as sr@xmath38ru@xmath0o@xmath39@xcite are categorized into quasi-@xmath24d @xmath23-orbital systems . for simplicity , we focus on sr@xmath0ruo@xmath1 ; the following properties remain qualitatively the same in other metallic ruthenates@xcite . first , the inplane resistivity is about @xmath40 times as small as the out - of - plane resistivity at low temperatures@xcite , and the almost cylindrical fermi surface ( fs ) is observed in the de haas - van alphen measurement@xcite . these indicate quasi-@xmath24d electronic conduction . moreover , according to several density - functional calculations@xcite in local - density approximation ( lda ) , conducting bands near the fermi level are formed by the antibonding bands of the ru @xmath23 and the o @xmath41 orbitals , and the @xmath23 orbitals mainly contribute to the density - of - states ( dos ) near the fermi level . since the topology of the fs obtained in the lda qualitatively agrees with experiments@xcite , the @xmath23 orbitals play dominant roles in discussing electronic properties at low temperatures . with the above background , i assume that the electronic structure obtained in the lda@xcite for sr@xmath0ruo@xmath1 is a good starting point to consider many - body effects beyond a mean - field approximation , and i use a @xmath23-orbital hubbard model on a square lattice as the effective model . thus , the hamiltonian is @xmath42 with @xmath43 and @xmath44 where @xmath45 is the energy dispersions , measuring from the chemical potential , @xmath46 , @xmath47 , @xmath48 , @xmath49 , and @xmath50 are intraorbital coulomb interaction , interorbital coulomb interaction , hund s rule coupling , and pair hopping term , @xmath51 is @xmath52 and @xmath53 is @xmath54 with the pauli matrices @xmath55 . since i do not consider the effects@xcite of the rotation and the tilting , we focus on electronic properties of sr@xmath0ruo@xmath1@xcite and some doped sr@xmath0ruo@xmath1@xcite without these distortions . note that the rotation is present in ca@xmath18sr@xmath19ruo@xmath1 around @xmath20@xcite . whose parameters are chosen so as to reproduce the results in the lda@xmath56 for sr@xmath0ruo@xmath1.,width=453 ] by considering some symmetrically possible hopping processes and the difference between the crystalline - electric - field energies of the @xmath33 and @xmath57 orbitals , @xmath58 , we can construct the tight - binding model whose @xmath45 is given by @xmath59 and otherwise @xmath60 . @xmath46 is chosen so that the electron number per a site , @xmath61 , is fixed . for the actual calculations , @xmath46 is determined by the bisection method using @xmath62,\label{eq : mu - int}\end{aligned}\ ] ] where the second term becomes zero without the interaction terms . here @xmath63 is @xmath64 @xmath65 is the fermi distribution function , @xmath66 is the noninteracting single - particle green s function with fermionic matsubara frequency , @xmath67 , @xmath68 and @xmath69 is the single - particle green s function whose determination is explained in sect . 2.3 . in eq . ( [ eq : mu - int ] ) , we put the chemical potentials in @xmath70 , @xmath71 , and @xmath72 the same to reduce the numerical error arising from the cut - off frequency . to reproduce the electronic structure obtained in the lda@xcite , i set @xmath73 ( ev ) and @xmath74 . actually , we see from fig . [ fig : fig1 ] that the total bandwidth , the topology of each fs sheet , and the location of the van hove singularity of the @xmath33 orbital agree with the lda results@xcite : the total bandwidth is about @xmath75 ev ; the fs consists of the quasi-@xmath34d hole - like @xmath11 and electron - like @xmath76 sheets of the @xmath57 orbital and the quasi-@xmath24d electron - like @xmath77 sheet of the @xmath33 orbital , and the @xmath77 sheet is located nearer the inner sheet in @xmath78 or @xmath79 line than in the experiment@xcite ; the van hove singularity is located above the fermi level . also , the occupation numbers of the @xmath57 and the @xmath33 orbital , being @xmath80 and @xmath81 , are consistent with the lda values@xcite . then , i set @xmath82 , @xmath83 , and @xmath84 , use @xmath47 as a parameter , and treat the effects of interactions in the flex approximation with the cvcs arising from the self - energy and mt term ; its detail is explained in sect . 2.3 . finally , we remark on suitability neglecting the spin - orbit coupling of ru ions for discussing many - body effects on the resistivity . a density - functional calculation@xcite for sr@xmath0ruo@xmath1 within local - spin - density approximation shows that the coupling constant is @xmath85 ev , and that the main effect on the electronic structure is the weak mixing between the bands of the @xmath57 and the @xmath33 orbital around @xmath86 and its symmetrically equivalent ones . this effect will not qualitatively change the results shown in sect . 3 since this coupling constant is smaller than the main terms in @xmath87 and that weak mixing will lead to small changes of the momentum dependence and value of the spin susceptibility from those without the spin - orbit coupling . thus , i believe that neglecting the spin - orbit coupling is suitable for qualitative discussions about many - body effects on the resistivity . i briefly review the formal derivation@xcite of the resistivity of a multiorbital hubbard model in a pm metallic phase . we first derive an exact expression of the longitudinal conductivity , the inverse of the resistivity , in the presence of electron correlation within the linear - response theory@xcite . then , we rewrite this exact expression in terms of the four - point vertex function by using the three - point vector vertex function . due to difficulty solving the exact expression , we derive an approximate expression by using the most - divergent - term approximation introduced by liashberg@xcite . to discuss the resistivity within the linear - response theory , we use the kubo formula@xcite for the longitudinal conductivity , @xmath88 ( @xmath89 ) , in @xmath90-limit and @xmath91 with @xmath92 being the transport relaxation time@xcite , which is of the order of magnitude of the qp lifetime . namely , @xmath93 is given by @xmath94 where @xmath95 is obtained by the analytic continuation of @xmath96 , @xmath97 with bosonic matsubara frequency , @xmath98 . here @xmath99 is @xmath100 , @xmath101 is the group velocity , @xmath102 , and @xmath103 is @xmath104 where @xmath105 is the reducible four - point vertex function . thus , the analytic continuation of @xmath103 is necessary to calculate @xmath93 . before the analytic continuation of @xmath103 , i remark on the important physical meanings of @xmath106 and @xmath91 the order of @xmath107 and @xmath108 is very important in discussing transport properties since the observable currents can be obtained by the dynamical and uniform field ( i.e. , @xmath106 ) but the static and non - uniform field ( i.e. , @xmath109 ) does not cause any observable currents due to the screening induced by the modulations of the charge distribution@xcite . also , the value of @xmath110 is very important since the adiabatic condition @xmath91 means the realization of local equilibrium due to the rapid relaxation compared with @xmath111 , a typical time scale of the field ; as a result of that relaxation , the electronic transports are governed mainly by the qps near the fermi level . for example , such importance of the inequality of @xmath112 is seen from the difference between the zero and the first sound@xcite . replacing @xmath113 and @xmath114 in eq . ( [ eq:2g - matsu ] ) by the corresponding contour integrals@xcite and doing several straightforward calculations@xcite with attention to the analytic properties@xcite of the single - particle green s function and four - point vertex function , we can carry out the analytic continuation of @xmath103 . as a result , @xmath95 is given by @xmath115,\label{eq : sum - analytic}\end{aligned}\ ] ] where @xmath116 is @xmath117 with @xmath118 being @xmath119 and @xmath120 and @xmath121 being @xmath122,\label{eq:4vc-11}\\ \mathcal{j}_{12;\{a\}}(k , k^{\prime};\omega ) = & \ \bigl(\tanh \frac{\epsilon^{\prime}+\omega}{2 t } -\tanh \frac{\epsilon^{\prime}}{2t}\bigr ) \gamma_{12;\{a\}}(k , k^{\prime};\omega),\label{eq:4vc-12}\\ \mathcal{j}_{13;\{a\}}(k , k^{\prime};\omega ) = & -\tanh \frac{\epsilon^{\prime}+\omega}{2 t } \gamma_{13\textrm{-i};\{a\}}(k , k^{\prime};\omega)\notag\\ & -\coth \frac{\epsilon+\epsilon^{\prime}+\omega}{2 t } \bigl[\gamma_{13\textrm{-ii};\{a\}}(k , k^{\prime};\omega)- \gamma_{13\textrm{-i};\{a\}}(k , k^{\prime};\omega)\bigr],\label{eq:4vc-13}\\ \mathcal{j}_{21;\{a\}}(k , k^{\prime};\omega ) = & \ \tanh \frac{\epsilon^{\prime}}{2 t } \gamma_{21;\{a\}}(k , k^{\prime};\omega),\label{eq:4vc-21}\\ \mathcal{j}_{22;\{a\}}(k , k^{\prime};\omega ) = & \ \bigl(\coth \frac{\epsilon^{\prime}-\epsilon}{2 t } -\tanh \frac{\epsilon^{\prime}}{2t}\bigr ) \gamma_{22\textrm{-ii};\{a\}}(k , k^{\prime};\omega)\notag\\ & + \bigl(\coth \frac{\epsilon^{\prime}+\epsilon+\omega}{2 t } -\coth \frac{\epsilon^{\prime}-\epsilon}{2t}\bigr ) \gamma_{22\textrm{-iii};\{a\}}(k , k^{\prime};\omega)\notag\\ & + \bigl(\tanh \frac{\epsilon^{\prime}+\omega}{2 t } -\coth \frac{\epsilon^{\prime}+\epsilon+\omega}{2t}\bigr ) \gamma_{22\textrm{-iv};\{a\}}(k , k^{\prime};\omega),\label{eq:4vc-22}\\ \mathcal{j}_{23;\{a\}}(k , k^{\prime};\omega ) = & -\tanh \frac{\epsilon^{\prime}+\omega}{2 t } \gamma_{23;\{a\}}(k , k^{\prime};\omega),\label{eq:4vc-23}%\\\end{aligned}\ ] ] @xmath123,\label{eq:4vc-31}\\ \mathcal{j}_{32;\{a\}}(k , k^{\prime};\omega ) = & \ \bigl(\tanh \frac{\epsilon^{\prime}+\omega}{2 t } -\tanh \frac{\epsilon^{\prime}}{2t}\bigr ) \gamma_{32;\{a\}}(k , k^{\prime};\omega),\label{eq:4vc-32}\end{aligned}\ ] ] and @xmath124.\label{eq:4vc-33}\end{aligned}\ ] ] in eq . ( [ eq : kpart - analytic ] ) , i have not explicitly written whether the frequency integral is the principal integral or not ; the integrals containing hyperbolic cotangent are the principal ones . also , in eqs . ( [ eq:4vc-11])([eq:4vc-33 ] ) the additional subscript of the four - point vertex function such as @xmath125 represents the relations among its three frequency variables , as shown in fig . [ fig : fig2 ] . since @xmath126 is determined by the bethe - salpeter equation , @xmath127 and @xmath128 is obtained by the method explained in sect . 2.3 , we can exactly calculate @xmath93 from eqs . ( [ eq : sigmaxx ] ) , and ( [ eq : sum - analytic])([eq:4vc - red - irred ] ) in principle . of that function.,width=226 ] then , to rewrite eq . ( [ eq : sum - analytic ] ) in a more compact form , we use the three - point vector vertex function instead of the four - point vertex function . the three - point vertex function in matsubara - frequency representation is defined as @xmath129 since the analytic continuation of the three - point vector vertex function can be carried out@xcite in a similar way to that used for @xmath103 , we obtain the three - point vector vertex function in real - frequency representation , @xmath130 here the additional subscript of the three - point vector vertex function , @xmath131 , denotes the conditions about its @xmath132 and @xmath90 : @xmath133 denotes @xmath134 and @xmath135 ; @xmath136 denotes @xmath137 and @xmath135 ; @xmath138 denotes @xmath137 and @xmath139 . combining eqs . ( [ eq : sum - analytic ] ) and ( [ eq : kpart - analytic ] ) with eq . ( [ eq:3-point - vc ] ) , we can rewrite @xmath95 as @xmath140.\label{eq : sum - analytic2}\end{aligned}\ ] ] since the exact expression is difficult to solve , we use the most - divergent - term approximation@xcite , based on the properties@xcite of the product of the single - particle green s functions in the limit @xmath141 in the presence of the qp peak . this is correct in the fl and remains appropriate in a correlated metallic system where perturbation expansion is satisfactory . as explained in sect . 1 , applicability of the fl theory differs from that of a microscopic perturbation theory . the microscopic perturbation theory is satisfactory to describe electronic properties of a correlated electron system if perturbation expansion has a good convergence or becomes an asymptotic expansion . let us recall , first , that a well approximate partial sum can be constructed even if its convergence is not good@xcite ; second , that perturbation expansion becomes an asymptotic expansion near a phase transition . before proceeding with the formal derivation of @xmath93 , we remark on the property@xcite of a pair of the single - particle green s functions in the limit @xmath142 . when the qp peak exists and the qp damping is much smaller than @xmath4 ( i.e. , cold - spot - type ) , only @xmath143 gives the most singular term being inversely proportional to the qp damping due to the merging of the poles of these green s functions in @xmath142 ; the others , the retarded - retarded and the advanced - advanced pair , become the higher order terms . in this discussion , we have considered only the contribution from the coherent part of the single - particle green s function since the incoherent part does not lead to such singular term in @xmath142 . also , i have used @xmath144=-2\pi i\delta(x - a)$ ] since the qp damping at momenta near the fermi level is assumed to be negligible at low temperatures . the singular contribution from the hot spot is less important than that from the cold spot since the qp damping at the hot spot remains non - negligible even on the fermi level . since the existence of the qp peak and cold - spot - type qp damping is appropriate at least for several momenta in a metallic phase near a magnetic qcp@xcite or a mott transition@xcite , the leading terms in @xmath145 with respect to the qp damping or @xmath90 at low temperatures and frequencies are given by @xmath146 [ \epsilon-\xi_{\beta}^{\ast}(\boldk)+i0+]},\label{eq : g1-fl}\\ g_{2;acdb}(k;\omega ) \sim & \ 2\pi i \textstyle\sum\limits_{\alpha,\beta } u_{a\alpha c ; d \beta b}(\boldk ) \dfrac{z_{\alpha}(\boldk)z_{\beta}(\boldk)\delta(\epsilon-\xi_{\alpha}^{\ast}(\boldk ) ) } { \omega-\xi_{\alpha}^{\ast}(\boldk)+\xi_{\beta}^{\ast}(\boldk ) + i[\gamma_{\alpha}^{\ast}(\boldk)+\gamma_{\beta}^{\ast}(\boldk)]},\label{eq : g2-fl}\end{aligned}\ ] ] and @xmath147 [ \epsilon-\xi_{\beta}^{\ast}(\boldk)-i0+]},\label{eq : g3-fl}\end{aligned}\ ] ] where @xmath148 with @xmath149 being the unitary matrix to obtain the qp bands [ not equal to @xmath150 . thus , the most divergent terms in the coherent limit arise from @xmath151 at cold spots . using the above property , we obtain an approximate expression of @xmath93 where we consider only the most divergent terms with respect to the qp damping in the coherent limit . to use the most - divergent - term approximation@xcite , we introduce two quantities , @xmath152 and @xmath153 , which are irreducible with respect to only a retarded - advanced pair of the single - particle green s functions , @xmath154 and @xmath155 using these two quantities with eqs . ( [ eq:4vc-12 ] ) , ( [ eq:4vc-21 ] ) , ( [ eq:4vc-23 ] ) and ( [ eq:4vc-32 ] ) and the exchange symmetry of the four - point vertex function among its momentum and frequency variables , we can rewrite eq . ( [ eq : sum - analytic2 ] ) as @xmath156\notag\\ & -\frac{1}{n } \textstyle\sum\limits_{\boldk } \textstyle\sum\limits_{\{a\ } } \lambda_{\nu;2;ba}^{(0)}(k;\omega ) \int^{\infty}_{-\infty}\dfrac{d\epsilon}{4\pi i } \bigl(\tanh \dfrac{\epsilon+\omega}{2 t } -\tanh \dfrac{\epsilon}{2t}\bigr)\notag\\ & \ \ \times g_{2;acdb}(k;\omega)\lambda_{\nu;2;cd}(k;\omega).\label{eq : sum - analytic3}\end{aligned}\ ] ] at this stage , this expression remains exact . then , since only the second term in eq . ( [ eq : sum - analytic3 ] ) contains a retarded - advanced pair and the leading term with respect to @xmath90 comes from @xmath157 , we can obtain an approximate expression of @xmath93 in the most - divergent - term approximation , @xmath158 from this and eq . ( [ eq : g2-fl ] ) , we can show that @xmath93 is inversely proportional to the unrenormalized qp damping@xcite . @xmath159 and @xmath160 in eq . ( [ eq : sigmaxx - approx ] ) are determined as follows . by combining eq . ( [ eq:3vc-0 ] ) with the ward identity@xcite , @xmath159 is given by @xmath161 in the present model , @xmath162 holds due to the even - parity and the time - reversal symmetry . since @xmath163 is satisfied due to eqs . ( [ eq:4vc-12 ] ) , ( [ eq:4vc-32 ] ) , and ( [ eq:4vc-0 ] ) , @xmath160 is given by @xmath164 thus , the present theory takes account of the cvcs due to the self - energy and irreducible four - point vertex function , neglected in the relaxation - time approximation@xcite . to calculate the resistivity , i use the flex approximation with the cvcs arising from the self - energy and mt term in a pm phase i first explain the flex approximation for a multiorbital hubbard model in a pm phase . next , i derive the irreducible four - point vertex function in this approximation ; as described in sect . 1 , i consider only the mt term and neglect the al term . then , i show the bethe - salpeter equation for the current by using the mt term as the kernel of the cvc . i also remark on the numerical treatment of the principal integral appearing in that cvc . finally , i discuss applicability of this method . i determine several single - particle or two - particle quantities using the flex approximation@xcite in a pm phase where only the electron - hole scattering processes of the bubble and the ladder diagrams are considered as the luttinger - ward functional@xcite , @xmath165 $ ] . since that is a conserving approximation@xcite based on the thermodynamic potential expressed in terms of the single - particle green s function , we can determine single - particle or two - particle quantities by using @xmath165 $ ] and its functional derivative . since constructing @xmath165 $ ] is equivalent to constructing the effective interaction , where the same kinds of the diagrams are considered , and the latter is easier , we formulate the flex approximation by the latter procedure as follows . first , considering the electron - hole scattering processes of the bubble and the ladder diagrams for @xmath166 , we obtain the effective interaction in the flex approximation , @xmath167\sigma_{s_{1 } s_{2}}^{0}\sigma_{s_{4 } s_{3}}^{0}\notag\\ \hspace{-4pt } -&\frac{1}{2 } \bigl [ u_{abcd}^{\textrm{s } } + \textstyle\sum\limits_{\{a\ } } u_{abab}^{\textrm{s}}\chi_{abcd}^{\textrm{s}}(\boldq , i\omega_{n})u_{cdcd}^{\textrm{s } } \bigr]\boldsigma_{s_{1 } s_{2}}\cdot \boldsigma_{s_{4 } s_{3}},\end{aligned}\ ] ] with @xmath168 and @xmath169 here we introduce the bare four - point vertex functions in spin and charge sector , @xmath170 and @xmath171 where @xmath172 is @xmath173 and the spin - flipping term , @xmath174 for @xmath175 , satisfies @xmath176 . also , we neglect the vertex corrections to the susceptibilities in spin and charge sector ; its effects are discussed later . then , the single - particle green s function is determined by the dyson equation , @xmath177 with the self - energy given by @xmath178 where @xmath179\notag\\ & + \dfrac{1}{2 } \bigl [ -u_{abcd}^{\textrm{c } } + \textstyle\sum\limits_{\{a\ } } u_{abab}^{\textrm{c } } \chi_{abcd}^{\textrm{c}}(\boldq , i\omega_{n } ) u_{cdcd}^{\textrm{c } } \bigr]\notag\\ & -\textstyle\sum\limits_{\{a\ } } u_{aabb}^{\uparrow\downarrow } \chi_{abcd}(\boldq , i\omega_{n } ) u_{ccdd}^{\uparrow\downarrow}.\label{eq : flex-6}\end{aligned}\ ] ] the last term in eq . ( [ eq : flex-6 ] ) is introduced to exclude the double counting of the topologically equivalent term in the self - energy . solving eqs . ( [ eq : flex-1])([eq : flex-3 ] ) and ( [ eq : flex-4])([eq : flex-6 ] ) with eqs . ( [ eq : mu - int ] ) , ( [ eq : g0 ] ) , ( [ eq : bares ] ) and ( [ eq : barec ] ) selfconsistently by iteration , we can determine the single - particle and the two - particle quantities in the flex approximation . it should be noted that the partial inclusion of mode - mode couplings@xcite for fluctuations , the interactions between fluctuations at different momenta , through the self - energy improves some unrealistic results obtained in the random - phase approximation , although the susceptibilities are determined by the random - phase approximation - type ( but renormalized ) equations . for example , in the flex approximation@xcite , the value of @xmath47 for a magnetic transition becomes about @xmath24 ev , the momentum dependences of the mass enhancement and fs deformation are taken into account , and the curie - weiss - like temperature dependence of the spin susceptibility is obtained near a magnetic qcp . in particular , the final improvement is powerful to describe electronic properties near a magnetic qcp . we also determine the irreducible four - point vertex function in the flex approximation in keeping conservation laws@xcite . in a conserving approximation , the irreducible four - point vertex function is given by@xcite @xmath180 for the actual calculations , we first calculate the right - hand side at @xmath181 and @xmath182 and then label momentum and frequency transfers correctly as the electron - hole scattering process among an electron of orbital @xmath183 with @xmath184 , a hole of orbital @xmath185 with @xmath186 , an electron of orbital @xmath187 with @xmath188 , and a hole of orbital @xmath189 with @xmath190 . after several straightforward calculations@xcite by using eqs . ( [ eq : flex-5])([eq : gamma1-consv ] ) , we obtain the irreducible four - point vertex function in the flex approximation , which is the sum of the mt and the al term@xcite . in this paper , i consider only the mt term , @xmath191 this treatment will be sufficient for a qualitative discussion about many - body effects on @xmath16 since the cvc arising from the al term gives the higher order contribution compared with that arising from the mt term@xcite . i have checked the validity of this statement by calculating the main terms of the al cvc@xcite . since it is necessary to calculate @xmath192 in eq . ( [ eq : lamb ] ) , we need to carry out the analytic continuation of eq . ( [ eq : mt - imfreq ] ) in region @xmath193-ii , region @xmath193-iii , and region @xmath193-iv . carrying out the analytic continuation@xcite of the mt term and using eq . ( [ eq:4vc-22 ] ) , we obtain @xmath194 substituting eq . ( [ eq : gamma1-flex - refreq ] ) into eq . ( [ eq : lamb ] ) , we obtain the bethe - salpeter equation for the current in the flex approximation with the mt cvc in a pm phase , @xmath195 we see the roles of the mt cvc are similar to those of the backflow correction@xcite . before discussing applicability , i explain how to treat the principal integral in eq . ( [ eq : bs - mtcvc ] ) for the numerical calculations . since both the numerator and denominator of the term containing @xmath196 in the mt cvc become zero simultaneously at @xmath197 due to @xmath198 , the principal integral can be calculated as follows : @xmath199 \bigl|_{\epsilon^{\prime}=\epsilon},\end{aligned}\ ] ] where the first term contains the contributions other than @xmath197 . finally , i discuss applicability of the flex approximation with the cvcs arising from the self - energy and mt term . first , the flex approximation is suitable to describe the electronic structure at low temperatures for moderately strong electron correlation . in a single - orbital hubbard model on a square lattice@xcite , the imaginary - time dependence of the single - particle green s function at several momenta in the flex approximation shows satisfactory ( but not perfect ) agreement with that in the quantum - monte - carlo calculation at @xmath47 being a half of the bandwidth ; the agreement becomes better near the af qcp than away from the af qcp . since the similar agreement will hold even in a multiorbital hubbard model on the same lattice and the flex approximation can treat the coherent part of the electronic spectrum satisfactorily@xcite , the electronic structure in metallic phases of the present model at low temperatures will be well described by the flex approximation at least qualitatively . actually , the flex approximation succeeded in reproducing the larger effective mass@xcite of the @xmath33 orbital than that of the @xmath57 orbital@xcite and its agreement with the experiment@xcite is better than the case of the dmft@xcite , as described in sect . 1 . in contrast , the flex approximation becomes unsuitable at high temperatures or near a mott transition for strong electron correlation . this is because local correlation plays important roles in such case@xcite and the effects of local correlation on the electronic spectrum are smeared out in the flex approximation@xcite . then , the magnetic properties at low temperatures for moderately strong electron correlation will be appropriately described by the flex approximation if the dominant correlation of the system is spin fluctuation whose largest contribution comes from a non - degenerate orbital . due to neglecting the vertex corrections to the susceptibilities , the enhancement of spin fluctuation arising from electron correlation is overestimated in the flex approximation compared with the enhancement of charge or orbital fluctuation . actually , in a two - degenerate - orbital hubbard model on a square lattice at small @xmath200 near an af qcp , the al vertex correction to the susceptibilities causes the enhancement of orbital fluctuation@xcite . however , i believe that in the present model the flex approximation is sufficient to describe the magnetic properties at least qualitatively since the @xmath33 orbital gives the largest contribution to spin fluctuation@xcite ; in this case , even if the mt and the al vertex correction to the susceptibilities are considered beyond the flex approximation , the magnetic properties will not qualitatively change and these corrections will modify the values of the susceptibilities since orbital fluctuation enhanced due to the al term does not dominate over spin fluctuation . actually , the strongest enhancement of spin fluctuation at @xmath201 in the flex approximation@xcite away from and near the af qcp agrees with the experiments in sr@xmath0ruo@xmath1@xcite and sr@xmath0ru@xmath2ti@xmath3o@xmath1@xcite , respectively . moreover , if the vertex corrections to the current arising from the self - energy and irreducible four - point vertex function due to electron correlation are added to the flex approximation , this method is suitable to describe the transport properties of a metallic phase due to low - frequency external field satisfying @xmath202 at low temperatures . in contrast to case of the vertex corrections to the susceptibilities , the vertex corrections to the current are essential for discussing the transport properties@xcite since the cvcs are vital to satisfy conservation laws and conservation laws play significant roles in transport phenomena . for example , the importance of the treatment holding conservation laws is known for a system without the lattice ( e.g. , the electron gas ) : only if the cvcs due to electron correlation are correctly taken into account , we can obtain the correct results such as the absence both of the resistivity@xcite and of the renormalization of the drude weight@xcite and electron cyclotron frequency@xcite . also , the cvcs due to electron correlation are important in a system with the lattice since these cvcs are necessary to obtain the correct effects of electron correlation on the transport coefficients in the presence of the umklapp scattering@xcite . another example showing the importance of the cvcs is the emergence of the curie - weiss - like temperature dependence of the hall coefficient near the af qcp due to the mt cvc in the flex approximation in the single - orbital hubbard model on a square lattice@xcite . in addition to the treatment holding conservation laws , the satisfactory treatment of the coherent part in the flex approximation is powerful to describe the transport properties of a metallic phase due to the low - frequency external field . this is because the dominant contributions to the response induced by that external field at low temperatures come from the contributions near the fermi level as a result of the energy derivative of the fermi distribution function in the response function . furthermore , this powerfulness of the flex approximation will hold even near a mott transition since it is shown in the dmft@xcite for a single - orbital hubbard model that the transport properties in @xmath91 are well described in the approximation where only the coherent part is considered . thus , the flex approximation with the cvcs arising from the self - energy and mt term is satisfactory to describe the transport properties of metallic phases of the present model at low temperatures . actually , this method@xcite near and away from the af qcp reproduced the temperature dependence of several transport properties of sr@xmath0ruo@xmath1@xcite and sr@xmath0ru@xmath2ti@xmath3o@xmath1@xcite , as pointed out in sect . 1 . in this section , i show the results of @xmath203 of some quasi-@xmath24d pm ruthenates in the flex approximation with the cvcs arising from the self - energy and mt term and compare these results obtained in more simplified cases than that method . in particular , we focus on the effects of the self - energy and mt term due to electron correlation and the role of each @xmath23 orbital . , for several temperatures at @xmath204 ( a ) @xmath205 and ( b ) @xmath206 ev , and ( c ) temperature dependences of @xmath207 at @xmath208 and @xmath206 ev . , width=472 ] i carried out the numerical calculations as follows . i set the @xmath209 meshes of the brillouin zone and @xmath210 matsubara frequencies and used the fast fourier transformation with the zero padding method@xcite . i obtained the single - particle green s function , self - energy , and mt term in the flex approximation by solving eqs . ( [ eq : flex-1])([eq : flex-3 ] ) and ( [ eq : flex-4])([eq : flex-6 ] ) with eqs . ( [ eq : mu - int ] ) , ( [ eq : g0 ] ) , ( [ eq : bares ] ) and ( [ eq : barec ] ) by iteration , where i assumed that convergence was reached when the difference between the self - energies before and after certain iteration was less than @xmath211 . to obtain the quantities as a function of real frequency , i used the pad approximation@xcite using the quantities at the lowest four matsubara frequencies . the real - frequency integrations were approximated by the integrations with the interval @xmath212 ev and the upper and lower values @xmath213 and @xmath214 ev . the current was determined by solving eq . ( [ eq : bs - mtcvc ] ) by iteration , where the convergence condition was assumed to be that the difference between the currents before and after certain iteration was less than @xmath211 . in the following , i consider cases at @xmath208 and @xmath206 ev as cases away from and near the af qcp , respectively . these correspondences are because of the pauli - pm temperature dependence of the spin susceptibility at @xmath208 ev and the curie - weiss - like temperature dependence of the spin susceptibility at @xmath215 at @xmath216 ev [ see figs . [ fig : fig3](a)[fig : fig3](c ) ] . the latter is characteristic of a magnetic qcp and causes the strong temperature dependence of the landau parameters with momentum transfer @xmath217 through the temperature dependence of the reducible four - point vertex function . the choice of @xmath47 is reasonable since the value estimated experimentally in sr@xmath0ruo@xmath1 is about @xmath24 ev@xcite . as explained below in detail , there are three main results : ( i ) @xmath16 of some quasi-@xmath24d pm ruthenates without the rotation and the tilting of ruo@xmath30 octahedra is determined almost by the conductions of the @xmath57 orbital due to the smaller unrenormalized qp dampings of the @xmath57 orbital than those of the @xmath33 orbital ; ( ii ) the crossover between the @xmath4-linear and the @xmath15 dependence of @xmath16 occurs away from the af qcp at about @xmath218 ev due to the temperature dependences of the unrenormalized qp dampings of the @xmath57 orbital ; ( iii ) the @xmath4-linear @xmath16 emerges near the af qcp due to the hot - spot structure of the qp dampings of the @xmath57 orbital at @xmath219 and its symmetrically equivalent ones . at @xmath208 ev . the definition of each case is described in the main text . the inset shows @xmath16 against @xmath15 below @xmath220 ev . , width=272 ] i begin with the temperature dependence of @xmath16 at @xmath208 ev in fig . [ fig : fig4 ] . three cases in that figure are defined as follows : in mt cvc case , the cvcs arising from the self - energy and mt term are included ; in no mt cvc case , the cvc arising from the self - energy is included ; in no all cvcs case , which is equivalent to the relaxation - time approximation@xcite , all the cvcs are neglected . there are three main remarks about fig . [ fig : fig4 ] . first , in all the three cases , decreasing temperature causes the crossover from the @xmath4-linear dependence to the @xmath15 dependence at about @xmath218 ev . the @xmath15 dependence at low temperatures is more clearly seen from the inset of fig . [ fig : fig4 ] . second , the value of @xmath16 in the no all cvcs case at each temperature is largest in the three cases . third , the value of @xmath16 increases from that in the no mt cvc case when the mt cvc is included . the first remark indicates that the cvcs little affect on the power of the temperature dependence of the resistivity . the second indicates that the value of the resistivity is overestimated in the relaxation - time approximation . the third indicates that the mt cvc enhances the resistivity as a result of the reduction of the current , which is similar to the effect of the backflow correction . thus , the main effects of the cvcs on the resistivity just change its value . at @xmath208 ev in ( a ) the mt cvc case , ( b ) the no mt cvc case , and ( c ) the no all cvcs case . the definition of each orbital - decomposed @xmath221 is described in the main text . , width=464 ] orbital at @xmath222 ( ev ) , and ( c ) temperature dependence of the unrenormalized qp damping of the @xmath57 orbital at @xmath219 at @xmath208 ev . the dashed line in panel ( b ) denotes @xmath220 ev . the inset in panel ( c ) shows the data against @xmath15 . , width=464 ] then , to understand the role of each @xmath23 orbital , i analyze the orbital - decomposed @xmath221 at @xmath208 ev . the orbital - decomposed @xmath221 for the @xmath57 and the @xmath33 orbital are obtained by replacing @xmath223 in eq . ( [ eq : sigmaxx - approx ] ) by @xmath224 and @xmath225 , respectively . those components are sufficient in the present model since the intraorbital components are much larger than the interobital components due to the large intraorbital hopping integrals compared with the interorbital ones . we see from figs . [ fig : fig5](a)[fig : fig5](c ) that the dominant contributions to the total of @xmath221 come from the component of the @xmath31 orbital , and that the component of the @xmath33 orbital is less than @xmath226 of the total . due to the rotational symmetry of the system , in case of @xmath227 , the component of the @xmath32 orbital gives the dominant contributions . thus , the inplane transport is governed mainly by the conductions of the @xmath57 orbital . the above orbital - dependent transport arises from the smaller unrenormalized qp dampings of the @xmath57 orbital than those of the @xmath33 orbital since @xmath93 is inversely proportional to the unrenormalized qp damping , as explained in sect . 2.2 . actually , we see from fig . [ fig : fig6](a ) that the unrenormalized qp dampings of the @xmath57 orbital are smaller . the similar orbital dependence holds at the other temperatures . also , we see from figs . [ fig : fig6](b ) and [ fig : fig6](c ) , respectively , that the qp dampings of the @xmath57 orbital remain the cold spot , and that the temperature dependence of the unrenormalized qp damping of the @xmath57 orbital at @xmath219 changes from @xmath15 to @xmath4-linear at about @xmath218 ev . thus , the latter is the origin of the crossover of the power of the temperature dependence of @xmath16 at about @xmath218 ev . at @xmath216 the definition of each case is described in the main text in sect . , width=272 ] i turn to the temperature dependence of @xmath16 at @xmath216 ev in the three cases , considered in sect . 3.1 . from fig . [ fig : fig7 ] , we see that the @xmath4-linear @xmath16 emerges in all the three cases . we also see the similar effects of the cvcs on the value of @xmath16 to those at @xmath208 ev , the overestimation in the relaxation - time approximation ( i.e. , the no all cvcs case ) and the increase from the value in the no mt cvc case due to the mt cvc . the first result indicates that the emergence of the @xmath4-linear @xmath16 near the af qcp arises from the temperature dependence of the unrenormalized qp damping . furthermore , those results and the corresponding results at @xmath208 ev suggest that the effects of the self - energy and mt term on the value of @xmath16 and power of the temperature dependence of @xmath16 are ubiquitous . orbital at @xmath228 ( ev ) , and ( c ) the temperature dependence of the unrenormalized qp damping of the @xmath57 orbital at @xmath219 against @xmath229 at @xmath216 ev . the dashed lines in panel ( b ) denotes @xmath220 ev . , width=464 ] also , comparing fig . [ fig : fig7 ] with fig . [ fig : fig4 ] , we see electron correlation enhances the value of @xmath16 at each temperature . this is due to an increase in the unrenormalized qp dampings of the @xmath57 orbital with increasing @xmath47 [ e.g. , see figs . [ fig : fig6](a ) and [ fig : fig8](a ) ] . at @xmath216 ev in ( a ) the mt cvc case , ( b ) the no mt cvc case , and ( c ) the no all cvcs case . the definition of each orbital - decomposed @xmath221 is described in the main text in sect . , width=464 ] then , in a similar way to that used at @xmath208 ev , i analyze the role of each @xmath23 orbital at @xmath216 ev . figures [ fig : fig9](a)[fig : fig9](c ) show the temperature dependences of the total and orbital - decomposed components of @xmath221 at @xmath216 ev in the three cases . the orbital - dependent transport holds even near the af qcp : the conductions of the @xmath57 orbital mainly contribute to the inplane transport . its mechanism is the same as that at @xmath208 ev , i.e. the smaller unrenormalized qp dampings of the @xmath57 orbital than those of the @xmath33 orbital [ see fig . [ fig : fig8](a ) ] . thus , this orbital - dependent transport is characteristic in some quasi-@xmath24d pm ruthenates . moreover , we find from figs . [ fig : fig8](b ) and [ fig : fig8](c ) , respectively , that the qp damping of the @xmath57 orbital at @xmath219 becomes the hot spot , and that the temperature dependence of the unrenormalized qp damping of the @xmath57 orbital at @xmath219 is roughly proportional to @xmath229 . since such @xmath229 dependence near the hot spot causes the @xmath4-linear dependence of the average of the unrenormalized qp dampings for states along the fs@xcite , the origin of the @xmath4-linear @xmath16 near the af qcp is the hot - spot structure of the qp dampings of the @xmath57 orbital at momenta connected each other by the characteristic spin fluctuations of this af qcp . in summary , i reviewed many - body effects@xcite of @xmath16 of the quasi-@xmath24d pm ruthenates away from and near the af qcp in the flex approximation with the cvcs arising from the self - energy and mt term or with the cvc arising from the self - energy or without all the cvcs . the temperature dependence of @xmath16 away from and near the af qcp qualitatively agree with experiments of sr@xmath0ruo@xmath1@xcite and sr@xmath0ru@xmath2ti@xmath3o@xmath1@xcite , respectively : in case away from the af qcp , the crossover between the @xmath4-linear and the @xmath15 dependence at about @xmath218 ev and the @xmath15 dependence at low temperatures are obtained ; in case near the af qcp , the @xmath4-linear dependence even at low temperatures is obtained . here the main effect of the ti substitution on the temperature dependence of @xmath16 is assumed to be pushing the system nearer the af qcp than sr@xmath0ruo@xmath1 . the obtained results reveal some important aspects of many - body effects on the resistivity of correlated electron systems . first , the overestimation of the value of @xmath16 in the relaxation - time approximation and the back - flow - like effect of the mt cvc on the value of @xmath16 are ubiquitous . it is also ubiquitous that the power of the temperature dependence of the resistivity is determined almost by the temperature dependence of the momentum- and orbital - dependent unrenormalized qp damping . the similar results have been obtained in a single - orbital hubbard model on a square lattice@xcite . moreover , the @xmath4-linear resistivity near the af qcp is similar to that obtained in those previous studies@xcite . however , i emphasize that the criticality of the resistivity , the power of its temperature dependence , does not always connect with the criticality of spin fluctuation enhanced near a magnetic qcp in multiorbital systems , while these are always the same in single - orbital systems . this characteristic property comes from the facts that the orbital whose unrenormalized qp damping is small mainly contributes to the resistivity , and that the main orbital of the characteristic spin fluctuation of a magnetic qcp has the large unrenormalized qp damping . since the momentum , the temperature , and the orbital dependence of the damping of a qp ( i.e. , the unrenormalized qp damping or the qp damping ) are overlooked in landau s fl theory and such momentum dependence is overlooked in the dmft@xcite , the obtained results highlight their importance in discussing the resistivity of correlated electron systems . i close this paper with several remarks about the remaining issues . first , it is necessary to study the transport properties of other ruthenates@xcite using the method i used and discuss the similarities and differences . in particular , the study for a quasi-@xmath24d ruthenate@xcite near a ferromagnetic qcp is highly desirable to understand the similarities and differences between many - body effects and role of each @xmath23 orbital near the af and the ferromagnetic qcp . furthermore , it is important to analyze the transport properties of @xmath35d ruthenates@xcite since comparison of the results in quasi-@xmath24d and @xmath35d ruthenates leads to a deep understanding of the dimensionality . other remaining issues are the applications to other correlated electron systems such as transition - metal oxides@xcite , organic conductors@xcite , cecoin@xmath230@xcite , and upt@xmath38@xcite since these studies are important to deduce ubiquitous properties of correlated electron systems and characteristic properties of multiorbital systems . then , it is intriguing to study the transport properties in a superconducting phase by extending the present method@xcite in a pm phase since in some cases@xcite the cvcs arising from not only spin fluctuations but also superconducting fluctuations play important roles . furthermore , another remaining issue is to clarify the role of each @xmath23 orbital in the thermal transport@xcite in the superconducting phase of sr@xmath0ruo@xmath1 on the basis of the method where the orbital dependence of the damping of a qp is satisfactorily considered . this is because the combination of my result@xcite and several previous studies@xcite suggest the existence of the difference between the main orbitals of the inplane transport and the superconductivity ; if this is correct , we should pay attention to the effects of the orbital - dependent damping of a qp on the thermal transport@xcite for correct understanding of the experimental results . i thank dr . phua kok khoo and dr . sun han for giving me the opportunity to review my recent study@xcite and waiting for the submission of this article with patience . all the numerical calculations were performed by using the facilities of the supercomputer center , the institute for solid state physics , the university of tokyo .

in a series of two papers , we test the primordial scenario of globular cluster formation using results of high - resolutions-body simulations . in this first paper we study the initial relaxation of a stellar core inside a live dark matter minihalo in the early universe . our dark - matter dominated globular clusters show features which are usually attributed to the action of the tidal field of the host galaxy . among them are the presence of an apparent cutoff ( `` tidal radius '' ) or of a `` break '' in the outer parts of the radial surface brightness profile , and a flat line - of - sight velocity dispersion profile in the outskirts of the cluster . the apparent mass - to - light ratios of our hybrid ( stars dark matter ) globular clusters are very close to those of purely stellar clusters . we suggest that additional observational evidence such as the presence of obvious tidal tails is required to rule out the presence of significant amounts of dark matter in present day globular clusters .

astro-ph0409605

the origin of globular clusters ( gcs ) is one of the important unsolved astrophysical problems . one can divide all proposed scenarios for gc formation into two large categories : 1 ) in situ formation ( when gcs are formed inside their present day host galaxies ) , and 2 ) pregalactic formation ( when gcs are formed in smaller galaxies which later merge to become a part of the present day host galaxy ) . pregalactic scenarios for gc formation can explain many puzzling properties of gcs in the milky way and other galaxies . this is especially true for so - called metal - poor gcs ( with @xmath2\lesssim -1 $ ] ) , which form a distinctive population in many galaxies . many of their properties are highly suggestive of their pregalactic origin : low metallicity , large age comparable to the age of the universe , isotropic orbits with the apocentric - to - pericentric distances ratio of @xmath3 ( * ? ? ? * based on data for 38 metal - poor galactic gcs ) which is consistent with the orbits of dark matter ( dm ) subhalos ( @xmath4 ) in cosmological @xmath5cdm simulations , and a very weak or no correlation with the properties of the host galaxy . one very interesting variation of the pregalactic picture is the primordial scenario of gc formation , which was first proposed by @xcite . in its modern interpretation , the primordial scenario assumes that gcs formed at the center of small dm minihalos in the early universe , before or during the reionization of the universe , which is believed to have ended around the redshift of @xmath6 @xcite . the primordial picture has been considered by many authors , including @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . it is usually assumed that during the reionization of the universe no star formation can take place inside small dm halos with the virial temperature below @xmath7 k ( or virial mass less than @xmath8 @xmath9 ) , as the gas in such halos should escape the shallow gravitational potential after being heated to @xmath7 k by the metagalactic lyman continuum ( lyc ) background @xcite . some authors @xcite suggested though , that the reionization of the universe can actually trigger star formation in small dm minihalos through different positive feedback mechanisms . in the scenario of @xcite , star formation in small gas - rich minihalos is triggered by radially converging radiation shock fronts caused by the external lyc radiation field . in the cosmological simulations with radiative transfer of @xcite , small - halo objects constitute the bulk of mass in stars until at least redshift @xmath10 because of the increased non - equilibrium fraction of free electrons in front of the cosmological h ii regions and inside the relic cosmological h ii regions , which results in more efficient h@xmath11 formation and hence cooling of the gas . despite the fact that many authors considered different hydrodynamic and radiative processes which can lead to formation of gc - like stellar clusters inside dm minihalos , there has been no detailed study on what happens to such hybrid ( gc @xmath1 dm halo ) object after the formation of stars from the point of galactic dynamics . fully consistent cosmological simulations of structure formation in the universe ( such as those of @xcite ) , and even higher resolution `` semi - consistent '' simulations ( like in @xcite ) lack orders of magnitude in spatial and mass resolution to be able to answer the following questions : 1 ) how does the presence of dm modify the observable properties of gcs ? 2 ) are there observable features from which the presence of dm in a gc can be inferred ? 3 ) will dm in `` hybrid gcs '' survive tidal stripping during the hierarchical assemblage of substructure leading to the formation of large galaxies such as milky way ? we try to answer the above questions in a series of two papers . in this first paper , we address the first and the second questions , with the last question being dealt with in the second paper ( * ? ? ? * hereafter paper ii ) . using the @xmath0-body tree - code gadget @xcite , we follow the relaxation of initially non - equilibrium stellar clusters inside live dm minihalos around the redshift of @xmath12 . our simulations are collisionless , and can be directly compared with dynamically unevolved gcs ; the impact of secular evolution ( core collapse ) on our results will be explored in paper ii . dm halos have either ( * ? ? ? * hereafter nfw ) or @xcite profiles , and have structural parameters taken from cosmological simulations . we use proto - gc model of ( * ? ? ? * hereafter ms04 ) to set up the initial non - equilibrium configuration of stellar clusters inside dm halos . in ms04 we showed that the collapse of homogeneous isothermal stellar spheres leads to the formation of clusters with the surface brightness profiles being very similar to those of dynamically young galactic gcs . in this model , all the observed correlations between structural and dynamic parameters of galactic gcs are accurately reproduced if the initial stellar density @xmath13 and velocity dispersion @xmath14 had the same universal values : @xmath15 @xmath9 pc@xmath16 and @xmath17 km s@xmath18 . this paper is organized as follows . in section [ model ] we describe our method of simulating the evolution of a `` hybrid '' gc and list the physical and numerical parameters of our models . in section [ results ] we show the results of the simulations . finally , in section [ discussion ] we discuss the results and give our conclusions . in our model , gcs with a stellar mass @xmath19 are formed at the center of dm halos with the virial mass @xmath20 at the redshift of @xmath12 . we fix the mass ratio @xmath21 to 0.0088 . the adopted value of @xmath22 is between the universal baryonic - to - dm density ratio @xmath23 @xcite and the fraction of baryons in gcs in the modern universe of @xmath24 @xcite . this leaves enough room for such effects as a less than 100% efficiency of star formation in proto - gcs , mass losses due to stellar evolution ( through supernovae and stellar winds ) , a decay of gc systems due to dynamic evolution of the clusters in the presence of tidal fields , and any biased mechanism of gc formation ( e.g. , when gcs are formed only in dm halos with a low specific angular momentum , like in @xcite ) . we consider a flat @xmath5cdm universe ( @xmath25 ) , with the following values of the cosmological parameters : @xmath26 and @xmath27 km s@xmath18 mpc@xmath18 @xcite . in a flat universe , the critical density can be written as @xmath28.\ ] ] the virial radius of a dm halo with a virial mass @xmath20 is then as follows : @xmath29 where the spherical collapse overdensity , @xmath30 , and the matter density of the universe in units of critical density at the redshift of @xmath31 , @xmath32 , are given by @xmath33 and @xmath34^{-1}\ ] ] @xcite . here @xmath35 . we consider two types of dm density profiles : nfw halos and burkert halos . the nfw model describes reasonably well dm halos from cosmological cdm simulations . it has a cuspy inner density profile with a slope of @xmath36 , and a steeper than isothermal outer density profile : @xmath37 where @xmath38^{-1}.\ ] ] here @xmath39 and @xmath40 are the scale radius and concentration of the halo . burkert halos , on the other hand , have a flat core . their outer density profile slope is identical to nfw halos ( @xmath41 ) . this model fits well the rotational curves of disk galaxies @xcite and has the following density profile : @xmath42},\ ] ] where @xmath43^{-1}.\ ] ] the concentration @xmath44 of cosmological halos has a weak dependence on a virial mass ( with less massive halos being more concentrated on average ) . @xcite give the following expression for a concentration of low mass dm halos in @xmath5cdm cosmological simulations at @xmath45 , which was obtained from the analysis of halos with the virial masses of @xmath46 @xmath9 : @xmath47 . combining this expression with the result of @xcite that the concentration scales with a redshift as @xmath48 , we obtained the following formula for a concentration of low mass halos at different redshifts : @xmath49 @xcite showed that cdm halos do not have concentrations smaller than @xmath50 , hence equation ( [ eqc ] ) becomes invalid for halos with @xmath51 . @xcite demonstrated that equation ( [ eqc ] ) describes very well concentrations of four dwarf disk galaxies from @xcite fitted by a burkert profile , so it can be used for both nfw and burkert halos . the initial non - equilibrium configuration of proto - gcs was modeled after ms04 as a homogeneous isothermal stellar sphere with isotropic maxwellian distribution of stellar velocities . stellar clusters of different mass @xmath19 initially had the same universal values of stellar density and velocity dispersion : @xmath52 @xmath9 pc@xmath16 and @xmath53 km s@xmath18 ( ms04 ) . as in ms04 , we use a mass parameter @xmath54 to describe different proto - gc models : @xmath55 where @xmath56 is the gravitational constant . the connection between the initial virial parameter @xmath57 and @xmath54 is @xmath58 , where @xmath59 and @xmath60 are initial kinetic and potential energies of the stellar cluster . we use models from ms04 to plot in figure [ beta_m ] a dependence of the fraction of stars , which form a gravitationally bound cluster after the initial relaxation phase of a proto - gc , on the mass parameter @xmath54 ( or the virial parameter @xmath57 ) . please note that this is the case of no dm ( bare stellar cores ) . as one can see , there are three different regimes of the initial relaxation phase : 1 ) `` hot '' collapse ( @xmath61 or @xmath62 ) results in a substantial loss of stars which initial velocity was above the escape speed for the system ; only the slowest moving stars collapse to form a bound cluster . 2 ) `` warm '' collapse ( @xmath63 or @xmath64 ) is very mild and non - violent , and results in virtually no escapers . 3 ) `` cold '' collapse ( @xmath65 or @xmath66 ) is violent , and leads to increasingly larger fraction of escapers for colder systems . in the last case , the nature of unbinding of stars is very different from the `` hot '' case . during a cold collapse , the radial gravitational potential is wildly fluctuating . at the intermediate stages of the relaxation , the central potential of the cluster becomes much deeper than the final , relaxed value . as a result , a fraction of stars are accelerated to speeds which will exceed the escape speed of the relaxed cluster . analysis shows that most of escapers are the stars that initially were predominantly in the outskirts of the cluster and moving in the outward direction . these three regimes of the initial relaxation of proto - gcs are of very general nature , and are not restricted to the gc formation scenario of ms04 . to illustrate that , let us write down an expression for the virial ratio of a homogeneous isothermal sphere , which is applicable to both pre - starburst gas phase of a forming gc , and the initial stellar configuration after the star burst takes place : @xmath67 here @xmath68 and @xmath69 are the mass and the radius of the system , and @xmath70 is the mean - square speed of either gas molecules or stars . assuming that newly - born stars have the same velocity dispersion as star - forming gas , equation ( [ nueq ] ) suggests three following physical scenarios resulting in our `` hot '' , `` warm '' , and `` cold '' stellar configurations ( in all scenarios we start with an adiabatically contracting gas cloud of a jeans mass , which corresponds to @xmath71 ) : 1 ) non - efficient stars formation with a subsequent loss of the remaining gas leads to our `` hot '' case , 2 ) almost 100%-efficient star formation results in the `` warm '' case , and 3 ) runaway cooling of the whole cloud on the time scale shorter than the free - fall time , leading to high - efficiency star formation , results in our `` cold '' case . cccccccccccccc model & @xmath54&@xmath72 & @xmath19 & @xmath73 & @xmath74&@xmath75&@xmath20 & @xmath44 & @xmath76 & @xmath39 & @xmath77&@xmath78 & @xmath79 + & & & @xmath9 & pc & myr & pc & @xmath9 & & pc & pc & pc & myr & + w@xmath80 & 0.4 & 0.54 & @xmath81 & 11.2 & 0.49 & 2.98 & @xmath82 & 4.88 & 885 & 181 & 395 & 52 & 4.7 + w@xmath83 & 0.4 & 0.54 & @xmath81 & 11.2 & 0.49 & 2.98 & @xmath82 & 4.88 & 885 & 181 & 436 & 61 & 117 + c@xmath80 & 1.4 & 0.12 & @xmath84 & 24.2 & 0.078 & 1.57 & @xmath85 & 4.06 & 1906 & 470 & 892 & 56 & 5.8 + c@xmath83 & 1.4 & 0.12 & @xmath84 & 24.2 & 0.078 & 1.57 & @xmath85 & 4.06 & 1906 & 470 & 997 & 66 & 173 + h@xmath80 & 0.6 & 2.5 & @xmath86 & 5.21 & 17 & 4.14 & @xmath87 & 5.87 & 411 & 70 & 174 & 48 & 3.7 + h@xmath83 & 0.6 & 2.5 & @xmath86 & 5.21 & 17 & 4.14 & @xmath87 & 5.87 & 411 & 70 & 190 & 55 & 78 + in this paper , we consider three following initial stellar configurations , which can be considered to be typical `` hot '' , `` warm '' , and `` cold '' cases ( see figure [ beta_m ] ) : @xmath88 ( model `` h '' , stands for `` hot '' ) , @xmath89 ( model `` w '' , stands for `` warm '' ) , and @xmath90 ( model `` c '' , stands for `` cold '' ) . we study relaxation of stellar cores in either nfw ( suffix `` @xmath91 '' ; e.g. , `` w@xmath80 '' ) or burkert ( suffix `` @xmath92 '' ) dm halos , which makes up a total of 6 different combinations of stellar cores and dm halos . the physical parameters of these 6 models are summarized in table [ tab1 ] . the stellar masses @xmath19 in our models cover the whole range of gc masses ( see table [ tab1 ] ) . the initial stellar radius @xmath73 is much smaller than the scale radius of the dm halo @xmath39 in all the models . the masses of dm halos @xmath20 range from @xmath87 to @xmath85 @xmath9 . from figure 6 of @xcite , these dm halos correspond to @xmath93 fluctuations collapsing at the redshift of @xmath12 . as one can see , halos in this mass range are populous enough at @xmath94 to account for all observed gcs . in all our models , stars initially dominate dm in the core ( @xmath95>1 $ ] , see table [ tab1 ] ) , which can be considered as a desirable property for a star formation to take place . it is instructive to check if the above condition ( @xmath96 ) holds for other plausible values of the gc formation redshift @xmath31 and stellar mass fraction @xmath22 . in figure [ m_min ] we show the redshift dependence of the minimum mass of a stellar core in a dm halo , satisfying the condition @xmath97 , for a few fixed values of @xmath22 0.0025 , 0.0088 ( our fiducial value ) , and 0.20 . both nfw ( thick lines ) and burkert ( thin lines ) cases are shown . as one can see , only in the extreme case of an nfw halo at @xmath98 with @xmath99 fraction of total mass in stars , the minimum mass of a dominant stellar core @xmath100 @xmath9 is approaching the mass range of gcs . the conclusion we arrive at is that , for all plausible values of @xmath31 and @xmath22 , stellar cores with the universal initial density @xmath52 @xmath9 pc@xmath16 and gc masses dominate dm at the center of dm halos . dm halos in our models are assumed to have an isotropic velocity dispersion tensor . to generate @xmath0-body realizations of our dm halos , we sample the following probability density function ( pdf ) @xcite : @xmath101 here @xmath102 and @xmath103 are the relative dimensionless potential and particle energy in units of @xmath104 ( for nfw halos ) and @xmath105 ( for burkert halos ) , @xmath106 is the phase - space distribution function , and @xmath107 is the dimensionless radial coordinate of the particle . the dimensionless potential @xmath102 is equal 1 at the center and 0 in infinity . for nfw halos , @xmath108 . for burkert halos , the corresponding expression is given in appendix [ app1 ] ( eq . [ [ psibur ] ] ) . we use the analytic fitting formulae of @xcite to calculate @xmath106 for nfw model , and derive our own fitting formulae for the burkert profile ( see appendix [ app1 ] ) . our models are truncated at the virial radius @xmath76 ( eq . [ [ rvir ] ] ) : @xmath109 for @xmath110 . as we will show in [ warm_sect ] , this truncation results in an evolution in the outer dm density profile ( with the profile becoming steeper for @xmath111 ) , which should not affect our results . we sample the pdf ( eq . [ [ pdfeq ] ] ) in two steps : 1 ) a uniform random number @xmath112 $ ] is generated . the radial distance @xmath69 is then obtained by solving numerically one of the two following non - linear equations : @xmath113\ ] ] for the nfw profile and @xmath114/2-\arctan r = & \nonumber\\ & = x\left[\ln(1+c)+[\ln(1+c^2)]/2-\arctan c\right]&\end{aligned}\ ] ] for the burkert profile . 2 ) . given @xmath69 , we can calculate @xmath102 . instead of equation ( [ pdfeq ] ) , we can now sample the following pdf : @xmath115 as @xmath106 is a monotonically increasing function and @xmath116 , the following inequality holds : @xmath117 . two uniform random numbers , @xmath118 $ ] and @xmath119 $ ] , are generated . if @xmath120 , we accept the value of the dimensionless energy @xmath103 ; otherwise , we go back to the previous step ( generation of @xmath121 and @xmath103 ) . the velocity module @xmath122 can be obtained from the equation for the total energy of a particle : @xmath123 . finally , two spherical coordinates angles @xmath124 and @xmath125 are generated for both radius and velocity vectors using the following expressions : @xmath126 and @xmath127 . ( here @xmath128 and @xmath129 are random numbers distributed uniformly between 0 and 1 . ) the above method to generate @xmath0-body realizations of either nfw or burkert halos does not use a `` local maxwellian approximation '' to assign velocities to particles . instead , it explicitly uses phase - space distribution functions , which was shown to be a superior way of setting up @xmath0-body models @xcite . stellar cores are set up as homogeneous spheres at the center of dm halos : @xmath130 , where @xmath131 $ ] is a uniformly distributed random number . we use equal mass stellar particles . velocities of the particles have a maxwellian distribution . the components of the radius and velocity vectors are generated in the same fashion as for dm particles . ccccccccc model & @xmath132 & @xmath133 & @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & note + & & pc & & pc & myr & myr & myr & + w@xmath139 & @xmath140 & 0.15 & & & 190 & 370 & 0.3 & 1 + w@xmath80s & @xmath141 & 0.30 & & & 95 & 190 & 0.2 & 2 + w@xmath80 & @xmath141 & 0.30 & @xmath87 & 1.5 & 200 & 250 & 0.2 & + w@xmath83s & @xmath141 & 0.30 & & & 95 & 190 & 0.2 & 2 + w@xmath83 & @xmath141 & 0.30 & @xmath87 & 1.5 & 150 & 250 & 0.2 & + c@xmath139 & @xmath140 & 0.32 & & & 56 & 92 & 0.3 & 1 + c@xmath80 & @xmath140 & 0.32 & @xmath142 & 4.7 & 130 & 190 & 0.2 & + c@xmath83 & @xmath140 & 0.32 & @xmath142 & 5.2 & 95 & 190 & 0.2 & + h@xmath139 & @xmath140 & 0.069 & & & 1300 & 1800 & 0.3 & 1 + h@xmath80 & @xmath141 & 0.15 & @xmath142 & 1.7 & 1500 & 3700 & 20 & + h@xmath83 & @xmath141 & 0.15 & @xmath142 & 1.8 & 1500 & 3700 & 20 & + in addition to 6 models from table [ tab1 ] which consist of a live dm halo plus a stellar core , we also ran 2 more models with a static dm potential ( marked with the letter `` s '' at the end ) . the numerical parameters for these 8 models are listed in table [ tab2 ] . in this table we also list parameters for the three properly rescaled stars - only models from ms04 ( h@xmath139 , w@xmath139 , and c@xmath139 ) , which will be used as reference cases . to run our models , we use a parallel version of the multistepping tree code gadget-1.1 @xcite . the code allows us to handle separately stellar and dm particles , with each species having a different softening length @xmath133 and @xmath135 , respectively . we slightly modified the code by introducing an optional static dm potential ( either nfw , or burkert ) . the values of the softening lengths were chosen to be comparable with the initial average interparticle distance . we used the following expression for the stellar particles : @xmath143 @xcite , where the half - mass radius @xmath144 was measured at the initial moment of time ( so @xmath145 ) . the same expression was used to calculate @xmath146 for h@xmath147 models , whereas for the models w@xmath147 and c@xmath147 we used half of the value given by the equation of @xcite to achieve a better spatial resolution within the stellar core . @xcite showed that collapsing homogeneous stellar spheres with a non - negligible velocity dispersion @xmath14 do not experience a fragmentation instability ( and as a consequence preserve orbital angular momentum of stars ) , if the following condition is met : @xmath148 ^ 3 $ ] , where @xmath132 is the number of stars . for our models , this adiabaticity criterion can be rewritten as @xmath149 as we discussed in ms04 , if real gcs indeed started off as isothermal homogeneous stellar spheres , they should have collapsed adiabatically . we also showed that if both the condition in equation ( [ eqbeta ] ) and @xmath150 are met , the results of simulations of collapsing stellar spheres do not depend on the number of particles @xmath132 and the softening length @xmath133 . ( here @xmath75 is the minimum half - mass radius of the cluster during the collapse . ) by comparing tables [ tab1 ] and [ tab2 ] one can see that the values of @xmath132 and @xmath133 we use in our simulations meet both above conditions . the evolution time @xmath137 in our runs is at least 3 times longer than the crossing time for dm @xmath78 , and hundreds and even thousands times longer than the crossing time for stellar particles @xmath74 ( see tables [ tab1 ] and [ tab2 ] ) . as we will see in the next section , this time is enough for stellar and dm density profiles to converge . on the other hand , it is short enough to avoid significant dynamic evolution in the stellar clusters caused by encounters between individual particles . the individual time steps in the simulations are equal to @xmath151 , where @xmath152 is the acceleration of a particle , and parameter @xmath153 controls the accuracy of integration . we used a very conservative value of @xmath154 and set the maximum possible individual time step @xmath138 value to either 20 or 0.2 myr ( see table [ tab2 ] ) . as a result , the accuracy of integration was very high : @xmath155% , where @xmath156 is the maximum deviation of the total energy of the system from its initial value . one has to keep in mind though that numerical artifacts are the most pronounced in the central densest area , where the frequency of strong gravitational interaction between particles is the highest . we estimate the severity of these effects by looking at the total energy conservation in our purely stellar models from ms04 : in models h@xmath139 , w@xmath139 , and c@xmath139 the values of @xmath157 are @xmath158% , @xmath159 , and @xmath160% ( for the model w@xmath139 we used a larger value of @xmath161 , hence the relatively large errors ) . as you can see , even though the numerical artifacts in the dense stellar cluster are more visible than in the dm @xmath1 a stellar core case , the magnitude of these effects is still reasonably low . for completeness sake , we also give the values of other code parameters which control the accuracy of simulations : errtoltheta=0.6 , typeofopeningcriterion=1 , errtolforceacc=0.01 , maxnodemove=0.05 , treeupdatefrequency=0.1 , and domainupdatefrequency=0.2 . ( please see the code manual for explanation of these parameters ) . the code was compiled with the option dbmax enabled , which allowed it to use a very conservative node - opening criterion . for each of our models , we generated 100200 individual timeframe snapshots . the last 2060% of the snapshots ( corresponding to the time interval @xmath162 , see table [ tab2 ] ) were used to calculate the properties of relaxed models , including radial profiles and different stellar cluster parameters listed in table [ tab3 ] . in all snapshots , we removed in an iterative manner the escapers ( particles whose velocity exceeds the escape velocity @xmath163^{1/2}$ ] , where @xmath164 is the local value of the gravitational potential ) . this procedure affected models from ms04 ( h@xmath139 , w@xmath139 , and c@xmath139 ) , and dm particles only in the models presented here . one of the reasons for removing escapers was to make the results of the simulations directly comparable with the results from paper ii where this procedure is applied to discount particles stripped by tidal forces . the radial profiles shown in this section were obtained by averaging the corresponding profiles from individual snapshots in the time interval @xmath165 . for each profile , we only show the part which is sufficiently converged ( the dispersion between different snapshots is very small ) . for projected quantities ( such as surface brightness @xmath166 and line - of - sight velocity dispersion @xmath167 ) we use the projection method described in appendix [ app2 ] which produces much less noisy radial profiles than a straightforward projection onto one plane or three orthogonal planes . this is crucial for the analysis of the properties of the outer , low density parts of stellar clusters , which is the main purpose of this study . the virial ratio @xmath57 for our models is consistent with unity within the measurement errors at the end of simulations . all the global model parameters ( listed in table [ tab3 ] ) converge to their final values by @xmath168 . similarly , radial density and velocity dispersion profiles converge to their final form over a wide range of radial distances . we explicitly checked that the evolution time @xmath137 is more than three crossing times at the largest radius for stellar density profiles shown in figures [ rho_w ] , [ rho_c ] , and [ rho_h ] . all the above let us conclude that the results presented here are collisionless steady - state solutions . in the absence of dm , an isothermal homogeneous stellar sphere with the mass parameter @xmath89 relaxes to its equilibrium state with virtually no stars lost ( see figure [ beta_m ] ) . this makes it an interesting case to test the result of @xcite , that a stellar cluster inside a static constant density dm halo can acquire a radial density cutoff similar to that expected from the action of tidal forces of the host galaxy , for the case of live dm halos with cosmologically relevant density profiles and initially non - equilibrium stellar cores . in figure [ rho_w ] we show the radial density profiles for relaxed stellar core and dm halo , both for nfw model ( panel _ a _ ) and burkert model ( panel _ b _ ) , for the case of a warm collapse ( `` w '' models ) . here long - dashed lines show the density profiles for stars in the absence of dm ( thin lines ) and dm in the absence of stars ( thick lines ) . solid lines , on the other hand , show the density profiles for stars and dm having been evolved together . we also show as dotted lines the stellar radial density profiles for the case of a static dm potential . as you can see from figure [ rho_w ] , in the absence of dm ( model w@xmath139 ) , the relaxed stellar cluster has a central flat core and a close to power - law outer density profile . the radial density profile has also a small `` dent '' outside of the core . it is not a transient feature ( in the collisionless approximation ) , as the virial ratio for the whole cluster is equal to unity within the measurements errors ( @xmath169 for @xmath170 ) and the evolution time is @xmath171 crossing times even at the very last radial density profile point . as we show in paper ii , collisional long - term dynamic evolution of a cluster will remove this feature , bringing the overall appearance of the density profile closer to that of the majority of observed gcs . the central stellar density in the model w@xmath139 is @xmath172 three orders of magnitude larger than the central density of dm in the undisturbed burkert halo , and becomes comparable to the density of the undisturbed nfw halo only at very small radii @xmath173 pc ( which is outside of the range of radial distances in figure [ rho_w]a ) . in the case of a live dm halo coevolving with the stellar core , the dm density profile is significantly modified within the central area dominated by stars ( see figure [ rho_w ] ) . in both nfw and burkert halos , dm is adiabatically compressed by the potential of the collapsing stellar cluster to form a steeper slope in the dm radial density profile . the slope of the innermost part of the dm density profile is @xmath174 for the nfw halo and @xmath175 for the burkert halo . the lack of resolution does not allow us to check if the slope stays the same closer to the center of the dm halos . in the outer parts of the dm halos ( beyond the scale radius @xmath39 , which is marked by vertical dotted lines in figure [ rho_w ] ) , the dm density profile becomes steeper than the original slope of @xmath41 . this can be explained by the fact that our dm halo models are truncated at a finite radius @xmath176 and hence are not in equilibrium . it should bear no effect on our results , as in all our models stellar clusters do not extend beyond @xmath39 . as you can see in figure [ rho_w ] , the impact of the presence of a live dm halo on the stellar density profile is remarkably small ( especially for the burkert halo ) . in the presence of dm , the central stellar density becomes somewhat larger ( by @xmath177% and @xmath178% for the nfw and burkert halos , respectively ; see table [ tab3 ] ) . the most interesting effect is observed in the outer parts of the stellar density profiles , where the slope of the profiles becomes significantly larger than that of a purely stellar cluster , which is similar to the result of @xcite . the density profile starts deviating from the profile for the model w@xmath139 somewhere between the radius @xmath179 and the radius @xmath180 ( where the enclosed masses for stars and dm become equal , see table [ tab3 ] ) . there is a simple explanation for the steepening of the stellar density profile seen in figure [ rho_w ] around the radius @xmath180 . in the absence of dm , warm collapse of a homogeneous stellar sphere ( our model w@xmath139 ) is violent enough to eject a number of stars into almost radial orbits , forming a halo of the relaxed cluster . in the case when the dm halo is present , dynamics does not change much near the center of the cluster where stars are the dominant mass component . around the radius @xmath180 the enclosed dm mass becomes comparable to that of stars . as a result , gravitational deceleration , experienced by stars ejected beyond @xmath180 , more than doubles , resulting in a significantly smaller number of stars populating the outer stellar halo at @xmath181 this should lead to significantly steeper radial density profile beyond @xmath180 , as observed in our w@xmath147 models . as in @xcite model , in our model the steepening of the stellar density profile is caused by the fact that at large radii the gravitational potential is dominated by dm . the principal difference between the two models is that in @xcite the stellar cluster is assumed to be isothermal and in equilibrium ; our clusters are not in equilibrium initially , and their final equilibrium configuration is not isothermal ( stars are dynamically colder in the outskirts of the cluster ) . to allow a direct comparison of the model results with observed gcs , in figure [ proj_w ] we plot both the surface brightness @xmath182 profile ( panel _ a _ ) and the line - of - sight velocity dispersion @xmath167 profile ( panel _ b _ ) for warm collapse models . the surface brightness @xmath166 in units of mag arcsec@xmath183 is calculated using the following formula : @xmath184 . \label{eqsigma}\ ] ] here @xmath185 is the absolute @xmath186-band magnitude of the sun , @xmath187 is the projected surface mass density in units of @xmath9 pc@xmath183 , and @xmath188 is the assumed @xmath186-band mass - to - light ratio for baryons in gcs . for @xmath188 we adopt the observationally derived averaged value of 1.45 from @xcite , which is also between the two stellar - synthesis model values ( 1.36 for the salpeter and 1.56 for the composite with a zero slope for stellar masses @xmath189 @xmath9 initial mass functions ) of @xcite . as you can see in figure [ proj_w ] , the presence of a dm halo introduces an outer cutoff in the surface brightness radial profile of stellar cluster , which makes the profile look very similar to a king model profile . interestingly , stellar cluster evolving in a live nfw dm halo acquires even steeper outer density and brightness profiles than for the case of a static dm potential ( see figures [ rho_w]a and [ proj_w]a ) . analysis of the line - of - sight velocity dispersion profiles for warm collapse models ( figure [ proj_w]b ) shows that the presence of a dm halo slightly inflates the value of @xmath167 in the core of the cluster ( by @xmath190% , see table [ tab3 ] ) . the outer dispersion profile either stays unchanged ( as in the case of the burkert halo ) , or becomes slightly steeper ( for the nfw model ) . the static dm models present profiles of an intermediate type . a slight increase of @xmath167 in the central part of the cluster can be misinterpreted observationally as a gc with no dm which has a bit larger value of the baryonic mass - to - light ratio @xmath191 . to quantify this effect , we apply a core fitting ( or king s ) method of finding the central value of @xmath191 in spherical stellar systems @xcite : @xmath192 here @xmath193 is the central surface brightness in physical units . we assume here that @xmath194 , where @xmath195 is the projected surface mass density at the center of the model cluster . the core fitting method assumes that the mass - to - light ratio is independent of radius , which is obviously not true for our stars @xmath1 dm models . in our models the stellar particles have the same mass , so there is no radial mass segregation between high and low mass stars caused by the long - term dynamic evolution of the cluster . as a result , the equation ( [ equpsilon ] ) should not be used to compare our models with real gcs . instead , we use it to check if there is a change in the apparent mass - to - light ratio @xmath191 between our purely stellar models and the models containing dm . we list the values of @xmath191 for different models in table [ tab3 ] . as you can see , in the case of nfw halo ( model w@xmath80 ) , the presence of the dm halo leads to @xmath196% increase in the value of the apparent mass - to - light ratio . such a small increase is well within the observed dispersion of @xmath191 values for galactic gcs @xcite . in the case of the burkert halo , the apparent central mass - to - light ratio is only @xmath197% larger than for the purely stellar model . a principal difference between warm and cold collapse models from ms04 is that in the cold collapse case a significant fraction of stars becomes unbound after the initial relaxation phase ( see figure [ beta_m ] ) . in our model c@xmath139 with the mass parameter @xmath90 the fraction of the stars lost is 30% . the radial density profile for c@xmath139 model is similar to that of w@xmath139 model ( see figure [ rho_c ] ) . one can intuitively expect the escapers from the model c@xmath139 to be trapped inside our models containing dm , forming a distinctive feature in the outer density and velocity dispersion profiles . as you can see in figures [ rho_c ] and [ proj_c ] , our cold collapse models do show such features . both density and surface brightness profiles become more shallow in the outer parts of the stellar clusters where dm dominates stars . this is valid for both nfw and burkert halos . for the nfw and burkert cases , the slope of the outer stellar density profile is @xmath198 and @xmath199 , respectively . for the purely stellar case ( model c@xmath139 ) the slope is @xmath200 , so the relative change in the slope is @xmath201 and @xmath202 for the nfw and burkert cases . this behavior is mimicked by the corresponding surface brightness profiles ( figure [ proj_c]a ) . it is interesting to note that , for the c@xmath80 model the radial @xmath166 profile exhibits a significant steepening of the slope in the outmost parts of the cluster , creating an appearance of a `` tidal cutoff '' ( similarly to the warm collapse cases ) . even more pronounced are features in the line - of - sight velocity dispersion profiles ( figure [ proj_c]b ) . in both nfw and burkert cases , there is a plateau in the radial @xmath167 profiles around the radius @xmath180 where dm becomes the dominant mass component . the apparent `` break '' in the surface brightness profile and accompanying flattening of the radial @xmath167 profile , seen in our models containing dm , can be misinterpreted observationally as a presence of `` extratidal '' stars heated by the tidal field of the host galaxy . similarly to warm collapse models , the apparent mass - to - light ratio @xmath191 for cold collapse models in the presence of a live dm halo is very close to the purely stellar case c@xmath139 ( see table [ tab3 ] ) . interestingly , the half - mass radii for warm and cold collapse models with dm are almost identical . this is consistent with the properties of the observed gcs , which have comparable half - mass radii for a wide range of cluster masses . dm density profiles for the cold collapse cases ( figure [ rho_c ] ) exhibit a very similar behavior to the warm collapse models in the stars dominated central region . for both nfw and burkert halos , the innermost slopes of the density profiles become steeper ( @xmath203 and @xmath204 , respectively ) in the presence of a stellar core . the nfw dm density profile shows a break around the radius @xmath180 . as you can see in table [ tab3 ] and figure [ rho_c ] , the final stellar cluster half - mass radius is smaller than the dm softening length @xmath205 pc . we reran models c@xmath80 and c@xmath83 with much smaller value of @xmath206 pc to test the possibility that our results were influenced by the fact that dm is not resolved on the stellar cluster scale . for both c@xmath80 and c@xmath83 , the relaxed stellar profiles are found to be virtually identical to the cases with larger @xmath135 . in particular , a `` kink '' seen in figures [ rho_c]a and [ rho_c]b around the radius @xmath135 is also present at the same location in the simulations with much smaller value of @xmath135 , and is definitely not a numerical artifact . in the case of nfw halo , the central stellar velocity dispersion becomes slightly larger , which results in somewhat larger value of @xmath207 . this could be in part because of artificial mass segregation , which should be more pronounced in the case of @xmath135 being significantly lower than the optimal value ( in our cold collapse models , dm particles are @xmath208 times more massive than stellar particles ) . as a result , the actual @xmath191 value could be even closer to the baryonic value . in the case of burkert halo with @xmath206 pc , the core mass - to - light ratio @xmath209 , which is identical to the case of no dm . we conclude that our choice of @xmath135 did not affect the main results presented in this section . it is generally assumed that in an equilibrium star - forming cloud no bound stellar cluster will be formed after stellar winds and supernova explosions expel the remaining gas if the star formation efficiency is less than 50% ( which corresponds to the initial virial parameter for the stellar cluster of @xmath210 ) . in ms04 we showed that it is not true for initially homogeneous stellar clusters which have a maxwellian distribution of stellar velocities . in such clusters with the initial virial parameter as large as 2.9 ( corresponding to the mass parameter @xmath54 as low as @xmath211 ) , a bound cluster , containing less than 100% of the total mass , is formed by the slowest moving stars . ( a similar conclusion was reached by @xcite for initially polytropic stellar spheres . ) in our model h@xmath139 ( @xmath88 , @xmath212 ) , @xmath213% of all stars stay gravitationally bound after the initial relaxation phase ( see figure [ beta_m ] ) . because in both cold and hot collapse models for purely stellar clusters a significant fraction of stars become unbound , one might expect that stellar density and velocity dispersion profiles to be similar for these two cases in the presence of dm . analysis of figures [ rho_h ] and [ proj_h ] shows that it is definitely not the case : h@xmath147 models show completely different behavior from c@xmath147 models . the explanation is simple : c@xmath139 model starts with the stellar density which is already larger than the density of dm within the stellar core in models c@xmath147 ( in other words , @xmath96 , see table [ tab1 ] ) , and after the relaxation phase reaches even higher density which is @xmath214 times higher than the initial density . the hot collapse model h@xmath139 , on the other hand , having started with @xmath96 , acquires a density which is @xmath215 times lower than the initial density after the expansion and relaxation phase , resulting in @xmath216 for the relaxed stellar cluster even for the burkert halo . under these circumstances , radial profiles for h@xmath80 and h@xmath83 models should be significantly different from the case with no dm ( h@xmath139 ) , which is what is seen in figures [ rho_h ] and [ proj_h ] . the only common feature between the models c@xmath147 and h@xmath147 is that in both cases all would - be escapers are trapped by the potential of the dm halo . in the case of the nfw halo , central stellar density and central velocity dispersion become larger in the presence of dm by factors of 35 and 3 , respectively . in the burkert halo case , the increase is not as dramatic , but still significant : central density and velocity dispersion become 7 and 2 times larger than those in the absence of dm . interestingly , as in the cold collapse cases , the presence of dm brings the half - mass radius of the stellar core in h@xmath147 models closer to the half - mass radius of the warm collapse models ( see table [ tab3 ] ) . the apparent central mass - to - light ratio @xmath191 is noticeably inflated by the presence of dm ( especially for the case of nfw halo ) by 76% and 36% for nfw and burkert models , respectively . as you can see in figure [ rho_h]a , in the model h@xmath80 stars have a comparable density to dm in the core , and become less dense than dm outside of the core radius @xmath217 . the situation is different ( and more in line with warm and cold collapse models ) for the h@xmath83 model ( figure [ rho_h]b ) , where stars are the dominant ( though not by a large margin ) mass component in the core . there are no obvious features in the density and surface brightness profiles caused by the presence of dm : the profiles look very similar to w@xmath139 and c@xmath139 models which do not have dm . similarly to warm and cold collapse cases , the presence of a stellar cluster makes dm denser in the central area . in the case of nfw halo , the innermost slope of the dm density profile is comparable to the slope of @xmath36 for undisturbed halo . for the burkert halo , the slope becomes steeper , reaching the value of @xmath218 at the innermost resolved point . the most interesting behavior is exhibited by the line - of - sight velocity dispersion profiles ( figure [ proj_h]b ) . in the presence of dm , the profiles become remarkably flat , changing by mere 0.2 0.3 dex over the whole range of radial distances ( out to @xmath219 apparent half - mass radii ) . the combination of the low mass , low central surface brightness , large apparent mass - to - light ratio , and almost flat velocity dispersion profiles seen in models h@xmath147 can be mistakingly interpreted as a gc observed at the final stage of its disruption by the tidal forces of the host galaxy . a significant hindrance to a wider acceptance of the primordial scenarios for gc origin is an apparent absence of dm in galactic gcs . many observational facts have been suggested to be evidence for gcs having no dm , including the presence of such features in the outer parts of the gc density profiles as apparent tidal cutoffs or breaks , relatively low values of the apparent central mass - to - light ratio @xmath191 which are consistent with purely baryonic clusters , flat radial distribution of the line - of - sight velocity dispersion in the outskirts of gcs believed to be a sign of tidal heating , and non - spherical shape of the clusters in their outer parts . here we present the results of simulations of stellar clusters relaxing inside live dm minihalos in the early universe ( at @xmath12 ) . we study three distinctly different cases which can correspond to very different gas - dynamic processes forming a gc : a mild warm collapse , a violent cold collapse resulting in a much denser cluster with a significant fraction of stars escaping the gc in the absence of dm , and a hot collapse resulting in a lower density cluster with many would - be stars - escapers . we show that gcs forming in dm minihalos exhibit the same properties as one would expect from the action of the tidal field of the host galaxy on a purely stellar cluster : king - like radial density cutoffs ( for the case of a warm collapse ) , and breaks in the outer parts of the density profile accompanied by a plateau in the velocity dispersion profile ( for a cold collapse ) . also , the apparent mass - to - light ratio for our clusters with dm is generally close to the case of a purely stellar cluster . ( the special case of a hot collapse inside a dm halo which produces inflated values of @xmath191 can be mistaken for a cluster being at its last stage of disruption by the tidal forces . ) we argue that increasingly eccentric isodensity contours observed in the outskirts of some gcs could be created by a stellar cluster relaxing inside a triaxial dm minihalo , and not by external tidal fields as it is usually interpreted . indeed , cosmological dm halos are known to have noticeably non - spherical shapes ; a stellar cluster relaxing inside such a halo would have close to spherical distribution in its denser part where the stars dominate dm , and would exhibit isodensity contours of increasingly larger eccentricity in its outskirts where dm becomes the dominant mass component . it is also important to remember that few galactic gcs show clear signs of a `` tidal '' cutoff in the outer density profiles @xcite . the `` tidal '' features of an opposite nature `` breaks '' in the outer parts of the radial surface brightness profiles in some gcs are often observed at or below the inferred level of contamination by foreground / background objects , and could be in many cases an artifact of the background subtraction procedure which relies heavily on the assumption that the background objects are smoothly distributed across the field of view . a good example is that of draco dwarf spheroidal galaxy . @xcite used simple non - filtered stellar counts from photographic plates followed by a background subtraction procedure to show that this galaxy appears to have a relatively small value of its radial density `` tidal cutoff '' @xmath220 arcmin and a substantial population of `` extratidal '' stars . @xcite used a more advanced approach of multi - color filtering of draco stars from the sloan digital sky survey images and achieved much higher signal - to - noise ratio than in @xcite . new , higher quality results were supposed to make the `` extratidal '' features of draco much more visible . instead , @xcite demonstrated that the draco s radial surface brightness profile is very regular down to a very low level ( 0.003 of the central surface brightness ) , and suggested a larger value for the king tidal radius of @xmath221 arcmin . we argue that the qualitative results presented in this paper are very general , and do not depend much on the the fact that we used ms04 model to set up the initial non - equilibrium stellar core configurations , and on the particular values of the model parameters ( such as @xmath13 , @xmath14 , and @xmath22 ) . as we discussed in section [ results ] , the appearance of `` tidal '' or `` extratidal '' features in our warm and cold collapse models is caused by two reasons : ( 1 ) at radii @xmath181 the potential is dominated by dm , whereas in the stellar core the potential is dominated by stars from the beginning till the end of simulations , and ( 2 ) the collapse is violent enough to eject a fraction of stars beyond the initial stellar cluster radius . as we showed in [ physical ] , for the initial stellar density @xmath222 @xmath9 pc@xmath16 ( from ms04 ) , the whole physically plausible range of stars - to - dm mass ratios @xmath22 and the gc formation redshifts @xmath31 satisfy the above first condition . for warm and cold collapse , this condition can be reexpressed as @xmath223>1 $ ] . one can easily estimate @xmath79 for other values of @xmath13 . the second condition is more difficult to quantify . in ms04 we showed that collapsing homogeneous isothermal spheres produce extended halos for any values of the initial virial ratio @xmath57 ( except for @xmath224 systems which are too hot to form a bound cluster in the absence of dm ) . @xcite simulated cold collapse for a wider spectrum of initial cluster configurations , including power - law density profile , clumpy and rotating clusters . in all their simulations an extended halo is formed after the initial violent relaxation . it appears that in many ( probably most ) stellar cluster configurations , which are not in detailed equilibrium initially , our second condition can be met . of course , not every stellar cluster configuration will result in a gc - like object , with a relatively large core and an extended halo , after the initial violent relaxation phase . @xcite demonstrated that a warm ( @xmath225 ) collapse of a homogeneous isothermal sphere produces clusters with large cores ( their models d and g ) . adiabatic collapse of homogeneous isothermal spheres produces clusters which surface density profiles are very close to those of dynamically young galactic gcs ( ms04 ) . @xcite concluded that any cold stellar system which does not contain significant inhomogeneities relaxes to a large - core configuration . the results of @xcite can also be used to estimate the importance of adiabaticity for core formation . indeed , their initially homogeneous models h and g span a large range of @xmath57 , and include both adiabatic cases ( @xmath226 for their number of particles @xmath227 , from eq.[[eqbeta ] ] ) and non - adiabatic ones ( @xmath228 ) . it appears that all their models ( adiabatic and non - adiabatic ) form a relatively large core . the issue is still open , but it appears that the adiabaticity requirement ( our eq.[[eqbeta ] ] ) is not a very important one for our problem . ( though this requirement is met automatically for real gcs for the values of @xmath13 and @xmath14 derived in ms04 . ) an important point to make is that the simulations presented in this paper describe the collisionless phase of gc formation and evolution , and can not be directly applied to gcs which have experienced significant secular evolution due to encounters between individual stars . in ms04 we showed that such collisionless simulations of purely stellar clusters describe very well the surface brightness profiles of dynamically young galactic gcs ( such as ngc 2419 , ngc 5139 , ic 4499 , arp 2 , and palomar 3 see fig . 1 in ms04 ) . in paper ii we will address ( among other things ) the issue of long - term dynamic evolution of hybrid gcs . we will demonstrate that , at least for the warm - collapse case , secular evolution does not change our qualitative results presented in this paper . in the light of the results presented in this paper and the above arguments , we argue that additional observational evidence is required to determine with any degree of confidence if gcs have any dm presently attached to them , or if they are purely stellar systems truncated by the tidal field of the host galaxy . a decisive evidence would be the presence of obvious tidal tails . a beautiful example is given by palomar 5 @xcite , where tidal tails were observed to extend over @xmath229 in the sky . even if a gc cluster is proven not to have a significant amount of dm , it does not preclude it having been formed originally inside a dm minihalo . in `` semi - consistent '' simulations of @xcite of a dwarf galaxy formation , proto - gcs were observed to form inside dm minihalos , with the dm being lost during the violent relaxation accompanying the formation of the dwarf galaxy . unfortunately , their simulations did not have enough resolution to clarify the fate of dm in proto - gcs . in paper ii we will address the issue of the fate of dm in hybrid proto - gcs experiencing severe tidal stripping in the potential of the host dwarf galaxy . the dimensionless potential @xmath102 of burkert halos with the density profile given by equations ( [ rho_b])([rho0_b ] ) is as follows : @xmath230 + \frac12 ( 1-r^{-1})\ln ( 1+r^2)\right\}.\ ] ] here @xmath107 is the radial distance in scale radius units , and @xmath102 is in @xmath231 units . the potential @xmath102 is equal to 1 at the center of the halo , and zero in the infinity . phase - space distribution function for a burkert halo with an isotropic dispersion tensor can be derived through an abel transform ( * ? ? ? * ; we skip the second part of the integrant , which is equal to zero for burkert halos ) : @xmath232 here @xmath103 is the relative energy in the same units as @xmath102 and @xmath233 is the density in @xmath234 units , which results in @xmath235 units for @xmath236 . the function @xmath237 for a burkert halo has the same asymptotic behavior for @xmath238 and @xmath239 as the model i of @xcite , which has the density profile @xmath240 : @xmath241 this allowed us to use the following analytic fitting formula of @xcite for the distribution function of burkert halo : @xmath242 here @xmath243 is a polynomial introduced to improve the fit . we solve equation ( [ fe ] ) numerically for the interval of radial distances @xmath244 . we obtained the following values for the fitting coefficients in equation ( [ wid ] ) : @xmath245 we checked the accuracy of the fitting formula ( [ wid ] ) with the above values of the fitting coefficients by comparing the analytic density profile of the burkert halo ( eq . [ [ rho_b ] ] ) with the density profile derived from the distribution function @xcite : @xmath246 the deviation of @xmath247 in equation ( [ rho_r ] ) , with @xmath106 given by equations ( [ wid])([fit ] ) , from the analytic density profile was found to be less than 0.7% for the interval of the radial distances @xmath244 . for @xmath0-body models of spherically symmetric stellar systems , such as gc models presented in the paper , one can produce radial profiles of projected observable quantities ( such as surface mass density or velocity dispersion ) by averaging these quantities over all possible rotations of the cluster relative to the observer . radial profiles created in this way preserve maximum information , which is very important for studies of the outskirts of the clusters and for clusters simulated with relatively small number of particles . let us consider an @xmath248-th stellar particle @xmath249 in a spherically symmetric cluster , located at the distance @xmath250 from the center of the cluster ( figure [ scheme ] ) . we want to find its averaged contribution to one projection bin , corresponding to the range of projected distances @xmath251 . for an arbitrarily rotated cluster , the probability @xmath252 to detect our particle inside the bin ( the shaded areas in figure [ scheme ] ) is equal to the fraction of the surface area of the sphere , containing the particle @xmath249 , cut out by two concentric cylinders with the radii @xmath253 and @xmath254 . it is easy to show that this probability is so the probability @xmath252 serves as a weight for scalar variables . for example , to calculate the square of the projected three - dimensional velocity dispersion @xmath261 , corresponding to the projected distances range @xmath256 , one has to set @xmath262 in equation ( [ eqq ] ) , where @xmath263 are the velocity components of the @xmath248-th particle . observationally , one can determine @xmath261 by measuring both line - of - sight velocities and two proper motion components for stars located within the interval @xmath256 of projected radial distances . one can show that the above method is superior to a straightforward projection onto one plane or three orthogonal planes by considering a contribution made by one particle to the bin @xmath256 . in case of the straightforward projection , there is a high probability that the particle will make no contribution to the projection bin . in our method , every particle with @xmath264 will make a contribution to the bin , which will make the radial profile of the projected quantity significantly less noisy . central value of the projected quantity is obtained by setting @xmath253 to zero in equations ( [ w_i])([eqq ] ) . in this case , all particles make contribution to the averaged value of the quantity . the situation with vector quantities is a bit more complicated . we designed a simplified procedure which is based on the assumption that the bin sizes are small ( so the number of bins is large ) . each particle with an associated vector quantity ( say , velocity vector ) is first rotated around the @xmath265 axis to bring the particle in the plane @xmath266 . next , the particle is rotated around the axis @xmath267 to make the projected radial distance of the particle equal to the radial distance of the center of the bin @xmath268 . ( the observer is assumed to be at the end of the axis @xmath265 . ) the detailed procedure is as follows . for a particle with coordinates @xmath269 , after the two rotations the associated vector @xmath270 is transformed into vector @xmath271 with the following components : if vector @xmath270 is velocity vector , then substituting @xmath285 in place of @xmath286 in equation ( [ eqq ] ) will give us the value of the square of the line - of - sight velocity dispersion @xmath167 , averaged over all possible rotations of the cluster , for the bin @xmath256 . similarly , replacing @xmath286 with @xmath287 and @xmath288 will produce the averaged values of the square of the radial and tangential components of the proper motion dispersion , respectively . we would like to thank volker springel for his help with introducing an external static potential in gadget . s. m. is partially supported by sharcnet . the simulations reported in this paper were carried out on mckenzie cluster at the canadian institute for theoretical astrophysics . aarseth , s. j. , lin , d. n. c. , & papaloizou , j. c. b. 1988 , , 324 , 288 barkana , r. & loeb , a. 1999 , , 523 , 54 barkana , r. & loeb , a. 2001 , physics reports , 349 ( issue 2 ) , 125 binney , j. , & tremaine , s. 1987 , galactic dynamics ( princeton : princeton univ . press ) beasley , m. a. , kawata , d. , pearce , f. r. , forbes , d. a. , & gibson , b. k. 2003 , , 596 , l187 becker , r. h. , et al . 2001 , , 122 , 2850 boily , c. m. & kroupa , p. 2003 , , 338 , 665 bromm , v. & clarke , c. j. 2002 , , 566 , l1 bullock , j. s. , kolatt , t. s. , sigad , y. , somerville , r. s. , kravtsov , a. v. , klypin , a. a. , primack , j. r. , & dekel , a. 2001 , , 321 , 559 burkert , a. 1995 , , 447 , l25 cen , r. 2001 , , 560 , 592 dinescu , d. i. , girard , t. m. , & van altena , w. f. 1999 , , 117 , 1792 hayashi , e. , navarro , j. f. , taylor , j. e. , stadel , j. , & quinn , t. 2003 , , 584 , 541 irwin , m. & hatzidimitriou , d. 1995 , , 277 , 1354 kazantzidis , s. , magorrian , j. , & moore , b. 2004 , , 601 , 37 mashchenko , s. & sills , a. 2004a , , 605 , l121 ( ms04 ) mashchenko , s. & sills , a. 2004b , , in press ( preprint astro - ph/0409606 ) ( paper ii ) mateo , m. , olszewski , e. w. , vogt , s. s. , & keane , m. j. 1998 , , 116 , 2315 mclaughlin , d. e. 1999 , , 117 , 2398 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 odenkirchen , m. , et al . 2001 , , 122 , 2538 odenkirchen , m. , et al . 2003 , , 126 , 2385 padoan , p. , jimenez , r. , & jones , b. 1997 , , 285 , 711 peebles , p. j. e. 1984 , , 277 , 470 peebles , p. j. e. & dicke , r. h. 1968 , , 154 , 891 pryor , c. & meylan , g. 1993 , asp conf . ser . 50 : structure and dynamics of globular clusters , 357 richstone , d. o. & tremaine , s. 1986 , , 92 , 72 ricotti , m. , gnedin , n. y. , & shull , j. m. 2002 , , 575 , 49 rosenblatt , e. i. , faber , s. m. , & blumenthal , g. r. 1988 , , 330 , 191 roy , f. & perez , j. 2004 , , 348 , 62 salucci , p. & burkert , a. 2000 , , 537 , l9 spergel , d. n. , et al . 2003 , , 148 , 175 spitzer , l. j. & thuan , t. x. 1972 , , 175 , 31 springel , v. , yoshida , n. , & white , s. d. m. 2001 , new astronomy , 6 , 79 sternberg , a. , mckee , c. f. , & wolfire , m. g. 2002 , , 143 , 419 trager , s. c. , king , i. r. , & djorgovski , s. 1995 , , 109 , 218 widrow , l. m. 2000 , , 131 , 39 zhao , d. h. , jing , y. p. , mo , h. j. , & b " orner , g. 2003 , , 597 , l9 zhao , d. h. , mo , h. j. , jing , y. p. , & b " orner , g. 2003 , , 339 , 12

we perform a study of the standard model ( sm ) fit to the mixing quantities , and in order to bound contributions of new physics to mixing . we then use this to explore the branching fraction of in certain models of new physics ( np ) . in most cases , this constrains np amplitudes for to lie below the sm component .

1102.0009

we report here on a study of new physics ( np ) predictions for @xmath4 . the standard model ( sm ) prediction for @xmath5 is currently smaller than the experimental branching fraction limit @xcite of @xmath6 by about a factor of @xmath7 . this presents a window of opportunity for observing new physics ( np ) effects in this mode . this topic is particularly timely in view of experimental indications of np effects in both the exclusive decay @xmath8 @xcite ( for recent cdf results , also see ref . @xcite ) as well as the inclusive like - sign dimuon asymmetry observed in @xmath9 @xcite . moreover , future work at lhc - b , @xmath10 super b - factories and ongoing cdf @xmath11 d0 measurements at fermilab ( see the discussion following eq . ( [ bmm ] ) ) is expected to markedly improve the current branching fraction bound . our strategy in this paper is somewhat reminiscent of our recent study @xcite noting that in some np models the @xmath12 mixing and @xmath13 decay amplitudes have a common dependence on the np parameters . if so , one can predict the @xmath13 branching fraction in terms of the observed @xmath14 provided that much or all of the mixing is attributed to np . this is a viable possibility for @xmath12 mixing because the standard model ( sm ) signal has large theoretical uncertainties and because many np models can produce the observed mixing @xcite . for @xmath0 the situation is very different . here , the sm prediction is in accord with the observed value ( _ e.g. _ see refs . and papers cited therein ) . in fact , the analysis described below ( _ cf . _ see eqs . ( [ av - np]),([av - onesig ] ) ) gives @xmath15 , which demonstrates just how well the sm prediction agrees with the experimental value of @xmath0 . in view of this , our sm expression for @xmath0 will be given at nlo whereas lo results will suffice for np models . as regards the corresponding width difference @xmath16 , the experimental and theoretical uncertainties are still rather significant ( _ viz _ sect . ii - c ) . in those np models where mixing and @xmath5 arise from a common set of parameters , the severe constraint on any np signal to @xmath2 mixing places strong bounds on its contribution to @xmath17 . in fact , we shall find the constraint can be so strong that for some np models the predicted @xmath5 branching fraction lies well below the sm prediction . the first step in our study ( _ cf _ section ii ) will be to revisit the sm predictions for mixing in the @xmath18-quark system by using up - to - date inputs . we carry this out for the two mixing quantities @xmath0 and @xmath1 . the former in turn yields phenomenological bounds on np mixing contributions which in certain models can be used to bound the magnitude of the @xmath4 decay mode . we also update the sm branching fraction for @xmath4 by using the observed @xmath2 mixing as input . then , in section iii we discuss general properties of np models with tree - level amplitudes . in section iv , we explore various np models such as extra @xmath19 bosons , family symmetry , r - parity violating supersymmetry , flavor - changing higgs models , and models with the fourth sequential generation . our concluding remarks appear in section v , and some technical details are relegated to the appendix . l|l @xmath20 mev @xcite & @xmath21 s @xcite + @xmath22 & @xmath23 @xcite + @xmath24 @xcite & @xmath25 @xcite + @xmath26 @xcite & @xmath27 @xcite + @xmath28 gev @xcite & @xmath29 mev @xcite + @xmath30 @xcite & @xmath31 @xcite + .05 in we begin by considering the sm predictions for @xmath2 mixing . this step is crucial to obtaining bounds on np contributions . we also use the @xmath2 mixing signal as input to a determination of the branching fraction for @xmath5 . the work in this section takes advantage of recent progress made in determining several quantities used in the analysis . we summarize our numerical inputs in table i , along with corresponding references . included in table i is an updated determination of the top quark pole mass @xcite @xmath32 which in turn is used to determine the corresponding running mass @xmath33 @xcite along with several decay constants and b - factors as evaluated in lattice qcd . for definiteness , we have used values appearing in ref . @xcite . this area is , however , constantly evolving and one anticipates further developments in the near future @xcite . our values for the cabibbo - kobayashi - maskawa ( ckm ) matrix elements @xmath34 and @xmath35 are taken from ref . similar values occur for the global fits cited elsewhere ( _ e.g. _ refs . ) . the pdg value for @xmath0 , m_b_s^(expt ) = ( 117.0 0.8 ) 10 ^ -13 gev , [ mexpt ] is a very accurate one the uncertainty amounts to about @xmath36% . the nlo sm formula , m_b_s^(sm ) = 2 g_f^2 m_w^2 m_b_s f^2_b_sb_b_s 12 ^2 |v_ts^ * v_tb|^2 _ b_s s_0(|x_t ) , [ delmsm ] is arrived at from an operator product expansion of the mixing hamiltonian . the short - distance dependence in the wilson coefficient appears in the scale - insensitive combination @xmath37 , where the factor @xmath38 is an inami - lin function @xcite ( with @xmath39 ) and @xmath40 is the running top - quark mass parameter in @xmath41 renormalization . in particular , we have @xmath42 which leads to @xmath43 . using the same matching scale , we obtain @xmath44 for the nlo qcd factor . our evaluation for @xmath45 then gives & & m_b_s^(sm ) = ( 125.2^+13.8_-12.7 ) 10 ^ -13 gev , [ me ] which is in accord with the experimental value of eq . ( [ mexpt ] ) . the theoretical uncertainty in the sm prediction of eq . ( [ me ] ) is roughly a factor of sixteen larger than the experimental uncertainty of eq . ( [ delmsm ] ) . the largest source of error occurs in the nonperturbative factor @xmath46 , followed by that in the ckm matrix element @xmath47 . the asymmetry in the upper and lower uncertainties in @xmath45 arises from the corresponding asymmetry in the value of @xmath47 cited in ref . @xcite . finally , we note in passing that for the ratio @xmath48 the experimental value is @xmath49 whereas the sm determination gives @xmath50 . this good agreement is not surprising since the ratio @xmath48 contains less theoretical uncertainty than @xmath51 or @xmath0 separately . the above discussion of @xmath45 sets the stage for analyzing np contributions to @xmath5 . there is , in principle , a second approach which instead utilizes @xmath53 . the pdg value for the @xmath2 width difference is @xmath54 . together with eq . ( [ mexpt ] ) , this gives implies @xmath55 , consistent with the value in eq . ( [ bsratio ] ) . ] r^(expt ) = 0.062^+0.034_-0.037 10 ^ 12 s^-1 ( 17.77 0.12 ) 10 ^ 12 s^-1 = ( 34.9 20.0 ) 10 ^ -4 . [ bsratio ] whereas the corresponding sm prediction from ref . @xcite is @xmath56 . in contrast to the mass splitting @xmath0 , the theoretical uncertainty in the ratio @xmath1 is much smaller than in the current experimental determination . nonetheless , this situation is expected to change once lhcb gathers sufficient data . as such , we would expect a highly accurate value of @xmath57 to eventually become available . we propose that it could be applied to the kind of analysis used in this paper as follows . we define a kind of mass difference @xmath58 as m_b_s _ b_s^(expt ) . [ ratio ] the point is that if np contributions are neglected in @xmath59 transitions , then @xmath60 is purely a sm effect . in addition , the ratio @xmath61 will be less dependent on hadronic parameters than either factor separately . this quantity is also important in the scenarios where np contributes a significant cp - violating phase to @xmath0 . in this situation , @xmath62 will be reduced compared to its sm value @xmath63 by a factor of @xmath64 , where @xmath65 is related to the relative phase between the sm and np contributions to @xmath0 @xcite . at the very least , the relation in eq . ( [ ratio ] ) would be of interest to analyze the np issue using both quantities @xmath0 and the above @xmath66 . pdg entries for @xmath67 are _ b_s ^+^-^(expt ) < 4.7 10 ^ -8 _ b_s e^+e^-^(expt ) < 2.8 10 ^ -7 , [ bmm ] with no experimental limit currently for the @xmath68 transition . data collected by the d0 and cdf collaborations will improve the above brancing fraction limit . for example , the d0 collaboration reports @xmath69 , with an anticipated limit of eleven times the sm prediction and similarly for the cdf collaboration @xcite . since the long distance ( ld ) estimate for the branching fraction of @xmath5 in the sm gives @xmath70 @xcite , we consider only the short distance ( sd ) component in the following . using eq . ( [ delmsm ] ) as input to the sd - dominated @xmath5 transition ( see also ref . @xcite ) we arrive at _ b_s ^+^-^(sm ) = m_b_s _ b_s 3 g_f^2 m_w^2 m_^2 4 _ b_s b_b_s ^3 ^1/2 _ y^2 y^2(|x_t ) s_0(|x_t ) , where @xmath71 is another inami - lin function @xcite . expressing @xmath72 in this manner serves to remove some of the inherent model dependence . numerical evaluation gives _ b_s ^+^-^(sm ) = ( 3.33 0.21 ) 10 ^ -9 . [ bmmsm ] the major sources of uncertainty , ordered by magnitude , arise from the factors @xmath73 , then the lifetime @xmath74 and finally the top - quark mass value . in this section , we first obtain a numerical ( 1@xmath75 ) bound on any possible new physics contribution to @xmath0 . we then use this to constrain couplings in a variety of np models and thereby learn something about the @xmath5 transition . as shown in ref . @xcite , new physics in @xmath59 interactions can _ in principle _ markedly affect @xmath76 . the logic is similar to that used in ref . @xcite regarding the possible impact of np on @xmath77 . since , however , in @xmath2 mixing such models are not easy to come up with , one can simply assume that @xmath59 processes are dominated by the sm interactions . thus we can write [ deltabnp0 ] & & m_b_s = m_b_s^(sm ) + m_b_s^(np ) , if the @xmath59 sector were to contain significant np contributions , then the above relation would no longer be valid due to interference between the sm and np components . as can be seen from eq . ( [ deltabnp0 ] ) , interference between the sm and np components may also occur in the presence of a cp - violating phase @xmath78 in the np part of the mixing amplitude @xcite . this large np phase could markedly affect @xmath57 even in the absence of a np contribution to the on - shell @xmath59 transitions ( recall that @xmath57 depends explicitly on the cosine of the cp - violating phase @xmath65 @xcite ; the explicit relation between @xmath78 and @xmath65 can be found in @xcite ) . it is therefore more reasonable to use @xmath57 in studying those scenarios with a large np phase . the appropriate strategy here would be to use @xmath57 and @xmath63 to extract the phase @xmath65 , eliminate @xmath79 from eq . ( [ deltabnp0 ] ) , and then extract @xmath80 in order to relate it to the rare leptonic decay rate . to do so , however , will require a significant reduction in the experimental uncertainty of @xmath57 . alternatively , cp - violating phases could be extracted at lhcb from the studies of @xmath81 transition @xcite . we shall defer those studies to a future publication @xcite . in this paper we shall assume that the phase in the np component of @xmath0 is sufficiently small ( although not necessarily negligible ) , [ deltabnp1 ] & & m_b_s = m_b_s^(sm ) + m_b_s^(np ) . accounting for np as an additive contribution , m_b_s^(expt ) = m_b_s^(sm ) + m_b_s^(np ) , we have from eqs . ( [ mexpt]),([me ] ) , & & m_b_s^(np ) = ( - 8.2_-12.7^+13.8 ) 10 ^ -13 gev . [ av - np ] the error in @xmath82 has been included , but it is so small compared to the theoretical error in @xmath83 as to be negligible . the @xmath84 range for the np contribution is thus & & m_b_s^(np ) = ( - 20.9 + 5.6)10 ^ -13 gev . [ av - onesig ] to proceed further without ambiguity , we would need to know the relative phase between the sm and np components . lacking this , we employ the absolute value of the largest possible number , & & |m_b_s^(np)| 20.9 10 ^ -13 gev , [ bnd ] to constrain the np parameters . new physics can affect both @xmath2 mixing and rare decays like @xmath5 by engaging in these two transitions at tree level . in this section we will , for generality , consider a generic spin-1 boson v or a spin-0 boson s with flavor - changing and flavor - conserving neutral current interactions that couple both to quarks and leptons . the bosons v and s can be of either parity . this situation is frequently realized , as in the interactions of a heavy @xmath85 boson or in multi - higgs doublet models without natural flavor conservation . _ spin-1 boson v _ : assuming that the spin-1 particle @xmath86 has flavor - changing couplings , the most general lagrangian can be written as will be @xmath87 . ] [ vecham ] h_v = g_v1^_l^__l v^+ g_v2^_r^__r v^ + g_v1 b_l _ s_l v^+ g_v2 b_r _ s_r v^+ . here @xmath88 is the vector field and the flavor of the lepton @xmath89 might or might not coincide with @xmath90 . it is not important whether the field @xmath88 corresponds to an abelian or non - abelian gauge symmetry group . using methods similar to those in ref . @xcite , we obtain [ dmv2 ] m_b_s^(v ) & = & f_b_s^2 m_b_s 3 m_v^2 re , where the superscript on @xmath91 denotes propagation of a vector boson in the tree amplitude . the wilson coefficients evaluated at a scale @xmath92 are related to the couplings @xmath93 and @xmath94 as l c_1 ( ) = r ( , m_v ) g_v1 ^ 2 , + c_2()= 2 r(,m_v)^1/2 g_v1 g_v2 , l c_3()= g_v1 g_v2 , + c_6()= r ( , m_v ) g_v2 ^ 2 , [ zwilsons ] where ( presuming that @xmath95 and @xmath96 ) , r(,m)&= & ( ) ^2/7 ( ) ^6/23 . [ wilson ] similar calculations can be performed for the @xmath97 decay . the effective hamiltonian in this case is _ bq^+^-^(v ) = , where the operators @xmath98 can be read off from those in ref . @xcite with the label changes @xmath99 and @xmath100 . this leads to the branching fraction , [ gammav ] b_b_s^0 ^+^-^(v ) = |g_v1 - g_v2|^2 |g_v1 - g_v2|^2 . clearly , eqs . ( [ dmv2]),([gammav ] ) can be related to each other only for a specific set of np models . _ spin-0 boson s _ : analogous procedures can be followed if now the fcnc is generated by quarks interacting with spin-0 particles . again , the most general hamiltonian can be written as [ sfcnc ] h_s = g_s 1^_l _ r s + g_s 2^_r _ l s + g_s 1 b_l s_r s + g_s 2 b_r s_l s+ . evaluation of @xmath101 at scale @xmath102 gives m_b_s^(s ) = re + [ dms2 ] with the wilson coefficients defined as [ scalarcoeff ] & & c_3 ( ) = - 2 r(,m_s ) ^-4 g_s 1 g_s 2 c_3 ( ) g_s 1 g_s 2 + & & c_4 ( ) = - g_s 2 ^ 2 c_4 ( ) g_s 2 ^ 2 + & & c_5 ( ) = g_s 2 ^ 2 c_5 ( ) g_s 2 ^ 2 [ cs1 ] + & & c_7()= - g_s 1 ^ 2 c_7 ( ) g_s 1 ^ 2 + & & c_8 ( ) = g_s 1 ^ 2 c_8 ( ) g_s 1 ^ 2 , where for notational simplicity we have defined @xmath103 . note that eq . ( [ dms2 ] ) is true only for the real spin-0 field @xmath104 . if @xmath104 is a complex field , then only operator @xmath105 will contribute to eq . ( [ dms2 ] ) . the effective hamiltonian for the @xmath106 decay via a heavy scalar s with fcnc interactions is then _ bs^+^-^(s ) = - , and from this , it follows that the branching fraction is [ gammas ] b_b_s^0 ^+^-^(s ) & = & |g_s 1 - g_s 2|^2 + & & . note that if the spin-0 particle @xmath104 only has _ fcnc couplings , _ i.e. _ @xmath107 , no contribution to @xmath97 branching ratio is generated at tree level ; the non - zero contribution to rare decays is instead produced at one - loop level . this follows from the _ pseudoscalar _ nature of the @xmath2-meson . let us now consider specific models where the correlations between the @xmath108 mixing rates and ( in particular ) the @xmath5 rare decay can be found . _ @xmath2 mixing : _ the @xmath2 mixing arising from the @xmath19 pole diagram has the same form as in @xmath12 mixing @xcite , m_b_s^(z ) = m_b_s f_b_s^2 b_b_s r_1(m_b , m_z ) 3 , where @xmath109 is a qcd factor which we take to be r_1(m_b , m_z ) 0.79 . this is a compromise between @xmath110 and @xmath111 . solving for the @xmath19 parameters , we have = 3 |m_b_s^(np)| m_b_s f_b_s^2 b_b_s r_1(m_b , m_z ) 3.02 10 ^ -11 gev^-2 upon using the constraint from @xmath2 mixing . _ @xmath5 decay : _ this has already been calculated for @xmath112 decay in ref . @xcite . inserting obvious modifications for @xmath113 , we have from the branching fraction relation eq . ( 39 ) of ref . @xcite , [ gammazpr ] b_b_s ^+^-^(z ) = g_zs|b^2 m_z^2 . upon inserting numbers , we obtain _ b_s ^+^-^(z ) 0.30 10 ^ -9 ( 1 tev m_z)^2 . this value is already below the corresponding sm prediction ( @xmath114 ) even if we take a @xmath19 mass as light as @xmath115 tev . one of the models of new physics that has a rich flavor phenomenology is r - parity violating ( rpv ) susy . the crucial difference between studies of rpv susy contributions to phenomenology of the up - quark ( see @xcite ) and down - type quark sectors is the possibility of tree - level diagrams contributing to @xmath2-mixing and @xmath116 decays @xcite . if one allows for r - parity violation , the following terms should be added to the superpotential , [ rpvsup ] w _ r= _ ijk l_i l_j e^c_k + _ ijk^l_i q_j d^c_k + _ ijk^ u_i^c d^c_j d^c_k . here @xmath117 and @xmath118 denote @xmath119 doublet quark and lepton superfields , and @xmath120 , @xmath121 and @xmath122 stand for the @xmath119 singlet up - quark , down - quark and charged lepton superfields . also , @xmath123 are generation indices . we shall require baryon number symmetry by setting @xmath124 to zero . also , we will assume cp - conservation , so all couplings @xmath125 and @xmath126 are treated as real . _ @xmath127 mixing _ : neglecting the baryon - number violating contribution , the lagrangian describing rpv susy contribution to @xmath127 mixing can be written as @xmath128 where @xmath129 is a generational index for the sneutrino . matching to eq . ( [ sfcnc ] ) implies that the only non - zero contribution comes from the operator @xmath105 . taking into account renormalization group running , we obtain for @xmath130 from the r - parity violating terms , & & m_b_s^(r ) = f_b_s^2 m_b_s f(c_3,b_3 ) _ i , where @xmath131 denotes the mass of the sneutrino of @xmath132th generation and the function f(c_3,b_3 ) = c_3 ( , m__i ) b_3 , is defined in terms of reduced wilson coefficient of eq . ( [ scalarcoeff ] ) and the b - factor is defined in table [ tab : bagconstants ] of the appendix . _ @xmath5 decay _ : in rpv - susy , the underlying transition for @xmath5 is @xmath133 via tree - level @xmath134-squark or sneutrino exchange . in order to relate the rare decay to the mass difference contribution from rpv susy @xmath135 , we need to assume that the up - squark contribution is negligible . this can be achieved in models where sneutrinos are much lighter than the up - type squarks , which are phenomenologically viable . employing this assumption leads to the predicted branching fraction [ y2 ] b^(r)_b_s^+^- & = & ( ) ^2 ( 1- ) + & & ( |_i|^2 + |_i|^2 ) . in order to relate @xmath5 to @xmath130 in the framework of rpv susy , we need to make additional assumptions . in particular , we shall assume that the sum is dominated by a single sneutrino state , which we shall denote by @xmath136 . in addition , we will assume that @xmath137 , which will reduce the number of unknown parameters . this assumption is not needed , however , if one wishes to set a bound on a combination of coupling constants directly from the experimental bound on @xmath138 . then , neglecting cp - violation , [ y2a ] b^(r)_b_s^+^- = k ( ) ^2 ( ) ^2 ( 1- ) , where @xmath139 if an assumption that @xmath137 is made , and @xmath140 otherwise . since no @xmath5 signal has yet been seen , we can use the experimental bound to obtain an updated constraint on the rpv couplings , [ y4 ] _ k22 ^_k32 5.5 10 ^ -6 ( ) ^2 . now , assuming @xmath137 , one can relate the branching ratio @xmath138 to @xmath141 , [ relate3 ] b^(r)_b_s ^+^- = ( ) ^2 ( 1- ) x_b_s^(r ) . it is possible to plot the dependence of @xmath138 on @xmath142 for different values of @xmath131 , which we present in fig . [ rpvgraph ] . [ tb ] as a function of rpv leptonic coupling @xmath142 and sneutrino mass @xmath143 gev , @xmath144 gev , and @xmath145 gev ( solid , dashed , and dash - dotted lines ) . the yellow shaded area represents excluded parameter space.,width=340 ] the gauge sector in the standard model has a large global symmetry which is broken by the higgs interaction @xcite . by enlarging the higgs sector , some subgroup of this symmetry can be imposed on the full sm lagrangian and the symmetry can be broken spontaneously . this family symmetry can be global @xcite as well as gauged @xcite . if the new gauge couplings are very weak or the gauge boson masses are large , the difference between a gauged or global symmetry is rather difficult to distinguish in practice @xcite . in general there would be fcnc effects from both the gauge and scalar sectors . here we study the gauge contribution . consider the family gauge symmetry group @xmath146 acting on the three left - handed families . spontaneous symmetry breaking renders all the gauge bosons massive . if the su(3 ) is broken first to su(2 ) before being completely broken , we may have an effective ` low ' energy symmetry @xmath147 . this means that the gauge bosons @xmath148 are much lighter than the @xmath149 . for simplicity we assume that after symmetry breaking the gauge boson mass matrix is diagonal to a good approximation . if so , the light gauge bosons @xmath150 are mass eigenstates with negligible mixing . the lh doublets @xmath151 transform as @xmath152 under @xmath147 , as do the lepton doublets @xmath153 and the right - handed fermions are singlets under @xmath147 . in the above , the superscript @xmath154 refers to the fact that these are weak eigenstates and not mass eigenstates . the couplings of fermions to the light family gauge bosons @xmath150 is given by @xmath155 \ \ , \ ] ] where @xmath156 denotes the coupling strength and @xmath157 are the generators of @xmath147 the fermion mass eigenstates are given by , first for quarks , @xmath158 and then for leptons , @xmath159 the four matrices @xmath160 and @xmath161 are unknown , except for @xmath162 where @xmath163 is the maki - nakagawa - sakata - pontcorvo lepton mixing matrix . the couplings of the gauge bosons relevant for the @xmath2 system in the mass basis are : @xmath164\end{aligned}\ ] ] the contribution to @xmath165 mixing is given by @xmath166 \label{famsymm}\ ] ] where @xmath167 \\ b = -{\rm re}\left[\left ( u_{b1 } \ u^*_{s2 } - \ u_{b2 } \ u^*_{s1 } \right ) ^2\right ] \label{abc } \\ c = { \rm re}\left[\left ( u_{b1 } u^*_{s1 } - u_{b2}u^*_{s2}\right)^2\right ] \end{array}\end{aligned}\ ] ] in a simple scheme of symmetry breaking @xcite , one obtains @xmath168 and the square bracket in eq . ( [ famsymm ] ) becomes @xmath169 \ \ .\ ] ] although the matrices @xmath170 in principle are unknown , it has been argued that a reasonable ansatz @xcite , which is incorporated in many models is @xmath171 . in this case one can simplify @xmath172 and @xmath173 further : @xmath174 thus the @xmath2 mixing becomes m_b_s^(fs ) f^2 m_1 ^ 2 1.610 ^ -3 , [ fammix ] so that , substituting experimental bound @xmath175 , . [ fm1 ] the same above ansatz also implies that @xmath176 and @xmath177 . then the coupling of the gauge bosons to muon pairs is given by @xmath178 \bar{\mu}_l \gamma_\lambda \mu_l \ \ . \label{gmm}\end{aligned}\ ] ] the branching ratio for @xmath179 is given by @xmath180 next we employ the approximation ( well - supported empirically ) that @xmath181 , where @xmath182 is the tri - bi - maximal matrix @xcite . then eq . ( [ gmm ] ) becomes @xmath183 \bar{\mu}_l \gamma_\mu \mu_l \ \ .\end{aligned}\ ] ] with this , the contribution to the branching ratio for @xmath184 becomes @xmath185 ^ 2 { 1 \over m_1 ^ 4 } \nonumber \\ & \simeq & \frac{m_{b_s } f^2_{b_s } m_\mu^2 f^4}{32 \pi \gamma_{b_{s}}}~ { 1.4 \times 10^{-4 } \over m_1 ^ 4 } \ \ . \label{brmumu}\end{aligned}\ ] ] the dependence on unknown factors in eq . ( [ brmumu ] ) ( _ i.e. _ @xmath186 ) can be entirely removed by using the bound in eq . ( [ fm1 ] ) to yield & & b_b_s ^+ ^-^(fs ) |m_b_s^(np)|^2 . [ fsbr ] from the bounds of eqs . ( [ av - np]),([av - onesig ] ) , we obtain & & b_b_s ^+ ^-^(fs ) 0.92 10 ^ -12 . many extensions of the standard model contain multiple scalar doublets , which increases the possibility of fcnc mediated by flavor non - diagonal interactions of neutral components . while many ideas exist on how to suppress those interactions ( see , _ @xcite ) , the ultimate test of those ideas would involve direct observation of scalar - mediated fcnc . consider a generic yukawa interaction consisting of a set of @xmath187 higgs doublets @xmath188 @xmath189 with sm fermions , _ y = _ ijn^u q_li u_rj h_n + _ ijn^d q_li d_rj h_n + _ ijn^e l_li e_rj h_n + , [ int1 ] where @xmath190 and @xmath191 ( @xmath192 ) are respectively the left - handed weak doublets of an @xmath132th - generation of quarks ( leptons ) . restricting the discussion to @xmath193 mixing and @xmath194 decay , we find that eq . ( [ int1 ] ) reduces to [ fcnchiggs ] h^h_y = _ 23n^d s_l b_r ^0_n + _ 32n^d b_l s_r ^0_n + _ 22n^e _ l _ r ^0_n + , where @xmath195 . bringing this to the form of eq . ( [ sfcnc ] ) and confining the discussion only to the contribution of the lightest @xmath196 and @xmath197 states , we obtain [ fcnchiggs2 ] h^h_y & = & b_r s_l ^0 + b_l s_r ^0 + _ l _ r ^0 + & - & i b_r s_l a^0 + i b_l s_r a^0 + i _ l _ r a^0 + ... + , where ellipses stand for the terms containing heavier @xmath196 and @xmath197 states whose contributions to @xmath0 and @xmath17 will be suppressed . if the matrix of coupling constants in eq . ( [ fcnchiggs2 ] ) is hermitian , _ e.g. _ @xmath198 , then we can identify the couplings of eq . ( [ sfcnc ] ) as [ scouplings ] & & g_s_1 = g_s_2 = , g_s_1^= g_s_2^= for _ scalar _ interactions and [ pcouplings ] & & g_s_1 = - g_s_2 = , g_s_1^= - g_s_2^= for _ pseudoscalar _ interactions . to proceed , we need to separate two cases : ( i ) the lightest fcnc higgs particle is a scalar , and ( ii ) the lightest fcnc higgs particle is pseudoscalar . the case of relatively light scalar higgs state is quite common , arising most often in type - iii two - higgs doublet models ( models without natural flavor conservation ) @xcite . _ @xmath199-@xmath200 mixing _ : given the general formulas of eq . ( [ dms2 ] ) , it is easy to compute the contribution to @xmath201 of an intermediate scalar ( @xmath78 ) with fcnc couplings , m_b_s^ ( ) & = & ( _ 32^d m _ ) ^2 , [ dmh2 ] + f_(c_i , m_b ) & & c_3(m_b ) b_3 - ( c_4(m_b ) b_4 + c_7(m_b ) b_7 ) + ( c_5(m_b ) b_5 + c_8(m_b ) b_8 ) , with reduced wilson coefficients @xmath202 given in eq . ( [ cs1 ] ) . _ @xmath203 decay _ : comparing eq . ( [ scouplings ] ) to eq . ( [ gammas ] ) , we can easily see that the branching fraction for the rare decay @xmath204 is zero for the intermediate scalar higgs , _ b_s^0^+^-^ ( ) = 0 . this is consistent with what was already discussed in sec . [ sec : treelevel ] and implies that the fcnc higgs model does not produce a contribution to @xmath205 at tree level . the non - zero contribution to @xmath205 decay is produced at one - loop level @xcite . the case of a lightest _ pseudoscalar _ higgs state can occur in the non - minimal supersymmetric standard model ( nmssm ) @xcite or related models @xcite . in nmssm , a complex singlet higgs is introduced to dynamically solve the @xmath92 problem . the resulting pseudoscalar can be as light as tens of gev . this does not mean , however , that it necessarily gives the dominant contribution to both @xmath206 mixing and the @xmath203 decay rate since there can be loop contributions from other higgs states . in the following , we shall work in the region of the parameter space where it does . _ @xmath199-@xmath200 mixing _ : the contribution to @xmath207 due to intermediate pseudoscalar with flavor - changing couplings can be computed using the general formula in eq . ( [ dms2 ] ) along with the identification given in eq . ( [ pcouplings ] ) , [ dma ] m_b_s^(a ) & = & ( _ 32^d m_a ) ^2 , + f_a(c_i , m_b ) & = & with ` reduced ' wilson coefficients @xmath208 again being defined in eq . ( [ cs1 ] ) . _ @xmath203 decay _ : the branching ratio for rare decay can be computed with the help of the general formula of eq . ( [ gammas ] ) , [ gammaa ] b_b_s^0 ^+^-^(a ) & = & ( 1-)^1/2 ( _ 32^d _ 22^e m_a^2 ) ^2 . we can now eliminate one of the three unknown parameters ( @xmath209 , @xmath210 , and @xmath211 ) which appear in eqs.([dma ] ) and ( [ gammaa ] ) . we choose to eliminate @xmath209 , so [ arel ] b_b_s^0 ^+^-^(a ) & = & ( 1-)^1/2 ( ) ^2 , where @xmath212 . as one can see , the unknown factors enter eq . ( [ arel ] ) in the combination @xmath213 . it is , however , more convenient to plot the dependence on @xmath211 for different values of @xmath210 , which we present in fig . [ agraph ] . [ tb ] as a function of pseudoscalar higgs mass @xmath211 . left : @xmath214 ( solid , dashed , dash - dotted lines ) . right : @xmath215 ( solid , dashed , dash - dotted lines ) . in each figure , the yellow shaded area represents excluded parameter space.,title="fig:",width=340 ] as a function of pseudoscalar higgs mass @xmath211 . left : @xmath214 ( solid , dashed , dash - dotted lines ) . right : @xmath215 ( solid , dashed , dash - dotted lines ) . in each figure , the yellow shaded area represents excluded parameter space.,title="fig:",width=340 ] it must be emphasized that the discussion above assumed the absence of large destructive interference of the np and sm contributions to @xmath127 mixing . concrete models where such interference is present ( and thus the new physics contribution is larger than the sm one ) can be constructed @xcite . in such models possible contribution to @xmath5 could be large . one of the simplest extensions of the standard model involves addition of the sequential fourth generation of chiral quarks @xcite , denoted for the lack of the better names by @xmath216 and @xmath217 . the addition of the sequential fourth generation of quarks leads to a 4@xmath2184 ckm quark mixing matrix @xcite . this implies that the parameterization of this matrix requires six real parameters and three phases . besides providing new sources of cp - violation , the two additional phases can affect the branching ratios considered in this paper due to interference effects @xcite . there are many existing constraints on the parameters related to the fourth generation of quarks . in particular , a fit of precision electroweak data ( s and t parameters ) @xcite implies that the masses of the new quarks are strongly constrained to be @xcite m_t-m_b ( 1 + ) 50 , with @xmath219 gev . here @xmath220 is the sm higgs mass , which we take for simplicity to be 120 gev . we also used updated constraints on ckm matrix elements @xcite . the relationship between @xmath0 and @xmath17 in the model with four generations of quarks has been previously studied in detail in @xcite . here we update their result . the branching ratio of @xmath5 can be related to the experimentally - measured from table [ tab : corr ] , as the separation of np and sm contributions used in the rest of this paper , @xmath221 , is not possible due to loops with both @xmath216 and @xmath222 , @xmath223 , or @xmath134 quarks . ] @xmath224 as @xcite _ b_s ^+^- = , where the parameter @xmath225 is a @xmath2-mixing loop parameter @xcite , ^= _ t s_0(x_t ) + _ t r_tt^2 s_0(x_t^ ) + 2 _ t^ r_tt s_0(x_t , x_t^ ) , and @xmath226 . @xmath227 can be obtained from table [ tab : corr ] . the definition of the function @xmath228 can be found in ref . the wilson coefficient @xmath229 is defined as c_10^tot ( ) = c_10 ( ) + r_tt c_10^t ( ) with @xmath230 obtained by substituting @xmath231 into the sm expression for @xmath232 @xcite . the results can be found in fig . [ fourgengraph ] . as one can see , the resulting branching ratios are still lower than the current experimental bound of eq . ( [ bmm ] ) , but for the values of the four - generation ckm matrix @xmath233 of about @xmath234 , disfavored by @xcite , but still favored by @xcite , can be quite close to it . [ tb ] as a function of the top - prime mass @xmath231 for different values of the phase @xmath235 ( solid , dashed , dash - dotted lines ) and @xmath236 @xcite ( see also @xcite ) . right : branching ratio of @xmath237 as a function of the ckm parameter combination @xmath238 with @xmath239 and different values of @xmath240 gev ( solid ) , @xmath241 gev ( dashed ) , and @xmath242 gev ( dash - dotted).,title="fig:",width=340 ] as a function of the top - prime mass @xmath231 for different values of the phase @xmath235 ( solid , dashed , dash - dotted lines ) and @xmath236 @xcite ( see also @xcite ) . right : branching ratio of @xmath237 as a function of the ckm parameter combination @xmath238 with @xmath239 and different values of @xmath240 gev ( solid ) , @xmath241 gev ( dashed ) , and @xmath242 gev ( dash - dotted).,title="fig:",width=340 ] experiment has determined @xmath0 exceedingly well . the standard model determination provides a consistent value , although with a markedly greater uncertainty ( due mainly to the dependence on the nonperturbative quantity @xmath243 and to a lesser extent on the ckm mixing element @xmath47 ) . we have argued that this fact can be used to constrain np predictions for other processes , such as the @xmath3 transition considered here . we expect this kind of correlation to be a rather general feature of new physics models , provided there is an overlap between the np parameters which describe @xmath0 and ( for our purposes here ) @xmath3 . however , given the abundance of new physics scenarios , each with its particular structure , it is not reasonable to expect any _ universal _ correlation between @xmath2-mixing and @xmath5 . instead , what we have done in this paper is to analyze several np models in detail . in each case , we have first determined the set of unknown np parameters and then , using dynamical assumptions , have been able to reduce ( or entirely eliminate ) the arbitrariness . analyzing specific np models this way has two purposes : to serve as an instructive example for further study and to see what kinds of numerical predictions these particular models yield . not surprisingly , the simplest model ( with a single @xmath19 boson ) provides a strong correlation between @xmath0 and @xmath3 in which the latter is determined in terms of @xmath244 . an even stronger prediction occurs in the particular version of the family symmetry model discussed earlier , where a clean determination of @xmath3 is obtained . in this instance , a set of reasonable assumptions allows for the initial presence of unknown parameters to be totally overcome . a similar , but not quite as fortunate , situation occurs for r - parity violating supersymmetry , wherein a reasonable assumption partially reduces the np parameter set . in this case , @xmath3 can be expressed in terms of a ratio of a coupling constant and sneutrino mass @xmath245 . the flavor - changing higgs model turns out to be less accommodating in that no set of assumptions known to us can reduce the original set of three unknown parameters . thus , the constraint from @xmath2 mixing still leaves one with two unknowns ( see fig . [ agraph ] ) . we also updated constraints on the models with fourth sequential generation of quarks . of course , additional np models are available for study , _ e.g. _ r - parity conserving supersymmetry @xcite , and work proceeds on these . finally , as discussed in sect . iii , it would be of interest to address the impact of np cp - violating contributions to @xmath2 mixing . indeed , we plan do so in a future project , but first await more accurate data on @xmath76 or studies of @xmath81 transition at lhcb . we give our warm thanks to aida el - khadra for useful remarks on the present status of various qcd - lattice predictions . the work of e.g. was supported in part by the u.s . national science foundation under grant phy0555304 , j.h . was supported by the u.s . department of energy under contract de - ac02 - 76sf00515 , s.p . was supported by the u.s . department of energy under contract de - fg02 - 04er41291 and a.a.p . and g.k.y . were supported in part by the u.s . national science foundation under career award phy0547794 , and by the u.s . department of energy under contract de - fg02 - 96er41005 . there are eight @xmath246 effective operators that can contribute to @xmath2-mixing . the operator basis we shall employ is [ setofoperators ] where quantities enclosed in parentheses are color singlets , _ e.g. _ @xmath247 . these operators are generated at a scale @xmath248 where the np is integrated out . a non - trivial operator mixing then occurs via renormalization group running of these operators between the heavy scale @xmath248 and the light scale @xmath92 at which hadronic matrix elements are computed . we need to evaluate the @xmath199-to-@xmath249 matrix elements of these eight dimension - six basis operators . this introduces eight _ non - perturbative _ b - parameters @xmath250 that require evaluation by means of qcd sum rules or qcd - lattice simulation . we express these in the form @xmath251 where @xmath252 is the @xmath2 meson decay constant and @xmath253 . ref . @xcite has performed a qcd - lattice determination ( quenched approximation ) of the b - parameters in an operator basis @xmath254 which is distinct from the @xmath255 of eq . ( [ setofoperators ] ) , @xmath256 three more operators @xmath257 can be obtained by substituting right - handed chiral projection operators with the left - handed ones @xmath258 in eq . ( [ setofoperators1 ] ) . the @xmath199-to-@xmath249 matrix elements of these operators have been parameterized in ref . @xcite as & & several of the quantities introduced above are scale dependent , _ i.e. _ @xmath262 , @xmath263 and @xmath264 . throughout this paper , we shall understand all these quantities to be renormalized at a common scale @xmath102 and to simplify notation , we shall denote them simply as @xmath265 , @xmath266 and @xmath267 . in particular , our evaluation at scale @xmath102 of the quantity @xmath268 yields r_s^2 = m_b_s^2/(|m_b ( |m_b ) + |m_s ( |m_b))^2 = 1.57_-0.10^+0.04 , [ rs ] where we have used the input values @xmath269 gev @xcite and @xmath270 gev @xcite . l||c|c list of @xmath275 & @xmath275 from lattice qcd & @xmath276 in mvs + ( in @xmath277 basis ) & ( from ref . @xcite ) & ( from eq . ( [ me_mvs ] ) ) + @xmath278 & @xmath279 & @xmath279 + @xmath280 & @xmath281 & @xmath282 $ ] + @xmath283 & @xmath284 & @xmath285 $ ] + @xmath286 & @xmath287 & @xmath288 + @xmath289 & @xmath290 & @xmath288 + alternatively , the b - parameters can be estimated using the ` modified vacuum saturation ' ( mvs ) approach , wherein all matrix elements in eq . ( [ me ] ) are written in terms of ( known ) matrix elements of @xmath291 and @xmath292 matrix elements @xmath293 and @xmath294 , @xmath295 \ , \\ \langle q_3 \rangle = f_{{\rm b}_s}^2 m_{{\rm b}_s}^2 b_{b_s } \left [ \displaystyle{1 \over 4 n_c } + \displaystyle{\eta \over 2 } \right ] \ , \\ \langle q_4 \rangle = - \displaystyle{2 n_c - 1 \over 4 n_c } f_{{\rm b}_s}^2 m_{{\rm b}_s}^2 b_{b_s}~ \eta \ , \end{array } \qquad \qquad \begin{array}{l } \langle q_5 \rangle = \displaystyle{3 \over n_c } f_{{\rm b}_s}^2 m_{{\rm b}_s}^2 b_{b_s } ~ \eta \ , \\ \langle q_6 \rangle = \langle q_1 \rangle \ , \\ \langle q_7 \rangle = \langle q_4 \rangle \ , \\ \langle q_8 \rangle = \langle q_5 \rangle \ \ , \end{array } \ ] ] where we take @xmath296 as the number of colors and define [ bbar ] & & r_s^2 b_b_s^(s ) = b_b_s . it is instructive to compare how well the mvs approximation estimates the recent lattice results . we provide such a comparison in table [ tab : bagconstants ] . m. bona _ et al . _ [ utfit collaboration ] , pmc phys . a * 3 * , 6 ( 2009 ) [ arxiv:0803.0659 [ hep - ph ] ] . ` measurement of @xmath297 at cdf ' , talk by louise oakes at flavor physics and cp violation conference 2010 , turin , italy ( may 25 - 29 , 2010 ) ; ` new measurement of @xmath2 mixing phase at cdf ' , talk by gavril giurgiu talk delivered at ichep 2010 ( paris , france ) . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . d * 82 * , 032001 ( 2010 ) [ arxiv:1005.2757 [ hep - ex ] ] . e. golowich , j. hewett , s. pakvasa and a. a. petrov , phys . d * 79 * , 114030 ( 2009 ) [ arxiv:0903.2830 [ hep - ph ] ] . e. golowich , j. hewett , s. pakvasa and a. a. petrov , phys . rev . d * 76 * , 095009 ( 2007 ) [ arxiv:0705.3650 [ hep - ph ] ] . a. j. buras , phys . lett . b * 566 * , 115 ( 2003 ) [ arxiv : hep - ph/0303060 ] . a. lenz and u. nierste , jhep * 0706 * , 072 ( 2007 ) [ arxiv : hep - ph/0612167 ] . a. j. buras , m. jamin and p. h. weisz , nucl . b * 347 * , 491 ( 1990 ) . j. urban , f. krauss , u. jentschura and g. soff , nucl . b * 523 * , 40 ( 1998 ) [ arxiv : hep - ph/9710245 ] . `` top quark physics '' , plenary talk by mousumi datta at the 2009 meeting of the division of particles and fields of the american physical society , wayne state university detroit , mi ( 7/26/09- 7/31/09 ) . s. bethke , eur . j. c * 64 * , 689 ( 2009 ) [ arxiv:0908.1135 [ hep - ph ] ] . j. laiho , e. lunghi and r. s. van de water , phys . d * 81 * , 034503 ( 2010 ) [ arxiv:0910.2928 [ hep - ph ] ] . k. melnikov and t. v. ritbergen , phys . b * 482 * , 99 ( 2000 ) [ arxiv : hep - ph/9912391 ] . aida el - khadra , private communication . t. inami and c.s . lim , prog . * 65 * , 297 ( 1981 ) [ erratum - ibid . * 65 * , 1772 ( 1981 ) ] . y. grossman , phys . b * 380 * , 99 ( 1996 ) [ arxiv : hep - ph/9603244 ] . casey , _ new upper limit on the decay @xmath5 from d0 _ , talk delivered at ichep 2010 ( paris , france ) ; see also v. m. abazov _ et al . _ [ d0 collaboration ] , phys . b * 693 * , 539 ( 2010 ) [ arxiv:1006.3469 [ hep - ex ] ] . g. burdman , e. golowich , j. hewett and s. pakvasa , phys . d * 66 * , 014009 ( 2002 ) [ arxiv : hep - ph/0112235 ] . a. badin , f. gabbiani and a. a. petrov , phys . b * 653 * , 230 ( 2007 ) [ arxiv:0707.0294 [ hep - ph ] ] ; see also a. badin , f. gabbiani and a. a. petrov , arxiv:0909.4897 [ hep - ph ] . e. golowich , s. pakvasa and a.a . petrov , phys . * 98 * , 181808 - 1 ( 2007 ) [ arxiv : hep - ph/0610039 ] . p. ball and r. fleischer , eur . j. c * 48 * , 413 ( 2006 ) [ arxiv : hep - ph/0604249 ] . e. golowich , j. hewett , s. pakvasa , a. a. petrov , and g. k. yeghiyan , to be published . a. kundu and j. p. saha , phys . d * 70 * , 096002 ( 2004 ) [ arxiv : hep - ph/0403154 ] . y. kao and t. takeuchi , arxiv:0910.4980 [ hep - ph ] . h. k. dreiner , m. kramer and b. oleary , phys . d * 75 * , 114016 ( 2007 ) [ arxiv : hep - ph/0612278 ] . j. p. saha and a. kundu , phys . d * 66 * , 054021 ( 2002 ) [ arxiv : hep - ph/0205046 ] ; we concur with a conclusion of ref . @xcite regarding the missing factor of 4 in this paper . m. sher , phys . rept . * 179 * , 273 ( 1989 ) . t. maehara and t. yanagida , lett . nuovo cim . * 19 * , 424 ( 1977 ) ; m. a. b. beg and a. sirlin , phys . lett . * 38 * , 1113 ( 1977 ) ; c. l. ong , phys . d * 19 * , 2738 ( 1979 ) ; f. wilczek and a. zee , phys . rev . lett . * 42 * , 421 ( 1979 ) ; a. davidson , m. koca and k. c. wali , phys . d * 20 * , 1195 ( 1979 ) , phys . * 43 * , 92 ( 1979 ) . v. a. monich , b. v. struminsky and g. g. volkov , phys . b * 104 * , 382 ( 1981 ) [ jetp lett . * 34 * , 213 ( 1981 ) ] [ zh . 34 * , 222 ( 1981 ) ] . bjorken , s. pakvasa and s.f . tuan , phys . d * 66 * , 053008 ( 2002 ) [ arxiv : hep - ph/0206116 ] . p. f. harrison , d. h. perkins and w. g. scott , phys . b * 530 * , 167 ( 2002 ) [ arxiv : hep - ph/0202074 ] . d. becirevic , v. gimenez , g. martinelli , m. papinutto and j. reyes , jhep * 0204 * , 025 ( 2002 ) [ arxiv : hep - lat/0110091 ] . l. j. hall and s. weinberg , phys . d * 48 * , 979 ( 1993 ) [ arxiv : hep - ph/9303241 ] . t. p. cheng and m. sher , phys . d * 35 * , 3484 ( 1987 ) . a. pich and p. tuzon , phys . d * 80 * , 091702 ( 2009 ) [ arxiv:0908.1554 [ hep - ph ] ] . v. d. barger , j. l. hewett and r. j. n. phillips , phys . d * 41 * , 3421 ( 1990 ) . d. atwood , l. reina and a. soni , phys . d * 55 * , 3156 ( 1997 ) [ arxiv : hep - ph/9609279 ] . r. a. diaz , r. martinez and c. e. sandoval , eur . j. c * 41 * , 305 ( 2005 ) [ arxiv : hep - ph/0406265 ] ; m. j. savage , phys . b * 266 * , 135 ( 1991 ) ; m. e. luke and m. j. savage , phys . b * 307 * , 387 ( 1993 ) [ arxiv : hep - ph/9303249 ] . a. e. blechman , a. a. petrov and g. yeghiyan , jhep * 1011 * , 075 ( 2010 ) [ arxiv:1009.1612 [ hep - ph ] ] . h. p. nilles , m. srednicki and d. wyler , phys . b * 120 * , 346 ( 1983 ) ; j. m. frere , d. r. t. jones and s. raby , nucl . b * 222 * , 11 ( 1983 ) ; j. p. derendinger and c. a. savoy , nucl . phys . b * 237 * , 307 ( 1984 ) . j. r. ellis , j. f. gunion , h. e. haber , l. roszkowski and f. zwirner , phys . d * 39 * , 844 ( 1989 ) . u. ellwanger , m. rausch de traubenberg and c. a. savoy , nucl . b * 492 * , 21 ( 1997 ) [ arxiv : hep - ph/9611251 ] . g. hiller , phys . d * 70 * , 034018 ( 2004 ) [ arxiv : hep - ph/0404220 ] . b. a. dobrescu , phys . d * 63 * , 015004 ( 2001 ) [ arxiv : hep - ph/9908391 ] . m. endo and n. yokozaki , arxiv:1012.5501 [ hep - ph ] . m. bobrowski , a. lenz , j. riedl and j. rohrwild , phys . d * 79 * , 113006 ( 2009 ) [ arxiv:0902.4883 [ hep - ph ] ] . v. a. novikov , l. b. okun , a. n. rozanov _ et al . _ , mod . lett . * a10 * , 1915 - 1922 ( 1995 ) . j. erler and p. langacker , phys . lett . * 105 * , 031801 ( 2010 ) [ arxiv:1003.3211 [ hep - ph ] ] . g. d. kribs , t. plehn , m. spannowsky and t. m. p. tait , phys . d * 76 * , 075016 ( 2007 ) [ arxiv:0706.3718 [ hep - ph ] ] . a. k. alok , a. dighe and d. london , arxiv:1011.2634 [ hep - ph ] . a. soni , a. k. alok , a. giri , r. mohanta and s. nandi , phys . d * 82 * , 033009 ( 2010 ) [ arxiv:1002.0595 [ hep - ph ] ] . a. j. buras , m. munz , phys . rev . * d52 * , 186 - 195 ( 1995 ) . [ hep - ph/9501281 ] .

we present a scalable method for generation of a cluster state for measurement - based quantum computing using van der waals or dipole - dipole interactions between neutral atoms or polar molecules in an optical lattice . nearest neighbor entanglement is accomplished by performing a phase gate using interaction of atoms in rydberg states or molecules in large dipole moment states . all nearest neighbors are sequentially entangled in a finite number of operations , independent of the number of qubits , producing a 1d cluster state . a universal 2d cluster state can be generated in several ms in a two - dimensional optical lattice by producing a series of 1d cluster states in one lattice direction , followed by application of the entangling operations in another lattice direction . we discuss the viability of the scheme with rb rydberg atoms .

1106.0713

entanglement plays a major role in quantum computing @xcite , quantum communication @xcite and quantum metrology @xcite . a special type of a multipartite entangled state , cluster state , represents a universal resource for the measurement - based quantum computing ( mbqc ) paradigm @xcite . in measurement - based quantum computing , quantum computations are carried out on the cluster state through individual qubit measurements in adaptive bases @xmath0 . the primary advantage of mbqc over other register - based architectures rests in the fact that all interactions , required _ e.g. _ for two - qubit gates , are performed in the initialization stages of the resource state . the computation may then be performed through simultaneous measurement of many individual qubits as warranted by the specific program being implemented . any one- and two - qubit gate can be realized by appropriate measurements , making mbqc equivalent to the standard quantum circuit model @xcite . moreover , since the actual computation is done by local measurements , it can be faster compared to the equivalent gates in the circuit model . the cluster state is realized by preparing individual qubits in an eigenstate @xmath1 of the pauli spin operator @xmath2 , and letting the nearest neighbor qubits interact via ( up to local operations ) an ising hamiltonian @xmath3 during a time @xmath4 so that @xmath5 . it clearly shows that the cluster state can be generated by applying a phase gate @xmath6 between all nearest neighbor qubits . a cluster state was experimentally realized with photons @xcite using linear optics , but this approach is difficult to scale to a large number of qubits . schemes to generate a cluster state using atom - cavity entanglement in the framework of cavity qed @xcite , cold ions via phonon - mediated spin - spin interactions @xcite , and distributed networks of collectively excited atomic ensembles @xcite have also been proposed . a naturally highly scalable system is neutral atoms in an optical lattice , where a 1d cluster state has been produced via ultracold s - wave collisions of atoms in a spin - dependent lattice @xcite . neutral atoms can also be entangled via long - range van der waals ( vdw ) or dipole - dipole interaction when excited to rydberg states . atoms in rydberg states can have huge dipole moments of several kilodebyes and interact strongly with each other @xcite , which was suggested as a tool to produce two - qubit gates in neutral - atom based quantum computing @xcite . recently , in a series of experiments @xcite rydberg ( dipole ) blockade between two atoms has been observed , followed by the demonstrations of a blockade - assisted cnot gate @xcite and entanglement between two atoms @xcite . it is interesting to explore the possibility to use the strong interaction in rydberg states to generate a cluster state . moreover , it would offer a way to produce a cluster state with polar molecules , coupled via dipole - dipole interaction , for which collision - based interactions can result in undesirable inelastic or chemical reaction losses . for simplicity we primarily discuss in this work neutral atoms interacting in rydberg states , the same scheme can be applied to molecules . we propose to realize a cluster state by applying a phase gate to all pairs of nearest neighbor atoms in an optical lattice . the phase gate can be realized using either direct or blockaded interaction in rydberg states . this can be accomplished in a scalable way , _ i.e. _ in a finite number of operations , not depending on the number of atoms in the lattice . in fact , to make pair excitation controllable and to minimize errors due to interactions of multiple atoms in rydberg states , it already suffices to excite every other pair : ( a ) to produce a 1d cluster state , four iterations are required and would be implemented as follows : using a periodic entangler , such as a standing wave , entangle positions 1 with 2 , @xmath7 with @xmath8 , etc . ; in the second step , entangle positions 3 with 4 , @xmath9 with @xmath10 , etc . ; in the third step entangle positions 2 and 3 , @xmath11 and @xmath9 , etc ; in the last step entangle positions 4 and 5 , @xmath10 and @xmath7 etc . ( b ) for 2d cluster state generation , first the rows in @xmath12-direction are entangled , then the entangling operations are applied to columns in y - direction . this can be done using optical standing wave excitation as is illustrated in fig.[fig : superlattices]a in the case of a 1d optical lattice . pairs of atoms at the maxima of the standing wave intensity will get entangled , while those at the minima will not . changing the phase of the standing wave fields , _ i.e. _ shifting the position of the intensity maxima and minima , will allow to perform the phase gate between all nearest neighbors , producing a cluster state . the paper is organized as follows . in section ii we describe how a 1d cluster state can be realized using interaction in rydberg states . in section iii we show how to generalize the technique used for the 1d cluster state generation to produce a universal 2d cluster state . finally , we discuss the main features of the scheme and conclude in section iv . we start by analyzing how a 1d cluster state can be generated by applying the phase gate between all neighboring atoms in a 1d optical lattice . we assume that initially atoms are loaded into the ground motional state of an optical lattice described by a potential @xmath13 , shown in fig.[fig : superlattices]a , which can be done using a superfluid - mott insulator transition . two ground state hyperfine sublevels @xmath14 encode qubit states @xmath15 , @xmath16 . as a final preliminary step atoms are transferred into the @xmath1 superposition by applying a @xmath17 pulse , resonant with the qubit transition . a phase gate between atoms in neighboring sites can be implemented using strong vdw or dipole - dipole interaction in rydberg states . there are two ways to perform the phase gate depending on whether atoms are individually addressable or not @xcite . standard rydberg blockade requires individual addressing of the atoms . individual addressing in an optical lattice has been demonstrated recently using a tightly focused laser beam and a spin - dependent lattice @xcite . however , sequential application of the phase gate to each pair of atoms is not scalable to a large number of qubits . in a good scalable approach the cluster state has to be generated in a finite number of operations which does not depend on the size of the system . below we analyze the possibility to generate the 1d cluster state assuming that atoms in each pair are not addressed individually . if one uses a standing wave excitation pulse with the rabi frequency @xmath18 , one has the maxima at _ e.g. _ even and minima at odd pairs sites ( see fig.[fig : superlattices]a ) . the atoms at the maxima are transferred from the qubit state @xmath16 to the rydberg state @xmath19 either directly or by a two - photon excitation , as shown in fig.[fig : superlattices]b , below we assume two - photon excitation . the phase gate is then realized in the limit @xmath20 ( @xmath21 is the interaction strength in the rydberg state @xmath22 ) as follows : 1 ) a @xmath23-pulse resonant to the @xmath24 transition is applied simultaneously to both atoms , exciting each to the @xmath19 state , since the shift of the doubly excited state @xmath22 is smaller than the rabi frequency ; 2 ) atoms in the @xmath22 state interact during time @xmath25 and accumulate a @xmath23 phase shift @xmath26 ; 3 ) a second @xmath23-pulse deexcites atoms back to their original qubit states . the standing wave therefore allows to control the rydberg excitation pattern , which would be more difficult if a spatially homogeneous excitation pulse is used . we also note that vdw and dipole - dipole interactions of rydberg atoms are long - range . as a result , when atoms are excited to @xmath19 , the interaction strength between atoms in neighboring pairs is comparable to the interaction strength between atoms within a pair , which will result in a phase error . the error is considerably reduced by exciting atoms only in every other pair using the standing wave . the interaction strength between atoms in closest excited even pairs is then smaller by a factor of @xmath27 for vdw and @xmath28 for dipole - dipole interaction than the interaction strength within a pair . here @xmath29 is the interaction strength between atoms in a pair and @xmath30 is the interaction strength between atoms in closest exited pairs . excitation to rydberg states of alkali atoms is typically a two - photon process via intermediate @xmath31 , @xmath32 states . using the level scheme shown in fig.[fig : level - scheme]a we can write the schrdinger equation for the amplitudes of the qubit states . if the qubit is initially in the @xmath33 state , the corresponding system of equations is @xmath34 where @xmath35 , @xmath36 and @xmath37 are the amplitudes of the states @xmath33 , @xmath38 and @xmath22 ( other states are far - detuned and not populated provided that @xmath39 ; @xmath40 is the detuning from the intermediate state , @xmath41 are the rabi frequencies of the excitation pulses ) . assuming @xmath42 and @xmath43 the solution of this system is given by @xmath44 which shows that for a pulse duration @xmath45 such that @xmath46 the state evolves into @xmath47 . next , the excitation pulses are switched off and the atoms interact in the doubly excited state for time @xmath25 so that @xmath26 , and the state flips sign @xmath48 . finally , we apply the same @xmath23-pulse for time @xmath45 , bringing the system into a state @xmath49 . if the initial state is @xmath50 the system evolution is governed by equations ( similar for the @xmath51 state ) @xmath52 which in the case @xmath53 gives the solution @xmath54 where @xmath55 is the hyperfine splitting of the atomic ground state . one can see that for @xmath56 the system returns to the state @xmath57 . finally , if the system is initially in the @xmath58 state , the wavefunction evolves as @xmath59 bringing the system into @xmath60 for @xmath56 if @xmath53 . as a result , the phase gate @xmath61 is implemented . to proceed with cluster state generation the phase gate has now to be performed with odd pairs of atoms , _ i.e. _ atoms 3 and 4 , @xmath9 with @xmath10 , etc . for that the phase @xmath62 of the excitation pulse is shifted by @xmath17 to become @xmath63 . at this point all atoms in even and odd pairs become entangled . as a next step the phase gate has to be applied to neighboring atoms , where one atom belongs to an even and another to an odd pair , _ i.e. _ to atoms 2 and 3 , @xmath11 and @xmath9 , etc . and atoms 4 and 5 , @xmath10 and @xmath7 , etc . which can be done by setting @xmath64 and , finally , to @xmath65 this will result in the phase gate applied to all nearest neighbors . in fig.[fig : level - scheme]b we show the level scheme of @xmath66rb which we use as an example . a qubit can be encoded into @xmath67 and @xmath68 states , providing long coherence lifetimes due to small sensitivity to magnetic field fluctuations @xcite . a rydberg state @xmath69 with @xmath70 can be used as @xmath19 , in this case atoms will interact via isotropic vdw @xmath71 interaction . atoms can be excited to @xmath19 using two @xmath72 polarized pulses via intermediate @xmath31 @xmath73 states . the fidelity of the phase gate averaged over all initial two - qubit states is calculated in appendix a in detail and is given by @xmath74 , \end{aligned}\ ] ] where the state after the imperfect phase gate @xmath75 is compared to the state after the ideal gate @xmath76 , @xmath77 , @xmath78 , @xmath79 . in appendix a we calculate several types of intrinsic errors and find that the main errors of this type of phase gate are caused by the finite width of the ground motional state wavefunction of each atom in a lattice site . this results in a finite spread of the rabi frequency of the standing wave excitation pulse . other significant errors are due to the finite ratio of the interaction strength to the two - photon rabi frequency and the decay of population in rydberg states . the corresponding gate fidelity is @xmath80 where @xmath81 is the corresponding gate error , and averaging is over the motional ground state wavefunction . in the fidelity calculation equal rabi frequencies of the pulses @xmath42 were assumed and a decay of a rydberg state with the rate @xmath82 was introduced . in order to find the error due to the finite width of the ground state motional wavefunction we approximate the potential at each site as harmonic with the oscillation frequency @xmath83 and the corresponding wavefunction width @xmath84 for @xmath85 , which we use for an estimate . here @xmath86 is the atomic recoil energy , @xmath87 is the atomic mass . we assume the rabi frequencies of the excitation pulses @xmath88 , where @xmath12 varies around potential minima of the n@xmath89 even pair of sites @xmath90 with the gaussian probability distribution @xmath91 . the contribution of the first error term is @xmath92 choosing @xmath93 mhz and @xmath94 mhz , the second , third and fourth error terms are @xmath95 , @xmath96 and @xmath97 for the rydberg state lifetime @xmath98 @xmath99s , _ i.e. _ the total error is of the order of @xmath100 . the high value of the two - photon rabi frequency @xmath94 mhz can be achieved using a two - photon excitation path @xmath101 with @xmath102 and @xmath103 nm lasers , respectively , due to the larger dipole moment of the @xmath104 transition @xcite , allowing to reach a high rabi frequency @xmath105 . at the same time the decay time of the @xmath106 state ( 125 ns ) allows to minimize the error due to the decay of the intermediate state . the probability of the decay during the gate is @xmath107 . choosing @xmath108 ghz the probability of the intermediate state decay is @xmath109 . there is also an error due to the undesirable interaction of atoms belonging to different excited pairs , @xmath110 ( see appendix a , section a ) . assuming that interacting atoms belong to the closest excited pairs , we have @xmath111 for vdw and @xmath28 for dipole - dipole interaction . as a result , the error is @xmath112 for vdw and @xmath113 for dipole - dipole interaction . this analysis assumes that the excitation from @xmath16 to @xmath19 and back is adiabatic with respect to the lattice motional frequency and there is no uncertainty in the interaction strength , analyzed in appendix a , section c. if the excitation - deexcitation is non - adiabatic , there is an additional gate error due to the uncertainty in the interaction strength @xmath114 , where @xmath115 is the distance between nearest neighbors in the lattice . this error was derived assuming @xmath116 , which is not the case in our estimates . if @xmath117 the error will be higher , and we use the first term @xmath118 as a lower bound for this type of error . the error derivation assumes that the trapping potential for the ground and rydberg state is the same , providing the same motional frequency . this can be achieved in a blue - detuned optical lattice at a `` magic '' wavelength , at which ground and rydberg state polarizabilities are equal @xcite . on the other hand , the error can be avoided if the trapping lattice is switched off when atoms are excited to rydberg states and the gate pulses are much shorter than the motional period . in this case atoms move only a small fraction of the ground motional state width during the gate and will be recaptuted in the ground motional state once the trapping lattice is switched back on . finally , there is a finite probability to excite atoms in `` inactive '' lattice sites , sitting in minima of the standing wave excitation pulse . the probability that the pair of atoms stays in the initial state after the gate , averaged over four possible initial states , is calculated in appendix a , section e. for a two - photon excitation the averaged probability @xmath119 . the error analysis shows that first , vdw interaction results in smaller errors caused by the interaction of atoms in different excited pairs compared to the dipole - dipole one . second , the excitation to rydberg states has to be adiabatic to avoid errors due to the uncertainty of the interaction strength . on the other hand , these errors could be avoided altogether if excitation to the @xmath22 state is not required . next we therefore analyze the phase gate based on rydberg ( dipole ) blockade , where only one atom is excited to a rydberg state , provided individual addressing of atoms in a pair is possible . if individual addressing of atoms in a pair is possible , one can use standard dipole blockade @xcite , and the phase gate is realized in the limit @xmath120 as follows : 1 ) a control atom is excited from one of the qubit states , _ e.g. _ a qubit state @xmath16 to a rydberg state @xmath19 by a @xmath23-pulse ; 2 ) a @xmath121-pulse of the same frequency is applied to a target atom . dipole - dipole or vdw interaction shifts the energy of the doubly excited state @xmath22 , as a result the @xmath121-pulse has no effect on the target atom ( the excitation is blockaded ) since the rabi frequency of the pulse is much smaller than the energy shift ; 3 ) the control atom is deexcited back to its original qubit state . we stress that we do not require individual addressing of atoms in the lattice , we only need atoms in each pair to be separately addressable , this is why we call it `` individual '' addressing . individual addressing in a pair can be realized using a polarization gradient optical lattice @xcite , where not only the intensity of the lattice field but also its polarization changes periodically in space . a lattice of double - wells was demonstrated in @xcite with the polarization being linear in one site and elliptical in the neighboring site of each well . the elliptical polarization results in a non - zero contribution from a vector part of polarizability , producing a state - dependent shift @xmath122 , where @xmath123 is the vector part of the polarizability , @xmath124 is the atomic angular momentum , and @xmath125 is the positive frequency part of the total electric field @xmath126 . the contribution from the vector part can , therefore , be viewed as a fictitious magnetic field @xmath127 . atoms can be loaded into a double - well optical lattice , formed by two standing wave fields of different polarizations . let us consider the lattice field @xmath128 this lattice can be produced by two pairs of laser beams , intersecting at @xmath23 and @xmath129 angles @xcite . the corresponding lattice potential is a sum of scalar and vector parts @xmath130 , where @xmath131 is the scalar polarizability . the scalar part @xmath132 ( where @xmath133 and @xmath134 ) , represents a double - well lattice , with a spatial period @xmath135 ; the minima of the n@xmath89 double - well are at @xmath136 . the vector part is given by @xmath137 and at the minima of a double - well @xmath138 . here @xmath139 is the projection of the total angular momentum @xmath124 on the direction of the fictitious magnetic field @xmath140 ( x - direction ) , _ i.e. _ @xmath141 ; the @xmath142 signs refer to the left and right sites of the well . the scalar part of the potential is typically larger than the vector part , since the vector polarizability is about an order of magnitude smaller than the scalar part for alkalis . assuming @xmath143 , @xmath144 and @xmath85 the vector shift at the well s minima is @xmath145 . we can encode a qubit into the states @xmath146 , @xmath147 , this qubit transition is insensitive to magnetic field fluctuations at a bias magnetic field @xmath148 g @xcite . as a result , if a magnetic field is applied , the @xmath15 and @xmath16 states will shift by the same amount preserving the qubit frequency . the @xmath149 term , acting as a fictitious magnetic field , will then shift states @xmath15 , @xmath16 in the right site with respect to the qubit states in the left site of the well @xcite ( see fig.[fig : double - well]b ) , allowing to selectively excite only one atom in each pair , while having another atom unaffected by the excitation pulse . for @xmath66rb @xmath150 khz at the lattice wavelength @xmath151 nm , and the shifts for the @xmath15 , @xmath16 states are @xmath152 khz . larger shifts of the order of @xmath153 khz @xcite can be realized with optimized parameters of the lattice . if the two - photon excitation rabi frequency is much smaller than the vector shift @xmath154 , the atoms in the left and right sites can be excited to the @xmath19 state selectively . the phase gate then can be realized using dipole blockade as : 1 ) a @xmath23-pulse resonant to the @xmath24 transition of a control atom is applied , exciting it to the @xmath19 state ; a target atom is off - resonant by @xmath155 , which is the relative shift of the qubit state @xmath16 in left and right site , and is not excited ; 2 ) a @xmath121-pulse resonant to the @xmath24 transition of the target atom is applied . the doubly excited @xmath22 state is shifted by the large interaction energy @xmath21 and is not populated . the control atom is off - resonant by @xmath155 and is not affected ; 3 ) finally , a @xmath23-pulse resonant to the control atom brings it back to its original qubit state . these steps are analyzed in detail in appendix b and are summarized in the table below , which shows the evolution of the two - qubit states : @xmath156 where @xmath157 , the first qubit corresponds to the control and the second to the target atom . one can see that the phase gate can be realized provided that @xmath158 . the averaged fidelity of the phase gate with dipole blockade was calculated in appendix b and is given by @xmath159 the error terms are due to the spread of the two - photon rabi frequency experienced by each atom @xmath160 decay of rydberg states @xmath161 , and the imperfect frequency selectivity between the control and target atoms @xmath162 . here we assumed @xmath163 , @xmath144 , @xmath84 for @xmath85 , @xmath164 @xmath99s , the two - photon rabi frequency @xmath165 khz , and the vector shift of the qubit states @xmath166 khz . it shows that the total error of the phase gate in this case is @xmath167 and the phase gate time is @xmath168 @xmath99s . let us also discuss the error due to undesirable excitation of atoms in `` inactive '' wells , where the rabi frequency is close to a minimum . it was calculated in detail in appendix b , section d , and the probability to find the pair of atoms in the initial state after the gate , averaged over all four initial states , is @xmath169 in the case of two - photon excitation to rydberg states . by changing the phase of the standing wave so that intensity maxima shift to odd double - wells the phase gate can be realized in odd pairs . at this stage all pairs in double - wells are entangled . in the next subsection we describe how the phase gate can be realized with neighboring atoms belonging to different double - wells . in the previous subsection we showed how a phase gate can be applied to pairs of atoms in double - wells . to proceed with the cluster - state generation the phase gate has to be realized with the neighboring atoms in different double - wells , _ i.e. _ between each atom in a right site of the @xmath170 double - well and an atom in a left site of the @xmath171 double - well . the phase gate operations described in the previous subsection can be applied if the atoms are brought to the same double - well . this can be achieved by adiabatically manipulating the lattice in the following way : ( i ) decreasing @xmath172 ( fig.[fig : superlatt - manip]b , left panel ) which raises the barrier in each double - well , ( ii ) ramping the phase @xmath62 from @xmath173 to @xmath17 ( fig.[fig : superlatt - manip]c , left panel ) , ( iii ) finally , increasing @xmath172 back to its initial value . as a result , the atoms that were in the right and left sites of neighboring double - wells end up in the same double - well ( fig.[fig : superlatt - manip]d , left panel ) . the lattice manipulation has to be adiabatic to avoid motional excitation of atoms . modeling the evolution of the atomic motional state during the lattice manipulation requires calculation of bloch bands and bloch functions for every configuration of the lattice . these are the eigenstates and eigenfunctions of the single - particle atomic hamiltonian , which can be found by solving the schrdinger equation with the potential @xmath174 . the potential @xmath174 is periodic , and can be written as a discrete fourier sum , containing terms with wavevectors @xmath175 , in the following way : @xmath176 allowing one to write the solutions of the schrdinger equation in the form of bloch functions @xmath177 . here @xmath178 is the quasi - momentum ( restricted to the first brillouin zone ) , @xmath179 is the band index , and @xmath180 with @xmath181 a suitable cutoff number . the resulting system of equations for the @xmath182 coefficients and eigenenergies @xmath183 @xmath184 is solved numerically by truncating the sum in eq.([eq : bloch - func ] ) at some @xmath181 providing a necessary precision for the eigenenergies . we used @xmath185 to calculate the lowest bloch energies and functions . different configurations of the double - well potential and the corresponding band structures are shown in fig.[fig : bloch - bands ] . the left panel demonstrates bloch bands corresponding to the three stages of lattice manipulation , while in the right panel of the figure the energies of the @xmath186 eigenstates of several lowest bands are shown . the energies of the lowest bands weakly depend on @xmath178 , as can be seen from fig.[fig : bloch - bands](a),(c),(e ) , and the energies and bloch wavefunctions corresponding to @xmath186 can be used . given the bloch functions @xmath187 , which are delocalized over the entire lattice , one can construct wannier functions which are localized at lattice sites @xmath188 @xmath189 in a double - well lattice the two lowest bands are separated in energy by much less ( @xmath190 ) than the typical motional excitation energy @xmath191 . as a result , even at ultracold temperatures both bands are going to be populated . in this situation to obtain a correct description of the system evolution generalized wannier functions are introduced @xcite , which are superpositions of the wannier functions of different energy bands . in our case the bloch functions for the first two bands are symmetric ( ground band ) and anti - symmetric ( second band ) around the center of a double - well ( shown in fig.[fig : wannier - functions]a ) . combining the wannier functions corresponding to the two bands as @xmath192 and @xmath193 , one can obtain generalized wannier functions localized in the left and right well , respectively ( see fig.[fig : wannier - functions]b ) . the lattice parameters have to be changed adiabatically to avoid undesirable motional excitations of the atoms , _ i.e. _ , slow compared to the lattice motional energies . we checked the adiabaticity of lattice manipulation by numerical modeling of the system evolution . we took into account first four bloch states to simplify the analysis and assumed that initially atoms are in the ground bloch state . as we already mentioned , the lowest bloch bands weakly depend on the quasi - momentum @xmath178 , and , as a result , one can approximate the corresponding wannier functions by the bloch functions @xmath194 and @xmath195 , restricted to a single site . the population evolution is shown in the right panel of fig.[fig : superlatt - manip ] , along with the time - dependence of the amplitude @xmath172 and phase @xmath62 . by optimizing the manipulation time we found that @xmath196 of the population stays in the ground bloch band . we note that this value can be further increased by using more complex ( optimized ) functions @xmath197 and @xmath198 . the total time required for the manipulation is @xmath199 @xmath99s . we can now estimate the time required to generate a 1d cluster state . in the case considered in section ii a , when atoms are not addressed individually , the phase gate has to be performed adiabatically with the gate time @xmath200 @xmath99s , as a result , the excitation - deexcitation pulses have to be longer than @xmath201 @xmath99s . we can assume the phase gate duration @xmath202 @xmath99s . the resulting time of cluster state generation , including four phase gate sequences is then @xmath203 @xmath99s . if atoms in a pair are individually addressable , the gate duration was found to be @xmath204 @xmath99s in section ii b. the total time including the lattice manipulation is then @xmath205 @xmath99s . the scheme described in the previous section can be extended to generate a 2d cluster state , which is required for universal quantum computation . first , atoms can be loaded in a 2d lattice with @xmath206 , which produces a regular lattice with a period @xmath207 in @xmath208 direction and a regular or double - well lattice in @xmath12 direction . the @xmath208-lattice period is assumed sufficiently large so that atoms in neighboring @xmath12-chains do not interact when excited to rydberg states . in this way , following the steps of sections ii a or ii b , we can produce a series of 1d cluster states in the @xmath12 direction . as a next step we adiabatically reduce @xmath209 to get a regular lattice in the @xmath12-direction with a period @xmath210 ( in the case of a double - well lattice ) , followed by stretching of the @xmath12-lattice . next , the @xmath208-lattice period is reduced to bring the @xmath12-chains closer and the regular or double - well lattice @xmath211 is switched on in the @xmath208 direction . next , the entanglement operations of sections ii a or ii b can be repeated for 1d chains in the @xmath208 direction , producing a 2d cluster state . let us estimate the time required to generate the 2d state . as we showed in the end of section ii c a 1d cluster state in the @xmath12-direction requires @xmath212 @xmath99s without and @xmath213 @xmath99s with individual addressing in a pair to be realized . adiabatic reduction of @xmath172 will take @xmath214 @xmath99s , as can be seen from fig.[fig : superlatt - manip](e ) in the right panel . next , we need to estimate the time required to stretch the @xmath12-lattice . the stretch has to be adiabatic for atoms to remain in the ground motional state of the lattice . we modeled the stretch of the lattice @xmath215 so that the @xmath216 vector was adiabatically changed from @xmath217 to @xmath218 . the period of the lattice @xmath219 at the end of the stretch is comparable to the distance @xmath220 between atoms which belong to every other excited pair in the double - well . as a result , the error due to the interaction of atoms in neighboring @xmath12-chains is smaller than the error due to the interaction of atoms in neighboring excited pairs in the same chain . we modeled the evolution of the system by calculating the bloch bands and bloch functions , and used the latter to construct single - site wannier functions of the lattice @xmath174 during the stretch . we again took into account four lowest bloch bands , assumed that initially the population was in the ground bloch band , and monitored the excitation to higher - energy bands . we found that the stretch can be performed rather fast in @xmath221 @xmath99s , where @xmath150 khz for @xmath66rb , while retaining @xmath222 of the population in the ground bloch band . the populations of the lowest four bands during the lattice stretch are shown in fig.[fig : latt - stretch ] . finally , we can estimate the total time required to generate the 2d cluster state in the blockaded case . the time required for 1d state generation @xmath205 @xmath99s ; ramping down the @xmath172 lattice takes @xmath214 @xmath99s and the lattice stretch in the @xmath12-direction takes @xmath223 @xmath99s ; the total time is @xmath224 ms . next , the lattice period in the @xmath208-direction has to be adiabatically reduced , which will similarly take @xmath223 @xmath99s , followed by ramping up the @xmath225 lattice in @xmath214 @xmath99s , and a sequence of phase gate operations applied to the @xmath208-chains , will take @xmath205 @xmath99s . therefore , the total time of the 2d cluster state generation in the blockaded case is @xmath226 ms . the time required for 2d cluster state generation in the no - blockade case can be found summing the time required to produce 1d states in @xmath12-direction , @xmath203 @xmath99s , the time required to stretch the lattice in @xmath12-direction and shrink it in @xmath208 direction @xmath227 @xmath99s=@xmath228 ms , followed by application of the phase gate sequence in @xmath208-direction , requiring time @xmath203 @xmath99s . the total time needed to produce a 2d cluster state is therefore @xmath229 ms . these times are much smaller than the qubit coherence times @xmath230 s @xcite , achievable in optical lattices . in sections ii.a.2 and ii.b.2 we showed that the phase gate between two neighboring atoms can be realized with an error @xmath231 and @xmath167 without and with dipole blockade in the pair , respectively , if two - photon excitation to rydberg states is used . the major contribution to the error in both cases comes from the variation of the two - photon rabi frequency experienced by each atom due to the spatial variation of the excitation field . this error can be reduced to @xmath232 , _ i.e. _ four times , if a single - photon excitation is used instead of a two - photon one . then , @xmath233 and @xmath234 in the cases without and with dipole blockade , respectively . it will reduce the total gate error to @xmath235 without and @xmath236 with dipole blockade . the error of the blockaded phase gate can be further reduced by increasing the relative qubit energy shift @xmath237 between left and right sites in a double well . namely , the error term @xmath238 ( here one - photon excitation is assumed ) can be reduced by increasing @xmath237 . one possible way to increase @xmath155 is to use hyperfine states @xmath14 with large @xmath239 for qubit encoding , since @xmath240 . for example , if cs @xmath241 and @xmath242 states are used as qubit states @xmath16 and @xmath15 , this will allow the increase up to @xmath243 khz . the @xmath155 can be increased even further using atoms with large hyperfine quantum numbers @xmath244 , such as the rare - earth ho , having @xmath245 in the ground @xmath246 state @xcite , which will allow the increase up to @xmath247 mhz . choosing one - photon rabi frequency @xmath248 khz , the blockaded gate errors become @xmath249 and @xmath250 . the error due to rabi frequency variation @xmath251 is harder to reduce , since it pretty much depends only on the ratio @xmath252 , which can not be increased much beyond @xmath253 in order to have motional states localized in left and right sites of a double - well . using a very deep optical lattice with @xmath254 and @xmath255 , which still results in motional states localized in left and right wells , allows to reduce the error to @xmath256 . the transition between @xmath241 and @xmath242 states of cs is sensitive to magnetic field at all field values , meaning that the qubit will experience decoherence in these states . to reduce its effect one can keep the qubit in field insensitive @xmath257 and @xmath258 states during storage and lattice manipulation time and transfer it to the @xmath241 , @xmath242 only when the phase gate is applied this will result in an additional phase gate error due to magnetic field fluctuations @xmath259 , where @xmath260 is the fluctuation of the qubit transition frequency due to the fluctuating magnetic field @xmath261 and @xmath262 is the electron s gyromagnetic ratio . in a recent study @xcite suppression of magnetic field fluctuations down to @xmath263 @xmath99 g was demonstrated , resulting in @xmath264 s@xmath265 . one phase gate operation requires @xmath266 @xmath99s in our setup , resulting in the dephasing error per gate @xmath267 . as a result , the total gate error in the blockaded case can be reduced to @xmath268 . at the same time there are errors due to the excitation of atoms in `` inactive '' latice sites , situated at the minima of the standing wave . with a two - photon excitation , the probability to find the pair of `` inactive '' atoms in the initial two - qubit state after the gate , averaged over all four initial states , is @xmath119 and @xmath169 in the cases without and with blockade , respectively . the probability in the former case can be increased using multi - photon , _ e.g. _ four - photon excitation to the rydberg states . with a four - photon exitation , the probability increases up to @xmath269 in the no - blockade case . at the same time the error due to the variation of the rabi frequency increases four times up to @xmath270 , giving the total gate error @xmath271 in the no - blockade case . on the other hand , single - photon excitation which reduces the gate error in the blockaded case , leads to the reduced probability to leave atoms in `` inactive '' sites unaffected @xmath272 . we can conclude from this analysis that in no - blockade case the multi - photon excitation , _ e.g. _ four - photon , leads to the optimal combination of the phase gate error ( @xmath271 ) and the error due to the excitation of atoms in `` inactive '' lattice sites ( @xmath273 ) . at the same time , in the blockaded case one - photon excitation seems to provide the optimal error combination : phase gate error @xmath274 which can be achieved in deep optical lattices and with atoms having large vector shifts . it is combined with the probability to have `` inactive '' atoms unaffected @xmath272 . we give an error summary for the no - blockade and blockaded cases in table i. [ cols="^,^,^",options="header " , ] [ table : errors ] it was shown in @xcite that fault - tolerant mbqc can be realized with a 3d cluster state using topologically protected gates . in particular , an error threshold for an entangling two - qubit gate during the preparation of the cluster state has been calculated as @xmath275 . recently the threshold was increased to @xmath276 @xcite . unfortunately , we can not directly compare the gate errors in our scheme with this threshold , since in @xcite the error was assumed to be partially depolarizing with the phase gate acting as @xmath277+(\epsilon/15)([\hat{i}_{a}\hat{x}_{b}]+ ... +[\hat{z}_{a}\hat{z}_{b}])$ ] . in our case the errors are due to leakage out of the computational subspace . we note , however , that the errors of the phase gate without and with blockade are higher than the threshold , but not too far from it . our analysis shows that the no - blockade and blockaded case have similar phase gate errors , but it requires a very deep lattice and atoms with large hyperfine numbers in the latter case . additionally , the probabilty to have `` inactive '' atoms unaffected is higher in the no - blockade compared to the blockaded case . on the other hand , the optimal error combination in the no - blockade case requires a four - photon excitation , while in the blockaded case one - photon excitation gives the optimal error combination . generation of a 1d cluster state in the no - blockade case requires almost an order of magnitude less time ( 80 @xmath99s ) compared to the blockaded case ( 700 @xmath99s ) due to the lattice change involved in the latter case . the times required to produce a 2d cluster state are , however , comparable : 1.62 ms without and 3.4 ms with dipole blockade . from the overall experimental complexity ( deep double - well lattice , lattice manipulation required ) the blockaded scheme seems more difficult to realize . the same scheme can be applied to generate the cluster state with polar molecules . for example , molecular states with small dipole moments can be used to encode a qubit while for the phase gate molecules can be excited to a state with a large dipole moment , such that in this state molecules can interact via dipole - dipole interaction . a good candidate is @xmath278co @xcite , where the ground electronic state has a rather small permanent dipole moment @xmath279 d and long - lived nuclear spin sublevels can be used to encode a qubit . it also has a metastable @xmath280 state with a significant permanent dipole moment @xmath281 d , in which molecules can interact via dipole - dipole interaction . polar molecules , that currently or in the near future can be cooled to ultracold temperatures and placed in an optical lattice , are limited to alkali di - atoms , having permanent dipole moments of @xmath282 debye ( @xmath283 debye=@xmath284 esu cm ) . it limits the dipole - dipole interaction strength @xmath285 khz between nearest neighbors with @xmath286 nm . the rather small interaction strength makes the no - blockade phase gate preferable for polar molecules . as an example , we can estimate the gate error for @xmath278co . assuming two - photon excitation to the large dipole moment state , the error due to the variation of the rabi frequency @xmath287 , with the lattice depth @xmath85 . the error due to the finite ratio of the interaction strength to the rabi frequency @xmath288 assuming @xmath289 debye , @xmath290 nm , resulting in @xmath291 khz and @xmath292 khz . the large dipole moment molecular state is metastable with the lifetime @xmath293 ms , giving @xmath294 s@xmath265 and the decay - induced errors @xmath295 , @xmath296 , resulting in the total gate error @xmath297 . as we discussed above for atoms , the gate pulses affect molecules in `` inactive '' sites if two - photon excitation is used . the probability to have the `` inactive '' molecules unaffected can be increased using multi - photon , _ e.g. _ four - photon excitation . in this case the error due to the rabi frequency variation is four times larger , and the total gate error becomes @xmath298 , while the probability for `` inactive '' atoms to stay in the initial qubit states becomes @xmath269 . this analysis shows that gate errors for molecules and for atoms are comparable . in conclusion , we propose and analyze generation of a cluster state for measurement - based quantum computing with neutral atoms and polar molecules in an optical lattice using van der waals and dipole - dipole interactions . we consider two schemes for implementation of a phase gate between pairs of nearest neighbors required to generate the cluster state : without and with individual addressing within a pair . we show that in the former case the gate error @xmath100 is feasible with two - photon excitation to rydberg states , provided the excitation - deexcitation to interacting states is adiabatic with respect to motional frequency of the lattice . two - photon excitation , however , leads to high error due to excitation of atoms in `` inactive '' wells at the minima of the standing wave . the optimal combination of the gate error and the `` inactive '' atoms excitation error can be realized using four - photon excitation to rydberg states , giving the gate error @xmath299 and the probability to have `` inactive '' atoms unaffected @xmath269 . in the second scheme individual addressing within a pair allows to implement the phase gate using rydberg ( dipole ) blockade . addressing of atoms within a pair can be realized in a polarization - gradient double - well lattice . the gate error in this case is @xmath167 , but can be reduced to @xmath268 using one - photon excitation to rydberg states , atoms with large hyperfine quantum numbers , such as cesium and holmium , and a very deep optical lattice . with one - photon excitation the probability of `` inactive '' atoms to stay in their initial state is @xmath272 . we also analyze the lattice manipulation required in the latter case to realize the phase gate between all nearest neighbors and show under which conditions the manipulation is adiabatic . the total time required to produce a 1d cluster state @xmath203 @xmath99s without and @xmath205 @xmath99s with individual addressing in a pair , where parameters of @xmath66rb were used for the estimate . finally , we show how the scheme for 1d cluster state generation can be extended to realize a universal 2d cluster state and estimate the total time required @xmath229 ms without and @xmath226 ms with individual addressing in a pair . the authors gratefully acknowledge financial support from afosr under the muri award fa9550 - 09 - 1 - 0588 . we calculate the fidelity of the phase gate @xmath75 averaged over all initial two - qubit states @xmath58 , @xmath50 , @xmath51 and @xmath33 as follows @xmath300,\nonumber\end{aligned}\ ] ] where we project the final state after the phase gate @xmath301 ( @xmath302 ) on the state , expected after the ideal phase gate @xmath76 , @xmath77 , @xmath78 , and @xmath79 . we assume that errors due to different mechanisms are small and can be analyzed separately and added together . we limit our analysis to intrinsic gate errors assuming that technical errors can be in principle eliminated . in this subsection we calculate the errors due to imperfect excitation from the @xmath33 to the @xmath22 state . in section iia we derived a system of equations for the amplitudes @xmath35 , @xmath36 and @xmath37 of the two - qubit states @xmath33 , @xmath38 , and @xmath22 in the case when the initial state is @xmath33 ( see eq.([eq : eqs_11 ] ) ) . the corresponding hamiltonian is @xmath304 assuming @xmath42 for the simplicity of the analysis and @xmath305 we can calculate the eigenvalues of the hamiltonian : @xmath306 with the corresponding eigenfunctions @xmath307 up to the second order in @xmath303 . the state @xmath33 evolves as @xmath308 where the coefficients @xmath309 are determined from the condition @xmath310 and are given by : @xmath311 the phase gate is realized in three steps : first , a @xmath23 pulse is applied to both atoms during the time @xmath45 such that @xmath46 . next , atoms interact in the @xmath22 state and accumulate the phase @xmath312 . finally , a second @xmath23 pulse of the same duration @xmath45 de - excites atoms back to the @xmath313 state . the wavefunction at the end of the gate is as follows @xmath314\times \nonumber \\ & & \times\left(-\frac{1}{\sqrt{2}}\left(1-\frac{3}{32}\left(\frac{v_{\rm int}}{\omega^{2}/\delta}\right)^{2}\right){\left| \psi_{1 } \right\rangle}e^{-i\frac{\pi}{4}\frac{v_{\rm int}}{\omega^{2}/\delta}}+\right . \nonumber \\ & & \left . + \frac{1}{\sqrt{2}}\left(1+\frac{5}{16}\frac{v_{\rm int}}{\omega^{2}/\delta}+\frac{23}{512}\left(\frac{v_{\rm int}}{\omega^{2}/\delta}\right)^{2}\right){\left| \psi_{2 } \right\rangle}\times \right . \nonumber \\ & & \left . \times e^{-i\frac{\pi}{8}\frac{v_{\rm int}}{\omega^{2}/\delta}-\frac{5\pi i}{128}\left(\frac{v_{\rm int}}{\omega^{2}/\delta}\right)^{2 } } \right . \nonumber \\ & & \left . -\frac{1}{\sqrt{2}}\left(1-\frac{5}{16}\frac{v_{\rm int}}{\omega^{2}/\delta}+\frac{23}{512}\left(\frac{v_{\rm int}}{\omega^{2}/\delta}\right)^{2}\right){\left| \psi_{3 } \right\rangle}\times \right . \nonumber \\ & & \left . \times e^{-i\frac{\pi}{8}\frac{v_{\rm int}}{\omega^{2}/\delta}+\frac{5\pi i}{128}\left(\frac{v_{\rm int}}{\omega^{2}/\delta}\right)^{2}}\right).\end{aligned}\ ] ] expanding the exponents in @xmath303 and keeping terms up to the second order we get @xmath315 as a result , @xmath316 , with the corresponding error @xmath317 . this error is present only for the initial state @xmath33 , since for the initial states @xmath50 , @xmath51 and @xmath58 as is seen from eqs.(3),(5 ) the state @xmath22 is not populated and no interaction is involved . the averaged over all initial states error is therefore @xmath318 . this analysis also allows to calculate the error due to the unwanted interaction of atoms belonging to different excited pairs . if we assume that @xmath319 , where @xmath30 is due to the interaction between closest pairs , the corresponding error averaged over all initial states is @xmath320 . in this subsection we calculate the error due to decay of rydberg states . we can calculate this error using eq.([eq : wavefunct_11 ] ) for the @xmath33 initial state assuming that the @xmath19 state decays with the rate @xmath82 . then , the @xmath33 state at the end of the gate becomes @xmath321 where we expanded the exponent using the smallness of @xmath322 , @xmath323 . as a result , @xmath324 , giving the error @xmath325 . if the initial state is @xmath50 the system evolves according to eq.([eq : wavefunct_01 ] ) . at the end of the gate @xmath326 the overlap with the desired @xmath327 state is then @xmath328 , resulting in the error @xmath329 . similar result is obtained for the @xmath51 state . there is no error due to the rydberg state decay for the @xmath58 initial state . the averaged over all initial states error is according to eq.([eq : fidelity ] ) given by @xmath329 . when two atoms in the @xmath33 internal state and in the ground motional states of the lattice @xmath330 are transferred to a doubly excited rydberg state @xmath22 , the state @xmath331 is no longer the eigenstate of the total hamiltonian , including vdw or dipole - dipole interaction . the interaction admixes higher - energy motional states to the original ground state as well as other rydberg states to a much smaller extent . as a result , new eigenstates are superpositions of several motional states . let us call the new ground motional state of two atoms in the presence of vdw or dipole - dipole interactions @xmath332 , while @xmath333 are the higher - energy motional states . if the interaction is not very strong only closest in energy states are admixed , @xmath334 in the first order perturbation theory , where @xmath335 is the j@xmath89 excited motional state of the i@xmath89 atom , and @xmath336 ( assuming @xmath337 ) . if the atoms are excited to @xmath22 adiabatically with respect to the energy separation between @xmath332 and higher - energy states ( which is of the order of the motional energy splitting ) , the state @xmath330 gradually evolves into @xmath332 and there is no uncertainty in the interaction strength . if , however , the excitation is non - adiabatic , the state @xmath330 does not change during the excitation , _ i.e. _ it is now a superposition of the new eigenstates @xmath338 , where @xmath339 . as a result , after the first @xmath23 pulse @xmath340 during the interaction time the state @xmath341 acquires a phase factor @xmath342 , where @xmath343 is the energy difference between the @xmath341 and @xmath332 states . after the interaction time the system is in the state @xmath344 the second @xmath23 pulse brings the system back to the @xmath33 state without changing the motional state . as a result , after the second @xmath23 pulse the state looks as follows @xmath345 the projection to the ideal @xmath346 is then given by @xmath347 . since @xmath348 , and @xmath349 , the perturbation theory gives @xmath350 . as a result , the projection to the ideal final state is @xmath351 , which gives the error @xmath352 . since this error is present only for the @xmath33 initial state the averaged error is @xmath353 . in this subsection we calculate the error caused by the variation of the rabi frequency experienced by each atom due to the finite width of the ground motional state . if the initial state is @xmath33 , the effect of the rabi frequency spread can be calculated from eq.([eq : wavefunct_11 ] ) , where the rabi frequency @xmath354 , and we again assume @xmath42 . assuming also that for the two @xmath23 pulses @xmath355 , the state at the end of the gate is given by @xmath356 \nonumber\end{aligned}\ ] ] expanding the exponents up to the second order in @xmath357 we have @xmath358 , which gives the projection to the desired @xmath313 state @xmath359 . if the initial state is @xmath50 the wavefunction is given by eq.([eq : wavefunct_01 ] ) @xmath360 assuming again @xmath354 and @xmath361 after the two @xmath23 pulses , the wavefunction becomes @xmath362 the projection to the ideal @xmath327 state is then @xmath363 giving the error @xmath364 . a similar result is obtained for the @xmath51 state . finally , if the initial state is @xmath58 , it evolves according to eq.([eq : eqs_00 ] ) . after the phase gate the state turns into @xmath365 . the overlap with the desired @xmath58 state is @xmath366 . the averaged over all initial states according to eq.([eq : fidelity ] ) error is then @xmath367 , where the averaging in the error expression is over the ground motional state wavefunction . we also need to estimate the probability that the atoms in `` inactive '' lattice sites , _ i.e. _ at the minima of the standing wave excitation pulse , are not affected . let us denote @xmath368 the two - photon rabi frequency at these sites . now , as can be seen from eq.([eq : wavefunct_11 ] ) the @xmath33 initial state after the gate becomes @xmath369 . \nonumber\end{aligned}\ ] ] the overlap with the initial @xmath33 state is then @xmath370 . \nonumber\end{aligned}\ ] ] the initial @xmath50 state ( the same for @xmath51 state ) , according to eq.([eq : wavefunct_01 ] ) after the gate becomes @xmath371 which results in the overlap with the initial @xmath50 state @xmath372 . the initial state @xmath58 becomes @xmath373 after the gate , according to eq.([eq : eqs_00 ] ) . the rabi frequency at the minima of the standing wave @xmath374 , at the maxima @xmath375 , as a result , the ratio @xmath376 . the probability to find the pair of atoms in the initial state after the gate , averaged over all initial states , is then @xmath377 . next , we calculate the errors of the phase gate with individual addressing in a pair of atoms . we start by calculating the error due to the finite ratio of the two - photon rabi frequency to the vector shift @xmath237 between atoms in the left and right wells and between the rabi frequency and the interaction strength @xmath29 , resulting in imperfect individual addressing . [ [ error - due - to - a - finite - ratio - of - the - rabi - frequency - to - the - vector - shift - omega2deltadelta_rm - vec - and - interaction - strength - omega2deltav_rm - int ] ] error due to a finite ratio of the rabi frequency to the vector shift @xmath378 and interaction strength @xmath379 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ first we consider the @xmath33 initial state . the phase gate is implemented in the following way : a @xmath23 pulse resonant to the @xmath24 transition of the control , _ e.g. _ the left atom is applied . the states @xmath33 , @xmath380 , @xmath381 and @xmath22 ( other states are far detuned and have much smaller amplitudes ) evolve according to the hamiltonian : we assume @xmath42 , @xmath383 , and @xmath384 . in the analysis we will , therefore , keep terms up to the first order in @xmath379 and up to the second order in @xmath378 . neglecting @xmath385 and @xmath386 as small , we have a system of equations for @xmath35 and @xmath387 governed by the hamiltonian : the corresponding eigenstates and eigenfunctions are @xmath389 and @xmath390 the initial state @xmath33 evolves as @xmath391 where @xmath392 after the first @xmath23 pulse the wavefunction , therefore , turns into @xmath393 next , a @xmath121 pulse resonant to the target , _ i.e. _ the right atom is applied . the system of equations for the amplitudes @xmath35 , @xmath394 , @xmath387 and @xmath37 is the same as eqs.([eq : eqs_ind_addr_11_1 ] ) with @xmath395 . the system of equations for @xmath35 and @xmath394 is also the same as eqs.([eq : eqs_ind_addr_11_2 ] ) with @xmath395 . as a result , during the @xmath121 pulse the state evolves as @xmath396 where @xmath397 are given by eqs.([eq : wavefunct_ind_addr_11 ] ) with @xmath395 and @xmath398 , @xmath399 are given by eqs.([eq : coefficients_11 ] ) . after the @xmath121 pulse the states @xmath33 and @xmath381 change as @xmath400 now we analyze the evolution of the @xmath50 state . during the first @xmath23 pulse , resonant to the left atom , the amplitudes of the states @xmath50 and @xmath407 ( other amplitudes are much smaller ) evolve according to the hamiltonian @xmath408 the initial state @xmath50 changes as @xmath409 where the eigenstates are given by @xmath410 during the second @xmath23 pulse the states @xmath50 and @xmath407 change in the following way @xmath414 where @xmath415 are given by eqs.([eq : psipm ] ) and @xmath416 , @xmath417 . after the second @xmath23 pulse the initial @xmath50 state becomes @xmath418 . \nonumber \end{aligned}\ ] ] the projection on the @xmath327 state is then @xmath419 assuming @xmath420 we have @xmath421 , and the resulting error is @xmath422 . during the first @xmath23 pulse the amplitudes @xmath423 and @xmath424 ( other states are far detuned and their amplitudes are much smaller ) change according to the hamiltonian @xmath425 during the first @xmath23 pulse the states @xmath50 and @xmath407 evolve as @xmath426 after the first @xmath23 pulse the states @xmath51 , @xmath427 become @xmath428 at the end of the @xmath121 pulse resonant to the right atom the state @xmath381 is given by @xmath441 finally , after the second @xmath23 pulse resonant to the left atom , @xmath442 as a result , at the end of the gate @xmath443 from the overlap @xmath444 the error is @xmath445 . the initial state @xmath451 after the first @xmath23 pulse evolves into @xmath452 . after the @xmath121 pulse @xmath453 . the state @xmath427 turns into @xmath454 after the second @xmath23 pulse . as a result , at the of the gate @xmath455 . this gives @xmath456 resulting in the error @xmath457 . finally , the initial state @xmath79 after the gate and there is no error due to the rydberg state decay . first , we analyze the @xmath33 initial state . during the first @xmath23 pulse , resonant to the left atom , the state evolves as @xmath459 again , we assume that @xmath354 and @xmath355 during a @xmath23 pulse . as a result , after the first @xmath23 pulse @xmath460 during the @xmath121 pulse resonant to the right atom , the states @xmath33 and @xmath381 evolve as @xmath461 so that at the end of the @xmath121 pulse @xmath462 the initial state @xmath33 after the @xmath121 pulse becomes during the @xmath121 pulse resonant to the right atom , the @xmath50 state changes as @xmath470 after the @xmath121 pulse the initial state becomes @xmath471 during the second @xmath23 pulse resonant to the left atom , the states @xmath50 and @xmath407 evolve as @xmath472 as a result , at the end of the gate the initial state @xmath50 turns into @xmath473 from @xmath474 the error is @xmath475 . the initial state @xmath51 during the first @xmath23 pulse evolves as @xmath476 after the @xmath23 pulse @xmath477 during the @xmath121 pulse resonant to the right atom the states @xmath51 and @xmath427 change as @xmath478 _ i.e. _ after the @xmath121 pulse @xmath479 at the end of the @xmath121 pulse the initial state becomes @xmath480 after the second @xmath23 pulse the states @xmath51 and @xmath427 turn into @xmath481 if @xmath420 , the state @xmath51 at the end of the gate becomes @xmath482 from @xmath483 the error is @xmath475 . the initial state @xmath33 after the first @xmath23 pulse turns into @xmath487 where @xmath488 and @xmath489 are the rabi frequencies at the lattice sites , corresponding to the minima and the maxima of the standing wave , and @xmath490 . after the @xmath121 pulse resonant to the right atom , the states @xmath33 and @xmath381 evolve as @xmath491 during the second @xmath23 pulse , resonant to the right atom , the states @xmath33 , @xmath381 and @xmath380 change as @xmath492 at the end of the gate the state @xmath33 becomes @xmath493 as a result , the overlap with the initial @xmath33 state is @xmath494 the probability that after the gate the atoms stay in @xmath33 is @xmath495 where we assumed @xmath420 . if the initial state is @xmath50 , it evolves into @xmath499 after the first @xmath23 pulse resonant to the left atom . after the @xmath121 pulse resonant to the right atom the @xmath50 state turns into @xmath500 during the second @xmath23 pulse resonant to the left atom , the states @xmath50 and @xmath427 change as @xmath501 at the end of the gate @xmath502 the overlap with the initial @xmath50 state is then @xmath503 , resulting in the probability to find the atomic pair in the initial state after the gate @xmath504 . the initial state @xmath51 during the first @xmath23 pulse resonant to the left atom evolves into @xmath505 during the @xmath121 pulse , resonant to the right atom , the states @xmath51 and @xmath427 change as @xmath506 during the second @xmath23 pulse the states @xmath51 and @xmath427 evolve as @xmath507 at the end of the gate the initial state @xmath51 becomes @xmath508 which gives the overlap with the initial @xmath51 state @xmath509 resulting in the probability to find the pair of atoms in the initial state @xmath510 , where we assumed that @xmath420 . the state @xmath58 after the gate bcomes @xmath511 , giving @xmath512 . the probability to find the pair of atoms in the initial state after the gate , averaged over all four initial states , is @xmath513 . p. walther , k. j. resch , t. rudolph , e. schenck , h. weinfurter , v. vedral , m. aspelmeyer , a. zeilinger , nature * 434 * , 169 ( 2005 ) ; c .- y . lu , x .- q . zhou , o. chne , w .- b . gao , j. zhang , z .- s . yuan , a. goebel , t. yang , j .- w . pan , nat . phys . * 3 * , 91 ( 2007 ) . e. urban , t. a. johnson , t. henage , l. isenhower , d. d. yavuz , t. g. walker , m. saffman , nat . phys . * 5 * , 110 ( 2009 ) ; a. gaetan , y. miroshnychenko , t. wilk , a. chotia , m. viteau , d. comparat , p. pillet , a. browayes , p. grangier , nat . * 5 * , 115 ( 2009 ) .

we use young s raising operators to introduce and study _ double theta polynomials _ , which specialize to both the theta polynomials of buch , kresch , and tamvakis , and to double ( or factorial ) schur-polynomials and-polynomials . these double theta polynomials give giambelli formulas which represent the equivariant schubert classes in the torus - equivariant cohomology ring of symplectic grassmannians , and we employ them to obtain a new presentation of this ring in terms of intrinsic generators and relations .

1410.8329

fix the nonnegative integer @xmath2 and let @xmath3 be the grassmannian which parametrizes isotropic subspaces of dimension @xmath4 in the vector space @xmath5 , equipped with a symplectic form . in 2008 , buch , kresch , and tamvakis @xcite introduced the theta polynomials @xmath6 , a family of polynomials in independent variables @xmath7 , defined using young s raising operators @xcite . when the @xmath7 are mapped to the special schubert classes , which are the chern classes of the tautological quotient vector bundle over @xmath8 , then the @xmath6 represent the schubert classes in @xmath9 . these theta polynomials give a combinatorially explicit , intrinsic solution to the _ giambelli problem _ for the cohomology ring of @xmath8 . this paper is concerned with giambelli formulas for the _ equivariant schubert classes _ in the equivariant cohomology ring @xmath10 , where @xmath11 denotes a maximal torus of the complex symplectic group . as explained in @xcite and @xcite , such formulas are equivalent to corresponding ones within the framework of degeneracy loci of vector bundles @xcite . in 2009 , the theta polynomials @xmath6 were extended to obtain representing polynomials in the torus - equivariant cohomology ring of @xmath8 , and more generally , of any isotropic partial flag variety , by the first author @xcite . the _ equivariant giambelli problem _ was thus solved in a uniform manner for any classical @xmath12 space , in terms of positive combinatorial formulas which are _ native to @xmath12_. see @xcite for an exposition of this work . it has been known for some time ( cf . @xcite ) that the equivariant schubert classes of the usual ( type a ) grassmannian and of the lagrangian grassmannian @xmath13 may be represented by jacobi - trudi type determinants and schur pfaffians which generalize the classical results of @xcite . the polynomials which appear in these formulas are called double ( or factorial ) schur @xmath0-polynomials and @xmath1-polynomials , respectively ( cf . @xcite ) . our aim here is to use raising operators to define _ double theta polynomials _ @xmath14 , which advance the theory of the single theta polynomials @xmath6 from @xcite to the equivariant setting , and specialize to both of the aforementioned versions of double schur polynomials . the @xmath14 represent the equivariant schubert classes on @xmath8 , but differ from the equivariant giambelli polynomials given in @xcite . it follows that the two theories must agree up to the ideal @xmath15 of _ classical relations _ ( [ relations ] ) among the variables @xmath7 . we discuss this in detail in [ compare ] and corollary [ comp ] . for the rest of this paper , @xmath16 and @xmath17 will denote two families of commuting variables . we set @xmath18 and @xmath19 for any @xmath20 . for any integers @xmath21 and @xmath22 , define the elementary and complete symmetric polynomials @xmath23 and @xmath24 by the generating series @xmath25 respectively . let @xmath26 , @xmath27 , and @xmath28 , where @xmath29 denotes the kronecker delta . furthermore , if @xmath30 then define @xmath31 . we will work throughout with integer sequences @xmath32 which are assumed to have finite support , when they appear as subscripts . for any positive integers @xmath33 and integer sequence @xmath34 , define the operator @xmath35 by @xmath36 a _ raising operator _ @xmath37 is any monomial in the basic operators @xmath35 . the sequence @xmath38 is a _ composition _ if @xmath39 for all @xmath40 , and a _ partition _ if @xmath41 for all @xmath42 . as is customary , we will identify partitions with their young diagram of boxes . fix the nonnegative integer @xmath2 . for any integers @xmath43 and @xmath44 , define @xmath45 and for any integer sequences @xmath38 and @xmath46 , let @xmath47 given any raising operator @xmath37 , define @xmath48 . it is important that the variables @xmath49 in the monomials @xmath50 are regarded as _ noncommuting _ for the purposes of this action . we say that a partition @xmath51 is _ @xmath2-strict _ if all parts @xmath52 which are strictly greater than @xmath2 are distinct . to any such @xmath51 , we attach a finite set of pairs @xmath53 and a sequence @xmath54 defined by @xmath55 following @xcite , consider the raising operator expression @xmath56 given by @xmath57 [ thdef ] for any @xmath2-strict partition @xmath51 , the _ double theta polynomial _ @xmath58 is defined by @xmath59 the single theta polynomial @xmath6 of @xcite is given by @xmath60 we next relate the double theta polynomials @xmath14 to the torus - equivariant schubert classes on @xmath8 . let @xmath61 denote the standard symplectic basis of @xmath5 and let @xmath62 be the subspace spanned by the first @xmath40 vectors of this basis , so that @xmath63 for @xmath64 . let @xmath65 denote the stabilizer of the flag @xmath66 in the symplectic group @xmath67 , and let @xmath68 be the associated maximal torus in the borel subgroup @xmath65 . the @xmath68-equivariant cohomology ring @xmath69 is defined as the cohomology ring of the borel mixing space @xmath70 . the schubert varieties in @xmath8 are the closures of the @xmath65-orbits , and are indexed by the @xmath2-strict partitions @xmath51 whose young diagram fits in an @xmath71 rectangle . any such @xmath51 defines a schubert variety @xmath72 of codimension @xmath73 by @xmath74 since @xmath75 is stable under the action of @xmath68 , we obtain an _ equivariant schubert class _ @xmath76^{t_n}:=[et_n\times^{t_n}x_{\lambda}]$ ] in @xmath77 . following @xcite and @xcite , we consider the _ stable _ equivariant cohomology ring of @xmath78 , denoted by @xmath79 . the latter is defined by @xmath80 where the inverse limit of the system @xmath81 is taken in the category of graded algebras . the surjections ( [ embeddings ] ) are induced from the natural inclusions @xmath82 of the weyl groups of type c , as in @xcite , @xcite , and [ trans ] of the present work . moreover , the variables @xmath83 are identified with the characters of the maximal tori @xmath68 in a compatible fashion , as in loc . one then has that @xmath79 is a free @xmath84$]-algebra with a basis of stable equivariant schubert classes @xmath85^{t_n},\ ] ] one for every @xmath2-strict partition @xmath51 . we view @xmath86 as a @xmath84$]-module via the natural projection map @xmath84\to { { \mathbb z}}[t_1,\ldots , t_n]$ ] . consider the graded polynomial ring @xmath87:={{\mathbb z}}[c_1,c_2,\ldots]$ ] where the element @xmath7 has degree @xmath43 for each @xmath88 . let @xmath89 $ ] be the homogeneous ideal generated by the relations @xmath90 and define the quotient ring @xmath91/{i^{(k)}}$ ] . we call the graded polynomial ring @xmath92 $ ] the _ ring of double theta polynomials_. [ mainthm ] the polynomials @xmath14 , as @xmath51 runs over all @xmath2-strict partitions , form a free @xmath84$]-basis of @xmath92 $ ] . there is an isomorphism of graded @xmath84$]-algebras @xmath93\to { { \mathbb h}}_t(\operatorname{ig}_k)\ ] ] such that @xmath14 is mapped to @xmath94 , for every @xmath2-strict partition @xmath51 . for every @xmath95 , the morphism @xmath96 induces a surjective homomorphism of graded @xmath84$]-algebras @xmath97\to \operatorname{\mathrm{h}}_{t_n}^*(\operatorname{ig}(n - k,2n))\ ] ] which maps @xmath14 to @xmath76^{t_n}$ ] , if @xmath51 fits inside an @xmath71 rectangle , and to zero , otherwise . the equivariant cohomology ring @xmath86 is presented as a quotient of @xmath92 $ ] modulo the relations @xmath98 and @xmath99 where @xmath100 denotes the partition @xmath101 with @xmath43 parts . theorem [ mainthm ] has a direct analogue for the torus - equivariant cohomology ring of the odd orthogonal grassmannian @xmath102 ; this follows from known results ( cf . * and 5.1 ) ) . in this case , the polynomials @xmath103 represent the equivariant schubert classes , where @xmath104 denotes the number of parts @xmath52 of @xmath51 which are strictly bigger than @xmath2 . theorem [ mainthm ] therefore generalizes the results of pragacz @xcite and ikeda and naruse @xcite to the equivariant cohomology of all symplectic and odd orthogonal grassmannians . let @xmath105 denote the universal exact sequence of vector bundles over @xmath78 , with @xmath106 the trivial bundle of rank @xmath107 and @xmath108 the tautological subbundle of rank @xmath4 . the @xmath68-equivariant vector bundles @xmath108 , @xmath106 , and @xmath109 have equivariant chern classes , denoted by @xmath110 , @xmath111 , and @xmath112 , respectively . define @xmath113 by the equation of total chern classes @xmath114 . we now have the following chern class formula for the schubert classes @xmath76^t$ ] . [ maincor ] let @xmath51 be a @xmath2-strict partition that fits inside an @xmath71 rectangle . then we have @xmath115^t = { \theta}_{\lambda}(e - e'-f_{n+{\beta}({\lambda } ) } ) = r^{\lambda}\ , c^t_{\lambda}(e - e'-f_{n+{\beta}({\lambda})})\ ] ] in the equivariant cohomology ring @xmath77 . the equivariant chern polynomial in ( [ genkaz ] ) is interpreted as the image of @xmath116 under the @xmath117-linear map which sends the noncommutative monomial @xmath118 to @xmath119 , for every integer sequence @xmath38 . formula ( [ genkaz ] ) is a natural generalization of kazarian s multi - pfaffian formula ( * ? ? ? 1.1 ) to arbitrary equivariant schubert classes on symplectic grassmannians . the double theta polynomials @xmath14 were defined in wilson s 2010 university of maryland ph.d . thesis @xcite , whose aim was to apply the raising operator approach of @xcite to the theory of factorial schur polynomials and equivariant giambelli formulas . of @xmath14 in the ring of type c double schubert polynomials of @xcite . ] in her dissertation , wilson gave a direct proof that the @xmath14 satisfy the _ equivariant chevalley rule _ in @xmath79 , i.e. , the combinatorial formula for the expansion of the products @xmath120 in the basis of stable equivariant schubert classes . this fits into a program for proving theorem [ mainthm ] , and its extension to equivariant quantum cohomology , by applying mihalcea s characterization theorem ( * ? ? ? 8.2 ) ( see [ trans ] ) . the proof of the chevalley rule for the @xmath121 which we present in [ prelim ] and [ cfdtp ] uses the technical tools for working with raising operators constructed in @xcite , but the argument is much simpler , since it avoids the general substitution rule employed in op . it was conjectured by wilson in @xcite that the @xmath14 represent the stable equivariant schubert classes on @xmath8 . ikeda and matsumura ( 1.2 ) established this result , which is the main ingredient behind the new presentation of the equivariant cohomology ring @xmath77 displayed in theorem [ mainthm ] . a key technical achievement in their proof was to show that the @xmath122 are compatible with the action of the ( left ) divided difference operators on @xmath92 $ ] . in section [ gdds ] , we explain how this fact may be combined with a formula for the equivariant schubert class of a point on @xmath78 , to obtain a proof of wilson s conjecture ; the argument found in @xcite is different , and uses localization in equivariant cohomology . it is easy to see that the polynomial @xmath14 may be written formally as a sum of schur pfaffians ( cf . proposition [ kstricts ] ) . this identity was used in @xcite in the proof of the compatibility of the @xmath122 with divided differences , allowing them to avoid the language of raising operators entirely . in section [ dds ] , we eliminate this feature of their argument , and work directly with the raising operator expressions @xmath56 . this makes the proof more transparent , and leads naturally to a companion theory of _ double eta polynomials _ @xmath123 , which represent the equivariant schubert classes on even orthogonal grassmannians ; see @xcite for further details . the role of the first author of this article has been mostly expository , extending the point of view found in @xcite to the present setting , and comparing the results with various earlier formulas in the literature . we emphasize that the double theta polynomials @xmath14 defined here are new , and are not a special case of the type c double schubert polynomials of @xcite , which represent the equivariant schubert classes on complete symplectic flag varieties . recently , hudson , ikeda , matsumura , and naruse @xcite proved an analogue of our cohomological formula in connective @xmath124-theory , and anderson and fulton @xcite obtained related chern class formulas for more general degeneracy loci . the latter work features a powerful geometric approach to the theory , which is uniform for all four classical lie types . this paper is organized as follows . section [ prelim ] establishes some preliminary lemmas which are required in our proof that the @xmath14 obey the equivariant chevalley rule ( theorem [ cvthm ] ) ; the latter is completed in sections [ cfdtp ] and [ trans ] . although theorem [ cvthm ] is a consequence of theorem [ mainthm ] , it is a much earlier result , and the independent proof given here exhibits alternating properties of the @xmath14 which are useful in other contexts . section [ firstexs ] is concerned with how the @xmath14 behave in special cases , and relates them to some earlier formulas . in section [ dds ] , we observe that the action of the divided differences on @xmath92 $ ] lifts to @xmath125 $ ] , and prove the key result about them and the @xmath14 ( proposition [ uniq ] ) using our raising operator approach . the proof of theorem [ mainthm ] and its corollaries is contained in section [ gdds ] . we thank andrew kresch and leonardo mihalcea for helpful discussions related to this project , and takeshi ikeda and tomoo matsumura for informative correspondence regarding their joint work . we begin with a purely formal lemma regarding the elements @xmath126 of @xmath125 $ ] . for @xmath30 we let @xmath127 . [ ctlem ] suppose that @xmath128 . _ ( a ) _ assume that @xmath129 . then we have @xmath130 _ ( b ) _ assume that @xmath131 . then we have @xmath132 for any positive integer @xmath133 , we have the basic equations @xmath134 if @xmath135 , it follows that @xmath136 and the result of part ( a ) is proved . for ( b ) , let @xmath137 and compute @xmath138 ( in the last equality , notice that we have added the term @xmath139 to the first sum , and added @xmath140 to the second sum . ) the result follows . let @xmath141 , equipped with the partial order @xmath142 defined by @xmath143 if and only if @xmath144 and @xmath145 . a finite subset @xmath146 of @xmath147 is a _ valid set of pairs _ if it is an order ideal in @xmath147 . an _ outer corner _ of a valid set of pairs @xmath146 is a pair @xmath148 such that @xmath149 is also a valid set of pairs . given a valid set of pairs @xmath146 and an integer sequence @xmath150 , we define sequences @xmath151 and @xmath152 by @xmath153 and the raising operator @xmath154 finally , let @xmath155 let @xmath156 denote the @xmath157-th standard basis vector in @xmath158 . the next result is the analogue of ( * ? ? ? * lemma 1.4 ) that we require here . [ tident ] let @xmath159 be an outer corner of the valid set of pairs @xmath146 , and suppose that the integer sequence @xmath150 satisfies @xmath160 . let @xmath161 for some integer @xmath162 and set @xmath163 . then we have @xmath164 in @xmath125 $ ] . in particular , if @xmath165 , then @xmath166 observe that since @xmath159 is an outer corner of @xmath146 , we have @xmath167 while @xmath168 . it follows that @xmath169 for any integer @xmath43 , lemma [ ctlem](a ) gives @xmath170 and hence @xmath171 notice that @xmath172 . therefore lemma [ ctlem](b ) gives @xmath173 and we deduce that @xmath174 since we have @xmath175 the first equality follows by combining ( [ qu ] ) with ( [ ccequ ] ) . to prove ( [ cancel ] ) , note that @xmath176 and @xmath177 , therefore @xmath178 . [ commutea ] let @xmath179 and @xmath180 be integer vectors . let @xmath146 be a valid set of pairs such that @xmath181 and for each @xmath182 , @xmath183 if and only if @xmath184 . then for any integers @xmath44 and @xmath185 , we have @xmath186 in @xmath125 $ ] . in particular , if @xmath187 , then @xmath188 . if @xmath189 is any integer sequence and @xmath190 , define @xmath191 the hypotheses on @xmath146 and @xmath157 imply that @xmath192 the argument is now the same as in the proof of ( * ? ? * lemma 1.2 ) ; the details are spelled out in ( * ? ? ? * lemma 10 ) . throughout this paper , we will often write equalities that hold only in the rings @xmath193 and @xmath92 $ ] , where we have imposed the relations ( [ relations ] ) on the generators @xmath7 . whenever these relations are needed , we will emphasize this by noting that the equalities are true in @xmath92 $ ] rather than in @xmath125 $ ] . [ r12lem ] given @xmath194 and @xmath195 , we have @xmath196 in @xmath92 $ ] . the relations ( [ relations ] ) readily imply that @xmath197 whenever @xmath198 ( see ( * ? ? ? ( 9 ) ) ) . for any raising operator @xmath37 , we have @xmath199 we deduce that @xmath200 lemma [ r12lem ] admits the following generalization . [ commutec ] let @xmath179 and @xmath180 be integer vectors . let @xmath146 be a valid set of pairs such that @xmath201 and for each @xmath202 , @xmath203 if and only if @xmath204 . if @xmath205 and @xmath206 , then we have @xmath207 in @xmath92 $ ] . in particular , if @xmath208 , then @xmath188 . the hypotheses on @xmath146 and @xmath157 imply that @xmath209 the rest of the proof follows ( * ? ? ? * lemma 1.3 ) , using lemma [ r12lem ] in place of ( * ? ? ? ( 9 ) ) ; see ( * ? ? ? * lemma 11 ) for the details . we begin this section with a basis theorem for the @xmath84$]-algebra @xmath92 $ ] . [ basisthm ] the monomials @xmath210 , the single theta polynomials @xmath6 , and the double theta polynomials @xmath14 form three @xmath84$]-bases of @xmath92 $ ] , as @xmath51 runs over all @xmath2-strict partitions . it follows from ( * ? ? ? 5.2 ) that the monomials @xmath210 and the single theta polynomials @xmath6 for @xmath51 a @xmath2-strict partition form two @xmath117-bases of @xmath193 . we deduce that these two families are also @xmath84$]-bases of @xmath92 $ ] . by expanding the raising operator definition of @xmath14 , we obtain that @xmath211 where @xmath212 $ ] and the sum is over @xmath2-strict partitions @xmath150 with @xmath213 in dominance order or @xmath214 . therefore , the @xmath122 for @xmath51 @xmath2-strict form another @xmath84$]-basis of @xmath92 $ ] . for the remainder of this section , @xmath215 will be used to denote @xmath14 . to state the chevalley formula for the double theta polynomials @xmath215 , we need to recall certain definitions from @xcite for the pieri products with the divisor class . we say that the box @xmath216 $ ] in row @xmath44 and column @xmath217 of a young diagram is _ @xmath2-related _ to the box @xmath218 $ ] if @xmath219 are @xmath2-related . -related boxes in a @xmath2-strict young diagram.,title="fig : " ] for any two @xmath2-strict partitions @xmath220 and @xmath150 , we write @xmath221 if @xmath150 may be obtained by ( i ) adding a box to @xmath51 or ( ii ) removing @xmath44 boxes from one of the first @xmath2 columns of @xmath51 and adding @xmath222 boxes to a single row of the result , so that the removed boxes and the bottom box of @xmath150 in that column are each @xmath2-related to one of the added boxes . if @xmath223 , let @xmath224 if @xmath225 and the added box in @xmath150 is not in column @xmath226 and is not @xmath2-related to a bottom box in one of the first @xmath2 columns of @xmath51 , and otherwise set @xmath227 . fix a @xmath2-strict partition @xmath51 and let @xmath228 denote the _ length _ of @xmath51 , that is , the number of non - zero parts @xmath52 . recall that the _ @xmath2-length _ of @xmath51 is @xmath229 [ cvthm ] the equation @xmath230 holds in @xmath92 $ ] , where the sum is over all @xmath2-strict partitions @xmath150 with @xmath231 . we begin by studying the partitions @xmath150 which appear on the right hand side of the product rule ( [ cr ] ) . it is an easy exercise to show that @xmath232 for all such @xmath150 ( a more general result is proved in ( * ? ? ? * lemma 2.1 ) ) . suppose that @xmath150 is a @xmath2-strict partition such that @xmath233 . we say that @xmath150 and the chevalley term @xmath234 are _ of type i _ if @xmath235 ; otherwise , @xmath150 and @xmath234 are said to be _ of type ii_. suppose that @xmath223 . it is then straightforward to check the following assertions : first , if @xmath150 is of type i , then @xmath227 . second , if @xmath150 is of type ii , then exactly one of the following three possibilities holds : ( a ) @xmath224 , in which case @xmath236 is a single box in some row @xmath237 , and @xmath238 consists of a single pair in row @xmath237 ; ( b ) @xmath236 is a single box in some row @xmath237 with @xmath239 , and the pairs in @xmath238 are all contained in column @xmath237 ; or ( c ) @xmath236 has @xmath222 boxes in some row @xmath237 and @xmath238 consists of @xmath44 pairs in row @xmath237 . we say that @xmath150 and the chevalley term @xmath234 are of type i , ii(a ) , ii(b ) , or ii(c ) , respectively . define @xmath240 . we represent the valid set of pairs @xmath241 as positions above the main diagonal of a matrix , denoted by dots in figure [ bgh ] . for @xmath242 , we let @xmath243 and @xmath244 ( by convention , we set @xmath245 ) . figure [ bgh ] illustrates these invariants . with @xmath246 , @xmath247 , and @xmath248 . ] [ lambdabg ] if @xmath249 then we have @xmath250 . the result is the same as ( * ? ? ? * lemma 3.5 ) , but we repeat the argument here for completeness . the inequality is clear if @xmath251 , as @xmath51 is @xmath2-strict and @xmath252 . if @xmath253 , then since @xmath254 and @xmath255 , we obtain @xmath256 for any @xmath257 , let @xmath258 . set @xmath259 and @xmath260 for any @xmath261 define the raising operator @xmath262 by @xmath263 we compute that @xmath264 and therefore @xmath265 we wish to replace @xmath266 by @xmath267 in ( [ initequiv ] ) ; note that this substitution only affects the term @xmath268 . since @xmath269 , we have @xmath270 and hence @xmath271 we examine the terms @xmath272 for @xmath273 $ ] , which appear on the right hand side of equation ( [ initequiv2 ] ) . if @xmath274 , then @xmath275 and @xmath276 . for any integer @xmath43 , lemma [ ctlem](b ) gives @xmath277 it follows that @xmath278 if @xmath279 , then @xmath280 and @xmath281 . lemma [ ctlem](a ) gives @xmath282 and hence @xmath283 finally , we have @xmath284 . we deduce that @xmath285 we claim that @xmath286 the proof of ( [ subs ] ) proceeds by showing that for every @xmath273 $ ] , the term @xmath287 is equal to ( i ) zero , or ( ii ) a chevalley term @xmath234 , or ( iii ) a sum @xmath288 of two distinct chevalley terms , exactly one of which is of type ii(a ) . the terms @xmath287 are decomposed exactly as in the non - equivariant setting of @xcite , using lemmas [ tident ] , [ commutea ] , and [ commutec ] as the main tools . however the analysis here is much easier , so we provide a short self - contained argument below . given any integer sequence @xmath150 , define a weight condition @xmath289 on @xmath150 by @xmath290 * case 1 . * terms @xmath287 with @xmath291 ( this condition always holds if @xmath292 ) . notice that we must have @xmath293 . * a ) * suppose that there is no outer corner @xmath294 of @xmath241 in column @xmath237 such that @xmath295 holds for @xmath296 . if @xmath297 , we have @xmath298 by lemma [ commutea ] with @xmath299 ( note that @xmath300 implies that @xmath301 , so the conditions of the lemma hold ) . otherwise , @xmath302 is a chevalley term of type i ( in the special case that @xmath303 we must have @xmath304 , so the result is still true ) . * b ) * if there exists an outer corner @xmath294 of @xmath241 in column @xmath237 such that @xmath295 holds for @xmath296 , then let @xmath305 be maximal such that @xmath306 holds for @xmath296 . observe that in this case , @xmath307 and hence @xmath308 or @xmath309 . notice also that we must have @xmath310 for @xmath311 . by definition , we have @xmath312 and @xmath313 . we claim that @xmath314 . indeed , we know that @xmath315 and @xmath316 , therefore @xmath317 . the claim now follows , since @xmath318 . let @xmath319.\ ] ] using lemma [ tident ] and the claim gives @xmath320 continuing in the same manner , we obtain @xmath321 since @xmath310 for @xmath311 , lemma [ commutec ] implies that all terms in the sum over @xmath157 vanish , except for the initial one . therefore , we have @xmath322 if @xmath297 , then lemma [ commutea ] implies that @xmath323 . this must be the case since @xmath324 implies that @xmath325 is not an outer corner of @xmath326 . if @xmath303 and @xmath327 , then @xmath328 , @xmath309 , and @xmath329 , by lemma [ commutec ] . the terms that remain after these cancellations are chevalley terms ; note that @xmath330 is of type ii(a ) , and @xmath331 is of type ii(b ) . * case 2 . * terms @xmath287 with @xmath332 ( and hence @xmath252 ) . notice that @xmath333 for all such @xmath237 , so @xmath334 . * a ) * suppose that there is no outer corner @xmath335 of @xmath241 such that @xmath336 holds for @xmath296 . if @xmath275 and @xmath337 , then we have @xmath298 by lemma [ commutec ] . in all other cases , @xmath302 is a chevalley term of type i. * b ) * suppose that there exists an outer corner @xmath335 of @xmath241 such that @xmath336 holds for @xmath296 . it follows that @xmath338 and @xmath339 . setting @xmath340 , we have @xmath341 , and hence @xmath342 . lemma [ lambdabg ] implies that @xmath343 we deduce that if @xmath344 , then @xmath345 . note that we have @xmath275 . let @xmath346 be maximal such that @xmath347 . define @xmath348 applying lemma [ tident ] repeatedly , and using @xmath339 , we obtain @xmath349 lemma [ commutea ] implies that all the terms in the sum in ( [ last ] ) are zero . therefore @xmath350 if @xmath351 , then we obtain two chevalley terms ; the first summand @xmath352 is of type ii(a ) , and the second @xmath353 is of type ii(c ) , with @xmath354 . if @xmath344 , then since @xmath345 , we conclude from lemma [ commutec ] that the last term @xmath355 vanishes . the claim ( [ subs ] ) now follows easily , by combining the above two cases . observe in particular that each of the chevalley terms @xmath234 of type ii(a ) appears twice in the decomposition of the left hand side of ( [ subs ] ) : once from a term @xmath287 with @xmath356 , in case 1(b ) , and once from a term @xmath287 with @xmath357 , in case 2(b ) . finally , equations ( [ basic ] ) and ( [ subs ] ) imply ( [ cr ] ) , since @xmath358 . let @xmath359 and consider the @xmath360-strict partition @xmath361 . the element of the weyl group @xmath362 associated to @xmath51 is @xmath363 ( @xmath364 is defined in [ trans ] ) . we have @xmath365 , @xmath366 , @xmath367 , and @xmath368 . the chevalley formula in this case is @xmath369 in the above rule , the partition @xmath370 is of type i , @xmath371 and @xmath372 are of type ii(a ) , @xmath373 is of type ii(b ) , and @xmath374 and @xmath375 are of type ii(c ) . the terms @xmath287 in the sum ( [ subs ] ) are expanded as follows : @xmath376 the weyl group for the root system of type @xmath377 is the _ hyperoctahedral group _ @xmath378 of signed permutations on the set @xmath379 . we adopt the notation where a bar over an integer denotes a negative sign . the group @xmath378 is generated by the simple transpositions @xmath380 for @xmath381 and the sign change @xmath382 . the natural inclusion @xmath383 is defined by adding the fixed point @xmath384 , and we set @xmath385 . the length of an element @xmath386 in @xmath387 is denoted by @xmath388 . an element @xmath389 is called _ @xmath2-grassmannian _ if we have @xmath390 for all @xmath391 ; the set of all @xmath2-grassmannian elements is denoted by @xmath392 . the elements of @xmath392 are the minimal length coset representatives in @xmath393 , where @xmath394 is the subgroup of @xmath387 generated by the @xmath395 for @xmath391 . there is a canonical bijection between @xmath2-strict partitions and @xmath392 ( see @xcite ) . the element of @xmath392 corresponding to @xmath51 is denoted by @xmath364 , and lies in @xmath378 if and only if @xmath51 fits inside an @xmath396 rectangle ; the set of all such @xmath2-strict partitions is denoted by @xmath397 . suppose that @xmath398 , so that @xmath399 and @xmath400 . then we deduce from ( * ? ? ? * and 4.4 ) that the bijection is given by the equations @xmath401 moreover , we have @xmath402 and @xmath403 we are now ready to justify that the formula displayed in theorem [ cvthm ] coincides with the equivariant chevalley formula for the products @xmath120 in the stable ring @xmath79 . it follows from ( * ? ? ? 1.1 ) and ( * ? ? ? * lemma 6.8 ) that @xmath404 where @xmath405 $ ] denotes the localization of the stable equivariant schubert class @xmath406 at the torus - fixed point indexed by @xmath364 . the polynomial @xmath407 may be computed e.g. as the image of the type c double schubert polynomial representing @xmath406 under the homomorphism @xmath408 of @xcite ; this gives @xmath409 the fact that the term ( [ inter ] ) coincides with the @xmath410-linear coefficient in equation ( [ cr ] ) amounts to the identity @xmath411 the equations ( [ css ] ) imply that @xmath412 since @xmath386 is a signed _ permutation _ , the equality ( [ agree ] ) follows . a theorem of mihalcea ( * ? ? ? 8.2 ) proves that the equivariant chevalley rule characterizes the @xmath84$]-algebra @xmath77 , when it can be verified for a given @xmath84$]-basis of the latter ring . however , an analogous characterization is not available for the stable equivariant cohomology ring @xmath79 . to complete the proof that the polynomials @xmath14 represent equivariant schubert classes , theorem [ cvthm ] could be combined with a _ vanishing theorem _ that determines their behavior under the natural projection @xmath413 of [ geomap ] to the finite ring @xmath77 . the vanishing is easy to show using lemma [ ctlem ] in the lagrangian case when @xmath414 , and similarly for maximal orthogonal and type a grassmannians . this method gives new , intrinsic proofs that double schur @xmath0- , @xmath1- , and @xmath415-polynomials represent equivariant schubert classes , which apply to equivariant _ quantum _ cohomology as well , along the lines of @xcite . it would be desirable to extend these arguments to arbitrary isotropic grassmannians ; for the non - equivariant case , see @xcite . in this section , we will prove some general facts about the double theta polynomials @xmath58 and study how they specialize and relate to earlier formulas . [ [ section ] ] let @xmath416 and @xmath417 be two integer sequences . consider the following schur type determinant : @xmath418 which specializes when @xmath419 to the _ schur @xmath0-polynomial _ @xmath420 furthermore , define the schur type pfaffian : @xmath421 ( the matrix in the pfaffian has size @xmath422 , where @xmath44 is the least even integer such that @xmath423 . ) this specializes when @xmath419 to the _ schur @xmath1-polynomial _ @xmath424 where the entries in the pfaffian satisfy @xmath425 let @xmath426 be a @xmath2-strict partition of length @xmath427 , and define @xmath428 and @xmath429 recall that a _ multiset _ is a set in which members are allowed to appear more than once , with finite multiplicity . the next result generalizes formulas found in ( * ? ? ? * prop . 5.9 ) and ( * ? ? ? * prop.2 ) to the case of double theta polynomials . [ kstricts ] for any @xmath2-strict partition @xmath51 of length @xmath427 , we have @xmath430 and @xmath431 in @xmath125 $ ] , where @xmath432 is the multiset of integer vectors defined by @xmath433 the equalities are an immediate consequence of the formal raising operator identities @xmath434 and @xmath435 [ [ extremecases ] ] we next examine the extreme cases when the parts of the partition @xmath51 are all small or all large , respectively , with respect to @xmath2 . if @xmath436 for all @xmath33 , then we have @xmath437 if @xmath438 for all @xmath42 , the determinant in ( [ schurdet ] ) is a _ double schur @xmath0-polynomial_. suppose that @xmath439 is another list of commuting independent variables . the specialization of @xmath440 which maps each @xmath441 to @xmath442 is called a factorial schur function , and was studied in @xcite and ( * ? ? ? * i.3 , ex . [ aprop ] suppose that @xmath436 for all @xmath33 . then we have @xmath443 in @xmath125 $ ] , where the sum is over all partitions @xmath150 contained in @xmath51 . set @xmath444 . for any composition @xmath38 , we have @xmath445 in @xmath87 $ ] , unless @xmath446 for some partition @xmath447 , in which case @xmath448 . for any raising operator @xmath37 , we have @xmath449 where the sums are over all compositions @xmath38 . we therefore obtain @xmath450 suppose now that @xmath451 for all nonzero parts @xmath52 . then we have @xmath452 in 2000 , kazarian @xcite proved a multi - pfaffian formula for those equivariant schubert classes @xmath76^t$ ] on isotropic grassmannians where all the subbundles @xmath109 involved in the defining conditions ( [ xlam ] ) are isotropic . it is easily seen that the pfaffian ( [ schurpf ] ) maps to kazarian s formula under the geometrization map @xmath413 of [ geomap ] . when @xmath414 , the pfaffian in ( [ schurpf ] ) becomes the _ double schur @xmath1-polynomial _ @xmath453 define the schur @xmath1-functions @xmath454 by the generating series @xmath455 then the specialization of @xmath456 which maps @xmath441 to @xmath457 for every @xmath22 may be identified with the factorial schur @xmath1-functions introduced by ivanov @xcite . the connection between ivanov s theory and the equivariant cohomology of the lagrangian grassmanian @xmath13 was pointed out by ikeda @xcite . [ cprop ] suppose that @xmath451 for @xmath458 . then we have @xmath459 in @xmath92 $ ] , where the sum is over all strict partitions @xmath447 with @xmath460 for @xmath461 . arguing as in the proof of proposition [ aprop ] , we obtain @xmath462 where the sum is over all compositions @xmath34 . observe that the polynomial @xmath463 vanishes unless @xmath464 for each @xmath157 , which implies that @xmath465 for all @xmath157 . therefore , according to ( * ? ? ? * lemma 1.3 ) , the @xmath1-polynomial @xmath466 is alternating in the indices @xmath467 . it follows that @xmath468 where the sums over @xmath469 are as in the statement of the proposition . [ [ compare ] ] the general degeneracy locus formulas of ( * ? ? ? ( 26 ) and thm . 3 ) produce many different answers for each indexing weyl group element @xmath386 , depending on a choice of two compatible sequences @xmath470 and @xmath471 , which specify the symmetries of the resulting expression . however , once this latter choice is fixed , these formulas are _ uniquely determined _ ( cf . we recall here the formula given in ( * ? ? ? 12(b ) ) for the equivariant schubert classes on symplectic grassmannians . the symmetric group @xmath472 is the subgroup of @xmath378 generated by the transpositions @xmath395 for @xmath381 ; we let @xmath473 be the corresponding subgroup of @xmath387 . for every permutation @xmath474 , let @xmath475 denote the type a schubert polynomial of lascoux and schtzenberger @xcite indexed by @xmath476 ( our notation follows @xcite ) . we say that a factorization @xmath477 in @xmath387 is reduced if @xmath478 . in any such factorization , the right factor @xmath479 is also @xmath2-grassmannian for some @xmath2-strict partition @xmath150 . consider the polynomial @xmath480 where the sum is over all reduced factorizations @xmath481 with @xmath474 . one can show that the partitions @xmath150 which appear in ( [ fstrep ] ) are all contained in @xmath51 and satisfy @xmath482 . it is furthermore proved in @xcite that @xmath483 represents the stable equivariant schubert class @xmath94 indexed by @xmath51 . theorem [ mainthm ] implies that @xmath483 must agree with @xmath121 up to the relations ( [ relations ] ) among the variables @xmath441 in @xmath193 ( see corollary [ comp ] ) . we give a direct proof of this below , in the two extreme cases considered in [ extremecases ] . if @xmath438 for all @xmath40 , then @xmath364 is a grassmannian element of @xmath484 , and hence is _ fully commutative _ in the sense of @xcite . there is a one - to - one correspondence between reduced factorizations @xmath485 and partitions @xmath447 . for each such @xmath150 , we have @xmath486 , and the schubert polynomial @xmath487 is a flagged skew schur polynomial . the latter may be computed using ( * ? ? ? 2.2 ) , which proves that @xmath488 . formula ( [ afacteq ] ) shows that in this case , @xmath489 as polynomials in @xmath125 $ ] . if @xmath490 for @xmath458 , then @xmath364 will not be a fully commutative element of @xmath387 , in general . however there is a one - to - one correspondence between reduced factorizations @xmath485 with @xmath474 and @xmath2-strict partitions @xmath447 with @xmath491 . for each such @xmath150 , we have @xmath492 , and can use e.g. the reduced words for @xmath364 and @xmath493 given in ( * ? ? ? * ex . 9 ) to deduce that @xmath476 is fully commutative . according to ( * ? ? ? 2.2 ) again , the schubert polynomial @xmath487 is equal to @xmath494 . we thus see that @xmath483 agrees with the right hand side of formula ( [ cfacteq ] ) , and hence latexmath:[$\omega_{\lambda}(c\ , let @xmath414 , so that the partition @xmath51 is strict , and @xmath6 and @xmath14 specialize to the single and double @xmath1-polynomials @xmath496 and @xmath497 , respectively . following ( * ? ? ? 12(b ) ) , we have @xmath498 summed over all strict partitions @xmath469 with @xmath499 . taking @xmath500 in ( [ qeq ] ) gives @xmath501 on the other hand , equation ( [ dblq ] ) above gives @xmath502 observe that @xmath503 , and thus the two polynomials are not equal in @xmath125 $ ] . however , since the relation @xmath504 holds in @xmath505 , we have agreement in @xmath506 $ ] , as expected . there is an action of @xmath387 on @xmath125 $ ] by ring automorphisms , defined as follows . the simple reflections @xmath395 for @xmath507 act by interchanging @xmath83 and @xmath508 and leaving all the remaining variables fixed . the reflection @xmath509 maps @xmath510 to @xmath511 , fixes the @xmath512 for @xmath513 , and satisfies , for each @xmath514 , @xmath515 the latter equation can be written in the form @xmath516 where @xmath476 is a formal variable with @xmath517 for each @xmath40 . it follows that @xmath518 and therefore that the braid relations for @xmath387 are satisfied on @xmath125 $ ] . for every @xmath519 , we define the _ divided difference operator _ @xmath520 on @xmath125 $ ] by @xmath521 the operators @xmath520 correspond to the _ left _ divided differences @xmath522 studied in @xcite and @xcite . however , the definition given here is more general than that found in op . , since the @xmath520 act on the polynomial ring @xmath125 $ ] rather than on its quotient @xmath92 $ ] . note that for each @xmath519 , the @xmath520 satisfy the leibnitz rule @xmath523 we next recall from @xcite some important identities satisfied by these operators , with statements and proofs in our setting . [ lem1 ] we have @xmath524 since @xmath126 is symmetric in @xmath525 when @xmath526 , the identity @xmath527 for @xmath528 is clear if @xmath507 . for @xmath129 , we have @xmath529 and we apply @xmath509 to both sides and use ( [ concise ] ) to prove that @xmath530 for all @xmath43 ; the argument when @xmath30 is similar . the rest of the proof is straightforward . [ lem2 ] suppose that @xmath128 . _ a ) _ for all @xmath519 , we have @xmath531 _ b ) _ for all @xmath42 , we have @xmath532 for part ( a ) , observe that if @xmath533 , then @xmath534 by lemma [ lem1 ] . if @xmath535 , then we compute easily using ( [ ee ] ) that @xmath536 from which the desired result follows . work similarly when @xmath537 or @xmath538 . for part ( b ) , we use the leibnitz rule and lemmas [ ctlem ] and [ lem1 ] to compute @xmath539 the following result was proved differently in @xcite , working in a ring @xmath540 which is isomorphic to @xmath92 $ ] , and not in the polynomial ring @xmath125 $ ] . [ uniq ] let @xmath51 and @xmath150 be @xmath2-strict partitions such that @xmath541 and @xmath542 for some simple reflection @xmath543 . then we have @xmath544 in @xmath125 $ ] . let @xmath545 , @xmath546 , and @xmath547 . following ( * ? ? ? * lemmas 3.4 , 3.5 ) , there are 4 possible cases for @xmath386 , discussed below . in each case , we have @xmath469 , so that @xmath548 for some @xmath88 and @xmath549 for all @xmath550 , and the properties listed are an easy consequence of equations ( [ equtn ] ) , ( [ cweq ] ) , and ( [ css ] ) . \(a ) @xmath551 . in this case we have @xmath552 , @xmath553 , @xmath554 , @xmath555 , while @xmath556 for all @xmath557 . \(b ) @xmath558 . in this case @xmath553 , @xmath554 , @xmath559 , and @xmath556 for all @xmath557 . \(c ) @xmath560 . in this case @xmath553 , @xmath561 and @xmath562 , and @xmath556 for all @xmath557 . \(d ) @xmath563 . in this case @xmath564 , where @xmath565 and @xmath566 . it follows that @xmath567 and @xmath568 . we see similarly that @xmath562 and @xmath569 , while @xmath556 for all @xmath570 . in cases ( a ) , ( b ) , or ( c ) , it follows using the leibnitz rule and proposition [ lem2](a ) that for any integer sequence @xmath189 , we have @xmath571 since @xmath572 , we deduce that if @xmath37 is any raising operator , then @xmath573 as @xmath574 , we conclude that @xmath575 in case ( d ) , it follows from the leibnitz rule as in the proof of proposition [ lem2](b ) that for any integer sequence @xmath189 , we have @xmath576 since @xmath572 , we deduce that if @xmath37 is any raising operator , then @xmath577 as @xmath578 , we conclude that @xmath579 let @xmath580 be the tautological sequence ( [ ses ] ) of vector bundles over @xmath78 , and define the subbundles @xmath109 of @xmath106 for @xmath581 as in the introduction . let @xmath582 denote the borel mixing space for the action of the torus @xmath68 on @xmath8 . the @xmath68-equivariant vector bundles @xmath583 over @xmath8 induce vector bundles over @xmath584 . their chern classes in @xmath585 are called _ equivariant chern classes _ and denoted by @xmath110 , @xmath111 , etc . recall that @xmath113 for @xmath514 is defined by the total chern class equation @xmath586 let @xmath587 for @xmath588 . following @xcite and ( * ? ? ? 3 ) , we define the _ geometrization map _ @xmath413 as the @xmath84$]-algebra homomorphism @xmath589 \to \operatorname{\mathrm{h}}^*_{t_n}(\operatorname{ig}(n - k,2n))\ ] ] determined by setting @xmath590 since @xmath591 are the ( equivariant ) chern roots of @xmath592 for @xmath593 , it follows that @xmath594 for @xmath595 . this can be extended to @xmath596 if we set @xmath597 for @xmath598 and @xmath599 for @xmath600 . for example , if @xmath51 is any partition in @xmath397 , then we have @xmath601 in agreement with formula ( [ genkaz ] ) . the embedding of @xmath378 into @xmath602 defined in [ trans ] induces maps of equivariant cohomology rings @xmath603 which are compatible with the morphisms @xmath413 . we thus obtain an induced @xmath84$]-algebra homomorphism @xmath604\to { { \mathbb h}}_t(\operatorname{ig}_k),\ ] ] which we will show has the properties listed in theorem [ mainthm ] . fix a rank @xmath605 and let @xmath606 be the @xmath2-strict partition associated to the @xmath2-grassmannian element of maximal length in @xmath378 . in this short section we discuss three different proofs that @xmath607^{t_n}.\ ] ] equation ( [ schurpf ] ) gives @xmath608 and by applying ( [ geomeq ] ) we obtain that @xmath609 observe now that the right hand side of ( [ kresult ] ) coincides with kazarian s multi - pfaffian formula ( * ? ? ? 1.1 ) for the cohomology class of the degeneracy locus which correponds to @xmath610^{t_n}$ ] ( compare with corollary [ maincor ] ) . therefore ( [ pteq ] ) follows directly from a known result , which is proved geometrically in op . cit . a second proof of ( [ pteq ] ) is obtained by using the equality @xmath611^{t_n}$ ] . note that the proof of this in op . uses the fact that the single @xmath1-polynomials @xmath612 in ( [ 2eq ] ) represent the corresponding cohomological schubert classes on @xmath8 . this last result is a special case of the main theorem of @xcite . finally , a third proof of ( [ pteq ] ) is provided by ikeda and matsumura in @xcite , starting from the pfaffian formula ( * ? ? ? 1.2 ) for the equivariant schubert class of a point on the complete symplectic flag variety @xmath613 . from this , using the left divided differences , they derive a pfaffian formula for the top equivariant schubert class on any symplectic partial flag variety . in particular , one recovers the pfaffian @xmath614 which represents the class @xmath610^{t_n}$ ] . we have shown in proposition [ basisthm ] that the @xmath14 for @xmath51 @xmath2-strict form a @xmath84$]-basis of @xmath92 $ ] . following @xcite , for any @xmath2-strict partition @xmath615 , write @xmath616 as a product of simple reflections @xmath617 in @xmath378 , with @xmath618 . since @xmath619 , we deduce from proposition [ uniq ] that @xmath620 holds in @xmath125 $ ] . the action of the operators @xmath520 on @xmath125 $ ] induces an action on the quotient ring @xmath92 $ ] . moreover , the corresponding left divided differences @xmath522 on @xmath621 from @xcite are compatible with the geometrization map @xmath622\to h_n$ ] . according to ( * ? ? ? 2.3 ) ( see also ( * ? ? ? ( 62 ) ) ) , we have @xmath623^{t_n})= [ x_{\mu}]^{t_n}$ ] whenever @xmath541 and @xmath542 for some simple reflection @xmath395 . it follows from this and equations ( [ pteq ] ) and ( [ iteratepar ] ) that @xmath624^{t_n}.\ ] ] the fact that @xmath625 whenever @xmath626 , or equivalently @xmath627 , is now a consequence of the vanishing property for equivariant schubert classes ( see e.g. ( * ? ? ? * prop . the induced map @xmath628\to { { \mathbb h}}_t(\operatorname{ig}_k)$ ] satisfies @xmath629 for all @xmath2-strict @xmath51 , and is a @xmath84$]-algebra isomorphism because the @xmath14 and @xmath94 for @xmath51 @xmath2-strict form @xmath84$]-bases of the respective algebras . we are left with proving the last assertion in theorem [ mainthm ] , about the presentation of the @xmath84$]-algebra @xmath630 . let @xmath631 be the ideal of @xmath92 $ ] generated by the @xmath14 for @xmath626 , and @xmath632 be the ideal of @xmath92 $ ] generated by the relations ( [ rel1 ] ) and ( [ rel2 ] ) . since @xmath633 , there is a surjection of @xmath84$]-algebras @xmath634/j_n \to c^{(k)}[t]/i_n\,.\ ] ] we have established that @xmath635/i_n$ ] , so it suffices to show that @xmath636 . consider the ideal @xmath637 of @xmath84 $ ] generated by the @xmath83 for all @xmath42 , and let @xmath638/j_n$ ] . when @xmath639 , we have the formal identity @xmath640 which is a specialization of ( [ schurdet ] ) . we deduce from ( [ ident ] ) and the presentation of @xmath641 given in ( * ? ? ? 1.2 ) that the polynomials @xmath6 for @xmath642 generate @xmath643 as a @xmath117-module . it follows from ( * ? ? ? * lemma 4.1 ) that the @xmath14 for @xmath642 generate @xmath644 as a @xmath84$]-module . now suppose that @xmath645 maps a general element @xmath646 of @xmath644 to zero , where @xmath647 $ ] and the sum is over @xmath615 . then @xmath648 . since @xmath631 is equal to the @xmath84$]-submodule of @xmath92 $ ] with basis @xmath14 for @xmath649 , we conclude that @xmath650 , for all @xmath51 . this completes the proof of theorem [ mainthm ] , and corollary [ maincor ] follows from equations ( [ kaz ] ) and ( [ pineq ] ) . according to @xcite , the polynomials @xmath651 of [ compare ] represent the stable equivariant schubert classes @xmath94 in @xmath79 under the geometrization map @xmath96 . the next result is therefore an immediate consequence of theorem [ mainthm ] . we remark that the right hand side of ( [ to ] ) is the unique expansion of @xmath14 as a @xmath117-linear combination in the product basis @xmath653 of @xmath92 $ ] , where @xmath150 ranges over all @xmath2-strict partitions and @xmath476 lies in @xmath484 . it is instructive to write equation ( [ to ] ) in the following way : @xmath654 v. n. ivanov : _ interpolation analogues of schur @xmath1-functions _ , zap . nauchn . sem . otdel . mat . inst . steklov . ( pomi ) * 307 * ( 2004 ) , teor . i algoritm . 10 , 99119 , 281282 ; translation in j. math . ( n. y. ) * 131 * ( 2005 ) , 54955507 . p. pragacz : _ algebro - geometric applications of schur @xmath0- and @xmath1-polynomials _ , sminare dalgbre dubreil - malliavin 1989 - 1990 , lecture notes in math . * 1478 * ( 1991 ) , 130191 , springer - verlag , berlin ,

we present next - to - leading order ( nlo ) perturbative - qcd calculations of the cross sections for and . the main feature of these processes is that the scattered lepton is not observed , so that the hard scale that makes them perturbative is set by the transverse momentum of the hadron or jet . kinematically , the two processes thus become direct analogs of single - inclusive production in hadronic collisions which , as has been pointed out in the literature , makes them promising tools for exploring transverse spin phenomena in qcd when the incident nucleon is transversely polarized . we find that the nlo corrections are sizable for the spin - averaged cross section . we also investigate in how far the scattering is dominated by the exchange of almost real ( weizscker - williams ) photons . we present numerical estimates of the cross sections for present - day fixed target experiments and for a possible future electron ion collider .

1505.06415

there has been growing interest recently , both experimentally @xcite and theoretically @xcite , in the processes @xmath0 and @xmath1 , the single inclusive production of a hadron or jet at large transverse momentum in lepton - nucleon scattering . in contrast to the far more customary process @xmath2 @xcite , for @xmath0 the scattered lepton in the final state is not observed , so that the process is truly one - hadron ( or one - jet ) inclusive . the reason for the interest in @xmath0 comes from the study of single transverse - spin phenomena in hadronic scattering processes . it is well known that large single - spin asymmetries have been observed @xcite for the process @xmath3 , where @xmath4 denotes a transversely polarized proton . to explain the large size of the asymmetries , and their persistence all the way from fixed - target to collider energies , has posed a major challenge to theory . although a lot has been learned , it is fair to say that a fully satisfactory understanding has yet to be obtained . measurements of corresponding asymmetries in the kinematically equivalent , but much simpler , processes @xmath5 , @xmath6 have the promise to shed new light on the mechanisms for single - spin asymmetries in qcd . first fairly precise experimental data for @xmath5 have recently been released by the hermes @xcite and jefferson lab hall a @xcite collaborations . we note that at first sight one might consider the related process @xmath7 ( which is just the standard inclusive deep - inelastic ( dis ) process ) to be equally suited for transverse - spin studies in lepton scattering . however , the analysis of the corresponding single - spin asymmetry is considerably more complex because higher order qed effects are required for the asymmetry to be non - vanishing @xcite . in the same spirit as @xmath5 , also the processes @xmath8 @xcite with longitudinal polarization of the lepton and @xmath9 @xcite with a transversely polarized @xmath10 hyperon have been considered in the literature recently . the proven method for analyzing single - inclusive processes such as @xmath11 or @xmath0 at large transverse momentum rests on qcd perturbation theory and collinear factorization . for single - transverse - spin observables , this involves a twist-3 formalism in terms of three - parton correlation functions of the nucleon or the fragmentation process @xcite . interestingly , the recent study @xcite suggests that the twist-3 fragmentation effects could be the dominant source of the observed large transverse - spin asymmetries in @xmath3 . an alternative approach for describing the single - spin asymmetry in inclusive hadron production in @xmath3 was devised in the context of a `` generalized '' parton model in which the dependence of parton distributions and fragmentation functions on transverse momentum is kept @xcite . although no such factorization in transverse momentum is known to be valid for a single - inclusive cross section , the approach has enjoyed considerable phenomenological success . both the collinear twist-3 approach and the generalized parton model have been used to obtain predictions for the spin asymmetry in @xmath5 . in ref . @xcite a leading order ( lo ) twist-3 analysis has been presented in terms of parton correlation functions that were previously extracted from data for @xmath3 . the results obtained in this way fail to describe the hermes data @xcite for the spin asymmetries in @xmath5 . a comparison of perturbative calculations to the corresponding jlab data @xcite is not possible as the data are for hadrons with transverse momenta below 1 gev . the lo generalized parton model approach , on the other hand , appears to give results quite consistent with the hermes data @xcite . in our view it is premature to draw any conclusions from these findings at lo . given the kinematics ( and the precision ) of the present data , one may expect higher - order qcd corrections to the cross sections and the asymmetry to be important @xcite for a meaningful comparison of data and theory . at least next - to - leading order ( nlo ) corrections should be included . we stress that the twist-3 formalism , although so far only developed to lo , offers a well - defined framework for a perturbative study of the transverse - spin asymmetry in @xmath5 . this is in contrast to the generalized parton model , for which there is likely no systematic way of going to higher orders in perturbation theory . that said , nlo calculations within the twist-3 formalism are technically very challenging , and only a few nlo calculations have been performed for the simpler drell - yan @xcite and semi - inclusive deep - inelastic scattering ( dis ) cases @xcite . in the present paper , we take a first step toward an nlo calculation of the transverse - spin asymmetry for @xmath5 by computing the nlo corrections to the spin - averaged cross section for the process , which constitutes the denominator of the spin asymmetry . we present analytical results for the nlo partonic cross sections . to our knowledge , despite the vast amount of work performed for lepton proton scattering in the literature ( see , for example @xcite ) , this calculation has not been presented so far . we also present similar nlo calculations for the process @xmath1 . we note that the process @xmath1 has also been extensively studied in terms of the concept of `` 1-jettiness '' @xcite . here one additionally writes the cross section differential in a variable @xmath12 that characterizes the hadronic final state that is _ not _ associated with the produced jet or the nucleon beam remnant . in ref . @xcite the full nlo corrections for the 1-jettiness were computed , where a fully numerical approach was adopted . in principle , it should be possible to recover our nlo results by performing a ( numerical ) integration over @xmath12 of the results of @xcite . because of the propagator of the exchanged photon , the cross section for @xmath5 will contain contributions for which the photon is almost on - shell . this is not yet the case at lo where the high transverse momentum of the produced hadron requires the photon to be highly virtual . starting from nlo , however , it may happen that the incoming lepton radiates the photon almost collinearly . this may then be followed by a @xmath13 scattering process of the photon with a parton in the nucleon , which is perfectly capable of producing the hadron at high @xmath14 . in processes where the scattered lepton is observed , such as @xmath2 , one can in fact select such contributions by requiring the scattered lepton to have a low scattering angle . the incoming lepton then effectively acts merely as a source of quasi - real photons , and the process may be very accurately described in terms of a ( perturbative ) distribution function for photons in leptons known as the weizscker - williams ( ww ) distribution @xcite . this approach has been widely used with much success in the hera physics program @xcite . in the context of our nlo calculation for @xmath0 it is therefore interesting to investigate whether also in this case the contributions by almost real photons dominate and the nlo corrections may be well approximated by a weizscker - williams type distribution . since it is much easier to compute the latter contribution than the full nlo correction , this would mean that one could also obtain approximate nlo results for the transversely polarized cross section within the twist-3 framework by simply considering real photons . given the complexity of a full nlo calculation for the twist-3 case , this would be a tremendous advantage . we note that the contributions to the spin - dependent cross sections for @xmath15 for real photons were discussed in @xcite , including the twist-3 contributions for the single - transverse spin case . actual lo calculations for the twist-2 longitudinal spin - dependent cross section were presented in ref . @xcite for quasi - real photons . we will closely examine the contributions by quasi - real photons also in our paper . their relevance will of course also depend on the lepton species that is used , because the lepton mass leads to a lower limit on the virtuality of the photon . our paper is structured as follows . in sec . [ nlocalc ] we present our nlo calculations for the partonic cross sections for @xmath5 and @xmath1 . we also discuss in some detail the weizscker - williams contribution and how the calculation can be done keeping a finite lepton mass . section [ pheno ] presents numerical predictions for the nlo cross section to be expected at various fixed - target experiments and at a future electron ion collider ( eic ) . finally , we summarize our results in sec . [ concl ] . in this section we present our derivation of the analytical nlo results for the processes @xmath0 and @xmath1 . the transverse momentum of the produced hadron or jet sets a hard scale , so that perturbative methods may be used for treating the cross sections . we first consider @xmath16 , where we have introduced our notation for the four - momenta . it is useful to introduce the mandelstam variables as @xmath17 , @xmath18 and @xmath19 . furthermore , we label the energy of the detected hadron as @xmath20 and its three - momentum by @xmath21 . in collinear leading - twist perturbative qcd the hadronic cross section is approximated by convolutions of hard partonic scattering cross sections and parton distribution / fragmentation functions . the momenta of the incoming parton , @xmath22 , and of the fragmenting parton , @xmath23 , which appear in the calculation of the partonic cross sections , are approximated as @xmath24 and @xmath25 , respectively . it is then convenient to work with the partonic mandelstam variables [ stu ] s=(k+l)^2=xs , t=(k - p)^2=t , u=(l - p)^2=. the general form of the factorized cross section for the inclusive hadron production process then is e_h & = & _ i , f _ 0 ^ 1 _ 0 ^ 1 f^i / n(x , ) + & & d^h / f(z,)^if(s , t , u , ) , [ invariantcs1 ] where @xmath26 is the parton distribution function ( pdf ) for the incoming parton @xmath27 in the nucleon @xmath28 and @xmath29 the corresponding fragmentation function ( ff ) for parton @xmath30 fragmenting into hadron @xmath31 , both evaluated at a factorization scale @xmath32 . we choose the factorization scales to be the same for the initial and the final state , and also equal to the renormalization scale . in eq . ( [ invariantcs1 ] ) , @xmath33 is the partonic cross section for the lepton - parton scattering process , @xmath34 , with @xmath35 an unobserved partonic final state . the sum in eq . ( [ invariantcs1 ] ) runs over the different species of partons , quarks , gluons and antiquarks . we note that the expression in eq . ( [ invariantcs1 ] ) holds up to corrections that are suppressed by inverse powers of the produced hadron s transverse momentum @xmath14 . the partonic cross sections @xmath33 in eq . ( [ invariantcs1 ] ) can be calculated in qcd perturbation theory . one may write their expansion in the strong coupling as ^if=^if _ + ^if_+ o(_s^2 ) . at lowest order ( lo ) only the tree - level process @xmath36 shown in fig . [ fig : lo ] contributes . the calculation of its cross section is straightforward . one finds _ ^qq = 2 _ ^2 e_q^2 ( s+t+u ) , [ cslo ] where @xmath37 is the fine structure constant and @xmath38 is the quark s fractional charge . at nlo , @xmath39 ) , both virtual ( fig . [ fig : nlovir ] ) and real - emission diagrams ( figs . [ fig : nlorealq2q][fig : nlorealg2q ] ) contribute . we will address these in turn in the following subsections . one can see from figs . [ fig : nlorealq2 g ] and [ fig : nlorealg2q ] that beyond lo there are also new contributions where a gluon fragments or where an initial gluon enters the hard scattering process . as is well known , all types of nlo contributions develop singularities at intermediate stages of the calculations , which we make manifest by using dimensional regularization with @xmath40 space - time dimensions . the subsequent treatment of the singularities is standard in pqcd calculations . the only non - standard feature arises for the incoming lepton . if we assume for the moment that we have an incoming quark instead of a lepton in the diagrams in figs . [ fig : nlorealq2q ] and an exchanged gluon instead of a photon , then the diagram would make an nlo contribution to , say , @xmath41 . being treated as massless , the initial quark would produce a singularity when it radiates the gluon collinearly . as is well understood , this singularity may be absorbed ( `` factorized '' ) into the proton s quark pdf , exactly in the same way as for the incoming quark at the bottom of the diagram . in case of an incoming lepton , on the other hand , the lepton s mass ensures that no collinear singularity arises when the lepton radiates a collinear photon that subsequently participates in the hard scattering . in fact , keeping the lepton mass @xmath42 , the cross section will develop a logarithmic term of the form @xmath43 , where @xmath10 represents a hard scale of the problem , and in the limit @xmath44 this logarithm precisely produces the required collinear singularity . in principle we should therefore perform the nlo calculation keeping the lepton mass finite . this is technically very cumbersome , and in fact not needed . we can adopt two different , and equivalent , approaches instead : in the first approach we neglect the lepton s mass and regularize the ensuing collinear pole in dimensional regularization . the pole is then subtracted ( for example , in the @xmath45 scheme ) and absorbed into a `` parton '' distribution function for photons in a lepton . this distribution may be evaluated perturbatively in first - order qed , giving rise essentially to the well - known `` weizscker - williams '' distribution . this approach may in principle be extended to higher order in qed . in the second approach , we calculate the cross section for a massive lepton , keeping however only the leading terms in @xmath42 which are of the form @xmath46 $ ] . this is justified by the fact that all terms beyond this approximation are suppressed as powers of @xmath42 over the hard scale and hence numerically tiny . we note that although the logarithm can become large ( as @xmath42 is small compared to typical qcd hard scales ) , the smallness of @xmath37 will usually make the term @xmath47 small enough to be regarded as a perturbative correction . we will present our main calculation for the case of massless leptons and comment on the use of a finite lepton mass in the calculation later . it is convenient to rewrite the @xmath35- and @xmath48-integrals in eq . ( [ invariantcs1 ] ) in terms of new variables @xmath49 and @xmath50 . using ( [ stu ] ) , we have x=,z= , and eq . ( [ invariantcs1 ] ) becomes e_h & = & ( ) _ i , f _ ^1 + _ ^1 + & & h^if(v , w)^if(v , w , ) , [ trafox ] where we have defined @xmath51 for ease of notation , we have kept the symbol @xmath33 also for the cross section when expressed in terms of the new variables . we note that the invariant mass of the unobserved recoiling partonic final state is given by @xmath52 . the function @xmath53 in the lo cross section ( [ cslo ] ) expresses the fact that at lo the recoil consists of a single parton . at the nlo level , the virtual contributions shown in fig . [ fig : nlovir ] contribute through their interference with the born diagram . the virtual contributions thus have born kinematics and are proportional to @xmath54 . since we are only interested in qcd virtual corrections , only the quark line is affected , and we may adopt the result directly from the corresponding calculation in ref . @xcite for the basic photon - quark scattering diagrams in dis . this gives ^qq _ & = & + & & ( ) ^(---8 ) ^qq _ , , [ virto ] where @xmath55 is the born cross section computed in @xmath56 dimensions . furthermore , @xmath57 , with @xmath58 the number of colors . the real diagrams have @xmath59 topology . to obtain the desired contribution to an inclusive - parton cross section we need to integrate over the phase space of the lepton and the `` unobserved '' parton in the final state . this can be done in @xmath56 dimensions using the standard techniques available in the literature @xcite . after phase space integration , the result for the real - emission contribution for the @xmath60 channel takes the form @xmath61 where both functions @xmath62 and @xmath63 carry a @xmath64-pole , but are well - behaved in the limit @xmath65 . obviously , the second term in ( [ realq2q1 ] ) requires special care in this limit since the denominator would lead to a non - integrable behavior for @xmath66 . we deal with this limit by means of the expansion ( 1-w)^-1 - 2&= & -(1-w)+-2()_+ + & & + ( ^2 ) , [ 1mwidentity ] where the plus distribution is defined in the usual way by _ 0 ^ 1dw f(w)_+ = _ 0 ^ 1dw g(w ) .[plus ] this expansion makes the singularities in @xmath64 explicit . when combined with the pole terms in @xmath63 , the term @xmath67 in ( [ 1mwidentity ] ) leads to a double pole term that cancels against the double pole in the virtual correction in eq . ( [ virto ] ) . this well - known behavior reflects the cancelation of infrared singularities in partonic observables . the channels @xmath68 and @xmath69 in figs . [ fig : nlorealq2 g ] and [ fig : nlorealg2q ] are infrared finite at nlo . after the cancelation of infrared singularities between real and virtual contribution , the partonic cross sections still exhibit single poles that reflect collinear singularities arising when an `` observed '' parton ( either the incoming one , or the one that fragments ) becomes collinear with the unobserved parton . the factorization theorem states that these poles may be absorbed into the parton distribution functions or into the fragmentation functions . this procedure may be formulated in terms of _ renormalized _ parton densities and fragmentation functions ( see , e.g. , ref . in fact , naive definitions of `` bare '' parton densities and fragmentation functions contain ultraviolet singularities that can be dealt with as well by using dimensional regularization . at nlo , the corresponding ultraviolet @xmath64-poles that appear can be removed in the @xmath45 scheme by introducing `` renormalized '' functions in the form @xmath70 & & + \frac{\alpha_s(\mu)}{2\pi}\frac{s_\varepsilon}{\varepsilon}\left(p_{qg}\otimes f^{g / n}_{\mathrm{ren}}\right)(x,\mu)+\mathcal{o}(\alpha_s^2)\ ; , \label{renpdf}\\[2 mm ] d^{h / q}_{\mathrm{bare}}(z,\mu ) & = & d^{h / q}_{\mathrm{ren}}(z,\mu)+ \frac{\alpha_s(\mu)}{2\pi}\frac{s_\varepsilon}{\varepsilon}\left(p_{qq}\otimes d^{h / q}_{\mathrm{ren}}\right)(z,\mu ) \nonumber\\[2 mm ] & & + \frac{\alpha_s(\mu)}{2\pi}\frac{s_\varepsilon}{\varepsilon } \left(p_{gq}\otimes d^{g / n}_{\mathrm{ren}}\right)(z,\mu)+\mathcal{o}(\alpha_s^2)\ ; , \label{renff}\end{aligned}\ ] ] where we have the usual splitting functions @xmath71\;,\label{pqq}\\[2 mm ] p_{qg}(y ) & = & t_r\left[y^2+(1-y)^2\right]\,,\label{pqg}\\[2 mm ] p_{gq}(y ) & = & c_f\ , \frac{1+(1-y)^2}{y}\,,\label{pgq}\end{aligned}\ ] ] ( with @xmath72 ) , and where the `` @xmath73''-symbol indicates the convolution @xmath74 the constant @xmath75 in ( [ renpdf ] ) and ( [ renff ] ) corresponds to the usual @xmath45 scheme . inserting the bare distributions into the lo expression for the hadronic cross section , we obtain additional @xmath76 contributions . these precisely cancel the collinear poles associated with the observed partons in the nlo partonic cross sections , for all three channels . even after this procedure , one type of collinear singularity remains . it is generated by a momentum configuration where the exchanged photon is collinear to the incoming lepton . as discussed at the beginning of this section , the presence of this singularity is an artifact of neglecting the lepton s mass . in the following two subsections we discuss our treatment of this issue . one approach for dealing with the collinear lepton singularity is to introduce bare and renormalized qed parton distributions for the lepton , much in analogy with the procedure that we discussed in the previous section for the nucleon s parton distributions . the only differences are that for leptons the partons are the lepton itself and the photon , and that we can safely compute their distributions in qed perturbation theory . to lowest order in qed , we have just @xmath77 , corresponding to the born contribution in fig . [ fig : lo ] . the hard process involving an incoming lepton will always require two electromagnetic interactions and hence be of order @xmath78 , as seen explicitly in eq . ( [ cslo ] ) . this is different for a hard process with an incoming photon such as @xmath79 , which is of order @xmath80 . this implies that at nlo in qcd ( at order @xmath81 ) there will be contributions generated by the photon acting as a parton of the lepton and participating in the hard process . a generic picture for such types of contributions , known as weizscker - williams contributions , is shown in fig . [ fig : ww ] . in essence , the lepton merely serves as a source of real photons for the contributions shown in the figure . like its nucleon counterpart , the corresponding photon - in - lepton distribution @xmath82 will require renormalization . following ( [ renpdf ] ) we may write f^/_(y , ) = f^/_(y , ) + ( p _ f^/_)(y , ) + [ fge ] where @xmath83 and the ellipses denote a term involving a photon - to - photon splitting that makes contributions beyond the order in @xmath37 we consider here . within the same reasoning , we can set @xmath84 in ( [ fge ] ) . the bare photon - in - lepton distribution @xmath85 in eq . ( [ fge ] ) can be defined analogously to the gluon distribution in a nucleon in terms of the matrix element ( see also @xcite ) @xmath86 \ , f^{\rho \mu}_{\mathrm{em}}(\lambda n ) |\ell \rangle , \nonumber\\[2 mm ] & = & \frac{-g_\perp^{\mu \nu}}{2(1-\varepsilon)}\;f^{\gamma/\ell}_{\mathrm{bare}}(y,\mu ) \;.\label{wwdistr}\end{aligned}\ ] ] in this definition @xmath87 is a light - cone vector conjugate to the lepton momentum @xmath88 , with @xmath89 and @xmath90 . furthermore , @xmath91 is the electromagnetic field - strength tensor , and we have inserted a ( straight ) wilson line @xmath92 $ ] that ensures the electromagnetic gauge invariance of the matrix element . the transverse projector in ( [ wwdistr ] ) is given as @xmath93 . since the matrix element in ( [ wwdistr ] ) contains electromagnetic fields and elementary leptons in the in- and out - states we can compute it to lo in qed . in this calculation we keep a non - vanishing lepton mass @xmath42 in order to obtain an infrared - finite result . to order @xmath94 we find , @xmath95+\mathcal{o}(\alpha_{\mathrm{em}}^2)\ , , \label{fwwbare}\ ] ] where , as before , @xmath75 . in close analogy to parton distributions of the nucleon we can perform an @xmath45-renormalization of the distribution and obtain , @xmath96 & = & \frac{\alpha_{\mathrm{em}}}{2\pi}\ ; p_{\gamma\ell}(y)\ ; \left[\ln \left(\frac{\mu^2}{y^2 m_\ell^2}\right ) -1\right]+ \mathcal{o}(\alpha_{\mathrm{em}}^2)\ ; .\label{fwwren}\end{aligned}\ ] ] this renormalized distribution is closely related to the ` classic ' weizscker - williams distribution @xcite . the logarithm in ( [ fwwren ] ) may be derived from an integration over the photon s virtuality @xmath97 ( where @xmath98 is the photon momentum ) . for the standard weizscker - williams distribution one performs this integration from the lower kinematic limit @xmath99 to an upper limit @xmath100 fixed by the experimental condition imposed on the scattered lepton . this gives rise to a term @xmath101 in the photon spectrum , which can be recovered by an appropriate choice of the scale @xmath32 in ( [ fwwren ] ) . for the contribution related to @xmath102 the photon virtuality is then neglected everywhere else in the hard scattering . one thus considers scattering diagrams with a real incoming photon . we thus write the generic factorized cross section for the contribution as @xmath103 & \times & f^{i / n}(x,\mu)\ ; d^{h / f}(z,\mu)\ ; f^{\gamma/\ell}_{\mathrm{ren}}(y,\mu)\ ; \hat{\sigma}^{\gamma i\to f},\label{csww}\end{aligned}\ ] ] with the cross sections @xmath104 describing the scattering @xmath105 of the photon off parton @xmath27 in the nucleon ( to be given below ) . at @xmath106 we encounter three channels with an incoming photon : @xmath107 , @xmath108 , and @xmath109 ( the partons in parentheses are not observed ) . the relevant diagrams are as those shown in figs . [ fig : nlorealq2q ] [ fig : nlorealg2q ] , but with the lepton lines removed and the virtual photon replaced by a real photon . being @xmath13 diagrams , their calculation is straightforward . inserting now the bare ww distribution we generate precisely the pole terms required to cancel the lepton collinear divergences discussed at the end of sec . this happens in the same way for all partonic channels . we note that the dependence on the scale @xmath32 associated with the lepton also disappears . this has to be the case , since for a finite lepton mass there would never be any lepton collinear divergences in the first place . as we noted earlier , the presence of collinear singularities associated with lepton - photon splitting is really an artifact of neglecting the lepton s mass . in principle we should therefore perform a full calculation in which the lepton s mass is kept finite . this is trivial for the virtual diagrams , since the qcd corrections do not affect the lepton line . however , inclusion of a lepton mass considerably complicates the phase space integrations for the real diagram . nevertheless , it is possible to compute the relevant integrals using the results given in ref . one may then expand the result in powers of the lepton mass and neglect terms suppressed by powers of @xmath111 . in this way , the `` would - be '' collinear singularity is regularized by the lepton mass and shows up as a term @xmath112 . terms independent of @xmath42 are also kept . all other parts of the calculation proceed as before , and the partonic cross section thus has the structure @xmath113 & & \hat{\sigma}^{i\to f}_{0}(v , w,\mu)+\mathcal{o}(m_\ell^2\ln(m_\ell^2 ) ) .\label{massive}\end{aligned}\ ] ] for each channel . we have checked explicitly for all three channels that our two approaches for treating the initial lepton are equivalent : the full result obtained using the ww contribution in the previous subsection agrees with that for @xmath110 , as long as we only keep the leading terms as discussed in eq . ( [ massive ] ) . the equivalence of the two approaches serves as an important check of our calculation and also explicitly demonstrates the universality of the ww - distribution . we now present our final results for the full partonic cross sections in analytic form . combining the cross section ( [ trafox ] ) for massless leptons with the weizscker - williams contribution ( [ csww ] ) , we may write the full nlo cross section as e_h & = & ( ) _ i , f _ ^1 + _ ^1 + & & h^if(v , w ) , [ trafox1 ] where @xmath114 has been defined in eq . ( [ shnx ] ) . the lo contribution , present only for the channel @xmath60 with an incoming quark that also fragments , was already given in ( [ cslo ] ) . for the nlo term in this channel we find @xmath115 & + & \frac{1}{(1-w)_+}\bigg\{b_{1}^{q\to q}\ln\left(\frac{1-v}{v(1-v(1-w))}\right)\nonumber\\[2 mm ] & + & b_{2}^{q\to q}\ln(1-v(1-w))+b_{3}^{q\to q}\ln\left(\frac{sv^2}{\mu^2}\right)\bigg\}\nn\\[2 mm ] & + & c_1^{q\to q}\ln(v(1-w))+c_2^{q\to q}\ln\left(\frac{(1-v)w}{1-vw}\right)\nonumber\\[2 mm ] & + & c_3^{q\to q}\ln\left(\frac{1-v}{(1-vw)(1-v(1-w))}\right)\nonumber\\[2 mm ] & + & c_4^{q\to q}\ln\left(\frac{s}{\mu^2}\right)+c_5^{q\to q}\bigg]\,,\label{resq2qnloreal1}\end{aligned}\ ] ] where the coefficients @xmath116 , @xmath117 , @xmath118 are functions of @xmath119 and @xmath120 and may be found in the appendix . the channels @xmath68 and @xmath69 have simpler expressions : @xmath121 & + & c_2^{q\to g}\ln\left(\frac{1-v}{(1-vw)(1-v(1-w))}\right)\nonumber\\[2 mm ] & + & c_3^{q\to g}\ln\left(\frac{v(1-w)s}{\mu^2}\right)+c_4^{q\to g}\bigg]\ , , \label{resq2gnloreal1}\end{aligned}\ ] ] @xmath122 & + & c_2^{g\to q}\ln\left(\frac{v(1-w)s}{\mu^2}\right)+c_3^{g\to q}\bigg]\ , . \label{resg2qnloreal1}\end{aligned}\ ] ] the coefficients @xmath123 and @xmath124 are again given in the appendix . we finally list the partonic cross sections for the weizscker - williams contributions : @xmath125 \hat{\sigma}_{\mathrm{lo}}^{\gamma q\to g}(v , w)&=&\frac{2\pi\,c_f \alpha_{\mathrm{em}}e_q^2}{s(1-v ) } \frac{1+(1-vw)^2}{1-vw}\,,\nn\\[2 mm ] \hat{\sigma}_{\mathrm{lo}}^{\gamma g\to q}(v , w)&= & \frac{2\pi\,t_r\alpha_{\mathrm{em}}e_q^2}{s(1-v)}\frac{v^2 w^2 + ( 1-vw)^2}{vw(1-vw)}\,.\end{aligned}\ ] ] having computed the inclusive hadron production cross section at nlo the extension to single inclusive jet production is straightforward . the cross section for @xmath126 may be written as @xmath127 & \times&\hat{\sigma}^{i\to \mathrm{jet}}\left(v=1+\frac{t}{s},w,\mu;r\right ) \label{jet}\,,\end{aligned}\ ] ] where @xmath128 and @xmath129 are the energy and three - momentum of the jet and the hadronic mandelstam variables are defined as before , now in terms of the jet momentum . the form of this expression follows from ( [ trafox ] ) by setting the fragmentation functions to @xmath130 . of course , beyond lo , the partonic cross sections @xmath131 for jet production differ from the ones for single - inclusive hadron production . this is evident from the fact that the latter are computed as `` inclusive - parton '' cross sections @xmath33 which , as we saw in subsection [ coll ] , require collinear subtraction . this is in contrast to a jet cross section which is by itself infrared - safe , as far as the final state is concerned . instead , it depends on the algorithm adopted to define the jet , as we have indicated by the dependence on a generic jet ( size ) parameter @xmath132 in ( [ jet ] ) . as was discussed in refs . @xcite , even at nlo one may still go rather straightforwardly from the single - inclusive parton cross sections @xmath33 to the @xmath131 , for any infrared - safe jet algorithm . the key is to properly account for the fact that at nlo two partons can fall into the same jet , so that the jet needs to be constructed from both . in fact , assuming the jet to be relatively narrow , one can determine the relation between @xmath33 and @xmath131 analytically @xcite . this `` narrow jet approximation ( nja ) '' formally corresponds to the limit @xmath133 but turns out to be accurate even at values @xmath134 relevant for experiment . we follow this approach in this work . in the nja , the structure of the nlo jet cross section is of the form @xmath135 ; corrections to this are of @xmath136 and are neglected . we note that to the order @xmath81 we consider in this paper , the weizscker - williams terms only contribute to the @xmath132-independent piece @xmath137 . this is because for almost real exchanged photons it is at this order not possible to have two coalescing partons in the final state . we now present phenomenological results for the nlo single - inclusive pion production cross section in lepton - proton scattering . as mentioned before , data on the transverse single - spin asymmetry for this reaction have been released by hermes @xcite and the jefferson lab hall a collaboration @xcite . unfortunately , corresponding cross sections were not presented , and we will therefore provide predictions for these . furthermore , we will also present predictions for compass at cern , for a future electron - ion - collider ( eic ) , and for experiments at jefferson lab after the cebaf upgrade to 12 gev beam energy . finally , at the end of this section we show some phenomenological results for the inclusive production of jets at the eic . as we saw in the previous subsections ( see eq . ( [ trafox1 ] ) ) , our nlo result can be formulated in such a way that it contains contributions involving the photon - in - lepton distribution @xmath102 and lo photon - parton cross sections . these represent the contributions by quasi - real photons to the cross section . an interesting question is whether this part of the cross section dominates the nlo corrections , at least for a suitable choice of the scale @xmath32 in ( [ fwwren ] ) . we recall that the logarithm in ( [ fwwren ] ) may be obtained by an integration over the photon s virtuality where only the @xmath138 propagator is kept for the photon , while @xmath139 is neglected everywhere else in the hard scattering . we now consider the cross section e_h & = & ( ) _ i , f _ ^1 + _ ^1 + & & h^if(v , w ) , + [ trafox2 ] which essentially corresponds to the full nlo one in ( [ trafox1 ] ) , but with the terms @xmath140 dropped . in other words , we use the lo term and add the weizscker - williams contribution . for the latter , we choose the upper limit on @xmath141 in the photon spectrum as a large scale in the problem , @xmath142 or even @xmath143 . this constitutes an attempt to obtain an approximation to the full nlo correction by assuming that the @xmath138-behavior of the hard cross sections is valid over most of the kinematical regime . in our studies we examine in this way the importance of the weizscker - williams contribution . as discussed in the introduction , _ if _ the contribution plays a dominant role for the nlo corrections , this opens the door to approximate nlo calculations also for the spin - dependent case . for all our calculations we use the cteq6.6 m @xcite set of parton distribution functions and the fragmentation functions of @xcite . figures [ fig : hermesxf ] and [ fig : hermespt ] present our results for @xmath144-production at hermes at @xmath145 gev . we fix the renormalization and factorization scales at @xmath146 . in order to match the conventions used in @xmath3 , hermes presents the spin asymmetry results in terms of the hadron s transverse momentum @xmath14 and feynman s @xmath147 , where @xmath148 is the @xmath48-component of the hadron momentum in the center - of - mass frame of the collision , and where positive @xmath149 is counted in the direction of the lepton beam . we have @xmath150 where @xmath151 . the hadronic mandelstam variables read @xmath152 u & = & -\frac{s}{2 } \left(\sqrt{x_f^2+x_t^2}-x_f\right)\,.\label{mandelstamxf}\end{aligned}\ ] ] figure [ fig : hermesxf ] shows the cross section as a function of @xmath149 , integrated over @xmath153 . this is the only @xmath14 bin used in ref . @xcite with @xmath154 . in fig . [ fig : hermespt ] we examine the @xmath14 dependence of the cross section for @xmath155 . in both cases we find large nlo corrections ; the nlo cross section is almost twice as large as the lo one . as discussed above , we also examine in how far the weizscker - williams contribution drives the nlo corrections , using eq . ( [ trafox2 ] ) with @xmath156 ( dotted ) and @xmath157 ( dot - dashed ) . as one can see from the figures , the weizscker - williams contribution does lead to an increase over lo , but provides only about @xmath158 to @xmath159 of the nlo correction . this is likely to be attributed to the fact that the overall c.m.s . energy is rather low . the result with @xmath157 provides a slightly better description of the full nlo , although the differences are minor . we note that the ww approximation appears to work better for smaller transverse hadron momenta @xmath14 and for larger @xmath149 . the latter feature perhaps is at first sight surprising since positive @xmath149 of the hadron imply on average backward scattering of the lepton , whereas the ww approximation should work better if the lepton is scattered in the forward direction . one can roughly understand this `` shift '' of the ww approximation towards positive @xmath149 from the fact that @xmath160 for @xmath161 in eq . ( [ mandelstamxf ] ) . since the dominant real - photon process @xmath162 in ( [ wwsigs ] ) has a @xmath163-behavior in contrast to the @xmath164-behaviour of the lo process , the ww aproximation favors the region @xmath165 . the full nlo partonic cross section inherits the @xmath164-behaviour of the lo one , so that the weizscker - williams contribution can approximate it well only for @xmath165 . our nlo predictions for the cross section for @xmath166 in 12 gev scattering at the jefferson lab are shown in figs . [ fig : jlabxf ] and [ fig : jlabpt ] . for the @xmath149 distribution on the left we have assumed a fixed transverse momentum @xmath167 . on the right we show the @xmath14 dependence of the cross section in the region @xmath168 . again , the renormalization scale is fixed to the transverse hadron momentum , @xmath146 . note that the rather modest c.m.s . energy available limits the possible size of @xmath14 severely . for collisions using the present 6 gev beam only transverse momenta outside the hard - scattering regime are possible , which is the reason why we can not present any results for this case . we again observe in figs . [ fig : jlabxf ] , [ fig : jlabpt ] that the nlo corrections are very large . the weizscker - williams contribution is clearly insufficient to match the nlo result here . the results of our nlo analysis for compass kinematics are shown in figs . [ fig : compasseta ] and [ fig : compasspt ] . compass uses a muon beam with energy 160 gev , resulting in @xmath169 gev . following the choice made by compass , we use here the c.m.s . pseudorapidity @xmath170 of the produced hadron rather than its feynman-@xmath149 . pseudorapidity is counted as positive in the forward direction of the incident muon . we have @xmath171 where the hadronic mandelstam variables read @xmath172 the compass spectrometer roughly covers the region @xmath173 . from the @xmath170 dependence shown in fig . [ fig : compasseta ] for a fixed transverse momentum @xmath174 we observe that the nlo corrections are significant but not as large as for hermes and jlab . they amount to an increase over lo of roughly 3040% . strikingly , the weizscker - williams contribution is very small here , even for the choice @xmath157 . this may be understood from the fact that the muon mass is about 200 times larger than the electron mass , resulting in a much smaller logarithm in the expression ( [ fwwren ] ) for the photon spectrum , which then is largely cancelled by the non - logarithmic term . for the @xmath14 spectrum shown in fig . [ fig : compasspt ] we also show the results for a different choice of the factorization and renormalization scales , @xmath175 . as one can see , the scale dependence decreases somewhat when going from lo to nlo but remains fairly sizable . we finally also discuss the cross section for single - inclusive pion production in electron - proton collisions at a proposed future eic @xcite with @xmath176 gev . thanks to the higher energy of an eic it will become possible to probe much larger transverse hadron momenta , where pqcd is expected to work better . fig . [ fig : eiceta ] shows the @xmath170 dependence of the cross section for a fixed transverse momentum @xmath177 gev . again we count positive @xmath170 in the forward direction of the incoming lepton . the @xmath14 dependence of the cross section is shown in fig . [ fig : eicpt ] , integrated over @xmath178 . the renormalization scale has again been fixed to the transverse hadron momentum , @xmath146 . as for compass we found that the scale dependence slighty decreases for eic kinematics when going from lo to nlo but remains relatively large . we again find sizable nlo corrections . overall , the weizscker - williams approximation works much better here than in the fixed - target regime . it describes the nlo cross section especially well when the hadron is produced in the electron forward direction . at mid - rapidity and negative rapidity the approximation tends to fall short of the full nlo result . from fig . [ fig : eicpt ] we observe that that the weizscker - williams also works better for smaller @xmath14 . given the high energy of an eic , also jet observables will be of much interest there @xcite . for example , combined analysis of data for the transverse - spin asymmetries for @xmath179 and @xmath180 from a future eic should allow for a clean separation of twist-3 parton correlations in the nucleon and in fragmentation . we therefore close this section by presenting predictions for the cross section for single - inclusive jet production , @xmath181 . here we use the nja formalism outlined in sec . [ jetpro ] to convert the single - hadron cross section into a jet one . we adopt the anti-@xmath182 jet algorithm of @xcite . in fig . [ fig : jets ] we present the dependence of the cross section on the jet pseudo - rapidity @xmath183 for a fixed transverse jet momentum of @xmath184 gev . we find once again that nlo contributions are large . we also observe that , compared to the case of hadron production considered in fig . [ fig : eiceta ] , the nlo cross section is much more peaked in the forward electron region . the reason is that at large positive pseudo - rapidity @xmath160 in eq . ( [ mandelstameta ] ) . since the minimal value for the incoming parton s momentum fraction is @xmath185 in ( [ jet ] ) rather small values of @xmath35 are probed at large pseudo - rapidity where in turn the nucleon s parton distributions are large . on the other hand the fragmentation process suppresses the forward and backward regions in hadron production due to the large @xmath48-values probed , whereas in jet production the forward electron region is enhanced due to the absence of fragmentation . in the figures [ fig : jets ] and [ fig : jets1 ] , we show results for two different jet size parameters , @xmath186 and @xmath187 . dependence on @xmath132 first occurs at nlo . as discussed at the end of sec . [ jetpro ] , the first - order weizscker - williams contribution does not depend on @xmath132 . it hence can not give an accurate approximation of nlo in general . as the figure shows , the ww result happens to be rather close to the result for @xmath186 ; this agreement , however , is essentially fortuitous . we have performed next - to - leading order calculations of the partonic cross sections for the processes @xmath0 and @xmath188 , for which the scattered lepton in the final state is not detected . we have derived our results for a finite lepton mass , neglecting terms that are suppressed as powers of the mass over a hard scale . the results have been obtained in two ways . we have first set the mass to zero . we have regularized the ensuing collinear singularity in dimensional regularization and then subtracted it by introducing a weizscker - williams type photon distribution in the lepton . the latter can be computed in qed perturbation theory and effectively reinstates the leading lepton mass dependence , which is logarithmic plus constant . in the second approach , we have kept the lepton mass in the calculation directly , expanding all phase space integrals in such a way that the leading mass dependence is obtained . both approaches give the same result . we have presented phenomenological nlo predictions for various experimental setups , from fixed - target experiments ( hermes , jlab , compass ) to collider experiments at an eic . we have found that the nlo corrections are large . we note that in the fixed target regime the bulk of the corrections comes from the plus distribution terms in eq . ( [ resq2qnloreal1 ] ) , especially at negative @xmath149 or rapidity . as is well known , the distributions are associated with the emission of soft gluons . since they recur with increasing power at every higher order of perturbation theory , it may be worthwhile for future work to address their resummation to all orders , similarly to what was done for the photoproduction case @xmath189 in @xcite . the rather large size of the corrections that we find suggests that also the cross section with transverse polarization of the initial nucleon may be subject to large nlo corrections . this would likely have ramifications for analyses of spin asymmetry data for @xmath190 in terms of twist-3 parton correlation functions . as full nlo calculations for transverse single - spin observables are difficult , we have also investigated in how far it is possible to match our full nlo result for the spin - averaged cross section by adding just the weizscker - williams contribution to the lo one . we have found that this simplified approach does not appear to work well quantitatively . in other words , the nlo corrections do not appear to be dominated by quasi - real photons . nonetheless , in order to obtain a first estimate of higher - order effects for the transverse - spin asymmetry , it may be worthwhile to use the weizscker - williams contribution for the case of transversely polarized nucleons , which is much simpler to do than the full nlo calculation and was already discussed in ref . @xcite . we again emphasize that our results suggest that contributions by quasi - real photons to the cross sections for the single - inclusive processes @xmath0 and @xmath1 are _ not _ the dominant contributions , at least for large transverse hadron momenta @xmath191 . in other words , an experimental setup where the final state lepton is not observed in lepton - nucleon collisions does not automatically imply that one measures an ( approximated ) quasi - real photoproduction process . however , although quasi - real photons do not dominate , they typically do play a non - negligible role for the nlo corrections . as is well known , high - energy real photons may also exhibit their own partonic structure , in which case they are referred to as `` resolved '' photons ( see ref . the corresponding resolved - photon contributions are formally of the same order as the weizscker - williams contribution we have considered here . they are typically suppressed in the fixed - target regime . it may be interesting to address this contribution in future work , also in order to study its impact on the transverse single - spin asymmetries . the concept of `` virtual photon structure '' may also prove useful in this context ( see , for example , ref . @xcite ) . we thank a. metz for interesting discussions that have initiated this project , and m. stratmann for helpful comments . we thank umberto dalesio for pointing out to us the inconsistencies in our results presented in an earlier version of this paper . this work was supported by the `` bundesministerium fr bildung und forschung '' ( bmbf ) grant 05p12vtctg . here we present the nlo coefficients in eqs . ( [ resq2qnloreal1 ] ) , ( [ resq2gnloreal1 ] ) , ( [ resg2qnloreal1 ] ) for the different channels : @xmath192 a_{1}^{q\to q } & = & 8 w \frac{1+v^2}{(1-v)^2},\nonumber\\[2 mm ] b_{1}^{q\to q } & = & 4w\frac{1-v(1-w)+v^2(1-w(1-w))}{(1-v)^2},\nonumber\\[2 mm ] b_{2}^{q\to q } & = & \frac{2w}{(1-v)^2(1-v(1-w))}\times\nonumber\\ & & \big[\left(1 - 2v(1-w)+v^2(1 - 2w+2w^2)\right)\times\nonumber\\ & & \left(2 - 2v(1-w)+v^2(1-w)^2\right)\big],\nonumber\\[2 mm ] b_{3}^{q\to q } & = & 4 w \frac{1+v^2}{(1-v)^2}.\label{ab1q2q}\end{aligned}\ ] ] @xmath193,\nonumber\\[2 mm ] c_{2}^{q\to q } & = & \frac{2(1+v^2(1 + 2w^2))}{(1-v)^2},\nonumber\\[2 mm ] c_{3}^{q\to q } & = & \frac{-2vw(3 - 2v(1-w)+v^2(1 - 2w+2w^2))}{(1-v)^2},\nonumber\\[2 mm ] c_{4}^{q\to q } & = & \frac{1}{(1-v)^2(1-vw)(1-v(1-w))}\big[2-w-2v(1 + 2w)\nonumber\\ & & + v^2(2 + 5w-6w^2+w^3)-2v^3(1-w+w^2 - 2w^3)\nonumber\\ & & + v^4w(2 - 2w-3w^2 + 4w^3)\nonumber\\ & & -2v^5w^2(1 - 3w+4w^2 - 2w^3)\big],\nonumber\\[2 mm ] c_{5}^{q\to q } & = & \frac{1}{(1-v)^2(1-vw)(1-v(1-w))}\big[2-w-2v(2-w)\nonumber\\ & & + v^2(4-w-2w^2+w^3)-2v^3+v^4w(2-w^2)\nonumber\\ & & -2v^5w^2(1 - 2w+2w^2-w^3)\big].\label{cq2q}\end{aligned}\ ] ] @xmath195 c_{2}^{q\to g } & = & \frac{vw(6 - 4vw+2v^2(1 - 2w(1-w)))}{(1-v)^2},\nonumber\\[2 mm ] c_{3}^{q\to g } & = & \frac{vw}{(1-v)^2(1-vw)^2(1-v(1-w))^2}\big[3 - 2v(3+w)\nonumber\\ & & + v^2(6 + 4w - w^2)-2v^3(3 - 3w+5w^2 - 2w^3)\nonumber\\ & & + v^4(3 - 4w+5w^2 - 2w^3)\nonumber\\ & & -2v^5w(2 - 6w+9w^2 - 7w^3 + 2w^4)\nonumber\\ & & + 2v^6(1-w)^2w^2(1 - 2w+2w^2)\big],\nonumber\\[2 mm ] c_{4}^{q\to g } & = & \frac{vw}{(1-v)^2(1-vw)^2(1-v(1-w))^2}\big[2 - 2v(5 - 3w)\nonumber\\ & & + v^2(16 - 3w-11w^2)-v^3(10 + 15w-27w^2 + 2w^3)\nonumber\\ & & + v^4(2 + 17w-23w^2 + 7w^3 - 3w^4)\nonumber\\ & & -v^5w(5 - 5w - w^2 + 3w^3 - 2w^4)\nonumber\\ & & + 2v^6(1-w)^2w^2(1-w+w^2)\big].\label{cq2g}\end{aligned}\ ] ] @xmath197 c_{2}^{g\to q } & = & \frac{1}{(1-v)^2(1-vw)^2}\big[2(1-w+w^2)\nonumber\\ & & -2vw(3 - 2w+2w^2)\nonumber\\ & & + v^2(2 - 4w+11w^2 - 2w^3 + 2w^4)\nonumber\\ & & -4v^3w(1 - 2w+3w^2)+3v^4w^2(1 - 2w+2w^2)\big],\nonumber\\[2 mm ] c_{3}^{g\to q } & = & \frac{1}{(1-v)^2(1-vw)^2}\big[1 + 4w-6w^2\nonumber\\ & & -2v(1 + 3w+w^2 - 6w^3)+\nonumber\\ & & v^2(1 + 9w+4w^2 - 8w^3 - 6w^4)\nonumber\\ & & -v^3w(3 + 9w-4w^2 - 6w^3)\nonumber\\ & & + v^4w^2(1 + 4w-4w^2)\big].\label{cg2q}\end{aligned}\ ] ] l. gamberg , z. b. kang , a. metz , d. pitonyak and a. prokudin , phys . rev . d * 90 * , 074012 ( 2014 ) [ arxiv:1407.5078 [ hep - ph ] ] . m. anselmino , m. boglione , j. hansson and f. murgia , eur . j. c * 13 * , 519 ( 2000 ) [ hep - ph/9906418 ] . m. klasen , rev . phys . * 74 * , 1221 ( 2002 ) [ hep - ph/0206169 ] . c. a. aidala , s. d. bass , d. hasch and g. k. mallot , rev . mod . phys . * 85 * , 655 ( 2013 ) [ arxiv:1209.2803 [ hep - ph ] ] . n. christ and t. d. lee , phys . * 143 * , 1310 ( 1966 ) . d. bdeker , phys . b * 292 * , 164 ( 1992 ) ; z. phys . c * 59 * , 501 ( 1993 ) ; g. kramer and s. g. salesch , z. phys . c * 61 * , 277 ( 1994 ) ; d. bdeker , g. kramer and s. g. salesch , z. phys . c * 63 * , 471 ( 1994 ) . s. frixione and g. ridolfi , nucl . b * 507 * , 315 ( 1997 ) [ hep - ph/9707345 ] ; d. de florian and s. frixione , phys . b * 457 * , 236 ( 1999 ) [ hep - ph/9904320 ] ; d. de florian and w. vogelsang , phys . d * 57 * , 4376 ( 1998 ) [ hep - ph/9712273 ] . b. jger , m. stratmann and w. vogelsang , phys . d * 68 * , 114018 ( 2003 ) [ hep - ph/0309051 ] ; eur . j. c * 44 * , 533 ( 2005 ) [ hep - ph/0505157 ] ; b. jger , phys . d * 78 * , 034017 ( 2008 ) [ arxiv:0807.0066 [ hep - ph ] ] . z. b. kang , s. mantry and j. w. qiu , phys . d * 86 * , 114011 ( 2012 ) [ arxiv:1204.5469 [ hep - ph ] ] ; z. b. kang , x. liu , s. mantry and j. w. qiu , phys . d * 88 * , 074020 ( 2013 ) [ arxiv:1303.3063 [ hep - ph ] ] . c. f. von weizscker , z. phys . * 88 * , 612 ( 1934 ) . e. j. williams , phys . * 45 * , 729 ( 1934 ) . a. c. bawa and w. j. stirling , j. phys . g * 15 * , 1339 ( 1989 ) . s. frixione , m. l. mangano , p. nason and g. ridolfi , phys . b * 319 * , 339 ( 1993 ) [ hep - ph/9310350 ] . a. afanasev , c. e. carlson and c. wahlquist , phys . b * 398 * , 393 ( 1997 ) [ hep - ph/9701215 ] ; phys . d * 58 * , 054007 ( 1998 ) [ hep - ph/9706522 ] ; phys . d * 61 * , 034014 ( 2000 ) [ hep - ph/9903493 ] . w. beenakker , h. kuijf , w. l. van neerven and j. smith , phys . d * 40 * , 54 ( 1989 ) . l. e. gordon and w. vogelsang , phys . d * 48 * , 3136 ( 1993 ) . j. c. collins , _ foundations of perturbative qcd _ , cambridge university press , cambridge , 2011 . t. kaufmann , a. mukherjee and w. vogelsang , phys . d * 91 * , 034001 ( 2015 ) [ arxiv:1412.0298 [ hep - ph ] ] . w. k. tung , h. l. lai , a. belyaev , j. pumplin , d. stump and c .- p . yuan , jhep * 0702 * , 053 ( 2007 ) [ hep - ph/0611254 ] . d. de florian , r. sassot , m. epele , r. j. hernandez - pinto and m. stratmann , phys . d * 91 * , 014035 ( 2015 ) [ arxiv:1410.6027 [ hep - ph ] ] . a. accardi , j. l. albacete , m. anselmino , n. armesto , e. c. aschenauer , a. bacchetta , d. boer and w. brooks _ et al . _ , arxiv:1212.1701 [ nucl - ex ] .

the recently proposed `` mini - seesaw mechanism '' combines naturally suppressed dirac and majorana masses to achieve light standard model neutrinos via a low - scale seesaw . a key feature of this approach is the presence of multiple light ( order gev ) sterile - neutrinos that mix with the standard model . in this work we study the bounds on these light sterile - neutrinos from processes like , invisible-decays , and neutrinoless double beta - decay . we show that viable parameter space exists and that , interestingly , key observables can lie just below current experimental sensitivities . in particular , a motivated region of parameter space predicts a branching fraction within the range to be probed by meg . neutrino mass and from a mini - seesaw michael duerr, damien p. george and kristian l. mcdonald + max - planck - institut fr kernphysik , + postfach 10 39 80 , 69029 heidelberg , germany + nikhef theory group , science park 105 , + 1098 xg amsterdam , the netherlands +

1105.0593

the discovery of neutrino mass has provided a wealth of information regarding the flavour structure of nature ( for reviews see @xcite ; for the latest global fit see @xcite ) . however , key questions regarding the neutrino sector remain unanswered and , in particular , the underlying mechanism responsible for neutrino mass remains elusive . the seesaw mechanism provides a simple explanation for the lightness of the known neutrinos @xcite . in its standard implementation one assumes @xmath7 , where @xmath8 is the neutrino dirac mass , @xmath9 is the singlet majorana mass and @xmath10 is the @xmath11 boson mass . the result is a light standard model ( sm ) neutrino with mass @xmath12 and a heavy sterile neutrino with mass @xmath13 . despite the simplicity of this approach , it offers little hope of being experimentally verified due to the invocation of the typically ( very ) large scale @xmath9 . a number of alternative mechanisms invoke new physics to explain neutrino masses without relying on a large uv scale . these offer the advantage of being experimentally verifiable and one may hope that the requisite new physics appears at the tev scale . this can happen in models with radiatively - generated neutrino mass @xcite , and the mechanism of mass generation can even be connected to the weak scale in models with coleman - weinberg symmetry breaking @xcite . of course , nature is likely not concerned with our ability to experimentally ascertain its inner workings . nonetheless it is interesting to explore alternative approaches to neutrino mass that may lie within experimental reach ; particularly if they enable a natural realization and/or provide insight into the observed flavour structures . beyond being sm singlets , right - handed neutrinos can form part of a more complicated hidden sector , and may provide a link to the dark / hidden sector of the universe . this feature is used to advantage in models that invoke non - standard properties of the gauge - singlet neutrinos in an effort to understand the origin of neutrino mass . for example , interesting exceptions to the standard seesaw picture arise if the right - handed neutrinos are bulk fields in an extra dimension @xcite or if they are composite objects within a strongly coupled hidden sector @xcite . a recent work has proposed a `` mini - seesaw mechanism '' in which naturally suppressed dirac and ( sterile ) majorana masses are combined to achieve light sm neutrinos via a low - scale seesaw @xcite . this approach borrows elements from both the extra - dimensional scenario and the composite right - handed neutrino approach : the sterile neutrinos are bulk fields in a truncated slice of @xmath14 but , via the ads / cft correspondence @xcite , the model has a dual 4d description in which the right - handed neutrinos are the lightest composites of a strongly coupled hidden sector . a key feature of this approach is the presence of a tower of light ( order gev ) sterile neutrinos that can be thought of as either kaluza - klein modes or higher fermionic resonances of the cft ( with mass gap ) . in this work we consider the detailed bounds on these light sterile - neutrinos . we show that viable parameter space exists in which sm neutrino masses can be successfully generated without recourse to large ( supra - tev ) scales . furthermore , we show that it may be possible to observe the effects of the sterile neutrinos in forthcoming experiments : the most stringent constraints come from the fcnc process @xmath0 and , as we will show , the interesting region of parameter space corresponds to @xmath15 . this is right below the current experimental bound and within the region to be probed by the meg experiment @xcite . the meg collaboration recently reported that the best value for the number of signal events in their maximal likelihood fit is three @xcite . interestingly , this corresponds to @xmath16 which , as we show , is obtained via the mini - seesaw with neutrino yukawa couplings approaching the @xmath17 level . before proceeding we note that bulk gauge - singlet neutrinos in randall - sundrum ( rs ) models @xcite were considered in @xcite and bulk sm fermions in @xcite . subsequent studies of neutrino mass appeared in @xcite and for an incomplete list of recent works in this active field see @xcite . related work on right - handed neutrinos within a strongly coupled cft was undertaken in @xcite . our implementation within a sub - tev scale effective theory differs from these previous works . other works have also considered a low - scale warped hidden sector @xcite and for another use of bulk sterile neutrinos see @xcite . light sterile - neutrinos have been studied extensively within the context of the @xmath18msm @xcite and for a discussion of the low - scale ( gev ) seesaw see @xcite . ref . @xcite contains a review on light steriles and a detailed analysis of the bounds on kk neutrinos in add models appeared in @xcite . a recent work has studied a viable realization of the meg signal in a susy model @xcite and other related recent works include @xcite . the layout of this paper is as follows . section [ sec : setup ] describes the model and outlines the basic neutrino spectrum . in section [ sec : mass_eigenstates ] the sterile neutrino spectrum is considered in more detail and the mixing with the sm is determined . the analysis of the bounds follows . we study the neutrinoless double beta - decay rate and invisible @xmath1-decays in sections [ sec : nubb ] and [ sec : hid_z_decay ] respectively , a number of other bounds on active - sterile neutrino mixing in section [ sec : other_bounds ] ( e.g. collider constraints ) , and the lepton number violating decay @xmath0 in section [ sec : mu_to_e_gamma ] . finally we discuss the prospects for observing invisible higgs decays in section [ sec : higgs_decay ] and conclude in section [ sec : conc ] . we consider a truncated rs model with a warped extra dimension described by the coordinate @xmath19 $ ] . a uv brane of characteristic energy scale @xmath20 is located at @xmath21 and an ir brane with characteristic scale @xmath22 is located at @xmath23 . the metric is given by ds^2 = ( _ dx^dx^ - dz^2)= g_mn dx^mdx^n , [ bulkmetric ] where @xmath24 ( @xmath25 ) are the 5d ( 4d ) lorentz indices and @xmath26 is the @xmath14 curvature . the characteristic ir scale is suppressed relative to the curvature , as is readily seen using the proper coordinate for the extra dimension . , where @xmath27 $ ] and @xmath28 , in terms of which the ir scale is exponentially suppressed , @xmath29 . ] when sourced by a bulk cosmological constant and appropriate brane tensions the metric of eq . is a solution to the 5d einstein equations @xcite . furthermore , the length of the space is readily stabilized @xcite . we take the sm to be localized on the uv brane , where the natural mass scale is @xmath30 , and accordingly take @xmath31 tev , where @xmath32 is the cutoff scale . in addition , we consider three singlet fermions propagating in the bulk . we label these by @xmath33 as the zero modes will be right - chiral fields that we identify as gauge - singlet neutrinos . the ir scale is nominally taken to be @xmath34 gev . a sketch of the setup is given in figure [ fig : nu_lit_throat ] . in analogy with the little rs model @xcite we can refer to this as a little warped space " ( lws ) . such a truncated spacetime may seem somewhat unusual at first sight . however , the setup can be thought of as an effective theory that describes the sub - tev scale physics of a more complete theory , enabling one to consider the effects of a light warped / composite hidden - sector without having to specify the supra - tev physics . in general the supra - tev effects will be encoded in uv localized effective operators . as with any sub - tev scale effective theory , effective operators that break approximate / exact symmetries of the sm must be adequately suppressed to ensure that problematic effects like rapid p - decay and excessive flavour violation do not occur . the action for a bulk fermion @xmath33 in the background ( [ bulkmetric ] ) is . ] @xmath35 where @xmath36 are the 5d dirac - gamma matrices , @xmath37 is the fnfbein and we write the dirac mass in units of the curvature @xmath26 . a kaluza - klein ( kk ) decomposition may be performed as n_r(x , z)=(kz)^2_n \{_l^(n)(x)f_l^(n)(z)+_r^(n)(x)f_r^(n)(z ) } , and the bulk wavefunctions @xmath38 for the chiral components of @xmath33 are readily found @xcite : @xmath39 where the order of the bessel functions is @xmath40 . the boundary conditions force one chirality to be odd and without loss of generality we take this to be the left - chiral field . the single massless mode in the spectrum then has right - chirality . the kk - expanded lagrangian is @xmath41 where @xmath42 is a dirac fermion with kk mass @xmath43 for @xmath44 and @xmath45 is the massless right - chiral zero mode , whose bulk profile is f^(0)_r(z)=(kz)^c . the dimensionless mass parameter @xmath46 is seen to control the localization of the zero mode . we will be interested in ir localization with @xmath47 , in order to reduce the wavefunction overlap of the zero mode with uv localized sm neutrinos and suppress the dirac mass below the weak scale . the kk masses are determined by enforcing the boundary conditions ( dirichlet / neumann for @xmath48 ) , and must satisfy @xmath49 , giving @xmath50 for @xmath44 . for @xmath51 these masses are @xmath52 the boundary values of the wavefunctions for the right - chiral modes will be important in the following ; for @xmath47 and @xmath53 these are f_r^(n)(k^-1)&&()^c , + f_r^(n)(r)&&(-1)^n . at this stage the spectrum consists of a single weyl neutrino and a tower of dirac neutrinos with masses @xmath54 . we now introduce lepton number violation in the form of a marginal operator on the ir brane : @xmath55 \begin{center } \includegraphics[width = 0.7\textwidth]{yukawa_vs_c.eps } \end{center } \caption{the neutrino yukawa coupling ( $ \lambda$ ) as a function the bulk mass parameter ( $ c$ ) for a fixed value of the sm neutrino mass ( $ m_\nu=0.1$~ev ) . the lines correspond to ( bottom to top ) : $ \lambda_{{\text{\tiny n}}}=0.1,\ 0.4,\ 0.8,\ 1.6$. the ir scale is fixed at $ r^{-1}=1$~gev and $ k = m_*/2=1.5$~tev . } \label{fig : yuk_vs_c_normal } \end{figure } this approach , referred to as a `` mini - seesaw mechanism''~\cite{mcdonald:2010jm } , generates naturally suppressed sm neutrino masses in a two - fold process . firstly , the effective 4d dirac and majorana mass scales are suppressed : the sub - tev majorana scale is generated by warping ( $ m_{00}\ll m_w$ ) while the dirac mass is suppressed by a small wavefunction overlap ( $ m^{{\text{\tiny d}}}_0\ll m_{00}\ll m_w$ ) . secondly , a low - scale seesaw operates between the lightest kk mode and the sm neutrino , further suppressing the sm neutrino mass . together these elements realize the order ev neutrino masses . we emphasize that , unlike most seesaw models with light sterile - majorana scales , the small neutrino masses are generated naturally and do not require tiny yukawa couplings \emph{a priori}. figure~\ref{fig : yuk_vs_c_normal } shows a plot of the yukawa coupling $ \lambda$ as a function of the bulk mass parameter\footnote{we restrict our attention to values of $ |c|$ no larger than those employed in~\cite{davoudiasl:2008hx}. } $ c$ for the fixed value $ m_\nu=0.1$~ev . it is clear from the plot that $ m_\nu$ can be readily suppressed below the weak scale to within the range of interest for the solar and atmospheric neutrino data , even for $ \mathcal{o}(1)$ values of the dimensionless couplings $ \lambda$ and $ \lambda_{{\text{\tiny n}}}$. for future reference we note that $ \lambda\in[10^{-2},0.1]$ for $ c\in[1.3,1.6]$ when $ \lambda_{{\text{\tiny n}}}=0.1$. the goal of this work is to determine the experimental bounds on the tower of low - scale sterile neutrinos . we will focus on values of $ r^{-1}\sim$~gev and investigate whether the bounds allow $ \mathcal{o}(1)$ values for the yukawa coupling $ \lambda$. before turning to these matters , we note that , via the application of the ads / cft correspondence to rs models~\cite{arkanihamed:2000ds } , the present model is considered dual to a 4d theory in which the uv localized sm fields are treated as fundamental objects external to a hidden strongly - coupled sector . the composite objects correspond to ir localized fields in the 5d picture ( roughly ; see e.g.~\cite{batell:2007jv } ) and the lightest fermionic composite is the right - handed neutrino ( or neutrinos in a three generation model ) . note that ir localization of the majorana mass term indicates that lepton number is broken within the strongly coupled sector . the dual theory is conformal for energies $ \mathrm{tev } \gtrsim e\gtrsim r^{-1}$ , with the conformal symmetry broken spontaneously in the ir ( dual to the ir brane ) and explicitly in the uv ( dual to the tev brane , itself associated with the scale of electroweak symmetry breaking ) . thus the little warped space allows one to model light right - handed neutrinos that are composites of a hidden cft with a sub - weak mass gap . the small yukawa coupling between the sm and the lightest right - chiral neutrino is also understood in the dual picture . for every bulk field in the 5d picture , there exists a cft operator in the dual 4d theory that is sourced by a fundamental field . the source field corresponds to the uv - brane value of the given bulk field in the 5d theory . the sm is external to the cft , but couples directly to the source field , $ n_r|_{z=1/k}$. writing this field in terms of the physical mass eigenstates introduces the aforementioned tiny mixing angle , the dependence on which ensures the effective yukawa coupling is highly suppressed . \section{mixing with a tower of sterile neutrinos\label{sec : mass_eigenstates } } having specified the model ingredients , we turn to a more detailed consideration of mass mixing in the neutral lepton sector . the neutrino mass lagrangian can be written as : \bea \mathcal{l}_\nu\supset -\frac{1}{2}\ \bar{\mathcal{v}}^c_{flav}\ \mathcal{m}\ \mathcal{v}_{flav } \ + \ \mathrm{h.c . } , \eea where the neutrino - space flavour basis vector is : \bea \mathcal{v}_{flav } & = & ( \nu_{\alpha l } , \nu_{0r}^{c } , \nu_{1r}^{c } , \nu_{1l},\nu_{2r}^{c } , \nu_{2l},\ \dots\ ) ^t\ , \eea and the subscript `` $ \alpha$ '' denotes that this is a sm flavour eigenstate . the mass matrix is given by \bea \mathcal{m}=\left(\begin{array}{cc}0 & \mathcal{m}_{d}^t\\\mathcal{m}_{d } & \mathcal{m}_h\end{array}\right)\ , \eea where the dirac mass matrix is \bea \mathcal{m}_{d}=(\ m_0^{{\text{\tiny d}}},\ m_1^{{\text{\tiny d}}},\ 0,\ m_2^{{\text{\tiny d}}},\ 0,\ \dots)^t\ , \eea and the mass matrix for the `` heavy '' kk modes is \bea \mathcal{m}_{h}= \left(\begin{array}{cccccc } -m_{00 } & -m_{01}&0 & -m_{02}&0&\cdots\\ -m_{01}&-m_{11}&m_1 & -m_{12}&0&\cdots\\ 0&m_1&0&0&0&\cdots\\ -m_{02 } & -m_{12}&0&-m_{22}&m_2&\cdots\\ 0&0&0&m_2&0&\cdots\\ \vdots&\vdots&\vdots&\vdots&\vdots&\ddots \end{array}\right)\ . \eea the dirac masses $ m_n^{{\text{\tiny d}}}$ and the majorana masses $ m_{nn'}$ were defined in the previous section . a mass matrix with this form was discussed recently in ref.~\cite{watanabe:2009br } and an exact expression for the inverse was obtained : \begin{eqnarray } \mathcal{m}_h^{-1 } \;=\ ; \left ( \begin{array}{cccccc } \frac{-1}{m_{{00 } } } & 0 & \!\!\!\frac{m_{{01}}}{m_{{00}}}\frac{-1}{m_{1}}\!\!\ ! & 0 & \!\!\!\frac{m_{{02}}}{m_{{00}}}\frac{-1}{m_{2}}\!\ ! & \cdots \\ 0&0 & \frac{1}{m_{1 } } & 0 & 0 & \cdots \\ \!\!\frac{m_{{01}}}{m_{{00}}}\frac{-1}{m_{1 } } & \frac{1}{m_{1 } } & 0 & 0 & 0 & \cdots \\ 0&0 & 0 & 0 & \frac{1}{m_{2 } } & \cdots \\ \!\!\frac{m_{{02}}}{m_{{00}}}\frac{-1}{m_{2 } } & 0 & 0 & \frac{1}{m_{2 } } & 0 & \cdots \\ \vdots & \vdots & \vdots & \vdots & \vdots & \ddots \\ \end{array } \right)\ . \label{inv}\end{aligned}\ ] ] we will use this result below . note that , in deriving the above expression for the inverse matrix , one makes use of relations between the elements @xmath56 which result from the underlying warped space derivation . though the context in which this matrix was considered in @xcite differs from that considered here , the same relations amongst the elements @xmath56 are present in our case . the one generation result presented in ref . @xcite also generalizes to three generations ; i.e. the elements of the matrix presented in @xcite ( and here ) can be treated as matrices for the three generation case @xcite . we focus on the one generation case but keep some expressions general . diagonalization of the matrix @xmath57 proceeds as follows . first we perform a standard seesaw rotation by defining the matrix @xmath58 as ( cc1 & _ d^t _ h^-1 + -(_d^t _ h^-1)^t & 1 ) , giving ^t ( cc- _ d^t _ h^-1_d&0 + 0 & _ h ) , in the basis @xmath59 . to diagonalize the heavy kk matrix we write it as @xmath60 , where the first ( second ) term contains the kk dirac ( majorana ) masses @xmath43 @xmath61 . we treat the majorana masses as a perturbation on the matrix @xmath62 . as shown in the previous section , the elements @xmath56 are all @xmath63 , while the kk masses grow like @xmath54 . thus for @xmath64 one has @xmath65 for @xmath66 , and the majorana masses merely perturb the spectrum of the higher kk modes . for the modes with @xmath67 one still has @xmath68 for @xmath64 , but the majorana masses can be more important as @xmath69 approaches and exceeds unity from below . for @xmath70 it is valid to treat the majorana masses as perturbations for the entire tower . for simplicity we focus on this region of parameter space but mention some differences where important . next we define the matrix @xmath71 as '' in ( [ eq : rotation ] ) is replaced by the pmns mixing matrix @xmath72 , which diagonalizes the light - sector matrix @xmath73 . ] = ( 1 , u_kk),[eq : rotation ] where @xmath74 diagonalizes the kk matrix @xmath75 : u_kk^t_h u_kk = ( m_0 , m_+1 , m_-1 , m_+2 , m_-2 , ) , and the leading order expression for @xmath74 can be found in appendix [ sec : mass_mixing ] . the mass eigenvalues are m _ & & ( kr)^-2c + m_0&&-m_00 + m_n&&m_n - m_nn , where we include the light mass eigenvalue @xmath76 for completeness . negative mass eigenvalues can be brought positive by an appropriate field redefinition , as per usual . we will leave the negative signs so we do nt need to keep track of phases in the mixing matrix . let us emphasize that the above expressions for the @xmath44 kk masses are correct to @xmath77 ; their precise form can change if @xmath56 is increased such that @xmath78 for some values of @xmath79 . however , the expression for the sm neutrino mass , @xmath80 , is exact to @xmath81 and does not depend on the relation between @xmath56 and @xmath43 . it is interesting that the leading order seesaw - suppressed mass depends only on the majorana mass of the zero - mode . the flavour basis can be expressed in terms of the mass basis : _ flav & = & _ mass + & & ( cc1&(_d^t _ h^-1 ) u_kk + -(_d^t _ h^-1)^t & u_kk ) _ mass , where the mass eigenstates are written as _ mass & = & ( _ i l , n_0r^c , n_+1l , n_-1l , n_+2l , n_-2l , ... ) ^t , and the subscript `` @xmath82 '' denotes a ( mostly ) sm mass eigenstate . we can write the sm flavour eigenstate in terms of the mass eigenstates as _ l= _ il+_0n_0r^c + _ n>0(_,+n n_+n + _ , -n n_-n ) _ l , where @xmath83 is the mixing matrix element between @xmath84 and the @xmath85th level kk mode . for much of what follows we will not need to differentiate the sm flavour eigenstates and accordingly refer to @xmath83 simply as @xmath86 . in the limit of a vanishing boundary majorana mass , @xmath87 , the two modes @xmath88 fuse together to form a single dirac neutrino with mass @xmath43 , and one has @xmath89 . treating the boundary majorana mass terms as a perturbation on the kk mass matrix , the mixing angles are approximately _ 0 & & - + _ n>0 + _ n & & ( 1 + ) + _ nn ( ) .we will make use of these expressions in what follows . note that the @xmath79-dependence of the majorana mass appearing in the sum term of @xmath90 is @xmath91 . consequently the sum piece in the expression for @xmath90 is found to be subdominant for @xmath92 , giving @xmath93 , which is of the usual seesaw form . similarly , the sum piece in the expressions for @xmath94 gives a sub - leading contribution . before turning to a detailed analysis of the bounds , we note that the hidden sector also contains a tower of kk gravitons with mass splittings of order @xmath34 gev . in the present work we follow ref . @xcite and give the bulk graviton a dirichlet bc on the uv brane . this projects the zero - mode graviton out of the spectrum and ensures that the higher modes do not couple directly to the uv - localized sm . in this case 4d gravity is not included and the model deals purely with particle physics . the spectrum also contains a physical 4d scalar ( the radion ) that can couple to the uv - localized sm . however , with the standard parametrization of the radion fluctuation @xcite we find that the radion @xmath95 couples to the sm fields like @xmath96 , with the dimensionful parameter being @xmath97 . radion interactions can thus be safely neglected for @xmath98 tev . in a more complete treatment , one could aim to retain 4d gravity by including a large uv - localized einstein - hilbert term and analyzing the metric fluctuations in the interval approach to braneworld gravity @xcite . the kk gravitons would be largely repelled from the uv brane so their coupling to uv - localized sm fields remains sufficiently weak to avoid experimental constraints . the radion coupling may also be modified by this procedure . we hope to study these matters in a future work . with the above information we have all we need to study the bounds on the mini - seesaw . the sterile kk modes provide additional channels for neutrinoless double beta - decay ( @xmath99-decay ) , the non - observation of which provides bounds on the parameter space . these bounds can be approximated by noting that the amplitude for @xmath99-decay from sm neutrinos goes like @xcite g_f^2 , where @xmath100 , @xmath101 is the fermi constant , and @xmath102 is the typical momentum exchanged during the process . the amplitude for @xmath99-decay induced by a single sterile majorana neutrino ( @xmath103 ) that mixes with the sm neutrino via @xmath104 goes like _ s g_f^2 . using the bound @xmath105 and assuming that the sterile neutrino contribution dominates , one obtains the bound @xmath106 @xcite . we can generalize this result for a tower of sterile neutrinos . the @xmath99-decay amplitude produced by the tower is _ kkg_f^2 + _ n>0 g_f^2 \{+ } , and one therefore requires 3|+_n>0 \{+ } | 510 ^ -8 ^-1 , where we make use of the fact that @xmath107 for the parameter space we consider . we also include a sum over three generations of sterile neutrinos , assuming that the active / sterile mixing is similar for each generation ( i.e. we do not assume any tuned cancellations of leading terms @xcite ) . with the information from the previous sections one can calculate the the above factor and determine the size of the @xmath99-decay amplitude . in general the resulting expressions are rather complicated and numerical evaluation is required . however , some simple leading - order expressions can be obtained for the regions of parameter space considered here . approximating the mixing angles by their leading terms : _ 0 & & - , + _ n & & ( 1+),and noting that _ n>0 \{+ } _ n>0 \{_+n^2 -_-n^2 + ( _ + n^2 + _ -n^2 ) + } , one has _ n>0 \{+ } _ n>0 ( ) ^2()+ . note that the @xmath79th term is proportional to the lepton number violating mass @xmath108 ; in the limit that the majorana mass vanishes the contributions to @xmath99-decay from the modes @xmath85 cancel each other out . for @xmath46 in the vicinity of unity the above may be expressed in terms of the light sm neutrino mass : in eq . ( [ nubb ] ) . more generally , the factor of `` @xmath109 '' is replaced by @xmath110 . ] & & 3[nubb ] + & & ( ) ( ) ^2 ( ) ^2| + 3_^4_n>0 ( ) ^2(c-2)|10 ^ -8 ^-1.the above expression can be evaluated numerically , with the sum over kk modes cutoff in the uv at @xmath111 . in practise , no significant error is introduced if one takes the rs2 limit and replaces the sum by a definite integral : @xmath112 . we plot eq . ( [ nubb ] ) as a function of the bulk mass parameter @xmath46 for a fixed value of the sm neutrino mass ( @xmath113 ev ) in figure [ fig : nubb_vs_c ] . for completeness we include the sub - leading terms in the mixing angles for the plot , though the result is well approximated by the above expression . the experimental bound is included in the plot for comparison and we take @xmath114 gev , @xmath115 tev and @xmath116 . observe that the effective @xmath99-decay parameter is below the experimental bound for most of the parameter space : all but the largest values of @xmath46 are viable , with some dependence on @xmath69 . it is instructive to comment on the @xmath69 dependence in the above expressions ( and the figure ) . for small values of @xmath69 the amplitude for @xmath99-decay is dominated by the zero - mode contribution . this is because the kk neutrinos @xmath88 form a pseudo - dirac pair for small @xmath69 and the @xmath99-decay contributions from the two modes tend to cancel each other out . the zero mode has no pseudo - dirac partner and for small @xmath69 its mass is the only one that `` badly '' breaks lepton number symmetry ; thus its contribution to the lepton - number violating @xmath99-decay rate is dominant . for small values of @xmath69 the overall @xmath99-decay rate diminishes with increasing @xmath69 due to the increase in the zero - mode mass . this behaviour is observed in the small @xmath46 regions in figure [ fig : nubb_vs_c ] . on the other hand , for larger values of @xmath69 the contribution from the kk modes is seen to dominate . the mass splitting between the pseudo - dirac kk pairs increases for increasing @xmath69 , @xmath117 . the kk modes @xmath88 thus become less pseudo - dirac - like and form a standard pair of majorana particles . accordingly , lepton - number symmetry is more severely broken in the higher kk sector and the @xmath99-decay signal increases . the cusps in figure [ fig : nubb_vs_c ] occur at the transition between zero - mode dominance and kk mode dominance ; one observes that , once the kk contribution is dominant , the overall rate increases with @xmath69 for a given value of @xmath46 . for perturbative values of @xmath69 , only the low - lying modes ( @xmath118 ) are expected to have sizeable mass splittings and thus contribute significantly to @xmath99-decay . the relative splitting goes like @xmath119 , @xmath120 , and the decreased splitting suppresses the contribution of the higher modes . we have considered the kk neutrino contribution to @xmath99-decay in this section . there is also a contribution from the sm neutrinos which , in general , will interfere with the zero - mode contribution ( these two form a standard seesaw pair and thus possess opposite cp parities ) . however , the contributions from the @xmath44 modes and the sm neutrinos can add constructively , so the total rate can be larger if @xmath46 is such that the kk contribution dominates the zero - mode . given the potential experimental reach for an inverted mass hierarchy @xcite , the additional contribution from the kk neutrinos can improve the prospects for @xmath99-decay observation . let us emphasize , however , that the sm contribution will typically dominate the kk contribution in the region of parameter space where the largest kk contribution comes from the zero - mode . it has been argued that , generically , in seesaw models the contribution from the sm neutrinos is larger than the sterile neutrino contribution @xcite . this holds in the present case for the @xmath99-decay contributions from the zero - mode and the sm neutrino , which essentially form a standard seesaw pair . indeed , for @xmath121 ev one obtains a contribution of @xmath122 to the quantity plotted in figure [ fig : nubb_vs_c ] , which always dominates the zero - mode contribution , but can be exceeded by the sum over higher kk modes for larger @xmath46 . we only plot the kk contribution in figure [ fig : nubb_vs_c ] to emphasize the behaviour of this new contribution . mixing between the sm neutrinos and sterile neutrinos induces an effective coupling between the @xmath1 boson and the sterile neutrinos ; if the sterile neutrinos are light enough , additional invisible @xmath1-decay channels exist @xcite . when a heavy sterile neutrino with mass @xmath123 mixes with the sm neutrinos , @xmath124 with @xmath125 , one has ( zn)=(z2 ) |_s|^2 ^2 . this result generalizes in an obvious way when a tower of kk neutrinos mixes with the sm : ( zn_n)=(z2)|_n|^2 ^2.[z_decay_width_one_kk ] note that @xmath126 for @xmath127 and one need only consider single production . the total non - standard contribution to the invisible width is thus ( z+)=9_n(zn_n ) , where three families of active and sterile neutrinos are included , and we again ignore the mixing of the three generations . neglecting the final state kk neutrino mass in ( [ z_decay_width_one_kk ] ) , this is approximately ( z+)9 ( z2 ) . the invisible @xmath1-width has been precisely measured as @xmath128 mev @xcite . this should be compared to the sm prediction of @xmath129 mev for @xmath1 decays to neutrinos . new contributions are severely constrained and should essentially lie within the experimental uncertainty , the @xmath130 value for which is of order mev . we plot the invisible @xmath1-width as a function of the bulk mass parameter @xmath46 in figure [ fig : invis_z_width_vs_c ] . the sm neutrino mass is fixed at @xmath113 ev and we take @xmath114 gev , @xmath115 tev and @xmath116 , while varying @xmath69 . one observes that the invisible width is well below the @xmath131 mev experimental bound for the entire region of parameter space . analysis of the bounds on a single sterile - neutrino with @xmath132 mass can be found in @xcite and , more recently , in @xcite . the lightest ( zero mode ) neutrinos in the present model must satisfy these standard bounds . the most robust constraints on an order @xmath133 mev sterile neutrino come from peak searches in the leptonic decays of pions and kaons @xcite . these constraints assume only that the heavy neutrino mixes with the sm and are therefore insensitive to the decay properties of the neutrinos . this differs from most collider bounds , which typically make assumptions about the neutrino decay properties to define the signal . for @xmath34 gev only the zero mode neutrinos can appear in the pion / kaon decays and one may use the data to constrain @xmath134 , which may itself be cast as : @xmath135 obtained from kaon decays and @xmath136 from pion decays ( see figure 3 in ref . @xcite ) are not severe . ref . @xcite has detailed a number of bounds for sterile neutrinos that mix with sm neutrinos . their focus was primarily on neutrinos with mass @xmath137 but some of their bounds are reliable for lighter masses . noting that mixing with sterile neutrinos in general introduces a correction to the muon decay constant they derive bounds from lepton universality , ckm unitarity and the invisible @xmath1-decay width . for @xmath137 the latter is affected by the ( slightly ) modified coupling of sm neutrinos to the @xmath1 ; in this work we have kk neutrinos satisfying @xmath138 and the invisible @xmath1 width is modified directly by the additional decay channels @xmath139 as detailed in section [ sec : hid_z_decay ] . the bounds of ref . @xcite can be summarized as _ e^2&<&7.110 ^ -3m_n>0.14 ( m _ ) , + _ ^2&<&1.410 ^ -3m_n>1.15 ( m _ ) , + _ ^2&<&1.710 ^ -2m_n>1.777 ( m_).[sterile_nardi_bounds ] here @xmath140 is an effective mixing angle between the sm neutrino with lepton flavour @xmath141 and sterile neutrinos with masses exceeding the quoted values . in the present context the effective angle is defined as @xmath142 , with the factor of three accounting for three generations of bulk neutrinos . for simplicity we do not differentiate the different generations in this work and assume that all mixing angles are of the same order . in this case a conservative bound of _ ^2=3\{_0 ^ 2+_n>0(_-n^2+_+n^2)}&<&7.110 ^ -3m_0 > m _ , is obtained . we plot the effective active - sterile mixing - angle squared as a function of the bulk mass parameter ( @xmath46 ) in figure [ fig : mixing_vs_c ] , along with the experimental bound . the figure shows that values of @xmath143 are consistent with the bounds for @xmath144 . light sterile neutrinos can also be created in collider experiments . kinematics permitting , sterile neutrinos can be singly produced on the @xmath1 peak and can decay via an off - shell gauge boson , @xmath145 with @xmath146 , or @xmath147 with @xmath148 . these decay signals have been sought at lep for gauge - singlet neutrinos in the mass range @xmath149 @xcite . the resulting bound is essentially constant at @xmath150 for @xmath151 in the range @xmath152 and decreases slowly ( rapidly ) for lighter ( heavier ) @xmath151 ; see figures 6 and 7 in ref . @xcite . in the present framework , each member of the kk tower can decay to sm fields via off - shell gauge bosons can also decay directly to on - shell vectors . ] and it appears that the lep bounds should be applied to all of the kk neutrinos . the lep bounds certainly apply to the zero mode neutrinos , which decay through their inherited sm couplings and behave essentially like a standard , light , sterile neutrino . one thus requires @xmath153 for @xmath154 $ ] , with the bound dropping to @xmath155 as @xmath156 decreases to @xmath157 mev . given that @xmath158 this bound is not restrictive . the higher modes ( @xmath44 ) can also decay to the sm . however , one should not apply the bounds of ref . @xcite directly to these modes . these bounds are obtained under the assumption that the sterile neutrinos _ only _ decay to standard model final states . this assumption does not hold for the @xmath44 kk modes , which can decay within the hidden sector via kk graviton production , @xmath159 , where @xmath160 is the @xmath161th kk graviton . the kk neutrinos and gravitons are both localized towards the ir brane and in general have large wavefunction overlaps . despite the vertex being gravitational in nature , the relevant coupling strength is of order @xmath162 and the hidden sector decays can not be neglected . graviton decays are allowed for @xmath163 and the partial widths go like @xmath164 . the modes are increasingly broad as one goes up the kk tower and , for a given value of @xmath79 , decays with @xmath165 are preferred . the total two - body hidden width for the @xmath79th mode is thus of order @xmath166 . decay of the @xmath79th mode to the sm is suppressed by a factor of @xmath167 so that , absent hierarchically small values of @xmath168 , decays within the hidden sector will dominate . once a given kk neutrino decays into the hidden sector , the daughter neutrino(s ) and graviton(s ) will further decay within the hidden sector ( when kinematically allowed ) . the graviton can decay via @xmath169 with width @xmath170 ( similar to rs @xcite ) and shows a preference for decays with @xmath171 . thus a hidden sector `` shower '' occurs , consisting of a series of cascade decays down the hidden kk towers until one has a collection of light kk modes with mass of order @xmath22 , which in turn decay back to the sm . the expected signal from single production of a higher kk neutrino is therefore very different to the signal searched for at lep @xcite . rather than applying the constraint @xmath172 directly to @xmath173 one should instead apply it to the quantity @xmath174 and @xmath175 for the parameters of interest here ( @xmath31 tev , @xmath34 gev and @xmath47 ) , the lep bound does not impose an additional constraint . bounds resulting from heavy neutrino searches at beam dump experiments by , e.g. , the wa66 collaboration @xcite , are similarly unrestrictive in the present context . indirect constraints on the parameter space for sterile neutrinos can be derived from searches for flavour changing neutral current processes like @xmath177 , @xmath0 and @xmath178 conversion in nuclei @xcite . for sterile neutrinos mixing with the sm , the induced @xmath0 branching fraction can be written as @xcite : br(e+ ) = |_n _ en ^*_n ( m_n^2/m_w^2)|^2 , where the function @xmath179 is ( x)=x[1 - 6x+3x^2 + 2x^3 - 6x^2(x)]/2(1-x)^4 , with @xmath180 $ ] for @xmath181 . present bounds require @xmath182 at the @xmath183 c.l . @xcite . we plot the @xmath0 branching ratio as a function of the bulk mass parameter @xmath46 in figure [ fig : mu_to_e_gamma_vs_c ] for three generations of bulk neutrinos with the sum again cut off at @xmath184 . the sm neutrino masses are fixed at @xmath113 ev and we use @xmath115 tev , @xmath114 gev and @xmath116 for the plot . the figure shows that @xmath185 constraints are more stringent than the constraints previously considered ; values of @xmath186 are already excluded for @xmath144 . recall that @xmath187 $ ] when @xmath188 $ ] for @xmath189 and @xmath190 ev . thus the interesting region of parameter space in which the yukawa coupling @xmath191 is @xmath17 corresponds to values of @xmath46 that are close to the upper bound . other flavour changing processes like @xmath192 conversion in nuclei and @xmath177 give similar constraints . the most accurate experimental result in this regard comes from searches for @xmath192 conversion in ti . the branching ratio with respect to the total nuclear muon capture rate is bound as @xmath193 @xcite . however , this bound turns out to be less severe than the @xmath185 constraint . for a single sterile neutrino , the resulting constraint is @xmath194 , which is weaker due to the @xmath195 dependence unless @xmath196 tev . we only consider @xmath197 tev in this work and thus find the @xmath185 bound more useful . in summary , we find that @xmath0 constraints on the present model are the most severe and that these typically require @xmath198 for @xmath199 . of interest is the fact that larger values of @xmath46 correspond to larger values of the yukawa coupling @xmath191 for a given value of @xmath76 ( see figure [ fig : yuk_vs_c_normal ] ) . the more interesting regions of parameter space thus correspond to values of @xmath200 just below current limits . in particular , yukawa couplings of @xmath201 correspond to @xmath202 . the meg collaboration is expected to report new results on the branching fraction for @xmath0 soon @xcite . given their targeted precision of measuring @xmath203 at the @xmath204 level , meg will have the potential to significantly improve the constraints on the present model . in existing releases the meg collaboration has reported that the best value for the number of signal events in their likelihood analysis is nonzero ( specifically , three events ) @xcite . interpreted as a signal , this corresponds to @xmath205 , which is right in the interesting range in the present context ( see figure [ fig : mu_to_e_gamma_vs_c ] ) . we eagerly await the release of more data by the meg collaboration to see if this potential signal holds up . the kk tower of sterile neutrinos has the potential to modify the higgs decay widths due to additional channels involving the sterile states . the higgs couples to the steriles via the uv localized yukawa interaction s&&- d^5x |l h n_r ( z - k^-1 ) , where @xmath206 is a lepton doublet and @xmath207 is the sm scalar doublet . in a general 4d theory , the sm higgs will decay into neutrino pairs @xmath208 whenever the sm neutrinos yukawa - couple to one or more light sterile neutrinos . however , the demand @xmath209 typically forces the yukawa couplings to be tiny as one expects @xmath210 for seesaw neutrino masses ; @xmath211 gev gives @xmath212 for @xmath213 ev and the corresponding modification to the higgs width is therefore negligible . in the present context , the yukawa couplings need not be tiny as @xmath76 is suppressed by warping ( or in the dual 4d theory , fundamental / composite mixing ) in addition to the usual seesaw suppression . the action can be written as s&&- d^5x |l h n_r ( z - k^-1 ) + & = & -_n d^4x h|_l ^(n)_r + + & = & - d^4x h|_l\{_0^n_0r + _ n>0_n^(n_+n l^c + n_-n l^c ) } + where we define @xmath214 , use the fact that @xmath215 , and write the effective yukawa coupling for the @xmath79th kk mode as _ n^ 2()^1/2 ( ) ^1/2()^c . if one adds a sterile neutrino @xmath216 with yukawa coupling @xmath217 to the sm , the decay width for the higgs to two neutrinos is : _ n(h n)&= & ( 1-)^2 where we neglect @xmath218 . neglecting the phase space factors , we can apply this result to the present model : & & ( h)9_n(h n_n)\{(_0^)^2+_n>02 ( _ n^)^2 } , where the sum is cut off in the uv at @xmath219 . this invisible width is plotted as a function of the bulk mass parameter @xmath46 in figure [ fig : higgs_width_vs_c ] . observe that it is significantly larger than that obtained in a standard low - scale seesaw with order gev sterile neutrinos ( e.g. , the @xmath18msm ) , for which one has @xmath220 gev for @xmath221 gev . noting that the width for sm higgs decays is @xmath222 $ ] gev for @xmath223 gev @xmath224 gev ] , the invisible branching fraction is @xmath225 $ ] for the values shown in the figure . for a lighter higgs mass with larger values of @xmath46 the invisible branching fraction is on the order of the two - photon width , @xmath226 . despite its small width , the two - photon channel is known to be useful if the higgs is light , due to the relatively clean signal . however , invisible decays are not as readily discerned from the data . the lhc can `` observe '' invisible higgs decays , but only for much larger branching fractions of roughly @xmath227 ( @xmath228 ) for @xmath229 ( @xmath230 ) of data @xcite . we conclude that , despite a significant increase relative to other models with a low - scale seesaw , the width for higgs decays to neutrinos remains too small to be observed at the lhc . the ilc would have a better discovery ( and limits ) reach for invisible higgs decays , possibly achieving discovery ( limits ) for invisible branching fractions of order @xmath231 ( @xmath232 ) @xcite . we have assumed `` natural values '' for the mixing angles in this analysis ; e.g. , the natural value for the mixing angle between a standard seesaw pair @xmath233 and @xmath103 is @xmath234 . however , the mixing between the sm neutrinos and the heavy neutrinos need not be as small as this natural value if there are cancellations amongst the parameters in the seesaw formula @xcite . an enhanced mixing angle means the dirac yukawa - coupling can be larger , leading to an enhancement of the hidden higgs decay width @xcite . such cancellations could be considered in the present context and the hidden higgs width can be similarly enhanced . we also note that the total invisible width can potentially be higher than the neutrino width . stabilization of the length of the extra dimension via the goldberger - wise mechanism , for example , could open additional invisible channels . the higgs can couple to , and possibly mix with , the goldberger - wise scalar and thus the overall invisible width could increase . this is beyond the scope of the present work but we hope to consider it in the future . the mini - seesaw mechanism achieves light sm neutrino masses by combining naturally suppressed dirac and ( sterile ) majorana mass - scales together in a low - scale seesaw mechanism . it employs a truncated ( `` little '' ) warped space that , via ads / cft , is dual to a 4d theory possessing conformal symmetry in some window , @xmath235 , with @xmath236 . the model generates light sm neutrino masses without recourse to a large uv scale , and thus offers hope that the mechanism of neutrino mass generation may be explored in feasible experiments . a key feature of this approach is the existence of a tower of light sterile - neutrinos that mix with the sm . we have considered the detailed bounds on these light neutrinos from processes like @xmath99-decay , @xmath0 and invisible @xmath1-decays . we find that the most stringent constraints come from fcnc processes like @xmath0 . nonetheless we have shown that viable parameter space exists in which the input yukawa couplings can be sizable . the model therefore provides a viable explanation for the lightness of the known sm neutrinos . furthermore , the @xmath185 branching fraction lies just below current experimental bounds for interesting regions of parameter space and , importantly , is within the region to be probed by the forthcoming meg experiment . we look forward to the meg results in the hope that ( indirect ) evidence for the origin of neutrino mass is revealed . the authors thank w. rodejohann and a. watanabe for helpful discussions . md was supported by the imprs - ptfs . dg was supported by the netherlands foundation for fundamental research of matter ( fom ) and the netherlands organisation for scientific research ( nwo ) . to @xmath77 the matrix that gives the heavy eigenstates is : u_kk= ( j. w. f. valle , arxiv:1001.5189 [ hep - ph ] ; a. strumia and f. vissani , arxiv : hep - ph/0606054 ; r. n. mohapatra and a. y. smirnov , ann . rev . nucl . sci . * 56 * , 569 ( 2006 ) [ arxiv : hep - ph/0603118 ] ; r. n. mohapatra _ et al . _ , rept . phys . * 70 * , 1757 ( 2007 ) [ arxiv : hep - ph/0510213 ] ; g. altarelli and f. feruglio , new j. phys . * 6 * , 106 ( 2004 ) [ arxiv : hep - ph/0405048 ] ; s. f. king , rept . phys . * 67 * , 107 ( 2004 ) [ arxiv : hep - ph/0310204 ] . t. schwetz , m. tortola , j. w. f. valle , arxiv:1103.0734 [ hep - ph ] . t. yanagida , in proceedings of the _ `` workshop on the unified theory and the baryon number in the universe '' _ , tsukuba , japan , feb . 13 - 14 , 1979 , edited by o. sawada and a. sugamoto , kek report kek-79 - 18 , p. 95 , and _ horizontal symmetry and masses of neutrinos _ , prog . . phys . * 64 * ( 1980 ) 1103 ; m. gell - mann , p. ramond and r. slansky , in _ supergravity _ ( north - holland , amsterdam , 1979 ) _ eds_. d. z. freedom and p. van nieuwenhuizen , print-80 - 0576 ( cern ) ; p. minkowski , phys . b * 67 * , 421 ( 1977 ) ; s. l. glashow , _ the future of elementary particle physics _ , in _ proceedings of the 1979 cargse summer institute on quarks and leptons _ ( m. lvy et al . eds . ) , plenum press , new york , 1980 , p. 687 ; r. n. mohapatra and g. senjanovic , phys . lett . * 44 * , 912 ( 1980 ) . a. zee , nucl . * b264 * , 99 ( 1986 ) ; k. s. babu , phys . * b203 * , 132 ( 1988 ) . k. a. meissner , h. nicolai , phys . * b648 * , 312 - 317 ( 2007 ) [ arxiv : hep - th/0612165 ] ; r. foot at al , phys . rev . * d76 * , 075014 ( 2007 ) [ arxiv:0706.1829 [ hep - ph ] ] ; r. foot et al , phys . * d77 * , 035006 ( 2008 ) [ arxiv:0709.2750 [ hep - ph ] ] ; s. iso , n. okada , y. orikasa , phys . * b676 * , 81 - 87 ( 2009 ) [ arxiv:0902.4050 [ hep - ph ] ] . n. arkani - hamed , s. dimopoulos , g. r. dvali , j. march - russell , phys . rev . * d65 * , 024032 ( 2002 ) [ arxiv : hep - ph/9811448 ] ; k. r. dienes , e. dudas , t. gherghetta , nucl . phys . * b557 * , 25 ( 1999 ) [ arxiv : hep - ph/9811428 ] . n. arkani - hamed and y. grossman , phys . lett . b * 459 * , 179 ( 1999 ) [ arxiv : hep - ph/9806223 ] ; t. okui , jhep * 0509 * , 017 ( 2005 ) [ arxiv : hep - ph/0405083 ] ; y. grossman and y. tsai , jhep * 0812 * , 016 ( 2008 ) [ arxiv:0811.0871 [ hep - ph ] ] ; y. grossman , d. j. robinson , jhep * 1101 * , 132 ( 2011 ) [ arxiv:1009.2781 [ hep - ph ] ] ; r. s. hundi , s. roy , arxiv:1105.0291 [ hep - ph ] . j. m. maldacena , adv . * 2 * , 231 ( 1998 ) [ int . j. theor . * 38 * , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] ; s. s. gubser , i. r. klebanov and a. m. polyakov , phys . b * 428 * , 105 ( 1998 ) [ arxiv : hep - th/9802109 ] ; e. witten , adv . theor . math . * 2 * , 253 ( 1998 ) [ arxiv : hep - th/9802150 ] . n. arkani - hamed , m. porrati and l. randall , jhep * 0108 * ( 2001 ) 017 [ arxiv : hep - th/0012148 ] ; r. rattazzi and a. zaffaroni , jhep * 0104 * , 021 ( 2001 ) [ arxiv : hep - th/0012248 ] . r. sawada , http://pos.sissa.it/archive/conferences/120/263/ichep%202010_263.pdf ; g. cavoto , arxiv:1012.2110 [ hep - ex ] . l. randall and r. sundrum , phys . lett . * 83 * , 3370 ( 1999 ) [ arxiv : hep - ph/9905221 ] . y. grossman and m. neubert , phys . lett . b * 474 * , 361 ( 2000 ) [ arxiv : hep - ph/9912408 ] . t. gherghetta and a. pomarol , nucl . b * 586 * , 141 ( 2000 ) [ arxiv : hep - ph/0003129 ] ; s. j. huber and q. shafi , phys . b * 498 * , 256 ( 2001 ) [ arxiv : hep - ph/0010195 ] . s. j. huber and q. shafi , phys . b * 544 * , 295 ( 2002 ) [ arxiv : hep - ph/0205327 ] ; s. j. huber and q. shafi , phys . b * 583 * , 293 ( 2004 ) [ arxiv : hep - ph/0309252 ] . t. gherghetta , phys . * 92 * , 161601 ( 2004 ) [ arxiv : hep - ph/0312392 ] . b. gripaios , nucl . b * 768 * , 157 ( 2007 ) [ erratum - ibid . * 830 * , 390 ( 2010 ) ] [ arxiv : hep - ph/0611218 ] ; t. gherghetta , k. kadota and m. yamaguchi , phys . d * 76 * , 023516 ( 2007 ) [ arxiv:0705.1749 [ hep - ph ] ] ; g. perez and l. randall , jhep * 0901 * ( 2009 ) 077 [ arxiv:0805.4652 [ hep - ph ] ] ; c. csaki , c. delaunay , c. grojean and y. grossman , jhep * 0810 * ( 2008 ) 055 [ arxiv:0806.0356 [ hep - ph ] ] ; f. del aguila , a. carmona and j. santiago , jhep * 1008 * ( 2010 ) 127 [ arxiv:1001.5151 [ hep - ph ] ] ; a. kadosh and e. pallante , jhep * 1008 * , 115 ( 2010 ) [ arxiv:1004.0321 [ hep - ph ] ] ; a. watanabe and k. yoshioka , arxiv:1007.1527 [ hep - ph ] . g. von gersdorff and m. quiros , phys . b * 678 * , 317 ( 2009 ) [ arxiv:0901.0006 [ hep - ph ] ] . t. flacke and d. maybury , jhep * 0703 * ( 2007 ) 007 [ arxiv : hep - ph/0612126 ] ; k. l. mcdonald and d. e. morrissey , jhep * 1005 * , 056 ( 2010 ) [ arxiv:1002.3361 [ hep - ph ] ] . k. l. mcdonald , d. e. morrissey , jhep * 1102 * , 087 ( 2011 ) [ arxiv:1010.5999 [ hep - ph ] ] . t. asaka , s. blanchet and m. shaposhnikov , phys . b * 631 * ( 2005 ) 151 [ arxiv : hep - ph/0503065 ] ; t. asaka and m. shaposhnikov , phys . b * 620 * , 17 ( 2005 ) [ arxiv : hep - ph/0505013 ] ; d. gorbunov , m. shaposhnikov , jhep * 0710 * , 015 ( 2007 ) [ arxiv:0705.1729 [ hep - ph ] ] a. ibarra , c. simonetto , jhep * 0804 * , 102 ( 2008 ) [ arxiv:0802.3858 [ hep - ph ] ] ; a. abada , c. biggio , f. bonnet , m. b. gavela , t. hambye , phys . rev . * d78 * , 033007 ( 2008 ) [ arxiv:0803.0481 [ hep - ph ] ] ; d. suematsu , t. toma , t. yoshida , phys . rev . * d79 * , 093004 ( 2009 ) [ arxiv:0903.0287 [ hep - ph ] ] ; j. hisano , m. nagai , p. paradisi , y. shimizu , jhep * 0912 * , 030 ( 2009 ) [ arxiv:0904.2080 [ hep - ph ] ] ; s. blanchet , t. hambye , f. -x . josse - michaux , jhep * 1004 * , 023 ( 2010 ) [ arxiv:0912.3153 [ hep - ph ] ] ; m. davidkov , d. i. kazakov , arxiv:1102.1582 [ hep - ph ] . w. d. goldberger and m. b. wise , phys . lett . * 83 * , 4922 ( 1999 ) [ arxiv : hep - ph/9907447 ] ; j. garriga and a. pomarol , phys . b * 560 * , 91 ( 2003 ) [ arxiv : hep - th/0212227 ] . h. davoudiasl , g. perez and a. soni , phys . b * 665 * , 67 ( 2008 ) [ arxiv:0802.0203 [ hep - ph ] ] ; h. davoudiasl , t. mcelmurry , a. soni , phys . rev . * d82 * , 115028 ( 2010 ) [ arxiv:1009.0764 [ hep - ph ] ] . d. i. britton _ et al . _ , * 68 * ( 1992 ) 3000 ; d. i. britton _ _ , phys . d * 46 * ( 1992 ) 885 . d. berghofer _ et al . inern . conf . on neutrino physics and astrophysics ( maui , hawaii , 1981 ) , 67 ( 1981 ) , eds . r. j. cence , e. ma and a. roberts , vol . ii ( university of hawaii , honolulu , hi , 1981 ) ; t. yamazaki _ et al . 22nd intern . conf . on high energy physics ( leipzig , 1984 ) , 262 ( 1984 ) , eds . a. meyer and e. wierczorek , vol . i ( akademie der wiessenachaften der ddr , leipzig , 1984 ) . e. nardi , e. roulet , d. tommasini , phys . . * b327 * , 319 - 326 ( 1994 ) [ arxiv : hep - ph/9402224 ] .

the 3c 273 and rx j1230.8 + 0115 sight lines probe the outskirts of the virgo cluster at physical separations between the sight lines of 200 - 50070 kpc . we present an analysis of available hst stis echelle and fuse uv spectroscopy of rx j1230.8 + 0115 , in which we detect five absorbers at virgo distances . one of these absorbers is a blend of two strong metal line absorbers at a recession velocity coincident with the highest neutral hydrogen column density absorber in the 3c 273 sight line , 350 70 kpc away . the consistency of the metal line column density ratios in the rx j1230.8 + 0115 sight line allows us to determine the ionization mechanism ( photoionization ) for these absorbers . while the low signal - to - noise ratio of the fuse spectrum limits our ability to model the neutral hydrogen column density of these absorbers precisely , we are able to constrain them to be in the range 10 2 . the properties of these absorbers are similar to those obtained for the nearby 3c 273 absorber studied by tripp and collaborators . however , the inferred line - of - sight size for the 3c 273 absorber is only 70 pc , much smaller than those inferred in rx j1230.8 + 0115 , which are 10 30 70 of kpc . the small sizes of all three absorbers are at odds with the > 70 kpc minimum transverse size implied by an application of the standard qso line pairs analysis . on the basis of absorber associations between these two sight lines we conclude that a large - scale structure filament produces a correlated , not contiguous , gaseous structure in this region of the virgo supercluster . these data may indicate that we are detecting overdensities in the large scale structure filaments in this region . alternatively , the presence of a galaxy 7170 kpc from a 3c 273 absorber may indicate that we have probed outflowing , starburst driven shells of gas associated with nearby galaxies .

astro-ph0303308

the structure and origin of `` forest '' absorbers remains uncertain despite decades of study with ground - based telescopes and more recently with the hubble space telescope ( hst ) . it has been argued that forest absorbers are related to the extended halos of bright galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ) , that they are associated with dwarf or low surface brightness galaxies ( e.g. , * ? ? ? * ) , or that they are one phase of the gaseous medium in spiral - rich galaxy groups @xcite . alternatively , it has been proposed that they are `` primordial '' overdensities associated with large scale structure , but not with any galaxy in particular @xcite . it is possible that all of these scenarios are represented . therefore , knowing the fraction of the forest associated with a given scenario , as a function of column density and redshift , is essential if we are going to use studies of the forest to understand the formation and early evolution of galaxies . the size and shape of these absorbing structures bears directly on the fraction of absorbers associated with a specific environment . the correlation between absorbers in quasar pairs is often used to determine statistically the sizes of absorbers , particularly at high redshift ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? these qso sight line pair experiments yield characteristic radii for higher column density ( @xmath3 14.5 2 ) absorbers of 300 70070 kpc , with some evidence for increasing sizes at lower redshifts @xcite . these very large size measurements are consistent with the expectations of cosmological n - body simulations @xcite as well as the expectations from an analytical jeans mass theory of gravitationally - bound , photoionized clouds @xcite . from a census of forest absorbers using hst / ghrs and stis observations , @xcite find that @xmath040% of all local baryons reside in the forest . this estimate , however , assumes a characteristic radii for spherical absorbing structures of 10070 kpc . it is uncertain whether this size , derived from high redshift absorber pairs , is an appropriate measurement of the absorber s contiguous size or merely the length scale over which the absorbers are associated with the same large scale structure ( dinshaw et al . 1998 ) . alternatively , the physical line - of - sight dimension of absorbers can be derived by assuming photoionization equilibrium , from measurements of h i column density , metal line strengths and the ratio of line strengths for different ionization states of the same species . the line strengths and ratios can be used to constrain metallicities and physical densities in these structures assuming photoionization equilibrium with the extragalactic ionizing flux @xcite . the calculation of hydrogen densities from the photoionization models can then be used to infer physical sizes along the line - of - sight . the sizes determined for absorbers with strong , low ionization metal lines like c ii , si ii , mg ii and fe ii @xcite , are much smaller than the characteristic sizes in the plane of the sky obtained from the qso pairs experiments . for example , the low ionization absorbers with relatively weak mg ii absorption ( @xmath40.3 ) have sizes derived from a photoionization equilibrium calculation which are @xmath4 1 kpc @xcite . in this paper we present the case of a pair of qsos , 3c 273 and rx j1230.8 + 0115 , which probe the outskirts of the virgo cluster at physical separations between the sight lines of 200 50070 kpc . both sight lines contain strong ( @xmath0 16 2 ) absorption systems which have relatively low ionization absorption line spectra . this low redshift pair provides us the opportunity to use both of the aforementioned size determination methods at redshifts where we can obtain a deep galaxy census . tripp et al . ( 2002 ; t2002 hereafter ) presented an analysis of the absorption along the line - of - sight to 3c 273 , 0.9 away from rx j1230.8 + 0115 on the sky . t2002 studied the two strong absorbers at virgo distances ( @xmath5=1011 and 1586 ) in a stis echelle ( 7 resolution ) spectrum of 3c 273 obtained by @xcite . these two absorbers had been detected and studied using the faint object spectrograph @xcite and the ghrs @xcite . more recently 3c 273 was observed with the far - ultraviolet spectroscopic explorer ( fuse ; * ? ? ? * ) allowing detections of the higher lyman lines in both of these virgo absorbers and o vi absorption in the 1011 absorber . the 1586 absorber contains only low ionization absorption lines of c ii , si ii and si iii . using a curve - of - growth analysis and a standard photoionization model t2002 constrained the column density (= 15.85 2 ) , @xmath6-value = 16.1 , metallicity ( [ c / h]=-1.2 and [ si / c]=+0.2 in solar units ) and the density ( log @xmath7= -2.8 @xmath8 ) ; these values determine the line - of - sight size of this absorber to be 70 pc . we discuss the hydrogen and metal absorption lines at virgo distances ( @xmath5=800 - 2400 ) that we have identified in the available stis echelle and fuse spectra of rx j1230.8 + 0115 ( 2 ) . in 3 we discuss in detail our inferences concerning the n@xmath9 , metallicity and ionization parameter for the @xmath5=1699 absorption complex based upon the hst and fuse data , and the assumption of photoionization equilibrium . in 4 we discuss the galaxy distribution around these absorbers ; in 5 we discuss the inferences based upon the coincident absorption line pairs in these two spectra . 6 contains both a summary of our observational results and a discussion of the model that best fits all of the information currently available on the low ionization , coincident velocity absorbers in these two sight lines . throughout this paper we use @xmath10=70 mpc@xmath11 and determine distances assuming a pure hubble flow ; the use of the tonry et al . ( 2000 ) virgocentric infall model reduces distances in this vicinity by @xmath010% for galaxies with @xmath12 . we can not rule out that redshifts yield a double - valued distance solution at this local supercluster location ( tonry et al . we describe the hst and fuse spectroscopy of rx j1230.8 + 0115 at recession velocities which correspond to virgo cluster distances , with primary emphasis on a very strong , two - component absorber at @xmath01700 . while the emission line redshift of this agn ( @xmath13=0.117 ) allows detection of absorbers at higher redshifts than virgo , they will not be discussed here . first we describe the and metal - line absorptions seen in an echelle mode spectrum obtained with stis , then show the higher - order lyman series lines for the strongest two absorbers at virgo distances obtained with fuse . an analysis of a ghrs first order grating spectrum of rx j1230.8 + 0115 was presented by penton , stocke & shull ( 2000a ) , in which 5 of the 6 absorbers tabulated here were detected . observations of rx j1230.8 + 0115 were obtained with the e140 m echelle mode of stis using the 0.2@xmath140.06 aperture on january 1 and january 13 , 1999 ( m. rauch , pi ) . the resolution of the spectrum is 7 over the spectral range of 1150 to 1710 . the data reduction of the spectrum was done using the standard stis extraction procedures . this reduction package returns heliocentric velocities which are estimated to be accurate to @xmath15 4 ( stis data handbook ) . the rx j1230.8 + 0115 spectrum has considerably lower signal - to - noise ( snr ) than the 3c 273 spectrum of @xcite because it is @xmath0 10 times fainter . nevertheless , the virgo distance absorbers are strong enough to be easily detected ( see also penton , stocke , & shull 2000a ) . figure [ fig : lya ] shows the region of absorption for virgo velocities ( @xmath16 2500 ) in the rx j1230.8 + 0115 sight line . we have binned these data by 3 pixels , giving a pixel width of 9.7 and removed a continuum that is the best by - eye fit to the damped profile . we get a best - fit value for galactic n@xmath17 2 scaled to a continuum . this fit is shown in figure [ fig : lya ] ( solid line ) . we also show a spline fit ( dashed line ) that was used to estimate line significance between 1218 and 1221 because the galactic damped fit appears to underestimate the true continuum in this region indicating that the agn continuum must be rising blueward of 1221 . also shown are the errors in the flux measurements . the absorption lines were fit with voigt profiles using vpfit @xcite . we identify 5 lines ( one of which has 2 components ) at 4-@xmath18 or greater significance , and one line at greater than 3.5-@xmath18 significance . because there are no galactic metal lines and no metal lines related to the associated absorbers , we identify all of the absorptions as intergalactic . table 1 contains the following information about these absorbers : ( 1 ) central wavelength of the voigt profile fit in , ( 2 ) central velocity of the fit ( heliocentric , non - relativistic , in ) ( 3 ) the redshift and ( 4 ) its associated error , ( 5 ) the @xmath6-value and ( 6 ) its associated error in , ( 7 ) the log of the h i column density and ( 8) its associated error in 2 these errors include a continuum placement error derived by adding and subtracting one - third of the rms from the continuum , ( 9 ) the rest frame equivalent width ( ew ) of the data ( not of the voigt profile fit ) measured in m , ( 10 ) the 4-@xmath18 ew limit integrated over the same velocity width for which the ew in column 9 was calculated ( typically 3 4 times the @xmath6-value in column 5 ) . all of these values , except for the lines at 1685 and 1721 , are single component fits determined from the absorptions alone . for these two components ( 1685 and 1721 ) we have fixed the redshifts ( based upon the observed metal line redshifts discussed in the next section ) and the column densities ( at = 16.2 and 16.6 2 as discussed in 2.4 ) , which is why there are no quoted errors . = 0.9 the 4-@xmath18 ew limits given in table 1 indicate that all of these absorption - line detections are stronger than 4-@xmath18 except for the line at 1219.27 which is 3.5-@xmath18 . our strongest absorbers , the @xmath19 ( 1699 ) blend and the @xmath20 ( 2302 ) absorption , are both highly saturated , which makes it much more difficult to obtain accurate measurements of column densities and @xmath6-values . in fact , for the 1699 blend , we have run an entire grid of h i column density models and found little discrimination based upon the fits alone over the range 10@xmath21 2 ( reduced @xmath22 values of 1.1 - 2.1 over this range ) . because the data are insufficient for this task , we discuss further constraints on the h i column density in the @xmath19 complex in 2.4 . we have examined the rx j1230.8 + 0115 stis echelle spectrum for metal lines associated with the h i absorptions discussed above and detect metal - line absorption in only the @xmath19 complex . two component metal lines of c ii , c iv , si ii , si iii , and si iv are found associated with this complex . we used the fuse data ( see 2.3 ) to search the region of c iii absorption at this redshift , but we do not fit the c iii absorption because the spectrum is noisy and the potential c iii absorptions are blended with galactic h@xmath23 . we also do not fit the si iii absorptions in the stis echelle data because they are far down on the damping - wing of galactic and , therefore in a region of very low s / n . the local continuum around the metal lines was determined with a linear least squares fit , as the regions were determined to be flat . the metal lines were then fit using vpfit as for the lines . figure [ fig:1666lines ] shows these metal lines and the vpfit results are summarized in table 2 . except for column ( 1 ) , which lists the identified ion , the columns have the same descriptors as table 1 . the final column gives 4-@xmath18 limits which are also appropriate for metal line non - detections in the other absorbers listed in table 1 . the statistical errors on the @xmath6-values listed in column ( 6 ) are errors returned on the specific line profile fits in figure 2 . = 0.9 in all cases where metal lines are detected , two components are detected well within the broad , saturated , profile shown at the top of figure [ fig:1666lines ] . because the velocities of the two components are consistent between species and ionization states , we have used the velocity of the strong c ii components to constrain the velocities of the 1685 and 1721 components in the lyman line analysis . the consistency of the column density ratios and the line profile shapes provides an excellent indication that the gas is photoionized and has a simple kinematic structure . a substantial contribution from collisionally - ionized gas is unlikely . the metal lines all have resolved line profiles that provide a measurement of the total @xmath6-value . these line widths are inconsistent with being due to pure thermal broadening since photoionization models predict temperatures below 20,000 k while a thermal broadening interpretation of the line widths implies temperatures of @xmath090,000 k. however , the line widths of the 1720 line ( the signal - to - noise of the 1685 line is too low to provide reliable b - value constraints ) are consistent within 2-@xmath18 with the same amount of bulk or turbulent motion for all of the ionization states . rx j1230.8 + 0115 was only observed in three brief `` snapshots '' totaling 4031 seconds with fuse . nevertheless , we examined the co - added snapshots to better constrain the h i column density of the @xmath19 complex . the spectrum has a snr of 1.1 for the sic1b channel and 1.5 for the sic2a channel . we have binned the data by 8 pixels for a resolution of 16 per pixel . the data were processed using calfuse version 2.1.6 . = 0.95 figure [ fig : fusefits ] shows the portion of the lyman series detected for this absorption complex . lyman-@xmath24 has not been included because it is severely blended with a galactic o vi line and lyman-@xmath25 has also been omitted because it is blended with galactic lyman-@xmath26 . various fits are shown spanning a plausible range in h i column density bounded by acceptable fits to the line which range from = 15 17.6 2 for the lower column density component . for all of these fits we have assumed that the two components , fixed at the metal line velocities , have comparable metallicities and , therefore , roughly consistent relative h i column densities . the lyman - limit does not place a significant constraint on the column density due to confusion with a plethora of galactic lines in the 915 - 920 range . figure [ fig : fusefits ] shows that even with the fuse data , it is very difficult to distinguish between these column densities . it does appear that the column density is larger than = 15.2 and 15.0 2 for the two components , and probably larger than 10@xmath27 2 in both components . an assumption that the metallicity is @xmath28% solar as in the t2002 absorber , implies = 16.2 , 16.6 2(see 3 ) is not a bad one . however , the fuse spectrum is inadequate to constrain the column densities at the high end . the saturation of the @xmath19 complex and the poor quality of the fuse spectrum make it difficult to determine accurate column densities , so we list our best - fits and constraints for these values , in order of most to least secure . * we place a firm lower limit on the h i column density in this line - of - sight by integrating the optical depth pixel by pixel within the line @xcite using the equation : + @xmath29\,{\rm d}v_i , \label{eq : app_col}\ ] ] + where @xmath30 is the observed flux and @xmath31 is the estimated continuum . if the observed flux is less than the rms in the flux , we assign the flux to be equal to the rms for that pixel . @xmath32 = 1215.67 , the rest wavelength for the transition , and f = 0.416 is the oscillator strength . the lower limit to the column density from this method is 3@xmath33 2 . * the consistency of the metal line column density ratios argues for a photoionization model while the lack of variation in the line profiles for different ions and species argues for a single phase for the gas . * while it is difficult to use the fuse data to place accurate constraints on these two absorbers , it seems probable that the ly-@xmath26 , ly-@xmath34 , and ly-@xmath35 absorptions requires @xmath36 16.0 2 . * we have used the redshifts of the c ii absorption lines in this complex to fix the positions of the h i components . for the saturated absorption complex , the best - fit ( @xmath37 ) from vpfit is found when = 15.0 2 for the @xmath38 component and = 15.2 2 for the @xmath39 component . but these column densities are not consistent with the strong , albeit noisy , higher lyman series absorption seen with fuse . further , as we discuss in 3 , if the gas is photoionized such low values of imply metallicities of order solar to account for the c ii and c iv column densities . metallicities greater than 0.3 solar are unlikely , as this corresponds to the metallicity of the intracluster medium at the center of the virgo cluster as derived from the x - ray observations @xcite . by comparison , the metallicity found in the 3c 273 strong absorber at this velocity is 6% solar ( t2002 ) . while we can derive acceptable line fits for all column densities tried up to = 17.8 2 for the @xmath38 component and = 18.0 2 for the @xmath40 component , for column densities exceeding 10@xmath41 2 the @xmath6-values derived to fit the equivalent widths are unrealistically low ( 8 - 14 ) for these absorbers . while such low measured values are possible if these absorbers have no turbulent motion @xcite , the presence of c iv and si iv at comparable strength to c ii and si ii make such low values very unlikely ; e.g. , the 3c 273 absorber studied by t2002 has no c iv and @xmath6= 16.1 . we conclude that the most likely column density range that fits all the data is n@xmath42=10@xmath1 2 . * if we assume that the metallicities of the components are the same as for the t2002 absorber in the 3c 273 sight line , we derive column densities of 1.6 @xmath43 2 and 4.0 @xmath43 2 . the fit has a @xmath44 = 2.04 ( probability = 0.005 ) , insignificantly different from the nominal best fits discussed above . these are the values shown in the middle panel of figure 3 and are clearly consistent with the fuse data . these values also have reasonable , although somewhat low , @xmath6-values for low-@xmath13 absorbers of 17 and 16 respectively . in conclusion , while we can not prove based upon the data alone that the column density of these two absorbers is = 16.2 and 16.6 2 and that the metallicities are a few percent of solar , these values are the most probable for this absorption complex . our `` best guess '' model has = 16.2 , 16.6 2 derived from the assumption that @xmath45=0.06 @xmath46 , as in the t2002 absorber , for the two components . because the highly saturated absorption profile and the poor quality of the fuse spectrum make it difficult to constrain the h i column density precisely , we use the metal line ratios to determine the properties of the gas before returning to analyze the h i absorption informed by the metal - line data . as summarized in table 2 , the metal lines are associated with two velocity components , one at @xmath47 , and one at @xmath48 . these velocity components show no differences in structure as a function of species or ionization state suggesting that the gas is in a single phase . several lines of argument favor a photoionization model for the gas : ( 1 ) for the @xmath48 kms@xmath11 component , limits on the line width constrain the temperature of the gas to be less than 105,000 k ; ( 2 ) the relative si ii and c ii column densities of both components are compatible with equal ionization fractions ( after accounting for the relative abundances of si and c ) ; ( 3 ) purely collisional ionization is inconsistent with the si ii / c ii ratio for temperatures above 20,000 k ( sutherland & dopita 1993 ) . however , at temperatures below 20,000 k extremely little c iv or si iv would be observed . thus , we consider the properties of the gas at each velocity in turn assuming : ( 1 ) all observed ionization stages of carbon and silicon are associated with the same gas , which is compatible with the ratio of column densities , as we will show ; ( 2 ) the gas is primarily photoionized , which is consistent with the predominance of lower ionization stages for carbon and silicon and with the consistency of the column density ratios ; ( 3 ) the gas is approximated as a plane parallel slab illuminated by a quasar - dominated spectrum @xcite with an intrinsic power law spectrum of spectral energy index @xmath49 . to model the physical properties of the gas we have constructed a large grid of photoionization models using the photoionization code cloudy @xcite . the calculated column densities of c ii , c iv , si ii , si iii , and si iv depend primarily on the metallicity , h i column density , and ionization parameter @xmath50 . relative to these parameters , the uncertainty in the spectral shape of the extragalactic background radiation contributes a negligible difference to the models . at this point , we examine the dependence of the n(c ii)/n(c iv ) , and n(si ii)/n(si iv ) column density ratios on the ionization parameter . for a given metallicity and h i column density , we can plot these ion ratios versus u , as shown in figure [ fig : modcolu ] . in this figure we assume the metallicity to be @xmath51 solar ( corresponding to = 16.2 and 16.6 for the 1685 and 1721 absorbers respectively ) , our `` best guess '' model based upon the constraints described in 2.4 . finally , closest agreement with the observed column densities of n(si ii ) and n(si iv ) is obtained by assuming log ( si / c ) @xmath52 , which is also compatible with the conclusions of t2002 for the absorber toward 3c 273 . the range in ionization parameter consistent with the carbon line observations is @xmath53 for the gas at @xmath54 km / s and @xmath55 for the gas at @xmath56 km / s . this range is also consistent with the n(si ii)/n(si iv ) column density ratio , so our one phase approximation likely remains a reasonable one . these ranges indicate the uncertainty in u , for a _ given _ metallicity or h i column density . to explore the sensitivity of u to the h i column density and metallicity , we step through column density within the range , = 14.7 17.5 2 , corresponding to a metallicity in the range 6 0.005 @xmath46 such that the absolute column density of each metal ion is always consistent with the values in table 2 . the range in u is fairly insensitive to and @xmath45 ; the expanded range becomes @xmath57 and @xmath58 for the gas at @xmath59 1685 and 1721 respectively over this range of . for a single - phase model , as suggested by the simplicity of the metal line profiles , the allowed range in ionization parameter may be translated into a range in density if the intensity of the extragalactic ionizing background is known . @xcite estimate a mean intensity of @xmath60 ergs s@xmath112 ster@xmath11hz@xmath11 at one rydberg by summing the contribution from quasars , which is consistent with @xmath61 calculated using more direct methods such as the low-@xmath13 proximity effect , truncation of h i disks , and limits on h-@xmath62 from extragalactic clouds @xcite . the implied range in hydrogen density for these absorbers is @xmath7 = 1 @xmath63 and 1 3 @xmath64 @xmath8 for the @xmath59 1685 and 1721 components respectively . these densities correspond to overdensities of @xmath65 600 3000 and 600 1800 for the two absorbers in units of the mean baryon density at @xmath66 . these densities are an order of magnitude lower than those inferred for the gas towards 3c 273 ( t2002 ) . = 0.9 the extent of the gas associated with the absorption , over the possible range of or metallicity is @xmath67 kpc . however , the requirement that our models reproduce the absolute c ii and c iv column densities , as well as their ratios , means that the and metallicity for successful models are highly correlated , as shown in figure 5 . since the line of sight extent of the gas in our models is derived by dividing by the mean hydrogen density , and since the range in mean hydrogen density compatible with line ratios varies slowly with there is also a good correlation of extent with assumed metallicity . figure [ fig : extent ] shows the line of sight extent of the gas as a function of the assumed metallicity . very large extents are possible only by invoking @xmath68 ; these low metallicities also require @xmath69 17 2 , which yield unrealistically low @xmath6-values ( see 2.4 ) . for the probable lower limit on @xmath70 16.0 2 suggested by the fuse detection of higher order lyman lines , the metallicity of these structures must be < @xmath7110 20% solar , in agreement with their location well away from the virgo core . = 0.9 = 0.9 for our `` best guess '' model , = 16.2 and 16.6 2 , derived from the assumption of @xmath0 6% solar abundances ( consistent with the nearby 3c 273 absorber ) we find sizes along the line of sight of @xmath010 30 kpc . the t2002 analysis of the 3c 273 absorber at the same velocity infers an even smaller line - of - sight depth of only 70 pc . we can not rule out that the column densities are an order of magnitude higher than our `` best guess , '' with corresponding metallicities of @xmath720.01@xmath46 , and much larger line of sight extents . however , the observations together with those analyzed in t2002 suggest that a very thin ( @xmath0 10 kpc ) `` ribbon or sheet '' of photoionized gas with abundances of a few percent of solar extends for several hundreds of kpc through this region on the outskirts of the virgo cluster or that there are several thin gaseous structures lying along a large scale filament in this region . given the expected velocity dispersion of galaxies in the outskirts of a cluster like virgo ( 100 - 300 ) , it is unlikely that a single thin `` ribbon '' would be dynamically stable for the length of time required to form it ( > @xmath210@xmath73 yrs ) . our knowledge of galaxies near the rx j1230.8 + 0115 sight line at virgo cluster velocities comes from four sources : ( 1 ) an augmentation to the center for astrophysics ( cfa ) redshift survey @xcite , ( 2 ) a pointed survey using las campanas observatory ( lco ) to obtain redshifts around 3c 273 @xcite , ( 3 ) two deep pointed surveys using multi - object spectroscopy from the wisconsin - indiana - yale - noao ( wiyn ) 3.5 m and from the las campanas observatory ( lco ) du pont 2.5 telescopes @xcite , and ( 4 ) the sloan digital sky survey ( sdss ) early data release ( edr , stoughton et al . 2002 ) . the pointed galaxy surveys were designed to investigate the relationship between the forest absorption systems along these sight lines and the galaxies . @xcite obtained spectroscopic data to augment the cfa redshift survey in the 3c 273 region so that the survey would be virtually complete down to @xmath74 15.7 ( m@xmath75 -16.2 at 24.3 mpc for @xmath5= 1700 ) in a region bounded by 11.5@xmath76 and -3.5@xmath77 + 8.5 . this region includes both the 3c 273 and rx j1230.8 + 0115 sight lines and extends several mpc around these sight lines at virgo distances . the sdss edr spectroscopic survey includes a strip around the north galactic cap at @xmath78 that extends up to @xmath79 , covering the declination of rx j1230.8 + 0115 but not the declination of 3c 273 . the nominal completeness limit for this spectroscopic survey is r @xmath4 17.8 , at least two magnitudes fainter than the augmented cfa survey in this region . @xcite also conducted a galaxy redshift survey around 3c 273 for which they claimed completeness to m@xmath8019 over a region with a 60radius in @xmath81 and 40 radius in @xmath82 from the quasar . while rx j1230.8 + 0115 is just beyond the southern edge of this survey region , it includes galaxies up to @xmath0 70070 kpc to the north and west of rx j1230.8 + 0115 . more recently , @xcite have conducted a multi - object spectroscopy survey in the region , including : ( 1 ) an lco slit - mask spectroscopy survey around rx j1230.8 + 0115 ; ( 2 ) an lco slit - mask survey around 3c 273 to augment the @xcite observations ; and ( 3 ) a wiyn / hydra fiber - fed spectroscopic survey in the region between 3c 273 and rx j1230.8 + 0115 centered at @xmath83 . a preliminary discussion of the observing setups , data reduction , and analysis of the lco and wiyn data can be found in @xcite . a more complete discussion of the techniques and results for all sight lines observed will be presented later @xcite . we review what has been done so far in the 3c 273/ rx j1230.8 + 0115 region by this on - going survey here : \(1 ) the lco observations surrounding rx j1230.8 + 0115 consist of spectra for 43 galaxies , or @xmath0 78% of the galaxies down to @xmath84 20 ( corresponding to @xmath85 -12 at the distance of the @xmath19 complex ) out to a radius of 5 ( 36 h@xmath11 kpc at this distance ) from the target and 21% of the galaxies out to 15 . no galaxies were found at the redshifts of any of the virgo cluster absorbers , although several galaxies associated with absorbers behind virgo were found . \(2 ) the lco slit mask spectroscopy around 3c 273 is 100% complete down to m@xmath8019.5 out to a 10 radius . we obtained redshifts for 5 galaxies that were missed by @xcite , probably due to a classification error made by the automated galaxy - finding software employed ( see * ? ? ? * ) despite their claim of 100% completeness over a large region . seventy - seven galaxies fainter than the @xcite magnitude limit were also observed . one of the five galaxies missed by @xcite is a relatively bright ( m@xmath86 ) galaxy 7170 kpc from 3c 273 ( 36070 kpc from rx j1230.8 + 0115 ) at the redshift of the strong h i absorber ( @xmath87=1589 ) analyzed by t2002 . this galaxy has the spectroscopic signatures of a `` post - starburst , '' a rare galaxy type in the local universe . the starburst phase might be expected to create a substantial wind that could be responsible for a high - column density , low - ionization absorber like the one present in the 3c 273 sight line @xcite . while this galaxy is too far from rx j1230.8 + 0115 to be related to the @xmath19 complex , the fact that this object was missed by @xcite due to a classification error , suggests that the region around rx j1230.8 + 0115 may not have been surveyed well enough to declare that similar luminosity galaxies have not been missed . no other galaxies were found in these observations at virgo distances . \(3 ) the wiyn / hydra observations concentrated on the region between rx j1230.8 + 0115 and 3c 273 , 0.9@xmath88 to the north . the fiber setup for these observations was centered at ra = 12:30:00 and dec = + 01:40:00 ( j2000.0 ) and redshifts were obtained for 81% of the galaxies within a 20 radius of this position . a total of 38 galaxies were observed successfully ; only one galaxy was observed at both wiyn and lco as we worked to prevent duplication as much as possible . thus , in the region north of rx j1230.8 + 0115 , towards 3c 273 , our galaxy survey is @xmath080% complete down to m@xmath8918.7 . no galaxies were found at virgo distances in the wiyn / hydra observations either . the galaxy survey work should provide some confidence that this region of space has been scrutinized exceptionally well for galaxies . there are 17 galaxies with @xmath90 2500 now known in the region within 1 mpc of either sight line . the sdss ( stoughton et al . 2002 ) and the @xcite galaxy survey work is almost completely overlapping in the rx j1230.8 + 0115 region to an absolute magnitude limit of m@xmath91 . the cfa and stocke et al . ( 2003 ) survey work suggests that all relatively bright galaxies ( m@xmath92 ) have been identified in the field and that only extremely faint ( m@xmath93 ) , extremely compact , or extremely low surface brightness galaxies could have been missed within 3670 kpc of the sight line . a summary of the nearest galaxy to each of the absorbers is included in table 3 , which lists by column : ( 1 ) target sight line ; ( 2 ) absorber redshift ( we list only the centroid of the absorption blend modeled here ) ; ( 3 ) the heliocentric absorber velocity in ; ( 4 ) the velocity difference in the sense of ( @xmath94 - @xmath95 ) in ; ( 5 ) the impact parameter in 70 mpc ; and ( 6 ) the name of the nearest galaxy to each absorber . the relationship between the absorbers and seven of these galaxies has been discussed extensively by @xcite . the galaxy leda 139866 is at 2329 , just out of the velocity range discussed by @xcite , and the new galaxy " is the post - starburst galaxy described briefly above . the galaxy labeled [ isi96]1228 + 0116 is a very low surface brightness dwarf irregular ( m@xmath96 ; impey et al . 1996 ) . ten of the 17 galaxies near rx j1230.8 + 0115 and 3c 273 have recession velocities within @xmath15300 of the @xmath19 complex , but all are at substantial impact parameters ( 280 52070 kpc ) . the nearest galaxy is an edge - on spiral of modest luminosity ( cgcg 014 - 064 ; m@xmath97 = -16.2 ) located 27970kpc away with a redshift nearly identical to that of one of the absorbers in this complex . if the galaxies responsible for absorption at large impact parameters are expected to be bright , there is a much brighter , face - on spiral ( ngc 4517a ; m@xmath97=-18.8 ) a little further away , @xmath98= 36570 kpc , with a larger velocity difference ( @xmath99 = 1530 ) . on the other hand , it has been argued that lsb galaxies may be responsible for much of the absorption that we see . there is an lsb galaxy , discovered by @xcite , in the field , but it is even further away in impact parameter ( 42370 kpc ) and velocity ( @xmath5 = 1473 ) . however , the velocity is similar to that of the absorber at @xmath5=1482 ( see table 1 ) . identification of this lsb galaxy with the 1482 absorber is problematic since there are three other galaxies within @xmath100 that are similar distance or closer to rx j1230.8 + 0115 on the sky . the only `` direct '' measurement of the physical sizes of absorbers comes from the statistical analysis of line pairs in adjacent sight lines ( see e.g. , * ? ? ? * ; * ? ? ? * ) . in summarizing recent results , @xcite showed that the statistics of line pairs suggest characteristic _ radii _ for high column density ( @xmath101 2 ; i.e. , higher column density than the bulk of the forest ) absorbers of 300 - 70070 kpc for spherical clouds , with a possible trend towards larger sizes at smaller redshifts . the lowest redshift previously investigated using this technique is one pair of sight lines at @xmath1020.5 @xcite . the maximum likelihood analysis used on these pairs provides a range of radii extending more than a factor of two on either side of the values quoted above . however , all of the radius estimates are comparable to , or larger than , typical distances between bright galaxies . thus , if these absorber radii are interpreted as being `` contiguous '' structures , it is unlikely that absorbers are physically or causally related to individual galaxies since their sizes are comparable to or larger than the distances between bright galaxies . these radii are more consistent with large - scale structure filaments found in numerical simulations @xcite . additionally , the observation that not all absorbers found in one sight line have corresponding absorption in the other sight line implies that these are elongated structures rather than enormous spherical halos if the structures are `` contiguous . '' an alternative explanation is that coincident line pairs indicate `` correlated '' structures of smaller systems as in the case of galaxy halos aligned along a large - scale structure filament ( see discussion in * ? ? ? * ; * ? ? ? this is suggested in several cases where substantial velocity and ew differences between coincident lines are noted @xcite . = 0.9 the pairs analysis of these sight lines is consistent with the inferences obtained from other sight line pairs @xcite . figure 7 shows the 3c 273 ( upper ) and rx j1230.8 + 0115 ( lower ) spectra with the locations all @xmath103 ( solid ticks ) and @xmath104 ( dotted tick ) absorbers marked as well as the velocity range for the two virgo supercluster `` clouds '' of galaxies in this direction . we note that the line pairs we have identified include absorbers with equivalent widths a factor of 4 less than those identified in the higher redshift studies of @xcite . using the @xmath105 150 criteria for `` coincident '' line pairs suggested by @xcite , two of the three 3c 273 absorbers are coincident with 4-@xmath18 absorbers in the other sight line and the third is coincident with a marginal detection . for rx j1230.8 + 0115 , four of five are coincident by the @xcite definition by counting the 1685 and 1721 absorbers as a single blended line and noting that the 1482 absorber is also coincident with the same absorber in the 3c 273 sight line ( thus , 3 `` coincidences '' if this sort of double counting is not justified ) . in either case , a large fraction of coincidences are present , suggesting that transverse sizes are large for at least some of these absorbers . the transverse distances between the sight lines at these redshifts is 200 - 50070 kpc , assuming a pure hubble flow in this direction , comparable to those investigated using other sight line pairs . the lowest velocity absorbers beyond virgo ( and thus at higher redshifts than shown in figure 7 ) have no coincident absorbers in the other sight line @xcite for transverse distances in the range 1.75 - 2.070 mpc . these coincidences indicate characteristic absorber _ radii _ of @xmath0 100 25070 kpc . unlike at higher redshifts , no chance pairings are expected in this velocity interval , which could confuse these issues . using the line densities found by @xcite at low h i column densities , we compute that there should be 0.14 @xmath15 0.08 ( a poisson probability of 4@xmath106 ) line pairs with @xmath105 150 created by chance in figure 7 . thus , all the absorber matches seen in these two spectra are real ( to 99.96% confidence ) statistical associations of absorbers . the only coincidence between the h i and o vi absorber at 1011 in the 3c 273 sightline and an rx j1230.8 + 0115 is with the marginal detection at 889 . while this is only a single incidence , a size limit of @xmath72 225 h@xmath11 kpc ( the distance between these two sightlines at this redshift ) supports the collisionally - ionized model for this absorber advocated by t2002 , because a photoionized model for this absorber requires a very large physical size ; i.e. , we would expect a very large photoionized absorber to be more uniform over these two sight lines . as shown by examples in tripp et al . ( 2001 ) and tripp and savage ( 2000 ) , the simultaneous presence of o vi and absence of c iv absorption may be a discriminator between collisionally - ionized and photoionized o vi absorbers . the characteristic radii derived above are based on using the @xmath105 150 criteria which was defined as a `` coincidence '' for much poorer resolution faint object spectrograph ( fos ) spectra ( @xmath107 25 times poorer than the present data ) . if these spectra were smoothed to the fos resolution ( tripp , private communication ) , only one line pair match and two mismatches would be found the lower equivalent width lines in figure 7 would not be detectable . the remaining coincidence includes the three metal line absorbers at 1586 1721 , suggesting a somewhat smaller but still substantial , characteristic size for these structures . while the difference in these results is indicative of the resolution and sensitivity dependent nature of this measurement , we would still conclude that these structures are large and elongated , comparable in size to the distance between these two sight lines . assuming that these absorbers reside in a gaseous filament with little turbulent motion , the hubble flow velocity differences between absorbers in the two sight lines should be in the range 30 60 . therefore , despite the better stis resolution and sensitivity , the @xmath108v @xmath109 criterion for absorber coincidences still seems reasonable over such large separations . while line pair matches within 150 suggest a relationship between these absorber pairs , the velocities , equivalent widths , and profile shapes are not so well correlated as to require `` contiguous '' structures stretching across these two sight lines ( see e.g. rauch et al . 2001 examination of sub - kpc separated sight lines at high-@xmath13 ) . however , we can not rule out the existence of a low column density ( @xmath110 @xmath111 ) contiguous structure , augmented at specific locations by higher column density gas ; e.g. , galaxy halos or winds embedded in a large - scale gaseous filament . the presence of a galaxy at similar velocity to the 1586 absorber and at a small impact parameter from 3c 273 ( t2002 and penton et al . 2002 ) may be evidence for this interpretation . the small number of absorbers found in these two sight lines and the possible `` special conditions '' of the virgo cluster make statistical inference based upon this one sight line pair insecure . however , the approximate velocity coincidence between the metal line systems studied in detail here and in t2002 strongly suggests that these absorbers are related . we discuss the sizes and shapes of the absorbers implied by these pairing further in the discussion section . the absorption line systems in the rx j1230.8 + 0115 and 3c 273 sight lines along the southern edge of the virgo cluster provide a low redshift laboratory in which to study the properties of these absorbers and their relationship with their environment . we have presented a detailed discussion of the physical properties of these absorbers , concentrating on the characteristic sizes of low-@xmath13 absorbers . we summarize these results in the next subsection . in 6.2 , we present an interpretation of the nature of these absorbers inferred from their properties and from the local environment of the absorption systems and present our conclusions about these absorbers . based upon a detailed photoionization equilibrium calculation , the two blended absorbers in the rx j1230.8 + 0115 sight line at 1685 and 1721 and the 1586 absorber in the 3c 273 sight line ( t2002 ) have inferred sizes along the line - of - sight of @xmath020 kpc and 70 pc respectively . these inferences are much more secure for the 3c 273 absorber because the hydrogen column density can be accurately determined due to available high quality fuse spectroscopy of the higher order lyman lines . for the 3c 273 absorber , t2002 find = 15.85 2 , log(@xmath7 ) = -2.8 @xmath150.3 @xmath8 , @xmath45 = 0.06 @xmath46 and an inferred thickness of @xmath070 pc . for the case of the two blended absorbers studied here , the combination of hst stis echelle and fuse spectroscopy are insufficient to provide similarly tight constraints . the fuse spectrum is sufficient to rule out column densities as low as 10@xmath112 2 and set a probable lower limit of = 10@xmath27 2 for each absorber due to significant detections of ly-@xmath113 and ly-@xmath114 in both components and possible detections of ly-@xmath26 , ly-@xmath25 and ly-@xmath35 in at least one component . additionally , the low @xmath6-values that would be required to fit rule out column densities above = 17.0 2 . we use the available stis echelle and fuse data to infer the physical conditions of these absorbers . the metal line ratios ( n(c iv)/n(c ii ) and n(si iv)/n(si ii ) ) constrain the ionization parameter ( -2.85@xmath4log @xmath115 - 2.60 ) for our best - guess model , assuming the gas is photoionized . assuming a standard value for the extragalactic ionizing radiation intensity , we can translate this to limits on the density of the gas . if we knew the metallicity of these structures , this would constrain the hydrogen column density and thus the absorber size along our line - of - sight . by assuming the same metallicity for these absorbers as for the 3c 273 absorber located some 35070 kpc away and column densities of = 16 17 , consistent with constraints from the stis and fuse data , we calculate line - of - sight sizes of 10 30 kpc . since these metallicities and column densities are similar to those obtained by t2002 , the larger physical sizes obtained here are due to the higher ionization ( lower density ) of the rx j1230.8 + 0115 absorbers ( c iv is detected here , but not in 3c 273 ) . the blended absorbers studied here have inferred line - of - sight sizes a factor of 7 300 times less than the minimum transverse size based upon detecting absorption in both sight lines at comparable redshifts , while the 1586 absorber in 3c 273 has a size that is a factor of @xmath04300 smaller than the minimum transverse size . this much larger inferred transverse size is more suggestive of objects associated with large scale structure than with individual galaxies , while the smaller line - of - sight sizes are more consistent with structures related to individual galaxies ( e.g. , very extended halos or outflowing winds ) . while only a few high column density absorbers similar to the ones in these two sight lines are known , they are not unique at low-@xmath13 . there is an absorber with a column density of @xmath016 2 at @xmath116 towards pks 0405 - 1219 @xcite and one with a column density of = 15.57 at @xmath117 towards h1821 + 643 @xcite . both of these systems very closely resemble the rx j1230.8 + 0115 @xmath19 complex in that they all contain a saturated absorber with a two component metal line system . these absorbers also have nearby galaxies , there is an 0.02 l@xmath118 spiral galaxy at 9070 kpc and a 1.2 l@xmath118 elliptical at 11070 kpc from pks 0405 - 1219 and there is a galaxy 11270 kpc away from the h1821 + 643 absorber . the difference between the rx j1230.8 + 0115 @xmath19 complex and these systems is the existence of higher ionization species detected with o vi and n v detected in pks 0405 - 1219 and o vi and si iii detected in h1821 + 643 . at somewhat higher redshifts , @xcite identify a class of weak mg ii systems which have similar physical properties to the absorbers discussed above including sizes on the order of 10 pc to several kpc . for the limited number of regions around the absorbers that have been surveyed optically , there are no galaxies down to 0.01l@xmath119 within 7070 kpc , but larger regions around the absorbers have not been studied . while the lower ionization weak mg ii absorbers are most similar to the absorbers studied here and in the 3c 273 sight line , there are also more highly ionized metal line absorbers ( i.e. , o vi detected ) that are found somewhat farther from galaxies , but that seem to be associated with galaxy groups . the pks 2155 - 304 sight line has 6 absorbers between 16,000 and 18,000 , which appear to be associated with a small group of spiral galaxies , although none of these are within 40070 kpc of the absorbers @xcite . two of these absorbers have been detected in o vi with fuse @xcite and there is a tentative detection of o viii with chandra @xcite . the two o vi absorbers in the pg 0953 + 415 sight line @xcite fall near peaks in the galaxy density , with the @xmath120 absorber having 4 galaxies within 130 and 3 mpc , the closest at a projected distance of 395 kpc @xcite . similarly , the h1821 + 643 @xmath121 absorber appears to fall within a galaxy group , with 7 galaxies within 500 and 3.1 mpc . these higher ionization absorbers have larger inferred line - of - sight sizes than those discussed above ( based on a photoionization model ) . however , since there are no adjacent sight lines , there are no constraints on their transverse sizes . we note that the column densities of the absorbers studied here are unusually high for the bulk of the local forest absorbers and so are probably not representative of the properties ( size , metallicity , etc . ) or the environment of the lower column density lines ( @xmath122 2 ; penton et al . 2002 ) . @xcite examined the hypothesis that the 3c 273 and rx j1230.8 + 0115 absorbers at virgo distance are due to extended galaxy halos and concluded that the evidence was inconsistent with the available data . based upon the known galaxy redshifts , the nearest galaxies to the strong absorbers in these sight lines were @xmath320070 kpc away and there seemed to be no correlation between impact parameter and the probability of detecting absorption as would be expected in a simple halo model . the galaxy redshift surveys have been improved and a dwarf galaxy has been found only 7170 kpc from the 3c 273 1586 absorber @xcite . finding this galaxy may indicate that the halos of dwarf galaxies or outflows from them are responsible for some absorbers . the `` post - starburst '' signature in the optical spectrum of this galaxy makes an outflow model for the 3c 273 absorber particularly attractive since it can explain the sub - solar metal abundances as well as the super - solar ( si / c ) ratio , which is indicative of sn type ii enrichment . the rx j1230.8 + 0115 @xmath19 complex could be associated with a gas - rich galaxy ( h i 1225 + 01 ) 400 and 27070 kpc away , or with the edge - on dwarf spiral ( cgcg 014 - 064 ) 28070 kpc and only 56 away . the fact that the dwarf galaxy 7170 kpc from 3c 273 was missed by @xcite provides a warning that the rx j1230.8 + 0115 region may not have been surveyed sufficiently well to conclude that there are no galaxies of similar luminosity near these blended absorbers , although there have been a number of relatively deep , independent surveys of this region . if another faint galaxy is found close to rx j1230.8 + 0115 at @xmath1231700 , then the model of outflowing winds from dwarf galaxies @xcite is an exceptional fit for both the 3c 273 and rx j1230.8 + 0115 @xmath1231700 absorbers @xcite . the higher than average galaxy density ( @xmath124 times the average ) in this region at the southern edge of the virgo cluster make this a reasonable possibility . however , this scenario may require very large amounts of h i associated with quite small galaxies ( see below and stocke et al . 2003 for further discussion ) . rx j1230.8 + 0115 is 0.9@xmath125 from 3c 273 , and both of these lines - of - sight pass through a filament of galaxies identified by @xcite as well as passing through the southern extension of virgo and near the w cloud ( the w cloud , discussed as part of this region by t2002 lies several degrees to the north ; * ? ? ? these galaxy clouds are thought to be independent of the virgo cluster . the w cloud is a large feature that may be another filament stretching to the north at higher velocities than are found in the @xcite filament . however , the se cloud is embedded in the northern end of this filament . @xcite showed that both of the strong 3c 273 virgo absorbers , all of the rx j1230.8 + 0115 virgo absorbers , plus the seven ( now eight ) known galaxies in the region all lie along a single large - scale structure filament that could extend for many mpc crossing these sight lines . the filamentary model for these absorbers is consistent with the abundance of common velocity ( @xmath15 150 ) absorber pairs seen in these two sight lines but _ not _ with the line - of - sight thicknesses of these structures inferred from photoionization models . either the methods used to estimate absorber sizes could be wrong , a very thin `` ribbon '' or `` sheet '' of gas could extend along this filament of galaxies for at least several hundred kpc , or , because the pairs analysis can not distinguish between `` contiguous '' or `` correlated '' structures as emphasized by its practitioners @xcite , these absorbers map out correlated rather than contiguous structures . given the difference in depth versus transverse size for these absorbers , and the lack of detailed correlations between the absorber properties ( i.e. , velocity differences greater than @xmath15 50 , which could be accounted for by hubble flow , as well as large equivalent width differences ; see 5 ) in these two sight lines , a correlated structure model seems much more plausible . one possibility is that the absorbers in this correlated structure are related to individual galaxies whose halos or outflowing winds have caused significant clumping and metal enrichment in small regions of the large gaseous structure in which they are embedded . if this is the case , we might expect to find one or more faint galaxies close to the rx j1230.8 + 0115 virgo absorbers . the high resolution stis echelle spectra , shown in figure 7 , indicate that there is no detailed correlation between the velocities or the equivalent widths of the absorbers in these sight lines . these differences in velocity and equivalent width structure contrast with the extremely detailed velocity and equivalent width correlations between the ly@xmath62 absorbers in the two gravitationally - lensed images of q1422 + 231 @xcite separated by 80h@xmath126 parsecs at @xmath127 . in the case of the q1422 + 231 absorbers there is a detailed agreement between these absorber properties , requiring a quiescent , contiguous gaseous structure @xcite . the differences between the q1422 + 231 sight lines and these sight lines may indicate the difference in scale over which we see contiguous structures ( tens of parsecs ) and correlated structures ( hundreds of kpc ) . if we assume that the absorber depths of 1 30 kpc calculated in 3 and in t2002 are indicative of the overall three - dimensional size of these absorbers , then , in order to obtain a covering factor of unity ( as would be required to detect three such absorbers in the two sight lines observed ) , there would need to be 10 to 10,000 of these structures in the filament ( assuming a filament thickness of @xmath0 100 kpc ) . charlton et al . ( 2002 ) also find that a large number of weak mg ii absorbing structures are required to explain their detected population found at higher redshifts ; i.e. , these absorbers must greatly outnumber bright galaxies . although not yet observed at mg ii , the three absorbers discussed here are very similar to those studied by @xcite . if these absorbers are associated with galaxies , we would expect to find numerous faint or low surface brightness galaxies throughout this filament . while we make the case in 6 that some galaxies may have been missed , this filament has been thoroughly surveyed for galaxies and only 8 have been found . it is very hard to believe that large numbers of star formation sites have been missed given the volume of galaxy redshift data available for this region ( 4 ) . one possible scenario is that while these absorbers have small thicknesses , they could be quite extended on the sky if the metal enriched gas is in the form of a thin shell surrounding a galaxy , perhaps driven by a wind . in this case , the area on the sky would be substantially larger than the thickness derived from photoionization ; e.g. , the radius of the shell could extend 50 100h@xmath11 kpc from the galaxy if the distance to the galaxy near the 3c 273 absorber is indicative of the size of these shells ( impact parameter @xmath128 kpc ; 4 and stocke et al . 2003 ) . in this case , only a few ( @xmath04 ) gaseous shells would be required to have the covering fraction of unity necessitated by the 3c 273/rx j1230.8 + 0115 observations . in the shell scenario , it would also be quite natural to find two absorbers closely spaced in velocity along a single sight line ( e.g. , rx j1230.8 + 0115 ) due to the interception of both sides of the expanding shell of gas . for a 75 kpc radius shell with an expansion velocity of 18 ( the separation between the rx j1230.8 + 0115 metal lines ) the epoch of star formation would have occured @xmath0 @xmath129 yrs ago expelling @xmath130 of gas ( given the average rx j1230.8 + 0115 gas density and shell thickness ) . thus , the galaxy putatively associated with the enriched gas in these absorbers does not need to be currently forming stars . this is the case for the dwarf near the 3c 273 absorber @xcite , which would have had to expel @xmath131 of gas in this scenario . this scenario requires large masses of gas to have been expelled from galaxies , but may be compelling if another galaxy is found near rx j1230.8 + 0115 . given the significant amount of optical spectroscopy already expended on this region , it may be more likely that the column density of the absorber is on the high end of our range , making the metallicity significantly smaller . in this lower metallicity scenario , these systems could represent more primordial overdensities expected to be associated with large scale structure @xcite . given the proximity of these absorbers to us , this region is the best available laboratory with which to test this hypothesis so we are presently surveying the region at the 21 cm line of h i to look for gas - rich galaxies as well as continuing our optical spectroscopy to answer this question definitively . jlr thanks steve penton for assistance extracting and calibrating the rx j1230.8 + 0115 stis spectrum and ken sembach for input on the fuse spectrum . the authors thank todd tripp for an extremely careful reading of the paper and helpful suggestions . jts , mlg and jlr acknowledge financial support from nasa - hst archival research grant ar-09221.01-a for this work . jts also acknowledges support for the cosmic origins spectrograph gto team through nag 512279 . jts thanks ray weymann and kevin mclin for permission to quote some results from our galaxy survey work in the rx j1230.8 + 0115 and 3c 273 region prior to publication . tripp , t. m. , jenkins , e. b. , williger , g. m. , heap , s. r. , bowers , c. w. , danks , a. c. , dav , r. , green , r. f. , gull , t. r. , joseph , c. l. , kaiser , m. e. , lindler , d. , weymann , r. j. , & woodgate , b. e. 2002 , , 575 , 697 . weymann , r. , januzzi , b. t. , lu , l. , bahcall , j. n. , bergeron , j. , boksenberg , a. , hartig , g. g. , kirhakos , s. , sargent , w. l. , savage , b. d. , schneider , d. p. , turnshek , d. a. , & wolfe , a. m. 1998 , , 506 , 1 crcccrccrc 1219.27 & 889 & 0.002964 & 0.000012 & 17 & 5 & 13.10 & 0.12 & 46 & 52 + 1221.68 & 1482 & 0.004945 & 0.000004 & 22 & 2 & 13.68 & 0.04 & 158 & 56 + 1222.50 & 1685 & 0.005621 & & 23 & 1 & 16.20 & & 497 & 54 + 1222.65 & 1721 & 0.005742 & & 17 & 1 & 16.60 & & 410 & 44 + 1223.11 & 1834 & 0.006120 & 0.000023 & 53 & 11 & 13.47 & 0.09 & 115 & 56 + 1225.01 & 2302 & 0.007680 & 0.000004 & 33 & 2 & 14.30 & 0.06 & 360 & 69 + lccccrcccrr c ii & & 1685 & 0.005621 & 0.000003 & 6 & 2 & 13.22 & 0.07 & 29 & 21 + c ii & & 1721 & 0.005742 & 0.000002 & 11 & 1 & 13.69 & 0.03 & 71 & 23 + c iv & & 1679 & 0.005600 & 0.000012 & 15 & 6 & 12.88 & 0.17 & 24 & 30 + c iv & & 1720 & 0.005738 & 0.000003 & 10 & 2 & 13.26 & 0.06 & 66 & 33 + si ii & & 1687 & 0.005628 & 0.000014 & 19 & 7 & 12.42 & 0.14 & 27 & 26 + si ii & & 1720 & 0.005738 & 0.000004 & 7 & 2 & 12.49 & 0.09 & 40 & 31 + si iv & & 1684 & 0.005617 & 0.000015 & 14 & 7 & 12.46 & 0.20 & 16 & 25 + si iv & & 1718 & 0.005731 & 0.000004 & 8 & 2 & 12.86 & 0.08 & 43 & 26 + rx j1230.8 + 0115 - 1 & 0.002963 & 888 & -218 & 0.117 & cgcg 014 - 054 + rx j1230.8 + 0115 - 2 & 0.003437 & 1030 & -76 & 0.117 & cgcg 014 - 054 + rx j1230.8 + 0115 - 3 & 0.004945 & 1482 & 184 & 0.271 & hi 1225 + 01 + rx j1230.8 + 0115 - 4 & 0.005668 & 1699 & 56 & 0.279 & cgcg 014 - 064 + rx j1230.8 + 0115 - 5 & 0.006148 & 1843 & 200 & 0.279 & cgcg 014 - 064 + rx j1230.8 + 0115 - 6 & 0.007680 & 2302 & 13 & 0.162 & [ isi96]1228 + 0116 + 3c 273 - 1 & 0.00337 & 1010 & -96 & 0.104 & cgcg 014 - 054 + 3c 273 - 2 & 0.00530 & 1589 & 33 & 0.071 & new galaxy + 3c 273 - 3 & 0.00745 & 2233 & -96 & 0.396 & leda 139866 +

hadwiger s conjecture from 1943 states that for every integer , every graph either can be-colored or has a subgraph that can be contracted to the complete graph on vertices . as pointed out by paul seymour in his recent survey on hadwiger s conjecture , proving that graphs with no minor are-colorable is the first case of hadwiger s conjecture that is still open . it is not known yet whether graphs with no minor are-colorable . using a kempe - chain argument along with the fact that an induced path on three vertices is dominating in a graph with independence number two , we first give a very short and computer - free proof of a recent result of albar and gonalves and generalize it to the next step by showing that every graph with no minor is-colorable , where . we then prove that graphs with no minor are-colorable and graphs with no minor are-colorable . finally we prove that if mader s bound for the extremal function for minors is true , then every graph with no minor is-colorable for all . this implies our first result . we believe that the kempe - chain method we have developed in this paper is of independent interest .

1606.05507

all graphs in this paper are finite and simple . a graph @xmath15 is a of a graph @xmath16 if @xmath15 can be obtained from a subgraph of @xmath16 by contracting edges . we write @xmath17 if @xmath15 is a minor of @xmath16 . in those circumstances we also say that @xmath16 has an @xmath15 . our work is motivated by the following hadwiger s conjecture @xcite , which is perhaps the most famous conjecture in graph theory , as pointed out by paul seymour in his recent survey @xcite . [ hc ] for every integer @xmath0 , every graph with no @xmath18 minor is @xmath1-colorable . hadwiger s conjecture is trivially true for @xmath19 , and reasonably easy for @xmath20 , as shown by dirac @xcite . however , for @xmath21 , hadwiger s conjecture implies the four color theorem . wagner @xcite proved that the case @xmath22 of hadwiger s conjecture is , in fact , equivalent to the four color theorem , and the same was shown for @xmath23 by robertson , seymour and thomas @xcite . conjecture remains open for @xmath24 . as pointed out by paul seymour @xcite in his recent survey on hadwiger s conjecture , proving that graphs with no @xmath3 minor are @xmath4-colourable is thus the first case of hadwiger s conjecture that is still open . it is not even known yet whether every graph with no @xmath3 minor is @xmath5-colorable . kawarabayashi and toft @xcite proved that every graph with no @xmath3 or @xmath25 minor is @xmath4-colorable . jakobsen @xcite proved that every graph with no @xmath26 minor is @xmath4-colorable and every graph with no @xmath27 minor is @xmath5-colorable , where for any integer @xmath28 , @xmath29 ( resp . @xmath30 ) denotes the graph obtained from @xmath13 by removing one edge ( resp . two edges ) . for more information on hadwiger s conjecture , the readers are referred to an earlier survey by toft @xcite and a very recent informative survey due to seymour @xcite . albar and gonalves @xcite recently proved the following : [ k7k8 ] ( albar and gonalves @xcite ) every graph with no @xmath3 minor is @xmath12-colorable , and every graph with no @xmath31 minor is @xmath32-colorable the proof of theorem [ k7k8 ] is computer - assisted and not simple . in this paper , we apply a kempe - chain argument ( see lemma [ wonderful ] below ) along with the fact that an induced path on three vertices is dominating in a graph with independence number two in order to give a much shorter and computer - free proof of theorem [ k7k8 ] . in addition , we generalize it to the next step by proving the following . [ main ] every graph with no @xmath6 minor is @xmath7-colorable , where @xmath33 . we want to point out that our proof of theorem [ main ] does not rely on mader s deep result on the connectivity of contraction - critical graphs ( see theorem [ 7con ] below ) . theorem [ main ] states that ( i ) every graph with no @xmath3 minor is @xmath12-colorable ; ( ii ) every graph with no @xmath31 minor is @xmath32-colorable ; and ( iii ) every graph with no @xmath34 minor is @xmath35-colorable . we prove theorem [ main ] in section [ kt ] . applying the method we developed in the proof of theorem [ main ] and mader s deep result ( theorem [ 7con ] ) , we then prove two new results theorem [ maink8- ] and theorem [ maink8= ] . [ maink8- ] every graph with no @xmath9 minor is @xmath10-colorable . [ maink8= ] every graph with no @xmath11 minor is @xmath12-colorable . our proofs of theorem [ maink8- ] and theorem [ maink8= ] are both short and computer - free and will be presented in section [ k8- ] and section [ k8= ] , respectively . to end this paper , we first propose a conjecture in section [ remarks ] . we then apply lemma [ wonderful ] to prove that if conjecture [ conj3 ] ( see section [ remarks ] ) is true , then every graph with no @xmath13 minor is @xmath7-colorable for all @xmath14 . our proof of the last result does not rely on the connectivity of contraction - critical graphs and the new idea we introduced yields a different / short proof of theorem [ main ] . to prove our results , we need to investigate the basic properties of contraction - critical graphs . for a positive integer @xmath1 , a graph @xmath16 is if @xmath36 and any proper minor of @xmath16 is @xmath37-colorable . lemma [ dirac ] below is a folklore result which is an extension of dirac s initial work @xcite on contraction - critical graphs . a proof of lemma [ dirac ] can be easily obtained from the definition of @xmath38-contraction - critical graphs . [ dirac](dirac @xcite ) every @xmath38-contraction - critical graph @xmath16 satisfies the following : for any @xmath39 , @xmath40 , where @xmath41 denotes the independence nunber of the subgraph of @xmath16 induced by @xmath42 ; no minimal separating set of @xmath16 is a clique . lemma [ wonderful ] below on contraction - critical graphs turns out to be very powerful , as the existence of pairwise vertex - disjoint paths is guaranteed without using the connectivity of such graphs . if two vertices @xmath43 in a graph @xmath16 are not adjacent , we say that @xmath44 is a of @xmath16 . one possible application of lemma [ wonderful ] is depicted in figure [ wonderfullemma ] . [ wonderful ] let @xmath16 be any @xmath38-contraction - critical graph . let @xmath45 be a vertex of degree @xmath46 with @xmath47 and let @xmath48 with @xmath49 be any independent set , where @xmath50 and @xmath51 are integers . if @xmath52 is not a clique , then for any @xmath53 , where @xmath54 , @xmath55 , the vertices @xmath56 are all distinct , and for any @xmath57 , the set @xmath58 consists of @xmath59 missing edges of @xmath52 with @xmath60 as a common end , then for each @xmath57 there exist paths @xmath61 in @xmath16 such that each @xmath62 has ends @xmath63 and all its internal vertices in @xmath64 $ ] for all @xmath65 . moreover , for any @xmath66 , the paths @xmath61 are vertex - disjoint from the paths @xmath67 . to @xmath68 with @xmath69 and @xmath70.,width=220 ] let @xmath16 , @xmath68 , @xmath71 and @xmath72 be as given in the statement . let @xmath15 be obtained from @xmath16 by contracting @xmath73 into a single vertex , say @xmath74 . then @xmath15 is @xmath75-colorable . let @xmath76 be a proper @xmath75-coloring of @xmath15 . we may assume that @xmath77 . then each of the colors @xmath78 must appear in @xmath52 , else we could assign @xmath68 the missing color and all vertices in @xmath71 the color @xmath79 to obtain a proper @xmath75-coloring of @xmath16 , a contradiction . since @xmath80 , we have @xmath81 for any two distinct vertices @xmath82 in @xmath52 . we next claim that for each @xmath83 and each @xmath84 there must exist a path between @xmath60 and @xmath85 with its internal vertices in @xmath64 $ ] . suppose not . let @xmath83 and @xmath86 be such that there is no such path between @xmath60 and @xmath85 . let @xmath87 be the subgraph of @xmath15 induced by the vertices colored @xmath88 or @xmath89 under the coloring @xmath90 . then @xmath91 . notice that @xmath60 and @xmath85 must belong to different components of @xmath87 as there is no path between @xmath60 and @xmath85 with its internal vertices in @xmath64 $ ] . by switching the colors on the component of @xmath87 containing @xmath60 , we obtain a @xmath75-coloring of @xmath15 with the color @xmath88 missing on @xmath52 , a contradiction . this proves that there must exist a path @xmath62 in @xmath87 with ends @xmath63 and all its internal vertices in @xmath92 $ ] for each @xmath83 and each @xmath84 . clearly , for any @xmath66 , the paths @xmath61 are vertex - disjoint from the paths @xmath67 , because no two vertices of @xmath93 are colored the same under the coloring @xmath90 . width6pt depth2pt + * remark . * if @xmath94 in the statement of lemma [ wonderful ] , we simply write @xmath95 , and so @xmath72 is a matching of missing edges of @xmath52 . in this case , the paths @xmath96 are pairwise vertex - disjoint if @xmath97 . if @xmath98 in the statement of lemma [ wonderful ] , we simply write @xmath99 . in this case , the paths @xmath100 have @xmath101 as a common end and are not necessarily pairwise internally vertex - disjoint if @xmath102 . theorem [ 7con ] below is a deep result of mader @xcite and will be used in the proofs of theorem [ maink8- ] and theorem [ maink8= ] . it seems very difficult to improve theorem [ 7con ] for small values of @xmath38 . for larger values of @xmath38 , some better results can be found . kawarabayashi @xcite has shown that any minimal non - complete @xmath38-contraction - critical graph with no @xmath103 minor is @xmath104-connected , while kawarabayashi and yu @xcite have shown that any minimal such graph is @xmath105-connected . chen , hu and song @xcite recently improved the bound further by showing that any minimal such graph is @xmath106-connected . [ 7con ] ( mader @xcite ) for @xmath107 , every @xmath38-contraction - critical graph is @xmath5-connected . we also need the following lemma in the proofs of theorem [ maink8- ] and theorem [ maink8= ] . [ lem:7conn2k6 ] for any @xmath5-connected graph @xmath16 , if @xmath16 contains two different @xmath108-subgraphs , then @xmath109 . let @xmath110 be two different @xmath108-subgraphs of @xmath16 with @xmath111 and @xmath112 . let @xmath113 . then @xmath114 . we may assume that @xmath115 for all @xmath116 if @xmath117 . assume that @xmath118 . then @xmath119 has a @xmath120-subgraph of @xmath16 . since @xmath16 is @xmath5-connected , it is easy to see that @xmath109 by contracting a component of @xmath121 into a single vertex . so we may assume that @xmath122 . then there exist @xmath123 pairwise disjoint paths @xmath124 between @xmath125 and @xmath126 in @xmath127 . we may assume that @xmath128 has ends @xmath129 for all @xmath130 . then @xmath131 is connected , so there must exist a path @xmath132 with one end , say @xmath133 , in @xmath134 , the other end , say @xmath135 , in @xmath136 , and no internal vertices in any of @xmath137 . we may assume that @xmath133 lies on the path @xmath138 . let @xmath139 be the subpath of @xmath140 with ends @xmath141 , and @xmath142 be the subpath of @xmath143 with ends @xmath144 . now contracting @xmath139 onto @xmath145 , @xmath146 onto @xmath147 , @xmath142 and @xmath148 onto @xmath149 , @xmath150 onto @xmath151 , and each of @xmath152 to a single vertex , together with @xmath153 if @xmath117 , yields a @xmath9 minor in @xmath16 , as desired . depth2pt + we need to introduce more notation . for a graph @xmath16 we use @xmath154 and @xmath155 to denote the number of vertices and minimum degree of @xmath16 , respectively . for a subset @xmath71 of @xmath156 , the subgraph induced by @xmath71 is denoted by @xmath157 $ ] and @xmath158 $ ] . the degree and neighborhood of a vertex @xmath135 in @xmath16 are denoted by @xmath159 and @xmath42 , respectively . by abusing notation we will also denote by @xmath42 the graph induced by the set @xmath42 . we define @xmath160=n(v)\cup \{v\}$ ] , and similarly will use the same symbol for the graph induced by that set . for @xmath161 , if @xmath157 $ ] is connected , then we denote by @xmath162 the graph obtained from @xmath16 by contracting @xmath157 $ ] into a single vertex and deleting all resulting parallel edges and loops . for @xmath163 , we say that @xmath164 is to @xmath165 if each vertex in @xmath164 is adjacent to all vertices in @xmath165 , and @xmath164 is to @xmath165 if no vertex in @xmath164 is adjacent to any vertex in @xmath165 . if @xmath166 , we simply say @xmath167 is complete to @xmath165 or @xmath167 is anti - complete to @xmath165 . the @xmath168 ( resp @xmath169 ) of two vertex disjoint graphs @xmath16 and @xmath15 is the graph having vertex set @xmath170 and edge set @xmath171 ( resp . @xmath172 ) . results on the extremal function for @xmath6 minors will be needed to prove theorem [ main ] . before doing so , we need to define @xmath173-cockade . for graphs @xmath110 and an integer @xmath38 , let us define an _ @xmath174-cockade _ recursively as follows . any graph isomorphic to @xmath175 or @xmath176 is an @xmath174-cockade . now let @xmath177 , @xmath178 be @xmath174-cockades and let @xmath16 be obtained from the disjoint union of @xmath177 and @xmath178 by identifying a clique of size @xmath38 in @xmath177 with a clique of the same size in @xmath178 . then the graph @xmath16 is also an @xmath174-cockade , and every @xmath174-cockade can be constructed in this way . if @xmath179 , then @xmath16 is simply called an @xmath180-cockade . the following theorem [ k7 ] was first shown by dirac @xcite for @xmath181 and by mader @xcite for @xmath182 . [ k7](mader @xcite ) for every integer @xmath183 , a graph on @xmath184 vertices and at least @xmath185 edges has a @xmath13 minor . the edge bound in theorem [ k7 ] is referred to as mader s bound for the extremal function for @xmath13 minors . jrgensen @xcite and later the second author and thomas @xcite generalized theorem [ k7 ] to @xmath186 and @xmath187 , respectively , as follows . [ k8](jrgensen @xcite ) every graph on @xmath188 vertices with at least @xmath189 edges either has a @xmath31 minor or is a @xmath190-cockade . [ k9](song and thomas @xcite ) every graph on @xmath191 vertices with at least @xmath192 edges either has a @xmath193 minor , or is a @xmath194-cockade , or is isomorphic to @xmath195 . it seems hard to generalize theorem [ k7 ] for all values of @xmath196 . in 2003 , seymour and thomas @xcite proposed the following conjecture . [ conj2](seymour and thomas @xcite ) for every @xmath197 there exists a constant @xmath198 such that every @xmath199-connected graph on @xmath200 vertices and at least @xmath185 edges has a @xmath13 minor . by theorem [ k9 ] , conjecture [ conj2 ] is true for @xmath201 . we next prove the following lemma [ k_tk1 ] , which can be obtained from the ( computer - assisted ) proof of lemma 3.7 in @xcite . here we give a computer - free proof of lemma [ k_tk1 ] so that the proof of theorem [ main ] is also computer - free . [ k_tk1 ] for @xmath202 , let @xmath15 be a graph with @xmath203 vertices and @xmath204 . then @xmath205 . suppose that @xmath15 has no @xmath206 minor . then @xmath207 . we claim that [ e : nok_t-3 ] ( ) @xmath208 . suppose that @xmath209 . let @xmath210 be isomorphic to @xmath211 . then @xmath212 . if @xmath213 contains an induced @xmath214-path , say @xmath215 , with ends @xmath216 , where a @xmath214-path is a path with three vertices , then every vertex of @xmath217 is adjacent to either @xmath218 or @xmath219 because @xmath204 . by contracting the path @xmath215 into a single vertex , we see that @xmath220>k_{t-2}$ ] and so @xmath205 , a contradiction . thus @xmath213 does not contain an induced path on three vertices . since @xmath204 , it follows that @xmath213 is a disjoint union of two cliques , say @xmath221 and @xmath222 . for @xmath223 , let @xmath224 . since @xmath204 , @xmath225 is complete to @xmath226 for each @xmath227 . thus @xmath228 is a clique for each @xmath229 and so either @xmath230 or @xmath231 is a clique of size at least @xmath232 , contrary to the fact that @xmath207 . height3pt width6pt depth2pt let @xmath233 and let @xmath234 be a vertex with @xmath235 . let @xmath236 $ ] . since @xmath204 , @xmath237 is a clique of size @xmath238 . by , @xmath239 and so @xmath240 . we next show that [ e:2cliques ] ( ) for any @xmath241 with @xmath242 , either @xmath243 $ ] contains two vertex - disjoint @xmath214-paths or @xmath244 $ ] is a disjoint union of two cliques . suppose @xmath244 $ ] is not a disjoint union of two cliques . then @xmath244 $ ] is connected because @xmath204 . we next show that @xmath243 $ ] contains two vertex - disjoint induced @xmath214-paths . by , @xmath244 $ ] is not a clique and thus contains an induced @xmath214-path , say @xmath245 , with ends @xmath246 . let @xmath247 , where @xmath248 . clearly @xmath243 $ ] contains two vertex - disjoint induced @xmath214-paths if @xmath249 $ ] is not a clique , since @xmath250 is an edge for @xmath251 . so we may assume that @xmath249=k_s$ ] . first assume that @xmath167 is complete to @xmath252 . by , @xmath253 is not complete to @xmath254 . we may assume that @xmath255 . clearly @xmath256 $ ] and @xmath257 $ ] are two vertex - disjoint @xmath214-paths if @xmath258 for some @xmath259 . so we may assume that @xmath260 . now either @xmath261 $ ] and @xmath262 $ ] ( if @xmath263 ) or @xmath264 $ ] and @xmath265 $ ] ( if @xmath266 ) are two vertex - disjoint @xmath214-paths . next assume that @xmath167 is not complete to @xmath252 . we may assume that @xmath267 . then @xmath268 because @xmath204 . by symmetry , we may assume that @xmath263 . then @xmath269 . now either @xmath270 $ ] and @xmath271 $ ] ( if @xmath272 ) or @xmath273 $ ] and @xmath262 $ ] ( if @xmath274 ) are two vertex - disjoint @xmath214-paths , as desired . height3pt width6pt depth2pt [ e : q = t-2 ] ( ) @xmath275 . suppose @xmath276 . by theorem [ k7 ] , @xmath277 , which yields that @xmath278 and @xmath279 . then @xmath15 is a graph on thirteen vertices . clearly , @xmath280 . let @xmath281 be such that @xmath282 is maximum . since @xmath283 , we have @xmath284 . if @xmath285 , then @xmath286=k_5 $ ] and @xmath287 . clearly @xmath288 if @xmath289{\backslash}z$ ] has two vertex - disjoint induced @xmath214-paths . by , @xmath289{\backslash}z$ ] is thus a disjoint union of two cliques , say @xmath290 . by , we may assume that @xmath291 and @xmath292 . let @xmath293 . by again , @xmath167 is not complete to @xmath294 and thus @xmath295 , contrary to the fact that @xmath296 . thus @xmath297 . let @xmath298 be the non - neighbor of @xmath219 . by the choice of @xmath219 , every vertex in @xmath299 has at least one non - neighbor in @xmath237 and so @xmath300 . since @xmath301 , @xmath302 . by , @xmath303 is not a clique and so @xmath304 is adjacent to at least one vertex , say @xmath74 , in @xmath303 , because @xmath204 . now the edge @xmath305 is dominating @xmath237 , i.e. , every vertex in @xmath237 is adjacent to either @xmath219 or @xmath74 . notice that @xmath306 . if @xmath289{\backslash}\{z , w\}$ ] contains two vertex - disjoint induced @xmath214-paths , say @xmath307 and @xmath308 , then @xmath288 by contracting the edge @xmath305 and the two @xmath214-paths @xmath307 and @xmath308 into three distinct vertices , respectively , a contradiction . thus @xmath289{\backslash}\{z , w\}$ ] does not contain two vertex - disjoint induced @xmath214-paths . by , @xmath309 is a disjoint union of two cliques , say @xmath310 and @xmath311 . since @xmath312 , we have @xmath313 . by , @xmath314 $ ] is not a clique . let @xmath315 be such that either @xmath316 or @xmath317 . since @xmath318 is adjacent to at most three vertices of @xmath237 , we see that @xmath319 , contrary to the fact that @xmath296 . depth2pt by , @xmath275 . if @xmath320 , then @xmath15 is a graph on nine vertices with @xmath321 . thus there exists a vertex @xmath322 such that @xmath323 and so @xmath324 $ ] contains a @xmath325-subgraph because @xmath326 , contrary to . hence @xmath327 . now @xmath328 $ ] is a clique of size @xmath329 and @xmath330 . clearly @xmath205 if @xmath331 contains two vertex - disjoint induced 3-paths , a contradiction . thus by , @xmath299 is a disjoint union of two cliques , say @xmath221 and @xmath222 . for @xmath223 , let @xmath332 : v \text { is not adjacent to some vertex in } a_{3-i}\}$ ] . since @xmath204 , @xmath225 is complete to @xmath226 for each @xmath227 . thus @xmath333 is a clique for each @xmath229 and so at least one of them is of size at least @xmath334 , contrary to . this completes the proof of lemma [ k_tk1 ] . depth2pt + we are now ready to prove theorem [ main ] . * proof of theorem [ main ] . * suppose the assertion is false . let @xmath16 be a graph with no @xmath6 minor such that @xmath16 is not @xmath7-colorable . we may choose such a graph @xmath16 so that it is @xmath335-contraction - critical . let @xmath336 be of minimum degree . since @xmath195 and each @xmath190-cockade are @xmath337-colorable , and every @xmath338-cockade is @xmath4-colorable , it follows from theorem [ k7 ] , theorem [ k8 ] and theorem [ k9 ] that @xmath339 . on the other hand , since @xmath16 is @xmath335-contraction - critical , by lemma [ dirac](i ) applied to @xmath340 , we see that @xmath341 . clearly , @xmath342 , otherwise @xmath343 $ ] is a clique of size at least @xmath344 , contrary to the fact that @xmath16 has no @xmath6 minor . thus @xmath342 and so @xmath345 . then @xmath346 . by lemma [ dirac](i ) applied to @xmath340 again , we have @xmath347 . we next show that [ e : nok_t-1 ] ( ) @xmath16 has no @xmath348-subgraph . suppose @xmath16 contains @xmath348 as a subgraph . let @xmath349 be isomorphic to @xmath348 . since @xmath350 , every vertex in @xmath15 is adjacent to at least one vertex in @xmath351 . then @xmath351 is disconnected , since otherwise @xmath352 by contracting @xmath351 into a single vertex , a contradiction . let @xmath177 be a component of @xmath351 . then @xmath353 is a minimal separating set of @xmath16 . in particular , @xmath354 is a clique , contrary to lemma [ dirac](ii ) . depth2pt [ e : cnbr ] ( ) for any @xmath355 , @xmath356 . suppose that there exists a vertex @xmath355 such that @xmath357 . since @xmath347 , @xmath340 contains a clique of size @xmath358|\ge t-2 $ ] and so @xmath343 $ ] has a @xmath348-subgraph , contrary to . depth2pt by lemma [ k_tk1 ] , @xmath359 . let @xmath360 be such that @xmath361 . clearly , @xmath218 is not adjacent to every vertex in @xmath362 , otherwise @xmath363>k_t$ ] , a contradiction . let @xmath364 $ ] , where @xmath365|$ ] . then @xmath218 is not adjacent to @xmath366 . by and , @xmath289\cap n(x)$ ] is not a clique . let @xmath367 be a missing edge in @xmath299 . by lemma [ wonderful ] applied to @xmath340 with @xmath368 , @xmath369 and @xmath370 , there exists a path @xmath128 with ends @xmath218 and @xmath371 and all its internal vertices in @xmath64 $ ] for each @xmath372 . note that the paths @xmath373 have @xmath218 as a common end . by contracting all @xmath374 onto @xmath218 , we see that @xmath375 , a contradiction . the primary purpose of this section is to prove theorem [ maink8- ] which states that every graph with no @xmath9 minor is @xmath10-colorable . we need the following results . theorem [ thm : k567-extremal ] was proved by dirac @xcite for the cases @xmath376 , and by jakobsen @xcite for the case @xmath377 . [ thm : k567-extremal](dirac @xcite , jakobsen @xcite ) for @xmath378 , if @xmath16 is a graph with @xmath379 vertices and at least @xmath380 edges , then @xmath381 , or @xmath16 is a @xmath382-cockade when @xmath383 , or @xmath377 and @xmath16 is a @xmath384-cockade . jakobsen @xcite also conjectured that theorem [ thm : k567-extremal ] extends to @xmath385 , which was confirmed by the second author @xcite : [ thm : k8-extremal](song @xcite ) if @xmath16 is a graph with @xmath188 vertices and at least @xmath386 edges , then @xmath109 or @xmath16 is a @xmath387-cockade . the extremal function for @xmath388 minors remains open for @xmath389 . the following lemma from @xcite will also be needed . [ lem : delta5nbrhd](song @xcite ) let @xmath16 be a graph with @xmath390 and @xmath391 . then either @xmath392 or @xmath16 is isomorphic to one of @xmath393 , @xmath394 , @xmath395 , @xmath396 , @xmath397 , or @xmath237 , where @xmath237 is the graph depicted in figure [ fig : j ] . in particular , all of these graphs are edge maximal ( subject to not having a @xmath398-minor ) with maximum degree @xmath399 . moreover , @xmath400 , @xmath401 , and @xmath402 . .,width=150 ] notice that of the counterexamples listed in lemma [ lem : delta5nbrhd ] , only the graph @xmath237 has ten vertices , and none has exactly nine vertices . we first prove the following lemma . [ lem:10vertexalpha2 ] let @xmath16 be a graph with @xmath403 vertices and @xmath404 . then either @xmath392 , or @xmath16 contains a @xmath405 subgraph , or @xmath16 is isomorphic to the graph @xmath237 depicted in figure [ fig : j ] . if @xmath406 , then by lemma [ lem : delta5nbrhd ] , either @xmath392 or @xmath16 is isomorphic to @xmath237 . so we may assume that @xmath407 . let @xmath408 be such that @xmath409 . since @xmath404 , one can easily see that @xmath410 if @xmath411 . hence we may further assume that @xmath412 . then @xmath64 $ ] must be a @xmath413 as @xmath404 . if @xmath343 $ ] also induces a @xmath413 , then @xmath16 contains a @xmath405 subgraph . otherwise , some edge is missing from @xmath340 , say @xmath414 with @xmath415 . then since @xmath404 , each vertex in @xmath64 $ ] must be adjacent to either @xmath218 or @xmath219 . thus @xmath416 , as desired . this completes the proof of lemma [ lem:10vertexalpha2 ] . depth2pt we are now ready to prove theorem [ maink8- ] . let @xmath16 be a graph with no @xmath9 minor . suppose for a contradiction that @xmath417 . we may choose such a graph @xmath16 so that it is @xmath32-contraction - critical . let @xmath336 be of minimum degree . since @xmath16 is @xmath32-contraction - critical and has no @xmath9 minor , by lemma [ dirac](i ) applied to @xmath340 , we see that @xmath342 and @xmath418 . on the other hand , since every @xmath419-cockade is @xmath5-colorable , by theorem [ thm : k8-extremal ] we see that @xmath420 . thus @xmath421 . by lemma [ dirac](i ) applied to @xmath340 , we have [ e : alpha ] ( ) @xmath422 . we next show that [ e : noj ] ( ) @xmath340 is not isomorphic to the graph @xmath237 . suppose that @xmath340 is isomorphic to the graph @xmath237 . let the vertices of @xmath237 be labeled as depicted in figure [ fig : j ] . by lemma [ wonderful ] applied to @xmath237 with @xmath423 and @xmath424 with @xmath425 , @xmath426 , there exist paths @xmath427 , @xmath428 , @xmath429 , @xmath430 , @xmath431 such that the paths @xmath427 , @xmath428 , @xmath429 , @xmath430 , @xmath431 have ends @xmath432 , @xmath433 , @xmath434 , @xmath435 , and @xmath436 , respectively , and all their internal vertices in @xmath437 $ ] . moreover , the paths @xmath427 , @xmath428 , @xmath429 , @xmath430 are vertex - disjoint from the path @xmath431 . by contracting @xmath438 onto @xmath439 , @xmath440 onto @xmath441 , and @xmath442 $ ] into a single vertex , we see that @xmath443 , a contradiction . depth2pt [ e : k5uk5 ] ( ) @xmath340 contains @xmath405 as a subgraph . suppose that @xmath340 does not contain @xmath405 as a subgraph . then by , and lemma [ lem:10vertexalpha2 ] , we see that @xmath444 . let @xmath445 be a vertex such that @xmath446 . clearly , @xmath218 is not adjacent to every vertex in @xmath362 , otherwise @xmath363>k_8 ^ -$ ] , a contradiction . let @xmath364 $ ] , where @xmath447|\ge1 $ ] . then @xmath218 is not adjacent to @xmath366 . clearly , @xmath448 is not a clique for all @xmath372 . by lemma [ wonderful ] applied @xmath196 times to @xmath340 with @xmath449 , @xmath450 and @xmath98 ( where @xmath451 for @xmath452 ) , there exists a path @xmath128 between @xmath218 and @xmath371 with its internal vertices in @xmath64 $ ] for each @xmath372 . note that the paths @xmath453 have @xmath218 as a common end . by contracting each @xmath374 onto @xmath218 , we see that @xmath454 , a contradiction . depth2pt by , @xmath68 belongs to two different @xmath108-subgraphs of @xmath16 . by theorem [ 7con ] , @xmath16 is @xmath5-connected . by lemma [ lem:7conn2k6 ] , @xmath109 . this contradiction completes the proof of theorem [ maink8- ] . we prove theorem [ maink8= ] in this section . the following result will be needed . theorem [ thm : k8=extremal ] for the cases @xmath455 is due to dirac @xcite , and theorem [ thm : k8=extremal ] for the cases @xmath456 is due to jakobsen @xcite . [ thm : k8=extremal](dirac @xcite , jakobsen @xcite ) for integer @xmath196 with @xmath457 , every graph with @xmath379 vertices and at least @xmath458 edges either contains a @xmath459-minor or is a @xmath460-cockade . we are ready to prove theorem [ maink8= ] . suppose the assertion is false . let @xmath16 be a graph with no @xmath11 minor such that @xmath461 . we may choose such a graph @xmath16 so that it is @xmath10-contraction - critical . let @xmath336 be of minimum degree . since @xmath16 is @xmath10-contraction - critical and has no @xmath11 minor , by lemma [ dirac](i ) applied to @xmath340 , we see that @xmath342 and @xmath462 . on the other hand , since each @xmath463-cockade is @xmath464-colorable , it follows from theorem [ thm : k8=extremal ] for @xmath186 that @xmath465 . thus @xmath466 , and so @xmath467 . it follows from theorem [ thm : k8=extremal ] for @xmath186 again that [ e : k8=:n9 ] ( ) @xmath16 contains at least 28 vertices of degree 9 . since @xmath16 has no @xmath11 minor , by lemma [ dirac](i ) applied to @xmath340 , [ e : k8=:alpha ] ( ) @xmath422 . we next show that [ e : k8=:deltanbr4 ] ( ) @xmath340 contains @xmath413 as a subgraph . suppose that @xmath340 does not contain @xmath413 as a subgraph . then @xmath468 and by , @xmath469 . we claim that @xmath470 . suppose that @xmath471 . by lemma [ lem : delta5nbrhd ] applied to @xmath340 , we see that @xmath472 . let @xmath445 be such that @xmath473 . clearly @xmath218 has at least two non - neighbors in @xmath474 , otherwise @xmath343>k_8^=$ ] , a contradiction . let @xmath475 be all non - neighbors of @xmath218 , where @xmath476|\ge2 $ ] . since @xmath468 , @xmath477 must have a missing edge , say @xmath44 . by lemma [ wonderful ] applied to @xmath340 with @xmath478 and @xmath479 , there exist @xmath480 paths @xmath481 such that each path @xmath128 has ends @xmath482 and all its internal vertices in @xmath437 $ ] . by contracting all the edges of each @xmath483 onto @xmath218 for all @xmath484 , we see that @xmath485 , a contradiction . this proves that @xmath470 , as claimed . let @xmath445 be such that @xmath218 has degree four in @xmath340 with @xmath486 maximum . let @xmath487 be the set of all neighbors of @xmath218 in @xmath340 . since @xmath468 , @xmath289 \cap n(x)$ ] is not complete . we may assume that @xmath488 . by , @xmath489 $ ] induce a @xmath325 subgraph . let @xmath490 $ ] . we next show that ( @xmath491 ) each of @xmath492 has at most one neighbor in @xmath493 . suppose , say @xmath494 , is adjacent to at least two vertices in @xmath493 . then the subgraph induced on @xmath495 has a @xmath496 minor and thus @xmath343>k_8^=$ ] if @xmath497 is adjacent to all vertices in @xmath493 ( by contracting the path @xmath498 into a single vertex ) , a contradiction . thus we may assume that @xmath497 is not adjacent to @xmath499 , where @xmath500 . by lemma [ wonderful ] applied to @xmath340 with @xmath501 and @xmath502 , there exist @xmath227 paths @xmath503 such that for each @xmath504 , the path @xmath505 has ends @xmath506 and all its internal vertices in @xmath437 $ ] . by contracting all @xmath507 onto @xmath497 and the path @xmath508 into a single vertex , we see that @xmath509 , a contradiction . this proves ( @xmath491 ) . depth2pt we next claim that @xmath289\cap n(x)=k_5 ^ -$ ] . suppose @xmath510 . by symmetry , we may apply ( @xmath491 ) to the missing edge @xmath511 in @xmath340 , and so we see that each of @xmath512 has at most one neighbor in @xmath493 . hence @xmath513 . on the other hand , since @xmath347 , each @xmath514 must be adjacent to at least one of the vertices in @xmath515 and @xmath516 , respectively , for all @xmath517 . thus @xmath518 , a contradiction . this proves that @xmath519 and thus @xmath340 does not have two independent missing edges . next if @xmath520 , then @xmath521 and @xmath289\cap n(x)=k_5^=$ ] . since @xmath468 , we may assume that @xmath522 . then @xmath523 must be adjacent to both @xmath524 and @xmath497 by . by applying ( @xmath491 ) to the missing edges @xmath525 and @xmath526 , we see that @xmath527 is anti - complete to @xmath528 and @xmath494 has at most one neighbor in @xmath493 . by ( 2 ) , @xmath529 is complete to @xmath530 . since @xmath494 has at most one neighbor in @xmath493 , we may assume that @xmath531 . now @xmath532 has degree four in @xmath340 with @xmath533\cap n(x)=k_5 ^ -$ ] , contrary to the choice of @xmath218 . n(x)=k_5 ^ -$ ] with @xmath534 the only missing edge , as claimed . since @xmath535 , each of @xmath529 and @xmath524 has at least one neighbor in @xmath493 . by , each of @xmath536 is adjacent to at least one of @xmath537 , and so either @xmath529 or @xmath524 has at least two neighbors in @xmath493 . by symmetry , we may assume that @xmath529 has more neighbors in @xmath493 than @xmath524 . on the other hand , each vertex in @xmath538 has at least one non - neighbor in @xmath493 as @xmath539 . thus , @xmath529 has either two or three neighbors in @xmath493 . we consider the following two cases . has exactly two neighbors in @xmath493.,width=225 ] first , assume that @xmath529 has exactly two neighbors in @xmath493 , say @xmath540 . then @xmath524 must have exactly two neighbors @xmath541 in @xmath493 . by ( @xmath491 ) , each of @xmath492 has at most one neighbor in @xmath493 . we may assume that @xmath494 is not adjacent to @xmath541 , and @xmath542 . by lemma [ wonderful ] applied twice to @xmath340 with @xmath501 and @xmath543 , there exist three vertex - disjoint paths @xmath544 and two vertex - disjoint paths @xmath545 such that the paths @xmath307 , @xmath308 , @xmath546 , @xmath547 , and @xmath548 have ends @xmath549 , @xmath550 , @xmath551 , @xmath552 and @xmath553 , respectively , and all their internal vertices in @xmath437 $ ] , as depicted in figure [ fig : case1 ] . notice that each @xmath128 is vertex - disjoint from @xmath554 for all @xmath555 and @xmath556 , @xmath307 and @xmath308 are vertex - disjoint but @xmath546 and @xmath308 are not necessarily vertex - disjoint . if @xmath546 and @xmath308 have only @xmath557 in common , then contracting @xmath558 onto @xmath218 , @xmath559 onto @xmath497 , and @xmath560 onto @xmath494 , and @xmath561 into two distinct vertices yields a @xmath11 minor in @xmath16 , a contradiction . thus @xmath546 and @xmath308 must have an internal vertex in common . let @xmath74 be the first vertex on @xmath546 ( when @xmath546 is read from @xmath218 to @xmath557 ) that is also on @xmath308 . then @xmath562 . let @xmath563 be the subpath of @xmath546 from @xmath218 to @xmath74 , @xmath564 be the subpath of @xmath308 from @xmath74 to @xmath557 . notice that @xmath565 is vertex - disjoint from @xmath308 but not necessarily internally disjoint from @xmath307 . now contracting @xmath566 , @xmath567 onto @xmath218 , @xmath568 onto @xmath557 , @xmath569 onto @xmath497 , @xmath560 onto @xmath494 , and @xmath561 into two distinct vertices yields another @xmath11 minor in @xmath16 , a contradiction . has exactly three neighbors in @xmath493.,width=225 ] it remains to consider the case when @xmath529 has exactly three neighbors , say @xmath570 in @xmath493 . then @xmath524 is adjacent to @xmath532 . by ( @xmath491 ) , we may assume that @xmath523 is not adjacent to @xmath492 , and @xmath571 . by lemma [ wonderful ] applied twice to @xmath340 with @xmath501 and @xmath572 , there exist vertex - disjoint paths @xmath573 and another path @xmath574 such that the paths @xmath575 have ends @xmath551 , @xmath576 , @xmath577 , and @xmath578 , respectively , and all their internal vertices in @xmath437 $ ] , as depicted in figure [ fig : case2 ] . notice that @xmath307 is vertex - disjoint from @xmath554 for @xmath579 , but that @xmath574 is not necessarily internally vertex - disjoint from either @xmath547 or @xmath548 . if @xmath574 and @xmath548 have only @xmath494 in common , then we obtain a @xmath11 minor by contracting @xmath307 and @xmath580 into two distinct vertices , @xmath581 , @xmath582 onto @xmath523 , and @xmath583 onto @xmath584 , a contradiction . thus @xmath574 and @xmath548 must have an internal vertex in common . let @xmath74 be the first vertex on @xmath574 ( when @xmath574 is read from @xmath523 to @xmath494 ) that is also on @xmath548 . then @xmath585 . let @xmath586 be the subpath of @xmath574 from @xmath523 to @xmath74 , @xmath587 be the subpath of @xmath548 from @xmath74 to @xmath494 . notice that @xmath588 is vertex - disjoint from @xmath548 but not necessarily internally disjoint from @xmath547 . now we obtain another @xmath11 minor by contracting @xmath307 and @xmath580 into two distinct vertices , @xmath581 , @xmath589 onto @xmath523 , @xmath590 onto @xmath494 , and @xmath591 onto @xmath584 , a contradiction . this completes the proof of . depth2pt by , every vertex of degree 9 belongs to some @xmath108-subgraph of @xmath16 . by , @xmath16 contains at least five different @xmath108-subgraphs . by theorem [ 7con ] , @xmath16 is @xmath5-connected and thus @xmath485 by lemma [ lem:7conn2k6 ] . this contradiction completes the proof of theorem [ maink8= ] . it seems very difficult to prove that every graph with no @xmath3 minor is @xmath5-colorable . we establish in @xcite the properties of @xmath12-contraction - critical graphs with no @xmath3 minor to shed some light on this open problem . as pointed out by robin thomas ( personal communication with the second author ) , one might be able to settle this open problem by using the key ideas in @xcite . however , it seems very hard to prove that every @xmath12-contraction - critical graph is @xmath12-connected , which means the truth of the seymour - thomas conjecture ( i.e. conjecture [ conj2 ] ) can not be applied easily to solve hadwiger s conjecture . theorem [ k7 ] is such a nice result . we believe that mader s bound for the extremal function for @xmath13 minors is true as follows : by theorem [ k9 ] , conjecture [ conj3 ] is true for @xmath201 . as mentioned earlier , lemma [ wonderful ] turns out to be very powerful . we believe that the application of lemma [ wonderful ] that we have developed in this paper is of independent interest . to end this section , we apply lemma [ wonderful ] along with a new idea ( namely , considering the chromatic number of @xmath340 ) to prove that the truth of conjecture [ conj3 ] implies that every graph with no @xmath13 minor is @xmath594-colorable for all @xmath14 . since conjecture [ conj3 ] is true for @xmath201 , we see that theorem [ final ] implies theorem [ main ] . suppose the assertion is false . by theorem [ k9 ] , we have @xmath595 . among all minimum counterexamples , we choose @xmath16 so that @xmath16 has no @xmath13 minor and @xmath16 is @xmath596-contraction - critical . let @xmath408 be such that @xmath597 . by the assumed truth of conjecture [ conj3 ] , @xmath598 . on the other hand , by lemma [ dirac](i ) applied to @xmath340 , we see that @xmath342 and @xmath599 . hence @xmath600 . by lemma [ dirac](i ) again , we have suppose that @xmath603 . let @xmath604 $ ] be isomorphic to @xmath605 . since @xmath606 , every vertex in @xmath15 is adjacent to @xmath607 vertices in @xmath351 . then @xmath351 is disconnected , otherwise @xmath352 by contracting @xmath351 into a single vertex , a contradiction . let @xmath177 be a component of @xmath351 . then @xmath353 is a minimal separating set of @xmath16 , in particular , @xmath354 is a clique , contrary to lemma [ dirac](ii ) . this proves that @xmath601 . by , we see that @xmath602.height3pt width6pt depth2pt suppose to the contrary that @xmath609 . by , it is clear that @xmath610 . thus @xmath611 for some @xmath612 . let @xmath613 be the color classes of any proper @xmath1-coloring of @xmath340 . we may assume that the color classes are ordered so that @xmath614 for @xmath615 and @xmath616 for @xmath617 . let @xmath618 . notice that @xmath619 induces a clique in @xmath340 , and so @xmath620 by . by , @xmath60 is adjacent to either @xmath621 or @xmath622 for each @xmath623 and each @xmath624 . notice that if @xmath625 , then @xmath626 . suppose that @xmath627 or that @xmath625 and @xmath628 have a common neighbor in @xmath629 , say @xmath630 . by lemma [ wonderful ] applied to @xmath340 with @xmath631 and @xmath632 , there exist @xmath633 pairwise vertex - disjoint paths @xmath634 such that each @xmath505 has ends @xmath635 and all its internal vertices in @xmath636 $ ] . by contracting each @xmath505 to a single vertex for all @xmath637 , together with @xmath638 if @xmath639 ; and together with @xmath640 if @xmath625 where @xmath641 is a common neighbor of @xmath628 , we obtain a clique minor on @xmath642 vertices in the former case and @xmath643 vertices in the latter case , a contradiction . thus @xmath625 and @xmath628 have no common neighbor in @xmath629 . since each of @xmath628 is adjacent to either @xmath645 or @xmath646 , by symmetry , we may assume that @xmath645 is adjacent to @xmath101 and @xmath647 , but not to @xmath648 . then @xmath646 is adjacent to @xmath648 . we may assume that @xmath646 is not adjacent to @xmath101 . for the worst case scenario , we may further assume that @xmath646 is not adjacent to @xmath647 . by lemma [ wonderful ] applied twice to @xmath340 with @xmath649 and @xmath650 , there exist pairwise vertex - disjoint paths @xmath651 such that each @xmath505 has ends @xmath635 and all its internal vertices in @xmath636 $ ] ; and paths @xmath652 with ends @xmath653 , @xmath654 , @xmath655 and @xmath656 , respectively , and all their internal vertices in @xmath636 $ ] . notice that each @xmath505 is vertex - disjoint from @xmath652 , and @xmath132 is vertex - disjoint from @xmath657 but not necessarily from @xmath658 . let @xmath74 be the first vertex on @xmath132 ( when read from @xmath648 to @xmath645 ) that is also on @xmath658 . note that @xmath74 could be @xmath645 . let @xmath659 be the subpath of @xmath132 between @xmath74 and @xmath648 , as depicted in figure [ fig : finala1a2a3 ] . by contracting each @xmath505 to a single vertex for all @xmath660 , @xmath661 and @xmath662 onto @xmath646 , @xmath663 onto @xmath648 , and @xmath664 onto @xmath645 , together with the vertices @xmath665 , we obtain a @xmath13 minor , a contradiction . depth2pt by and , we have @xmath667 . suppose there exists a vertex @xmath360 such that @xmath218 has exactly @xmath668 neighbors in @xmath340 . then @xmath489 $ ] is a clique of @xmath340 with @xmath668 vertices . furthermore , by , @xmath669 must have some missing edge , say @xmath44 . then every vertex of @xmath489 $ ] is adjacent to at least one of @xmath133 and @xmath135 . thus , by contracting the path @xmath670 to a single vertex , we can see that @xmath671 . now we can assume without loss of generality that @xmath445 is such that @xmath672 . clearly @xmath218 is not adjacent to every vertex in @xmath673 , or else @xmath674 > k_p$ ] , a contradiction . let @xmath675 $ ] , where @xmath0 . again , by , @xmath299 must have some missing edge , say @xmath44 . by lemma [ wonderful ] applied to @xmath340 with @xmath676 and @xmath677 , there exist paths @xmath678 such that each @xmath128 has ends @xmath482 and all internal vertices in @xmath636 $ ] . now by contracting each @xmath679 onto @xmath218 , we see that @xmath680 , a contradiction . depth2pt + by and , @xmath340 does not contain @xmath681 as a subgraph and @xmath666 . by , @xmath682 . let @xmath683 be the color classes of any proper @xmath684-coloring of @xmath340 . we may assume that the color classes are ordered so that @xmath685 and @xmath616 for @xmath686 . by , @xmath101 must be complete to some color class @xmath687 , say @xmath688 . by again , @xmath647 and @xmath689 must have one common neighbor in some color class @xmath690 , say @xmath691 . we may further assume that @xmath648 is adjacent to both @xmath692 . by symmetry , we may assume that @xmath693 is adjacent to @xmath647 . by lemma [ wonderful ] applied up to @xmath340 with @xmath694 and @xmath695 , there exist paths @xmath696 and pairwise vertex - disjoint paths @xmath697 such that @xmath696 have ends @xmath698 and @xmath699 , respectively ; each @xmath554 has ends @xmath635 and all such paths have their internal vertices in @xmath636 $ ] . by contracting @xmath700 onto @xmath693 , the edge @xmath701 onto @xmath648 , and each @xmath554 into a single vertex for @xmath702 , we see that @xmath680 , a contradiction . b. albar , d. gonalves , on triangles in @xmath703-minor free graphs , arxiv:1304.5468 . g. chen , z. hu , f. song , a new connectivity bound for linkages and its application to the hadwiger s conjecture , submitted . g. a. dirac , a property of @xmath464-chromatic graphs and some remarks on critical graphs , j. london math . 27 ( 1952 ) 8592 . g. a. dirac , homomorphism theorems for graphs , math . ( 1964 ) 6980 . g. a. dirac , trennende knotenpunktmengen und reduzibilitt abstrakter graphen mit anwendung auf das vierfarbenproblem , j. reine agew . 204 ( 1960 ) 116131 . h. hadwiger , ber eine klassifikation der strechenkomplexe , vierteljschr . ges zrich 88 ( 1943 ) 133142 . i. t. jakobsen , a homomorphism theorem with an application to the conjecture of hadwiger , studia sci . ( 1971 ) 151160 . i. t. jakobsen , on certain homomorphism properties of graphs i , math . 31 ( 1972 ) 379404 . i. t. jakobsen , on certain homomorphism properties of graphs ii , math . 52 ( 1983 ) 229261 . l. k. jrgensen , contractions to @xmath31 , j. graph theory 18 ( 1994 ) 431448 . k. kawarabayashi , b. toft , any @xmath5-chromatic graph has a @xmath3 or @xmath704 as a minor , combinatorica 25 ( 2005 ) 327353 . k. kawarabayashi , on the connectivity of minimum and minimal counterexamples to hadwiger s conjecture , j. combin . theory ser . b 97 ( 2007 ) 144150 . k. kawarabayashi , g. yu , connectivities for @xmath38-knitted graphs and for minimal counterexamples to hadwiger s conjecture , j. combin . theory ser . b 103 ( 2013 ) 320326 . w. mader , homomorphiestze fr graphen , math . ( 1968 ) 154168 . ber trennende eckenmengen in homomorphiekritischen graphen . ( 1968 ) 243252 . n. robertson , p. seymour and r. thomas , hadwiger s conjecture for @xmath108-free graphs , _ combinatorica * 13 * _ ( 1993 ) 279361 . m. rolek , z - x . song , properties of @xmath12-contraction - critical graphs with no @xmath3 minor , in preparation . p. seymour , hadwiger s conjecture , manuscript . song , the extremal function for @xmath9 minors , j. combin . theory ser . b 95 ( 2005 ) 300317 . z - x . song , r. thomas , the extremal function for @xmath193 minors . j. combin . theory , ser . b , 96 ( 2006 ) 240252 . b. toft , _ a survey on hadwiger s conjecture _ , in : _ surveys in graph theory _ ( edited by g. chartrand and m. jacobson ) , congr . 115 ( 1996 ) , 249283 . k. wagner , ber eine eigenschaft der ebenen komplexe , math . ( 1937 ) 570590 .

the problem of the fermi edge singularity in a one dimensional tomonaga luttinger liquid is reconsidered . the backward scattering of the conduction band electrons on the impurity like hole in the valence band is analyzed by mapping the problem onto a coulomb gas theory . for the case when the electron electron interaction is repulsive the obtained exponent of the one dimensional fermi edge singularity appears to be different from the exponent found in the previous studies . it is shown that the infrared physics of the fermi edge singularity in the presence of backward scattering and electron electron repulsion resembles the physics of the kondo problem . specialpagefalse oddhead evenheadoddhead

cond-mat9601021

large optical singularities in the absorption and emission spectra have been observed by calleja et . @xcite in semiconductor quantum wire structures in the extreme quantum limit when only the lowest one dimensional ( 1d ) subband is occupied by electrons . the pronounced features in the optical spectra have been interpreted as strong fermi edge singularities ( fes ) of the 1d electron gas . the fes arises ( e.g. for absorption ) , because the optical process is accompanied by a multiple scattering of the conducting electrons on a hole in the valence band , which is created at the absorption of the electro magnetic wave and acts as an impurity like center . in the case of 1d geometry the scattering of the conducting electrons by the hole is limited to scattering in the forward or backward directions . another special feature is related to the electron electron interaction in 1d . in contrast to higher dimensions , the 1d electron gas can not be treated as an ordinary fermi liquid even when the electron electron interaction is small . the experiment @xcite initiated a number of papers discussing the fes in 1d metallic systems for the forward scattering @xcite and the backward one @xcite in the presence of the electron electron interaction . in the present paper the effect of the backward scattering on the fes in 1d will be reconsidered . the exponent of the 1d fes obtained here appears to be different from the one that was found in ref.s @xcite . it is shown that the infrared physics of the fermi edge singularity in the presence of backward scattering together with electron electron repulsion resembles the physics of the kondo problem . this aspect of the problem was missed in the preceding studies . fes or mahan singularity is a power law singularity in the electro magnetic wave absorption ( or emission ) coefficient when the frequency is close to the fermi energy @xcite . that is in contrast to a naive expectation that the absorption coefficient is zero bellow the fermi energy and is proportional to the density of states of the conduction band above the fermi energy . since in common metals the fes is observed in the @xmath0ray range , it is also called sometimes the @xmath0ray singularity . the singularity arises because the absorption is accompanied by the infrared catastrophe phenomena . when the electro magnetic wave is absorbed , an electron from a deep level is excited to the conduction band and , correspondingly , a hole is created deep in the valence band . the abrupt appearance of the scattering potential of the deep hole leads to excitation of an infinite number of low energy electron hole pairs in the conduction band . besides , the electron excited from the deep level to the conduction band scatters multiply on the hole . the latter process , which includes an exchange with other electrons of the conduction band , leads to enhancement of the absorption coefficient . the physical reason of this enhancement is the increase of the effective density of states near the deep hole due to its attraction of electrons . on the other hand the creation of the electron hole pairs in the conduction band reflects the fact that the final state of the electrons in the presence of the deep hole potential is orthogonal to the initial state . this process is called the orthogonality catastrophe of anderson @xcite and it leads to a reduction of the absorption coefficient . the value of the exponent of the fes is determined by these two competing effects . the effects related to the physics of the fes can be found in various situation where a sudden change of a scattering potential happens . the well known example is the kondo problem describing scattering of conduction electrons on a magnetic impurity . when the impurity spin is flipped the impurity potential on which the conduction electrons are scattered is abruptly altered . therefore , the kondo problem can be treated as a response of the electrons on a sequence of sudden changes of the local scattering potential . the problem can be mapped on a classical coulomb gas theory of alternating positive and negative charges ( spin flips ) with logarithmic interaction . in the present paper it will be shown that the fes problem in 1d can also be analyzed by mapping it on a coulomb gas theory , similarly to the kondo problem . to calculate the exponent of the fes it is needed to sum a series of infrared divergent logarithmic terms . a corresponding theory was developed for three dimensional metals ( 3d ) by nozires and collaborators in a series of papers @xcite . the recoil of the hole , due to its finite mass , together with auger processes lead to a natural cutoff of the infrared singularities . it is usually assumed , for simplicity , at the discussion of the fes that these effects can be neglected . the other common assumption is that the potential created by the deep hole is a short range one . that is because the electron screening develops on short scales , while at studies of infrared phenomena large distances are essential . since in 3d metals the fermi liquid parameters are determined by the short scale physics , it is also generally assumed that one can use the fermi liquid description for the conduction electrons at the consideration of the fes . at studies of the fes in 1d the role of these common assumptions may change . it has been noticed @xcite that in 1d systems the recoil of the deep hole does not suppress the fes , at least for the forward scattering . here we will ignore the recoil effects completely as well as the auger processes . in this paper we will also not discuss the problem of screening of the potential of the deep hole , despite that it may be essential for the fes in modulation doped quantum wires . since we are mainly interested now in the effects of the @xmath1backward scattering , it will be assumed that the potential of the deep hole is local . the most intriguing aspect of fes in 1d is related to the role of the electron electron interaction . in some sense the fermi surface is weakened in 1d as a result of the interaction of electrons . in the case of the tomonaga luttinger model the jump in the occupation numbers of electrons at the fermi energy is replaced by a singularity , and there are no single electron quasi particles @xcite . since the fes is a fermi surface effect , the investigation of the role of the electron electron interaction in this problem is of clear interest . the effective way to investigate the infrared properties of the 1d electron systems is the bosonization technique . in fact , the important step in the developing of this technique was made just in connection with the problem of the fes . schotte and schotte extended the approach of ref . @xcite and developed a very compact and economical description of the fes for @xmath2-wave scattering in 3d @xcite . in the present paper the consideration of the electrons interaction will be confined to the tomonaga luttinger model where it is assumed that the electrons do not scatter backward in the course of the electron electron interactions . since tomonaga luttinger model has a simple solution in the bosonization technique , it is natural to apply that technique for finding the fes in 1d in the presence of interactions between the electrons . in the case when at the scattering on the potential of the hole in the valence band the electrons are scattered only in the forward direction , the fes can be found @xcite following the schotte and schotte method . for completeness of the presentation and with the purpose of introducing notations we have reconsidered this case in sec . [ se : fs ] . when the backward scattering on the hole potential is included , it is shown in sec . [ se : bs ] that the problem can be mapped on a gas of charged particles , similarly to the coulomb gas mapping in the kondo problem . the role of the spin flip in the kondo problem acts now the change of the electron motion from left to right and visa versa at the backward scattering on the hole . in comparison with the kondo problem the coulomb gas describing the 1d fes is different in two aspects . in the @xmath3 kondo problem successive flips should follow alternatively , while in the case of fes there is no limitations on the order of left right and right left scattering . ( in the end of sec . [ se : ri ] a certain relationship of the non alternating coulomb gas with the two channel kondo problem is outlined . ) the other difference is due to the creation of the additional electron in the conduction band at the absorption of the electro magnetic wave . this gives rise to two additional charges located at the ends of the coulomb gas system . in the case of repulsion of electrons the gas of the charged particles is in the hot plasma phase . therefore a new scale the screening radius of the plasma is generated in this problem as a combined effect of the backward scattering and the interaction . the shape of the absorption line for frequencies close enough to the fermi edge is controlled by this scale . the reduction of the 1d fes problem to the classical gas theory yields a value of the fes exponent in the vicinity of the fermi edge , which is different from the one obtained in preceding studies @xcite . in section [ se : nl ] the fes is analyzed using fields , which where first introduced in ref . @xcite . in that representation the scattering is simplified , but at a price that the electron electron interaction is described by a non local theory of self dual fields . a special treatment is needed in case of such a theory . the renormalization group analysis of the problem using these variables is elaborated in sec . [ se : rg ] . in addition , an iteration procedure is developed in sec . [ se : qtsp ] for the analysis of the fes problem . this procedure is useful for determining the screening length of the coulomb gas in the hot plasma phase . to understand the obtained exponent of the fes , in sec . [ se : ri ] the result is interpreted in terms of the phase shift theory of nozires and de dominicis @xcite . it is argued that the discrepancy between the value of the fes exponent obtained in the present paper and the result of ref.s @xcite is because in these papers the asymptotic regime of the problem is described by a weak link junction . however , two weakly connected wires represent a system with strong depletion of the density of electrons at the defect center . it is shown in the end of sec . [ se : ri ] that the use of such system as the asymptotic limit description of a weak impurity scatterer , as it was suggested in the theory of ref . @xcite , is in contradiction with the friedel sum rule . in fact , it is concluded here that the infrared asymptotic regime of the fes resembles to the kondo resonance . in this section we show that in 1-d systems the correlation function @xmath4 determining ( after the fourier transform ) the fermi edge singularity can be presented as a product of two terms . the first one is related to the forward scattering of the conduction electrons by the hole , which is created in the valence band at the absorption of the external electro magnetic wave . this term can be calculate directly , while the second one describing the backward scattering by the hole , is analyzed by mapping onto a coulomb gas . this approach allows us to find the behavior of @xmath4 at asymptotically large time . for compactness we confine the consideration bellow to a spinless case , and the spin degrees of freedom of the conduction electrons are included in the appendix ( [ app ] ) . the hamiltonian describing the 1-d electron liquid is given by @xmath5 where @xmath6 are the field operators of fermions that propagate to the right with wave vectors @xmath7 , and @xmath8 are the field operators of left propagating fermions with wave vectors @xmath9 ; @xmath10 are the electron density operators ; the spectrum of the electrons is linearized near the fermi points and @xmath11 is the fermi velocity ; @xmath12 describes the density density interaction with momentum transfers much smaller than @xmath13 . hamiltonian ( [ eq : ffh ] ) corresponds to the tomonaga luttinger model , which describes the 1-d electron liquid when the backward scattering amplitude of the electron - electron interaction may be ignored . ( [ eq : ffh ] ) is a fixed point hamiltonian for a broad class of 1-d systems . after the absorption of the external electro magnetic wave a hole is created in the valence band together with an additional electron in the conduction band . it will be assumed hereafter that the position of the hole is fixed at @xmath14 . the scattering of the conduction band electrons by this hole is given by @xmath15 where @xmath16 are the fourier transform amplitudes of the hole potential @xmath17 , and @xmath18 . thus , @xmath19 is the forward scattering amplitude and @xmath20 is the backward one . in order to find the electro magnetic wave absorption line shape @xmath21 one needs to calculate the fourier transform of the correlation function @xcite @xmath22 where @xmath23 refers to an electron which is created in the conduction band at the absorption , and @xmath24 is the hamiltonian describing the electron liquid after the creation of the hole . it has been shown in the seminal paper of schotte and schotte @xcite that bosonic representation of the fermion operators in eq . ( [ eq : f ] ) gives an illuminating approach for understanding the fes in the case of @xmath25wave scattering by the hole . it is natural to apply this approach for the analysis of the fes in 1-d conductors . it is well known @xcite that the bosonization technique ( for review see @xcite ) allows to reduce the tomonaga luttinger hamiltonian to a quadratic form in terms of operators of bosonic fields @xmath26 and @xmath27 : [ eq : dpd ] @xmath28,\\ \tilde\phi_0(x)&=&\frac{-i}{\sqrt{4\pi } } \frac{2\pi}{l } \sum_p \frac{\exp\left({-\eta |p|/2-ipx}\right)}{p } \left [ \rho_r(p)-\rho_l(p ) \right ] , \end{aligned}\ ] ] where @xmath29 is an ultraviolet cutoff , which is of the order of the conduction band width , and @xmath30 is the system length . the fields @xmath26 and its dual partner @xmath27 are conjugate variables , i.e. , @xmath31=i\bbox{\delta}(x - y).\ ] ] after rescaling the operators @xmath32 the bosonized representation of hamiltonian ( [ eq : ffh ] ) becomes @xmath33 where @xmath34 in the bosonization technique @xcite the bosonic representations of the operators @xmath35 and @xmath36 are given as [ eq : fermion ] @xmath37 } \\ \psi^b_l(x)&= & i\frac { e^{-ik_fx } } { \sqrt{2\pi\eta } } \exp{\left[- \frac{i}{2 } \left ( \frac { 4\pi } { \beta } \tilde\phi - \beta \phi \right ) \right]}.\end{aligned}\ ] ] then , the scattering by the hole may be written in terms of the @xmath38-field as @xmath39 where the dimensionless amplitudes @xmath40 are introduced and @xmath41 . in eq . ( [ eq : bih ] ) the term related to the forward scattering is linear in @xmath38 . therefore , a canonical transformation that shifts at the point @xmath14 the operator @xmath42 can be exploited to exclude the forward scattering . since @xmath43 the transformed hamiltonian @xmath44 will not contain the forward scattering term , if @xmath45 . in the absence of the backward scattering the correlation function @xmath4 , after carrying out the unitary transformation @xmath46 , becomes : @xmath47 the calculation of @xmath48 can be reduced to a gaussian like integral if the bosonic representation ( [ eq : fermion ] ) for the operators @xmath49 , @xmath50 is applied . let us choose the component @xmath51 for the operator @xmath49 ; the contribution of the other component is equal . then , the use of the bosonic representation gives : @xmath52 where @xmath53 , and the operators @xmath54 and @xmath55 depend on time as @xmath56 . now , the standard application of the baker - hausdorff formula @xmath57 } $ ] together with the gaussian averaging @xmath58 yield @xmath59 where @xmath60 is the green function of the @xmath38operators , and @xmath61 is defined similarly . it has also been assumed here that @xmath62 , what agrees with eqs . ( [ eq : dpd ] ) . from eqs . ( [ eq : dpd ] ) and ( [ eq : bfh ] ) it can be obtained for asymptotically large time @xmath63 finally one gets : @xmath64 and correspondingly the absorption line shape @xmath65 , where @xmath66 and @xmath67 here the first term reflects the luttinger liquid behavior of the 1d electrons in the presence of an electron electron interaction @xcite . note that contrary to the noninteracting case the last term in eq . ( [ eq : fw ] ) , which corresponds to the anderson orthogonality catastrophe , is not the half of the square of the second term . the above procedure was elaborated by schotte and schotte @xcite for consideration of the fes in 3-dimensional case for @xmath2-wave scattering . for the 1d case the function @xmath68 , including the spin degrees of freedom , has been obtained recently in ref . @xcite ( see bellow eq . ( [ eq : betas ] ) for comparison ) . in the presence of the backward scattering the correlation function @xmath4 will be studied in the interaction representation with respect to @xmath69 : @xmath70 where @xmath71 is the time ordering symbol and the backward scattering term is @xmath72 ( the action of the electron electron interaction , unlike the backward scattering term , is not limited to the interval @xmath73 $ ] . therefore , electron transitions not only within this time interval contribute to the fes . fortunately , in the case of the tomonaga luttinger model the renormalization of the fields @xmath38 and @xmath74 by eq . ( [ eq : rphi ] ) takes that complications into consideration , and eq . ( [ eq : fint ] ) holds . ) when an expansion of the @xmath71-exponential in @xmath75 is performed , the result can be written as a sum of all possible time ordered products of the exponentials of the operators @xmath76 $ ] . the averaging procedure of each term in this sum can be easily performed using repeatedly the baker hausdorff formula . this is a standard way of a reduction of the quantum problem of such kind to the physics of classical charged gases ; see for example refs . . now a subtle point related to the application of the baker hausdorff formula should be commented . eq . ( [ eq : cr ] ) determines the commutation relations of @xmath38 and @xmath77 only up to a constant : @xmath78= i\int^x \bbox{\delta}(x^\prime)dx^\prime = i\frac 12 \mbox { sign } ( x ) + c.\ ] ] this constant may be fixed by the requirement that the representation ( [ eq : fermion ] ) should ensure the fermion commutation relations of the operators @xmath79 and @xmath80 . that requirement leads to @xmath81 . in eq . ( [ eq : bih ] ) the cosine of the backward scattering term has been written with @xmath82 . now , using the fact that in @xmath4 the creation and annihilation operators of the right and left moving electrons appear in equal numbers for non vanishing terms , one can check in a formal way that @xmath83 does not influence @xmath4 . for that reason it will be assumed hereafter that @xmath84 = 0,\ ] ] what is in correspondence with a naive treatment of this commutator with representation ( [ eq : dpd ] ) . consequently , when the gaussian averaging is performed in the calculation of the correlation function @xmath4 it may be assumed that @xmath85 usually the correlation function of @xmath38 and @xmath74 operators controls the relative phase of certain operators when they exchange their coordinates , because @xmath86 in some problems the angle appearing in this equation has been interpreted as the aharonov bohm phase @xcite . the peculiarity of the problem at hand is in the fact that the operators in the correlation function @xmath4 are at a single point @xmath14 . consequently , in the @xmath87 plane all the operators stay along a straight line , and for that reason the aharonov bohm phases can be chosen to be zero . finally this procedure yields @xmath88 where @xmath89}}&.\end{aligned}\ ] ] here @xmath90 , while @xmath91 describe the contribution of the operators @xmath23 and @xmath49 , and @xmath92 means that only neutral configurations with @xmath93 are allowed . the signs of @xmath94 and @xmath95 depend on a particular combination of the operators @xmath96 and @xmath97 in eq . ( [ eq : fint ] ) . there are altogether four terms . if a diagonal term @xmath98 is considered , then @xmath99 ( @xmath100 ) . however , if @xmath101 is averaged , then @xmath102 ) and here non - vanishing configurations should contain a compensation charge from @xmath103 . these compensating charges give rise to factors @xmath104 in the odd power terms of @xmath105 . all the four combinations have the same dependence on the @xmath106-field . since , as it has been discussed , the fields @xmath107 and @xmath108 do not interfere , @xmath4 factorizes in the form given by eq . ( [ eq : f1 ] ) . the fields @xmath74 and @xmath38 are related correspondingly to the forward and the backward scattering channels . the factorization implies a decoupling of these channels . another way to prove this decoupling is a perturbative analysis , in which both the backward and the forward scattering terms are treated as perturbations . then the decoupling is a direct consequence of the vanishing of @xmath109 at @xmath110 . the independence of the forward and the backward scattering processes has been also realized in a different way in refs . @xcite . the asymptotic behavior of @xmath105 at large time is analyzed by performing analytical continuation to the euclidean time @xmath111 . then , @xmath112 acquires a clear physical interpretation it becomes a grand partition function of a classical gas of particles staying on a line of the length @xmath113 and interacting via logarithmic coulomb potentials @xmath114 is the fugacity of this gas , while @xmath115 acts as an effective temperature . in addition to the gas particles two half charged particles are fixed to the ends of the system , and only totally uncharged configurations contribute . the physics of such gases has been well studied . there are two phases separated by the kosterlitz thouless @xcite transition at a critical temperature @xmath116 . at low temperatures , when @xmath117 , the particles form dipoles and therefore the interactions of the charges at the ends are not screened . in the hot phase , when @xmath118 , the dipoles dissociate and the gas is in the plasma state where logarithmic interactions between particles are screened off at distances exceeding the radius of screening @xmath119 . the phase diagram of the gas can be obtained by the renormalization group analysis . it has been shown by bulgadaev @xcite that in @xmath112 , unlike the systems studied in refs . , the effective temperature is not renormalized , while the fugacity scales as @xmath120 where the logarithmic variable @xmath121 . for the dissociation of the dipoles @xmath122 plays the role of the activity of the products . if the electron electron interaction in the tomonaga luttinger model is attractive ( @xmath123 ) , @xmath122 scales to zero and , hence , at large distances , there are no free particles for screening . on the contrary , for the case of repulsion @xmath122 increases in the course of renormalization . at a distance @xmath124 the fugacity becomes @xmath125 , and there are enough particles for screening . the numerical constant @xmath2 can not be determined from eq . ( [ eq : rg ] ) alone . we expect it to be of order 1 , because the interpolation procedure discussed at the end of sec . [ se : nl ] indicates that @xmath126 . to analyze the effect of the backward scattering on the fes , it is convenient to represent the grand partition function @xmath127 as @xmath128 where @xmath129 is the grand partition function of the gas without the additional half charges at the ends , while @xmath130 is the correlation function of the end charges . for free electrons @xmath130 is determined by a series of logarithmically divergent terms . however , the situation alters entirely for repulsive electron electron interaction , because in this case the logarithmic coulomb interaction of the charged particles is screened off . as it will be shown bellow the effectively screened coulomb interaction decays like a power law of @xmath113 for large enough @xmath113 . the latter fact allows us to represent the correlation function @xmath130 at @xmath131 as @xmath132.\ ] ] here @xmath133 is the self energy of each of the end charges @xmath134 separately , while @xmath135 is their effective mutual interaction ; @xmath133 describes the change of the free energy of the gas due to the disturbance of the vicinity of one of the ends , which extends over a distance @xmath136 . the factor 2 appears because the change of the energy of the system due to the end charges does not depend on the sign of these charges . the prefactor @xmath137 is a result of the factor @xmath138 in eq . ( [ eq : fphi ] ) ( see also appendix b for a discussion on the effects of @xmath139 ) . let us discuss now the screening properties of the gas when it is in the plasma phase . since the logarithmic interaction is a two dimensional coulomb potential , the system can be viewed as a gas of charged particles confined to a line , while their electric field extends over the plane . in such a system , despite that a probe charge is compensated by the charge of the screening cloud , the screening is not complete , i.e. , the potential is not exponentially decaying . the equation that determines the debye screening of the classical coulomb gas in the hot plasma phase is given by @xmath140 here the lengths are measured in units of @xmath119 , i.e. , the scale above which the classical plasma description can be applied . the function @xmath141 describes the potential at a point @xmath142 of a probe charge located at @xmath143 in a system that extends from @xmath144 to @xmath113 , and it has been used in eq . ( [ eq : debye ] ) that the density of the screening charge @xmath145 . when the probe particle is inside an infinite system one can easily obtain that the screened potential falls off at large distances as @xmath146 where @xmath147 is the distance from the probe particle @xcite . ( in this case the probe charge @xmath148 is in the middle of the screening cloud . for the one dimensional geometry this leads to the configuration of charges : @xmath149 . this configuration gives a potential that decays like @xmath150 . ) for a finite system when the probe particle is placed at the end ( @xmath151 ) the screening is less effective . infinite line the potential falls off like @xmath152 , where @xmath153 . the condition @xmath154 insures the convergence of the integral over the density of the screening cloud , which is equal to the charge of the probe particle . it can be shown ( we also checked it numerically ) that @xmath155 in the case of a finite interval @xmath156 there is an ending effect which leads to a certain increase of the potential @xmath157 . nevertheless , one can check that the potential @xmath135 still falls off faster than @xmath158 . to see that let us confine the integration on @xmath159 in eq . ( [ eq : debye ] ) to the point @xmath160 , neglecting the contribution of particles located between @xmath160 and @xmath113 . the effective screening potential obtained in this way gives an upper bound for the screening potential , because the contribution of a part of the system to the screening is ignored . this procedure reduces the integral equation to the one of volterra type , which can be solved by the laplace transformation . for @xmath151 this estimate yields @xmath161 the fact that @xmath162 decays faster than @xmath163 was also confirmed numerically . since the polarization @xmath164 decays rather fast , like @xmath165 , the self energy of the charge @xmath166 attached to one of the ends , @xmath167 , is determined by finite distances @xmath168 , and it does not contain any singular contributions . the effective interaction of the end charges , @xmath135 , also does not lead to @xmath169 terms in @xmath130 as it was discussed above . therefore , in the hot phase , in contrast to the marginal non interacting case , there are no logarithmic terms in @xmath130 and it renormalizes @xmath112 only by a constant factor . it is remained now to analyze the contribution of the function @xmath129 to the fes . it is reasonable to write @xmath129 , which is the grand partition function of the gas , in the form of the mayer ursell linked cluster expansion @xmath170 since @xmath171 is directly related to the thermodynamic potential of the gas , @xmath172 in the thermodynamic limit ( @xmath173 ) , where @xmath174 is the gas pressure . ( when the analytical continuation back to @xmath175 is performed , the term linear in @xmath113 describes the change of the position of the absorption threshold . it arises due to a change of the energy of the electrons in the backscattering potential of the hole . ) to determine the fes exponent , it is necessary to extract from @xmath171 a singular sub leading term @xmath176 , if it exists in the problem under consideration . since the interaction between the charged particle is effectively short range due to the screening , it is reasonable to expect that for @xmath131 the thermodynamics of the coulomb gas behaves in the conventional way . in particular , one may expect that the effects related to the existence of the two ending points are not singular in that case , i.e. , that sub leading term @xmath177 does not exist . to determine the sensitivity to the ending points , let us consider @xmath178 if a @xmath179 term exists in @xmath171 , it should reveal itself in @xmath180 as @xmath181 term . taking the derivative in this equation fixes one of the points in a diagram of the linked cluster expansion to the left end of the interval @xmath156 . let us denote by @xmath182 the most distant point in the right direction in each of the diagrams , see fig . [ fg : delta ] . then , in each of the diagrams the integration over the coordinates is limited to the interval @xmath183 . because of the charge neutrality of the system , the left and right parts of the diagrams can not be connected by only one line describing the interaction , but should be connected by at least two interaction lines . since the coulomb interaction is long range , these interaction lines should be dressed in order to reproduce the debye hckel screening . by construction , the coordinates of the polarization operators in these lines are limited to the interval @xmath183 . therefore , the effective screened potential corresponding to the interaction lines which connect the left and right parts of the diagram is @xmath184 . the shaded bubbles at the left and right ends of the diagram in fig . [ fg : delta ] are equal to a certain constant which is determined by short scales @xmath136 . as a result @xmath185 since @xmath186 decays faster than @xmath187 , a term @xmath188 does not appear in @xmath180 . the diagrams in which the left and right ending parts can not be disconnected by cutting two interaction lines are not essential , because they decay with @xmath113 faster than the estimate of eq . [ eq : con ] . therefore , we conclude that @xmath171 does not contain a logarithmic contribution . finally , after continuing analytically back to real time @xmath189 , one obtains that in the asymptotic limit @xmath190 and it does not contain any power law decaying preexponential factor . ( this consideration does not exclude the possibility that some terms , which are exponentially smaller than the main one @xmath191 , contain a preexponential power law factor . however , after the analytical continuation such terms will not be at the fermi edge threshold frequency . ) thus , it has been obtained that in the asymptotic limit , when @xmath192 , the absorption line singularity is given by [ eq : betac ] @xmath193 where the exponent of the fes @xmath194 this expression is obtained for the spinless case . in the case of spin the result is given by eq . ( [ eq : alphas ] ) . the exponent @xmath195 is non universal and it depends on the interaction parameter @xmath196 even when forward scattering is absent ( @xmath197 ) . in the limit of @xmath198 the fes exponent becomes @xmath199 this expression differs from the result of refs . @xcite where @xmath200 has been obtained instead of @xmath201 . although the difference is relatively small , it arises as a consequence of entirely different physics at the asymptotic limit , i.e. when @xmath202 . an interpretation of the above result is presented in the discussion at sec . [ se : ri ] bellow . the possibility of mapping the problem onto a coulomb gas with a characteristic length ( @xmath119 ) indicates a similarity with the kondo problem , where in the asymptotic regime a resonance singlet is formed . in the kondo problem the inverse of the characteristic length is the width of the resonance level at the fermi energy , i.e. , @xmath203 . the spin charge separation in the kondo problem is similar to the @xmath38@xmath74 channels decoupling here , while the analogue of the spin singlet ( kondo resonance ) is in the considered case a neutral mixture of the left and right moving electrons . in the preceding section the analysis of the correlation function @xmath4 in the asymptotic limit has been discussed . in this section @xmath4 will be studied bellow the asymptotic regime . we exploit as a starting point the field theoretical variables , which has been introduced in ref . @xcite . in these fields the decoupling of the forward and the backward channels is transparent . however , at a price of that the problem is described by a non local theory of self dual fields . a special treatment is needed to handle the theory in the correct way . the renormalization group analysis of the problem using these variables is elaborated in sec . [ se : rg ] . finally a novel iteration procedure , which allows to find the behavior of @xmath204 at @xmath205 , is presented in sec . [ se : qtsp ] . the separation of the forward and backward channels can be obtained in a more transparent way , if instead of the fields @xmath208 , @xmath74 different fields @xmath209 are employed , which will be defined now . in order to introduce these fields one has to construct new operators of @xmath210 and @xmath211 fermion fields : @xmath212 it is easy to check that the operators @xmath213 have the standard anticommutation relations . with the use of these fields the hamiltonian of noninteracting electrons acquires the form @xmath214 + \\ \int dx \bigl [ u(0 ) \bigl ( \rho_e(x ) + \rho_o(x ) \bigr ) \bbox{\delta}(x ) - \end{array}\ ] ] note , that the momenta and the phase dependences in @xmath215 are adjusted to describe a local scattering problem @xmath216 in the most simple representation . now the bosonization procedure for the @xmath210 and the @xmath211 fields will be applied . the density operators will be defined in the usual way @xmath217 both @xmath218- and @xmath219-operators have the standard commutation relations of _ right _ movers : @xmath220 = \frac { q l}{2 \pi } \delta_{q , q^{\prime } } , \;\;\;\ ; \left [ \rho_{e}(q),\rho_{o } ( q^{\prime } ) \right ] = 0.\ ] ] ( this is because in the definition of the operators @xmath221 the momentum @xmath222 is used for the component @xmath36 . ) now , the bosonic representation of the fermion fields can be performed in the standard way ( see e.g. @xcite ) : @xmath223 where @xmath224 to complete the decoupling procedure of the forward and backward scattering processes , two fields are introduced @xmath225 that satisfy self dual commutation relations : @xmath226= 4\pi i \partial_y \bbox{\delta}(x - y ) ; \;\ ; \left [ \phi^\prime_+(x ) , \phi^\prime_- ( y ) \right]= 0\ ] ] ( here @xmath227 ) . these commutation relations become evident , if one notes that @xmath228 now , the hamiltonian @xmath229 will be rewritten in terms of the self dual fields @xmath230 . the bosonization of the free hamiltonian ( [ eq : doe ] ) does not cause any problem , and it remains to consider the nondiagonal part of the electron electron interaction @xmath231 . from eqs . ( [ eq : idoe ] ) and ( [ eq : deod ] ) it follows that @xmath232 and @xmath233 using the bosonic representation for the last terms one obtains that the hamiltonian @xmath234 can be represented as a sum of two commuting parts @xmath235\\ & h_- = \frac{v_f}{4 \pi } \int dx \bigl [ \frac{1}{2 \left ( 1 - \gamma \right ) } \phi_-^{\prime}(x)^2 - 2 \delta_- \phi_-^\prime ( x ) \bbox{\delta}(x ) & + \frac{2\gamma}{\eta^2 \left ( 1- \gamma \right ) } \left . \cos \phi_- ( x ) \cos \phi_- ( -x ) \right ) \bigr].\nonumber\end{aligned}\ ] ] a certain care was needed here in order to treat the commutation relations of the fermion operators properly . the splitting of @xmath236 into two independent parts reminds the decoupling of the fields @xmath237 at a single point ( see eq . ( [ eq:0cr ] ) ) , which was exploited in sec . [ se : phiphidual ] . now the decoupling of the forward and the backward scattering channels is performed on a deeper level it is obtained for the hamiltonian @xmath238 , but not only as a factorization of the correlation function @xmath4 in eq . ( [ eq : f1 ] ) . to check the relationship between the fields @xmath239 , @xmath240 and @xmath241 , @xmath242 the contribution of the @xmath241 field to the function @xmath4 will be calculated . the hamiltonian @xmath243 is quadratic and it is possible to perform its diagonalization explicitly . this can be achieved by the new variables @xmath244 where @xmath245 . next , the linear term related with @xmath246 can be removed by a canonical transformation @xmath247 ( here the self duality of the fields @xmath242 displays itself see for comparison eq . ( [ eq : tp ] ) ) . now applying for the calculation @xmath4 the bosonic representation ( [ eq : beo ] ) one can verify the fact that @xmath241 channel provides in @xmath4 exactly the same factor as the field @xmath74 in eqs . ( [ eq : fb2 ] ) and ( [ eq : f1 ] ) . thus , there is a direct relationship between @xmath248 and @xmath241 and , correspondingly , between @xmath249 and @xmath242 . in the variables @xmath241 , @xmath242 both the backward and forward scattering terms are equally simple . that is the main advantage of this representation . however , the problem under discussion displays a nontrivial element of the so called dimensional transmutation , namely the creation of a new scale the screening length @xmath119 , which alters the behavior of the system at large scales . this aspect of the problem is not transparent now . it originates from the interplay of the backward scattering term and the electron electron interaction , while the latter acquires a nontrivial form in the @xmath250fields representation . to demonstrate the relation of the @xmath242field theory with the coulomb gas , the renormalization group analysis of the hamiltonian @xmath251 will be performed . since the @xmath242fields posses a nonvanishing commutation relations , the nonlocality of the interaction term in @xmath251 demands a certain care . having this in mind , the interaction term will be rewritten in the normal ordered form : the components of the @xmath242 field with positive momenta ( see eq . ( [ eq : dp ] ) ) should be placed to the left of the components with negative momenta . the former are related with the creation operators of the chiral phonons , while the latter with the annihilation operators . as a result @xmath252 where @xmath253 and @xmath254 denotes the normal ordering of an operator @xmath255 . the @xmath256-dependent factor @xmath257 arises here because the interaction term has a nonlocal and nonlinear form in fields which do not commute . only the derivative like part in the product of two sines has been kept above , because the renormalization of the @xmath258 term will be studied . the transformation from a hamiltonian to a functional integral describing quantization of the field with the self dual commutation relation ( [ eq : cphi ] ) can be done following floreanini and jackiw @xcite . finally after passing to the euclidean space the effective action is given by @xmath259&,\end{aligned}\ ] ] where the imaginary time units is rescaled by the factor @xmath260 . at this stage the regular renormalization group procedure can be applied . the variables @xmath261 in the lagrangian density will be divided to fast ( @xmath262 ) and slow ( @xmath2 ) components @xcite . here this procedure will be done only with respect to variation in the @xmath256coordinate space , because the local impurity which is now under consideration is static . in the fourier expansion of the slow component the momentum space will be cut off by @xmath263 ; see eq . ( [ eq : dp ] ) . the renormalization group transformation will be carried out by progressively integrating out the fast component and obtaining an effective functional for the slow component : @xmath264 where @xmath265 and @xmath266 . to obtain the renormalization of the amplitude @xmath258 , the exponent in eq . ( [ eq : rg1 ] ) will be expanded in powers of the last two terms of @xmath267 , and the product of these two terms will be kept . then , for this product the fast field @xmath268 will be integrated out with the weight @xmath269 . since the integration of the fast variables leads to a short range kernel , it is enough to keep only the quadratic term for the cosine in @xmath270 . to carry out the integration it is convenient to use the fourier transform of the fields @xmath271 . as a result one gets @xmath272 where @xmath273 and @xmath274 denote correspondingly the slow and the fast momenta , and @xmath275 is the fourier transform of the factor @xmath257 . since @xmath276 , it follows that @xmath277 where @xmath278 . exponentiating @xmath279 back , one finds that the structure of the @xmath258term has been reproduced , and the renormalized amplitude @xmath280 obeys the equation @xmath281 which for small @xmath196 is identical to the renormalization group equation of the coulomb gas theory ( see eq . ( [ eq : rg ] ) ) . actually , the renormalization group procedure described above has not benefited much , in regard to technicalities , by using the @xmath242fields . now , an alternative procedure which relies on the fact that in this representation the backward scattering term acquires a linear form will be developed . a canonical transformation , @xmath282 , generalizing ( [ eq : ct ] ) , @xmath283 will be applied to simplify the hamiltonian @xmath251 , which is taken as : @xmath284 \ ; .\ ] ] here the factor @xmath285 has been omitted for brevity , and again only the derivative like part of the nonlinear term has been kept . the aim of the present consideration is to determines the factor @xmath286 in such a way that as a result of the transformation ( [ eq : ct1 ] ) the scattering term @xmath258 will be removed from the hamiltonian , i.e. , @xmath287 for @xmath288 the linear term generated after the transformation ( [ eq : ct1 ] ) by the quadratic term in @xmath251 will cancel out the impurity term . however , because of the nonlinear term , this is not the end of the story , since @xmath289 here the cosine was expanded with respect to the @xmath290term . in the fourier components the new term generated in @xmath251 is given as @xmath291 expanding @xmath292 , one finally gets @xmath293 the essential point here is that in this term the combination @xmath294 displays the structure of the original @xmath258 term . the factor @xmath286 that solves eq . ( [ eq : trans ] ) can be found by iterations . for @xmath295 the last integral in eq . ( [ eq : ha1 ] ) gives @xmath296 . so , the new @xmath258like scattering term has been generated with the amplitude @xmath297 . repeating successively the described procedure , one obtains @xmath298 we are ready now for the calculation of the correlation function @xmath4 . let us write @xmath4 as @xmath299 where @xmath300 correspond to the correlators of the fields @xmath301 respectively . as it has already been discussed in sec . [ se : phi+- ] the factor @xmath302 can be easily found . it is remained to find the correlator @xmath303 { \cal u}^{-1}_a e^{-\frac{i}{2 } \phi_- ( 0 ) } \right>,\ ] ] where @xmath282 are determined with @xmath304 given by eq . ( [ eq : ak ] ) . it will be assumed here that the essential part of the contribution of the nonlinear @xmath196term has been already taken into account through the dependence of @xmath286 on @xmath196 . therefore , for small @xmath196 the averaging over the @xmath242field in eq . ( [ eq : f- ] ) will be done keeping in @xmath251 the free term only . as a result one obtains @xmath305.\ ] ] in sec . [ se : phiphidual ] the function @xmath4 was obtained as a product of two correlators of the fields @xmath38 and @xmath74 . since @xmath302 is equal to the correlator of the @xmath74field , it follows from eq . ( [ eq : f1 ] ) that the correlator @xmath306 coincides with @xmath105 . therefore , from the analysis at the end of sec . [ se : phiphidual ] one can conclude that at asymptotically large time @xmath307 in the solution ( [ eq : logft1 ] ) ( obtained for small enough time ) the dependence on @xmath308 saturates when @xmath309 vanishes , i.e. , when @xmath310 this happens when @xmath311 . on that basis it has been supposed that in eq . ( [ eq : tscr ] ) the parameter @xmath126 . the described procedure was based on the expansion of the cosine in eq . ( [ eq : cosexp ] ) after carrying out the canonical transformation @xmath282 . that expansion is justified when @xmath312 . for small @xmath196 it can be shown that @xmath313 . at the calculation of @xmath4 one is interested in momenta @xmath314 . hence , the procedure is safe until @xmath315 , i.e. , when @xmath316 thus , @xmath306 has been found for asymptotically large time @xmath317 when the physics of the screened coulomb gas develops , and also for short time when @xmath318 . therefore , the point determined above from eq . ( [ eq : logft1 ] ) where @xmath306 saturates is , in fact , only an interpolation estimate : the behavior of @xmath306 has been found in two limits , and the two solutions are matched at a time which should determine @xmath119 . to interpret the obtained result for the exponent @xmath195 one can try to apply the nozires and de dominicis @xcite ( nd ) theory of the fes for one dimension . then a serious problem arises , because the nd theory uses the scattering theory description . this implies the existence of quasi - particles . however , as it is known from studies of the tomonaga luttinger model @xcite , quasi - particles do not exist in one dimension when @xmath196 is finite . therefore , it has sense to discuss the physical meaning of @xmath195 in terms of the scattering theory description only in the limit of vanishing @xmath196 , i.e. for eq . ( [ eq : alphac0 ] ) . at this limit the screening length @xmath319 goes to infinity , and the following discussion corresponds to @xmath320 in the spinless case the nd theory yields @xmath321 , where @xmath322 is the phase shift of the spherical harmonic component @xmath323 and @xmath324 is the phase shift of the state of the exited electron . in the case of one dimension the even and odd combinations replace the partial wave expansion @xcite . hence in the one dimensional case the nd theory gives @xmath325 when the even state is excited ( see appendix [ ap : odd ] for a discussion on the odd state ) . here @xmath326 are the eigenvalues of the @xmath327 unitary scattering matrix of the one dimensional system . the relation of the phase shifts with the transmission ( @xmath328 ) and the reflection ( @xmath329 ) scattering amplitudes can be easily found : @xmath330 where @xmath331 is the argument of the amplitude @xmath332 . to determine the phase shifts for the asymptotic limit ( [ eq : limits ] ) let us discuss the scattering of particles with a linearized spectrum , the appropriate schrdinger equation is @xmath333 the eigenfunctions of that equation can be easily found @xmath334 where @xmath335 , @xmath336 and @xmath337 . the phase shifts of this solution are given by @xmath338 ( here the decoupling of the forward and backward channels reveals itself in the fact that @xmath339 . it should be mentioned , however , that the use of the operator @xmath340 for the linearized spectrum of the electrons neglects at the description of the forward scattering the presence of the cutoff in the momentum space . when one solves eq . ( [ eq:22dif ] ) by the standard scattering formalism with a finite cutoff in the spectrum the decoupling does not occur . nevertheless , it is reasonable to follow the present treatment of the scattering since from the experience of studies of the @xmath0-ray absorption problem @xcite and the kondo problem it is known that the bosonization procedure provides the correct mapping on the coulomb gas if one replaces the born phase shift by the full one . ) substitution of the phase shifts ( [ eq : ephases ] ) into eq . ( [ eq : fesep ] ) and comparison with eq . ( [ eq : alphac0 ] ) yield @xmath341 from the general formalism of the scattering theory it is known that when a resonance exists in a channel the phase shift of this channel is close to @xmath342 . the result of refs . @xcite was based on a different assumption . comparison of eqs . ( [ eq : ephases ] ) and ( [ eq : selfphases ] ) gives @xmath343 in the course of the renormalization the strength of the backward scattering amplitude increases . therefore , in view of eq . ( [ eq : rdelta ] ) it is tempting to accept that the asymptotic limit corresponds to the total reflection . then , according to eq . ( [ eq : rdelta ] ) , @xmath344 , and from eqs . ( [ eq : fesep ] , [ eq : ephases ] ) it can be obtained @xmath345 . precisely that result was given in refs . @xcite . in fact , it has been assumed in ref . @xcite as a starting point that the reflection coefficient is one , and the appropriate boundary condition has been applied . the same assumption has been also employed in ref . @xcite . indeed , following eq . ( [ eq : rdelta ] ) it is difficult to imagine how the limit of a total reflection as the asymptotic one could be escaped . the answer was formulated in the first paragraph of this discussion . the point is that the interpretation in terms of the scattering phases can be applied only in the asymptotic limit of free particles . at the intermediate scales this simple interpretation is not valid . the same effects of electron electron interaction which lead to the renormalization of the backward scattering make the use of the theory of scattering not applicable at the intermediate scale . therefore , one can not apply eq . ( [ eq : rdelta ] ) during the course of the renormalization procedure . consequently , the argument that one can not get a value of the fes exponent larger than @xmath346 without crossing the regime corresponding to the total reflection can not be used against the result for @xmath195 obtained in eq . ( [ eq : alphac ] ) . another point , which is worth to be discussed , is the intimate relation of the result of refs . @xcite with the work of kane and fisher @xcite . in the latter it has been claimed that , since in the course of renormalization flow the strength of the impurity effectively increases , the final fixed point can be modeled by two disconnected semi infinite lines ( wires ) . that kind of fixed point leads to a total reflection and is in accordance with the results of refs . @xcite for the fes . however , two weakly connected wires represent a system with severely broken particle hole symmetry at the impurity site because electron density vanish at the ends of the wires . to break the particle hole symmetry a depletion of electrons must occur at the impurity area . the total depleted charge , @xmath347 , according to friedel sum rule @xcite , is equal to the sum of the phase shifts , i.e. according to eq . ( [ eq : ephases ] ) @xmath348 therefore the term responsible for the particle hole asymmetry is the amplitude @xmath19 , which in contrast to @xmath349 does not increase in the course of the renormalization flow . thus , two disconnected semi infinite lines can not be a fixed point to which a weak impurity system flows . it has been attempted @xcite to justify the conjecture of two disconnected semi infinite lines as the fixed point limit by mapping the problem for a particular value of the interaction coupling constant on an exactly solvable model of a semi infinite spin chain @xcite . to make this mapping onto a semi - infinite geometry possible , the authors had to impose an additional constraint on the field @xmath350 . it was needed to set @xmath351 ( see the discussion of eq . ( 8.3b ) in ref . @xcite ) , what implies that the charge fluctuations are frozen out at the location of the impurity center . indeed , this situation can be realized in a weak link junction . however , the vanishing of the charge density fluctuations is , in fact , not a result of the dynamics of the original problem , but is a direct consequence of the additionally imposed constraint . it has been shown that the asymptotic behavior of the 1d electron backward scattering in case of electron electron repulsion resembles the physics of the kondo resonance . a similarity with the kondo problem is seen in the possibility of mapping the considered above problem onto a coulomb gas with a characteristic length @xmath119 . in the considered problem a localized mixture of the left and right moving electrons acts the role of the kondo singlet . in eq . ( 44 ) of ref . @xcite a hamiltonian quadratic in fermion operators is presented , which is similar to the hamiltonian of a resonant level model . the partition function of this model corresponds to a non alternating coulomb gas with @xmath352 . it is also claimed in ref . @xcite that at low energies this resonant level hamiltonian is equivalent to the one describing the two channel kondo problem in the toulouse limit . thus , the fes problem in interacting one dimensional electron gas analyzed above by mapping on a non alternating coulomb gas has indeed a certain relationship with a kondo resonance physics . however , it should be emphasized that the interpretation of the obtained result for the fes in terms of the phase shifts should not be taken too literally . at the consideration of other physical quantities the phase shifts of eq . ( [ eq : resonancephase ] ) can not be used straightforwardly , because in one deimension an electron gas with interactions can not be described by the fermi liquid theory . the absorption of the electro - magnetic wave in one - dimensional electron systems has been analyzed near the absorption edge . it has been found that a new time scale , @xmath119 , is generated as a result of the combined effect of the backward scattering of electrons on an impurity like center created in the valence band at the absorption and a repulsive interaction of the conduction electrons . the shape of the absorption line is given by the fourier transform of the correlation function @xmath4 , the long time asymptotic behavior of which is determined by @xmath119 . consequently for frequencies close enough to the fermi edge the absorption line is controlled by this scale . the infrared physics of the fermi edge singularity in the presence of backward scattering together with electron electron repulsion resembles the physics of the kondo problem . this aspect of the fermi edge singularity in 1d systems was missed in the preceding studies of the question . the approach of the present paper may not be confined to the fermi edge singularity problem only . it may be useful in the studies of tunneling effects in quantum wires and also for description of effects related to the quantum hall edge states . we are grateful to g. kotliar , a. kamenev and d. orgad for useful and stimulating discussions . a. f. is supported by the barecha fund award . the work is supported by the israel academy of science under the grant no . 801/94 - 1 . when the spin degrees of freedom of the conduction electrons are included , the main features of the theory of the fes in 1d do not change . the forward scattering contribution can be found by a shifting operator , while the backward scattering contribution can be described by a reduction to the coulomb gas theory with a characteristic screening length . due to the spin , the charged plasma contains two types of particles . the fields that describe the forward and backward channels remain to be decoupled . the final expression for the fes exponent have the same structure as in the case of spinless electrons , but with slightly different coefficients . the electron electron interaction in the presence of spin is taken as @xmath353 where @xmath354 and @xmath355 . the forward and backward scattering terms are given by @xmath356\ ] ] and @xmath357 now the bosonization representation for up and down spins will be applied in a way similar to the spinless case ( see eqs . ( [ eq : dpd ] ) and ( [ eq : fermion ] ) ) . with the use of the conventional combinations @xmath358 and @xmath359 [ def : psdps ] @xmath360 one arrives to the bosonized hamiltonian @xmath361 : @xmath362,&\end{aligned}\ ] ] where @xmath363 , @xmath364 and @xmath365 . in the backward scattering term the @xmath358 and @xmath359 channels appear to be mixed . in the absence of the backward scattering the fes can be found by shifting @xmath366 with the use of the exponent of the dual operator @xmath367 . the forward scattering appears only in the @xmath358 channel , and in comparison to the spinless case it acquires an additional factor of @xmath368 . finally , the expression for the fes exponent when spin is included , but only the forward scattering exists . is @xmath369 this result coincides with the result of ogawa et . @xcite , if in the corresponding expression of ref . @xcite one substitutes @xmath11 by @xmath370 , @xmath371 by @xmath372 and @xmath373 by @xmath374 . the renormalization corrections arise , because the interaction between the electrons moving in the same direction have been included here . the correlation function @xmath4 in the presence of the backward scattering will be treated analogously to sec . [ se : bs ] . one should calculate an expression equivalent to formula ( [ eq : fint ] ) , but for the discussed case it has a slightly more complicated form : @xmath375 & \left . e^{-i \frac{1}{\sqrt{2 } } \frac{\delta_+}{\pi } \frac{\beta_\rho}{\alpha_\rho}\tilde \phi_\rho(0 ) } e^{i \frac 12 \left ( \sqrt { 4\pi } \tilde \phi^0_\uparrow(0 ) + q\sqrt{4 \pi } \phi^0_\uparrow(0 ) \right ) } \right>. & \end{aligned}\ ] ] here the dots inside the square brackets denote the time integrals of a sum of all possible time ordered products of the exponentials of the operators @xmath376 . only those terms which correspond to equal number of creation and annihilation operators of electrons , separately for each spin species , give non vanishing contributions . since the fields @xmath107 and @xmath108 do not interfere , @xmath4 is factorized as @xmath377 where @xmath378 , @xmath379 and @xmath380 . the correlator @xmath381 can be obtained similarly to eq . ( [ eq : fphi ] ) : @xmath382}}&.\end{aligned}\ ] ] here indices @xmath383 denote the spin projections @xmath384 ; @xmath385 and @xmath386 . the sum @xmath381 repeats the structure of @xmath105 in eq . ( [ eq : fphi ] ) with the only difference that now there are two sorts of green functions . correspondingly @xmath387 means that only neutral configurations for each spin species separately are allowed . now the discussion following eq . ( [ eq : fphi ] ) in sec . [ se : bs ] that leads to a coulomb gas theory can be repeated . in the case of spin this gas in addition to charges has two flavors . charges of the same flavor interact through @xmath388 , while charges with different flavor interact through @xmath389 . the screening properties of the plasma of such gas are similar to those of the single component case . as a result , in the asymptotic limit when @xmath390 the exponent describing the absorption singularity is given by @xmath391 let us define creation operators of the even and odd eigenmodes according to eq . ( [ eq : ef ] ) now let us consider @xmath393 corresponding to activation of the states @xmath394 at the light absorption , instead of the combination @xmath395 which has been studied in eq . ( [ eq : fint ] ) . for all configurations which give non vanishing contributions to @xmath112 the factors depending on the phase @xmath139 are canceled out when the eigenmodes operators @xmath394 are used . then @xmath396 corresponds to a maximum , while @xmath397 to a minimum , of the function @xmath398 , when one continues it analytically to the euclidean time @xmath399 and studies @xmath112 as a function of @xmath139 . this is because in the coulomb gas expansion of @xmath400 all terms become positive , while in the @xmath401 sum the terms corresponding to configurations with odd number of gas particles have opposite sign . therefore , the fes corresponding to the excitation of the odd channel is less singular . this is in accordance with eq . ( [ eq : zphi ] ) , because the odd combination corresponds to @xmath402 when the factor @xmath403 vanishes . this is in agreement with the well known result for the energy levels correlation function of random matrices ( ) . as was shown by dyson ( ) there is a direct correspondence between the statistics of energy levels of complex systems and the statistical mechanics of a logarithmic coulomb gas .